Reliability Design of Mechanical Systems - A Guide for Mechanical

and Civil Engineer

Seong-woo Woo

Reliability Association of Korea, 146-102 Sunyoo-ro, Yeongdeungpo-gu, Seoul 150-103, Korea

Preface

In the beginning of the twentieth century, new sophisticated mechanical systems such as bridges,

rockets, automobiles, airplanes and space shuttles were designed and built for people to live

comfortable lives through the engineering design processes. Typical design process can be broadly

summarized as 1) define the problems, 2) develop the product–prototype, design & testing, 3)

production. Due to the frequent occurrence of disasters for new products, product reliability has

become one of increasingly important factors (to consider) because of cost, competition, public

demand and adaptation of new technology. The most effective way to protect the reliability disaster is

to develop the reliability-embedded design process including its methodology in parallel with the

established design process.

As products with multiple modules require higher performance and material cost-reduction, the

reliability design of product have become more complex and increases the risk of product failure. The

studies of reliability engineering have been deepened to prevent the reliability disasters in the past

century. Even though there are a large number of concepts, theory, and texts on reliability, an up-to-

date book for emphasizing the new methodology of reliability design is still required to prevent the

reliability disasters of the mechanical/civil system.

From the standpoint of economics, company will decrease the operation profit for a failure in its

expected product lifetime because of Product Liability Law in the global market. All products from tires

to electric components are fabricated from the structure (or materials) that will tend to degrade or

break down abruptly by random loads. The mechanical system can eventually fracture due to fatigue

which can result from cyclical stresses (or loads). When products are subjected to random loads, they

start the void in material (or design defects), propagate, and rupture it. If failure for new product

happens, the product may no longer meet the established specifications for proper product

functionality. To avoid product failure in lifetime, product should be designed to robustly withstand a

variety of loads.

The main objectives of writing this book are focused on explaining the development necessity of the

reliability-embedded design process and its methodology. As reliability methodology, we will suggest

the new parametric accelerated life testing (ALT) that meets those market requirements - higher
performance, material cost-reduction, and higher reliability in field. The reliability-embedded design

process consists of parametric ALT plan, failure mechanism and design, acceleration factor, sample

size equation, and the parametric ALT. It produces the reliability quantitative test specifications (RQ)

in accordance with the reliability target. A parametric ALT method therefore will assess the reliability of

product subjected to repetitive stresses.

Based on the market data, parametric ALT plan will set up the reliability target of product and its

modules. Mechanical system in field subjected to loads arise how to design product for the failure

mechanisms – fatigue and fracture. The accumulated damage in system like palmer miner rule can be

represented at the time-to-failure model. The acceleration factor with a new effort concept (or loads)

was derived from a generalized life-stress failure model. So the new sample size equation with the

acceleration factor enabled the parametric ALT to quickly evaluate the expected lifetime of product.

This parametric ALT should help an engineer to uncover the missing design parameters affecting

reliability during the design process of new product.

Consequently, if applied in the established design process, new parametric ALT helps companies to

improve new product reliability and avoid the recalls of product failures in field. As the improper design

parameters in the design phase are identified by this reliability design method, the product will

improve the reliability that will be measured by the increase in lifetime, LB, and the reduction in failure

rate, . Product will meet the reliability target in industry. This book will help to prevent the reliability

disaster through the parametric ALT. We also provide a lot of parametric ALT examples that are

effective to be understood in the mechanical/civil field.

This book is composed of nine chapters. Chapter 1 presents the present aspect and need of reliability

engineering in the advance of modern technology. Chapter 2 reviews the historical reliability disasters

and their root cause within the past century. It will explain the significance of reliability assessment

and its methodology need to prevent reliability disasters in the design process. Chapter 3 will explains

the most important fundamental definitions of statistics and probability theory, the mathematical

essentials of reliability engineering, and the most significant aspects of reliability engineering

developed within the past century. It will help one to understand the basic concepts of reliability

methodologies that will be discussed in Chapter 8. Chapter 4 through Chapter 6 present load

analysis, the stress concept, and a brief overview of the typical reliability failure mechanism of product
– fatigues and fractures. Chapter 7 will be the fundamental concepts of the parametric ALT in product

that will be the core of this book. Chapter 8 will also present its case studies that are useful in a

variety of engineering area. Chapter 9 will cover the future aspects of parametric ALT in mechanical

product that will be developed as system engineering.

This book is intended to introduce the prerequisite concepts of the parametric ALT for senior level

undergraduate and graduate students, professional engineers, college and university level lecturers,

researchers, and design managers of the engineering system. We hope this noble methodology

explained in this book will help to prevent the reliability disasters of new product in field. The authors

would also like to thank for the publishing of Spring Co., especially Castro, Mayra, Springer DE. With

their help, this book has been published.

CONTENTS

1 Introduction to Reliability Design of Mechanical/Civil System

2 Reliability disaster and of its assessment Significance

2.1 Introduction

2.2 Reliability disasters

2.3 Development of Reliability Methodologies in History

3 Modern definitions in reliability engineering

3.1 Introduction

3.2 Fundamentals in Probability Theory

3.3 Reliability lifetime metrics

3.4 Statistical Distributions

3.5 Weibull distributions and its applications

4 Failure mechanics, Design and reliability testing

4.1 Introduction

4.2 Failure mechanics and designs

4.3 Failure mode and effect analysis (FMEA)

4.4 Fault Tree Analysis (FTA)

4.5 Robust Design (or Taguchi methods)

4.6. Reliability Testing

5 Load Analysis
 5.1 Introduction

5.2 Modeling of Mechanical System

5.3 Bond graph modeling

5.4 Load Spectrum and Rain-flow Counting

6 Mechanical System Failure

6.1 Introduction

6.2 Mechanism of Slip

6.3 Facture failure

6.4 Fatigue failure

6.5 Stress–strength analysis

6.6 Failure Analysis

7 Parametric Accelerated Life Testing in Mechanical System

7.1 Introduction

7.2 Reliability Design in Mechanical System

7.3 Reliability block diagram and its connection in product

7.4 Reliability Allocation of Product

7.5 Failure Mechanics, Design and Reliability Testing

7.6 Parametric Accelerated Life Testing

7.7 The reliability design of mechanical system and its verification

7.8 Testing equipment for quality and reliability

8 Parametric ALT and its Case studies

8-1 Failure analysis and redesign of ice-maker

8-2 Residential Sized Refrigerators during Transportation

8-3 Water dispenser lever in a refrigerator

8-4 Refrigerator compressor subjected to repetitive loads

8-5 Hinge kit system in a Kimchi refrigerator

8-6 Refrigerator drawer system

8-7 Compressor suction reed valve

8-8 Failure analysis and redesign of the evaporator tubing

8-9 Compressor with redesigned rotor and stator

8-10 French refrigerator drawer system

9 Parametric ALT: A Powerful Tool for future engineering development

Chapter 1

Introduction to Reliability Design of Mechanical/Civil System

Seong-woo Woo

Email: [email protected]

Reliability Association of Korea, Seoul, Korea (Republic of)

Abstracts: From the standpoint of system engineering, this chapter will be discussed with the

necessities of new reliability-embedded design process to catch up with incredible technology

innovations for new product. The modern products should survive to compete other products in global.

They are often required to have higher performance and reliability for the necessary intended

functions, though the product cost and developing time has to reduce it gradually. Because the new

product hardly satisfies the requirements within the limited development time, there is the presence of

risks on reliability disasters at all times. New assessment methodology of reliability in the reliability-

embedded design process is required to protect the massive recall in lifetime. It will therefore be

possible to discover the design defects in the design phase through new reliability methodology that

could enable product to satisfy the reliability target of product.

Keywords: Reliability-embedded design process, Reliability testing methodology, Design

requirements, Performance, Cost-down.

1.1 Introduction

As the frequent recalls for the new product occur globally, the term of product reliability seems to

friendly be used to everyday life. The product quality and its reliability seem to become important

requisites to ensure the continued success in the current global competitive marketplace. If the

company doesn’t satisfy the product quality, it will be expelled in global market. Thus, it is important
for the product design team to understand customer expectations or voices. The product reliability is

to create a product that can properly work the required intended functions under all environmental

conditions in operation lifetime of product. To achieve the reliability requirements in field, numerous

concepts – bathtub, MTBF and failure rate have been established in the last century. They also

require the fundamental knowledge of the probability and statistics. As they are put to use, reliability

could quantitatively describe the failure data from the marketplace.

Fig.1.1 Customer Requirements of Product (or Core Parts).

As seen in Figure 1.1, the product development in the field of the mechanical/civil engineering system

is continually confronting to be satisfied with the end user requirements – high control performance,

high response, energy efficiency, low noise, high reliability, long life time, the latest hardware design,

contamination resistance, low price, compact and highly-portable weight, and precision control for

wide frequency range. To survive the competitive global environments, the company should

manufacture the high-performance products that meet the customer expectations or their

specifications.
Engineers, however, wonders if product development satisfies the requirement of these attributes in

reality. Ironically, to get those attributes in the product design such as automotive and cell phones, the

product development times are continually decreasing. On the other hands, product reliability in

marketplace is highly required due to the recall costs. Thus, new product is hard to match the market

requirements of product – cost reduction, the shortening developing time, higher performance and

reliability. From the standpoint of system engineering, companies are asked to establish the design

process of satisfying the product requirements in the short development time. For example, while

product development time - automobile continued to shorten from 65 months to 24 months, reliability

required increases from 0.9 to 0.99. These declines mean that companies have reliability

methodology tools closely tied to the development process itself (Figure 1.2).

Fig.1.2 Historical time line for product quality

As market is requiring, a myriad of technology innovations are constantly emerging and disappearing.

People owing to a state-of-the-art technical renovations broaden their lives and widen their

boundaries. On the other hand, they also experience frequent malfunctions as new product has been

released in the marketplace. They ask to replace the problematic product with new one. To satisfy the
end users requirements, most global companies have to be established in the product developing

process that can find out the problematic design. As a matter of fact, they have traditional

methodologies to achieve high product quality as qualitative - FMEA (Failure Mode Effective Analysis)

and FTA (Fault Tree Analysis). But there is no quantitative methodology - reliability testing.

Competitive company in high technology industries only can prosper in markets whose customers

satisfy extreme needs, such as safety-critical mechanisms (aircraft) or high technology military

armaments.

The established developing process of product in company can be largely classified as Research &

Development (R&D) and Quality Assurance (QA). R&D is a core part of the modern company

because major design decisions in firms are made based on its technical level. As companies define

the design requirements from customer needs or past experiences, they start to develop new product

that satisfies those specifications. R&D activities also are conducted by departments with person

specialized in technique. They design architectural structures, proper materials, and robust systems

while considering the limitations - practicality, regulation, safety, and cost. A professional engineer can

apply the scientific methods to solve out engineering problems. They also use advanced

manufacturing processes, expensive safety certifications, specialized embedded software, computer-

aided design software, electronic designs and mechanical subsystems. The design process of product

embedded in reliability concept can be briefly defined as qualitative design process and quantitative

design process. It will briefly flow down the product planning, concept design, basic design, prototype,

detail design, and production (See Figure 1.3).

 Fig. 1.3.New developing process of product embedded in reliability concept

For a detailed design of product, QA will determine if the product is satisfactory to each company

specifications. In other words the quality of product may be defined as the product specifications that

are summarized as perception of the degree or the end-user's expectations. Quality verifications in

these forms were initially established by National Aeronautics and Space Administration (NASA), the

military and nuclear industries in the 1960s. The specification-oriented development process was

designed to develop better products that have no design modifications or technical innovations at that

time. And it was focused on manufacturing, testing and quality control, rather than design. At that time

the typical design tools – design review, FMEA and FTA are to qualitatively accomplish the

specifications of product quality but from a standpoint of quantitative quality, there is no design

process and reliability methodology to achieve the reliability targets.

Because the traditional methodology in the design process cannot find the chronic problems for

design issues of new technologies, products always have inherent design problems in marketplace

that might unceasingly give rise to massive recalls. For instance, the Boeing 787 Dreamliner

experienced some problems due to new design elements—the fuselage of carbon-fiber-reinforced

plastic (CFRP) and the electrical system incorporating lithium-ion batteries, which ultimately resulted

in grounding.

From a standpoint of reliability engineering, why do the historical reliability disasters such as the
explosion of challenger happen continually? They might come from the faulty components that have

the missing design parameters not found in the design process. The suspected components mounted

in product determine the lifetime of product when they are subjected to the wearout stress or

overstress under the end user operating or environmental conditions. So to find out the problematic

parts mounted in product, new reliability methodology is continually required. New reliability-

embedded developing process with new reliability methodology should be suggested in 1) product

reliability target/allocation/prediction, 2) reliability testing and Weibull analysis, 3) finding the design

problem of the suspected parts, and 4) the analysis of the field failure data.

In the relationship between failure costs and product life cycles, we know that the earlier reliability in

the design process is applied, the greater the profit is obtained. Total cost of a product is determined

by its design and its value is approximately 70%. For example, if cost $1 is required to rectify a design

defect in the prior design stage, the cost would increase to $10 after final engineering stage, $100 at

the initial production, and $1000 at the mass product stage (Figure 1.4).

Fig.1.4.Relationship between failure costs and product life cycles

Before releasing the product to the marketplace, the key factors of quality activities are to discover the

missing design parameters that have not been found in the design process. As a result, new

quantitative reliability methodologies in the reliability-embedded developing process should search out

the problematic components and prevent reliability disasters in the design phase that traditional R&D

developing process can’t solve. As Failure Analysis (FA), they should be closely looked in the product

design phase that failures in field may happen. In the reliability-embedded process engineer

generates reliability quantitative specifications that can fit to a newly developed product and increase

the lifetime of product by correcting the missing design parameters.

As reliability testing, fatigue failure due to design defects were traditionally assessed from fatigue

testing. Fatigue testing is critical, but it has many limitations – 1) requires many physical prototypes, 2)

difficult to achieve realistic tests, 3) slow and expensive difficult to conduct early in the design

process, 4) requires many tests and statistical interpretation. However, the product still has the

inherent design errors because they reveal to use in field soon or late.

Consequently, companies are required to develop new reliability methodology that make up for the

weak points and find them in the design process. As reliability quantitative test specifications (RQ),

new reliability methodology that will be discussed in Chapter 7 is mainly focused on the fundamental

concepts of reliability and parametric ALT. First, after reliability disasters are reviewed, reliability

assessment tools developed in history will explain its strengths and weaknesses. We will look over the

concepts of failure mechanism, design and reliability testing in the mechanical/civil systems. The

parametric ALT in Chapter 7 is a core part of the reliability-embedded product developing process. It

consists of parametric ALT plan, failure mechanism and design, acceleration factor, and sample size

equation. As a quantitative method, the parameter ALT will be helpful to set up the reliability target and

establish the verification specifications over the full range of functions fitted to each product. A variety

of parametric ALT case studies in Chapter 8 will also be suggested to clearly understand the

methodology of the parametric ALTs.

Chapter 2

Reliability disasters and its Assessment Significance

Seong-woo Woo

Email: [email protected]

Reliability Association of Korea, Seoul, Korea (Republic of)

Abstracts: This chapter will review the historical reliability disasters including natural hazard and the

methodology of its reliability assessment that were developed in the last century. Most of reliability

disasters come from the complexity of product intended functions due to the customer requirements

and its inheritance design defects as new technology introduces. As countermeasures against

reliability disasters, methodology of reliability assessment like bathtub curve and Weibull analysis has

been developed in the previous century. For instance, the frequent derail accidents of railroad in the

early of nineteen century started the research for its root cause and made the S-N Curve. The chronic

failed vacuum tube in the WWII created the bathtub curve. The FMEA, FTA and Weibull analysis for

reliability testing today have been widely used in company as NASA developed for the space shuttle

program in the mid-sixties. Now Physics Of Failure (POF) become more important tools to analyze the

failure mechanics since the introduction of Integrated Circuit (IC), transistor radio and TV in the late of

1960s. However, in the field of mechanical/civil system, representative POFs were still fracture and

fatigue but not introduced to find the problematic parts by reliability testing method.

Keywords: Reliability disasters, Natural hazard, Bathtub curve, Physics Of Failure (POF).

2.1 Introduction

A disaster - oil spill, nuclear plant accident and the others is a deep-felt functional failure of the

product accompanying catastrophic human, economic or environmental impacts, which has no

predicting ability of the community or society to manage its own resources. Thus, peoples often have

been learning the lessons and setting up countermeasures because they can be prevented if its root

causes were known previously. For instance, the RMS Titanic in 1912 had approximately 2,200

people on board. It took two hours and forty minutes to sink and drowned to deaths of more than

1,500 people. When the crews sighted the iceberg, Titanic was unable to quickly turn and collided the

floating ice in right side. There was no plan for rescue, though the ship was sinking fast.

The sinking of the titanic was caused primarily by the brittleness of the steel used to construct the hull

of the ship. In the icy water of the Atlantic even a small impact between iceburger and ship could have

caused a large amount of damage. The impact of an iceberg on the ship's hull resulted in brittle

fracture of the bolts that were holding the steel plates together (Figure 2.1).

Fig.2.1. Sinking picture of the RMS Titanic Illustration for ‟Die Gartenlaube‟ magazine by Willy Stöwer,

1912

After that, every ship has to have an evacuation plan in danger. When disasters have been studied for

more than 40 years, disasters in history might have been seen as the result of inappropriately risk
management or mutual combination of hazards and vulnerability. Since most of disasters result from

human-made carelessness or proper management measures can partially prevent it from developing

into a disaster.

In addition, a natural hazard will briefly be explained before going into the reliability disasters.

Developing countries - the Philippines, Nepal and the others are suffering from natural hazards that

are caused into more than 95 percent of all deaths, and they are 20 times greater than that of

industrialized countries. They are all natural hazards that kill thousands of people and destroy billions

of dollars of habitat and property each year. Because there is no single root cause from natural

hazards, they are more common in developing countries that those countries have no emergency

systems. Typical examples are flood, transport accidents, nuclear explosions/radiation, and an

earthquake that causes a tsunami, resulting in coastal flooding. Recorded in magnitude 9.2, the 1964

Alaskan earthquake occurred in March 27 and resulted in 139 deaths. Anchorage experienced great

destruction or damage to many houses, buildings, and infrastructures like roads, particularly in the

several landslide zones along Knik Arm.

Recently the population growth in the world and its environmental effects has increased the severity of

natural hazards due to the global warming, depletion of the ozone layer, and the Central Pacific El

Nino phenomenon. The well-known several reasons - the tropical climate and unstable land forms,

coupled with the deforestation of the Amazon, unplanned growth proliferation, non-engineered

barbaric constructions – in the worldwide make the natural hazard areas more vulnerable (Figure 2.2).

Typical countermeasures against natural hazards can be classified into (1) research into the scientific

aspects of disaster prevention, (2) the reinforcement of the disaster prevention system, its facilities

and equipment, and other preventive measures, (3) construction projects - dam and

meteorological[weather] observations - designed to enhance the country's ability to defend against

disasters, (4) and emergency measures and recovery operations. Developing countries suffer

chronically from natural hazards, though several preventions for natural hazard.

After the Kobe earthquake in 1981 claimed some 5,100 lives, Japan updated its building guidelines,

added fresh fuel to another round of research on earthquake safety and disaster management. In

2000, the country's building codes with specific requirements and mandatory checks were revised.
From 1979 to 2009, Shizuoka prefecture poured more than $4 billion into improving the safety of

hospitals, schools and social welfare facilities. Though Japanese cities often shake, they rarely topple.

Because Japan is located in the Pacific rim, one of the Earth's most violent earthquake and volcano

zones, they are still vulnerable.

Fig.2.2.Occurrence of reported natural disasters by continent (1950 - 2011)

Sources: Reported Natural Disasters 1950-2011 from CRED

2.2 Reliability disasters

Reliability disasters are the consequence of technological risks due to product failures or human-

induced damages. Typical examples include transport accidents, industrial accidents, oil spills and

nuclear explosions/radiation. Deliberate terrorism, like the September 11 attacks, may also be put in

this category. For example, when the basic causes of reliability disasters are considered, there might

have been product complexity as demanded by customers. Today a typical Boeing 747 jumbo jet

airplane is made of approximately 4.5 million parts including fasteners, multiple modules, and

subsystems. An automobile is made of more than 25,000 parts, multiple modules, and subsystems. In

1935 a farm tractor was made of 1200 critical parts and in 1990 the number increased to around

2900. Even for relatively simpler products such as bike, there has been a significant increase in

complexity with respect to parts. Consequently, the product design such as automobile is becoming to

require these parts to withstand the environmental and user loading conditions (See Figure 2.3).

Fig.2.3. Breakdown of passenger automobile with multi-modules

Together with product complexity, there are possibilities for the inherent design problems of the parts.

A study performed by the U.S. Navy concerning parts failure causes attributed 43% of the failures to

design, 30% to operation and maintenance, 20% to manufacturing, and 7% to miscellaneous factors

[1]. While the design cost occupies only 5%, the cost influence holds 70%. Thus, we know that quality

cost could be saved if the design factors of faulty parts are known before mass production (Figure

2.4).
Fig.2.4. Leverage in product design: total cost of product is determined by its design (approximately 70%)
Thus we will have suggested typical examples - space shuttle challenger, Chernobyl nuclear reactor,

Point Pleasant Bridge, and the others. They also might have been prevented if reliability in product

design had been considered seriously. It will help to understand why the reliability concept of modern

product is critical.

 Space Shuttle Challenger: This debacle occurred in 1986, in which all crew members lost

their lives. Sadly, many Americans are suffering from low self-esteem because of failure. The

main reason for this disaster was design defects of rubber O-rings under cold winter (Table

2.1).

Table 2.1. Summary of Space Shuttle Challenger Disaster

Reliability Disaster

Phenomenon

Structure

Failure of two rubber O-rings

Root Cause
(Environment conditions: A Cold Launch Day)

 Chernobyl Nuclear Reactor Explosion: This disaster occurred in 1986, in the former Soviet
 Union, in which 31 lives were lost. The debacle was the result of design defects such as

faulty switch in reactor design (Table 2.2).

Table 2.2. Summary of Chernobyl Nuclear Reactor Explosion

Reliability Disaster

Phenomenon

Structure

Reactor exposion
Root Cause
(Reactor is jumped to around 30,000 MW thermal)

 Point Pleasant Bridge Disaster: Bridge located on the West Virginia/Ohio border collapsed in

1967. The disaster resulted in the loss of 46 lives and its basic cause was the metal fatigue of
 a critical eye bar (Table 2.3).

Table 2.3. Summary of Point Pleasant Bridge Disaster

Reliability Disaster

Phenomenon

Structure

Root Cause Metal fatigue of a critical eye bar

And in the following sections we can suggest the numerous other cases for reliability disasters that

will help to understand their root causes of modern products.

2.2.1 Versailles rail accident in 1842

The Versailles rail accident in 1842 occurred on the railway between Versailles and Paris. Following

King Louis Philippe I's celebrations at the Palace of Versailles, a train returning to Paris derailed at

Meudon. After the leading locomotive broke an axle, the carriages behind piled into it and caught fire.

With approximately 200 deaths including that of the explorer Jules Dumont d'Urville, the first French

railway accident and the deadliest in the world recorded,. Because most of passengers wearing the

seat belt were died, the accident led to abandon the practice of locking passengers in their carriages.

It started the study of metal fatigue subjected to repetitive loads like S-N curve (See Figure 2.5).

Fig.2.5. Versailles rail accident (1842) from Wikipedia

 Root Cause: Metal fatigue of rail was poorly understood at the time and the accident is linked

to the beginnings of systematic research into the failure problem

2.2.2 Tacoma Narrows Bridge in 1940

The Tacoma Narrows Bridge is a pair of twin suspension bridges that connect the city of Tacoma to

the Kitsap Peninsula. It went past State Route 16 over the strait and was collapsed by a wind-induced

natural frequency on November 7, 1940. The collapse of the bridge had no loss of human life. As

recorded on film, this film as a cautionary tale has still been well-known to engineering, architecture,

and physics students (See Figure 2.6).

Fig.2.6. Tacoma Narrows Bridge (1940) from Wikipedia

 Root Cause: without any definitive conclusions, three possible failure causes are assumed

1) Aerodynamic instability by self-induced vibrations in the bridge structure

2) Periodic eddy formations in bridge

3) Random turbulence effects - the random fluctuations by wind velocity of the bridge.

2.2.3 De Havilland DH 106 Comet in 1953

The de Havilland DH 106 Comet was the first commercial jet engine airplane that replaced the

propeller plane and could have a transatlantic flight. The Comet prototype first had an

aerodynamically design with four turbojet engines in two wings, an aerodynamic fuselage, and large

square windows. It first flew on 27 July, 1949. For the 1952 appearance, it offered a quiet and

comfortable passenger cabin.

One year later the Comets began to suffer the design problems that three airplanes were breaking up

in flight. Due to airframe metal fatigue, the Comet eventually discover the design flaws at the corners

of the square windows subjected to repetitive stresses. As a result, the Comet was redesigned with

oval windows, structural reinforcement, and other changes (See Figure 2.7 and Figure 2.8).

Fig.2.7.De Havilland DH 106 Comet (1954) from Wikipedia

 Fig.2.8.Stress-strain curve and the modified oval window in airplane

 Root Cause: fine cracks near the fixed nails of large square windows → repeated

pressurization and decompression in airplane → spreading cracks → limit crack → blast in

air by the broken window of airplane.

2.2.4 G Company and M Company Rotary Compressor recall in 1981

In 1981, market share and profits in G company appliance division were falling. For example, making

refrigerator compressor required 65 minutes work of labor in comparison to 25 minutes for

competitors in Japan and Italy. Moreover, labor costs of G Company were higher than that of Japan

companies. The alternatives were to purchase compressors from a better design model of Japan or

Italy. By 1983, G Company was decided to build a new rotary compressor in-house along with a

commitment for a new $120 million factory. G Company and a rival M company had invented the

rotary compressor technology that had been using it in air-conditioners for many years.

A rotary compressor had the less weighted part because of one third fewer and was more energy

efficient than the conventional reciprocating compressors. The rotary compressors took up less

space, thus providing more room inside the refrigerator and better meeting customer requirements.

The rotary compressor for refrigerator was nearly identical to that used in air conditioners.
However,

in a refrigerator, the coolant flows only one-tenth as fast and the unit runs about four times longer in

one year than an air conditioner. Two small parts inside the compressor were made out of powdered

metal rather than the hardened steel and cast iron used in air conditioners because this material could

be much closer tolerances and reduce the machining costs. The design engineers did not consider

the critical failure in early product until the noise claims of domestic house in 1987.

When a rotary compressor was abnormally locking in 1987, G Company and M Company

experienced massive recalls of the rotary compressor. As the oil sludge in the refrigeration system

blocked the capillary tube, the cooling capacity of the refrigerator decreased. In the compressor

development process, reproducing this failure mode and preventing the blocking of this tube were

very important to the reliability of the refrigerator. However, reliability testing methods such as the

parametric ALT, was not used at that time (Table 2.4).

Table 2.4.G Company and M Company Rotary Compressor recall summary

G Company M Company

Product Household Refrigerator

Unit Rotary Comp (Sealed Refrigerant Compressor)

Production Date 1986.3 1985.1

Issued date 1987.7 1991.10

Failed Cost 450Million $ 560Million $

Failed Amount 1.1 Million 1 Million

Abnormal Wear Out Wear Out

(Sintered Iron) (Lubrication at High Temp)
Failure Mechanism
Oil Reaction/ Oil Reaction/
Sludge Imbedding Sludge Imbedding

User Environment Worst Case Worst Case

After Disaster Withdraw Comp BIZ Lock Out factory

 Root Cause: Abnormal wear out at sintered iron (new parts)

2.2.5 Firestone and Ford tire in 2000

In the early of 2000 the Firestone and Ford tire experienced an unusually tire failures on the Ford

Explorer equipped with Firestone tires. The Ford Motor Company had a historically good relationship

with Firestone. As Firestone became a subsidiary of Japanese tire manufacturer Bridgestone in 1988,

they drifted apart. The U.S. National Highway Traffic Safety Administration (NHTSA) contacted Ford in

May 2000 and asked about the high incidence of Firestone tire failure on Ford Explorers model.

Immediately Ford found that it had very high failure rates from 15-inch Firestone tires models (See

Figure 2.9).

(a) Ford Explorer and Firestone Tires

(b) Firestone fallout

Fig.2.9.Firestone and Ford tire controversy from Wikipedia

Firestone recalled the millions of tires including 2.8 million Firestone Wilderness AT tires. A large

number of lawsuits have been filed against both Ford and Firestone that there had been over 240

deaths and 3,000 catastrophic injuries. The actual accidents come from separating a kind of tire tread

when cornering on cloverleaf interchange in high speed.

 Root Cause: Remove air from the tires (Minor design change)→ Tire heat up → Damage the

tire → Interaction of steel and rubber tire → Tread separation.

2.2.6 Toshiba Satellite notebook and battery overheating problem in 2007

As approximately 41,000 Toshiba laptops were reported for more than 100 cases of melting laptop

and minor injuries, Toshiba had to fell massive recalls in 2007. The basic cause might overheat and

expose a burning to consumers. Heat will generate when processors and batteries run. Laptops are

designed to provide adequate airflow for the fan and eliminate the overheating from the case.

However, due to the requirements of slim, less weight and compact design, notebooks will push heat-

generating components into a smaller space (See Figure 2.10).

Fig.2.10.Toshiba Satellite T130 notebook and battery overheating problem

  Root Cause: pushing so much processing power and battery into such a small space (design

problem)

2.2.7 Toyota Motor Recalls in 2009

The recalls of automobiles by Toyota Motor Corporation occurred in 2009 for approximately 5.2 million

vehicles – the pedal entrapment/floor mat problem, and for 2.3 million vehicles – the accelerator pedal

problem. As Toyota widened the recalls to include 1.8 million vehicles in Europe and 75,000 in China,

total recall number of cars in the world were considerable 9 million. The U.S. National Highway Traffic

Safety Administration (NHTSA) reached to conclusion that pedal misapplication was found

responsible for most of the incidents (See Figure 2.11).

 Root Cause: the pedal entrapment/floor mat problem

(a) Toyota crash (b) Pedal design problem

Fig.2.11. Recalls of automobiles by Toyota Motor Corporation

2.3 Development of Reliability Methodologies in History

2.3.1. In the early of 20’s century – starting reliability studies

The modern concept for reliability was beginning in 1816. The word “reliability” was first coined by

poet Samuel Taylor Coleridge [1]. At that time reliability in statistics was defined as the consistency of
a set of measurements to express a test. A test is reliable if the same result is repeated. For instance,

if a test is designed to measure special marks, the results should be approximately nearly identical to

the one. Reliability was a common concept that had been perceived as an attribute of a product.

Before taking up the main subject, the milestones of the historical reliability technology in the past

century might be briefly summarized in Table 2.5.

Table 2.5. History Summary of Reliability Technology

Wilhelm Albert publishes the first article on fatigue (1837)
A. Wöhler summarized fatigue test results on rail-road axles (1870)
O.H. Basquin proposes a log-log relationship for S-N curves (1901)
~1950 -
John Ambrose Fleming invented vacuum tubes in 1904
Griffith’st theory of fracture (1921)
A.M. Miner introduces a linear damage hypothesis (1945)

WW II Germany V - I, V - II Rocket development (R. Lusser’s Law)

Reliability of the Electron Power Tube
WW II US
(Aircraft electronic devices failure in the WW II)
Surveys and studies on the Electron Power Tube Reliability
1954 Japan
in the Vacuum Committee of the Institute of Electrical Engineers

US DOD formed the Advisory Group on the Reliability of Electronic

1952~1957 US Equipment (AGREE).
AGREE suggest vacuum tube follows the bathtub curve.

1954 US First National Symposium on Reliability and Quality Control, New York
Several conferences began to focus on various reliability topics (e.g.,
1950s US
1955 Holm Conference on Electrical Contacts)
1961 Italy The Rome Air Development Center (RADC) introduced a PoF program
Launched the Apollo program(FMEA & FTA),
1962 US
First Reliability and Maintainability Conference
1962 US First Symposium on Physics of Failure in Electronics, Chicago

1965 IEC Reliability and Maintainability Technical Committee, TC 56, Toy

1968 - Tatsuo Endo introduces the rain-flow cycle count algorithm

1971 Japan First Reliability and Maintainability Symposium

In the early times, reliability disasters were the rail accident that France Versailles frequently occurred

in 1842. August Wöhler investigated the causes of fracture in railroad axles and started the first

systematic studies of S-N Curve (or Wöhler Curve) [2,3]. To prevent the railroad disasters, S-N curve

of materials can be used to minimize the fatigue problem by lowering the stress at critical points in a
component. Griffith during World War I developed fracture mechanics to explain the failure of brittle

materials. He suggested that the low fracture strength observed in experiments was due to the

presence of microscopic flaws in the bulk material that can be still useful (Figure 2.12) [4].

Fig.2.12. an edge crack (flaw) of length a in a material

σ f √ a≈C (2-1)

whereσ f is the failure stress

Failure occurs when the free energy attains a peak value at a critical crack length.

2 Eγ
C=
√ π (2-2)

where E is the Young's modulus of the material and γ is the surface energy density of the material.

Invented in 1904 by John Ambrose Fleming, vacuum tubes were a basic component for electronics -

the diffusion of radio, television, radar, sound reinforcement, sound recording and reproduction, large

telephone networks, analog and digital computers, and industrial process control. The invention of the
vacuum tube made modern technologies of product applicable. By 1916, radio with vacuum tubes

was used to begin in the public (Figure 2.13). The concept of reliability by the problematic vacuum

tubes began in earnest to develop.

Karl Pearson first mentioned “negative exponential distribution” in 1895. His exponential distribution

had a number of interesting properties that were available in the 1950’s and 60’s. That is, one

property of serial system is the ability to add failure rates of different components in product. Simply

adding it was rather easily applicable at the time when using mechanical and later electric systems.

R ( t ) =R 1 ( t )⋅R 2 ( t ) ⋯R n ( t ) (2-3)

−λ 1 t −λ2 t − λn t
R ( t ) =e ⋅e ⋯e
(2-4)

−( λ1 +λ2 +⋯+ λn ) t
R ( t ) =e (2-5)

where R is reliability function,  is the failure rate, and t is the use time

Fig.2.13.British engineer John Ambrose Fleming and his vacuum tubes patents [5]
As automobiles came into more common use in the early 1920s, product improvement by the

statistical quality control was introduced by Walter A. Shewhart at Bell Laboratories. He developed the

control chart in 1924 and the concept of statistical control. Statistics as a measurement tool would

become connected with the development of reliability concepts. While designers were responsible for

product quality and reliability, technician took care of the failures. In the 1930s Quality and process

measures in automobile were still growing (Figure 2.14 and Figure 2.15).

Fig.2.14. A popular automobiles in the early 1920s [6]

Fig.2.15. Typical Control Chart [7]

 Fig.2.16. A simple spark-gap transmitter with HV capacitor and output tuning coil.

In the 1940s W. Edwards Deming stressed management responsibility for quality in the military short

lecture. He expressed that most of quality problems are actually due to system design errors, not

worker error [8]. For instance, an initial reliability concept was applied to the spark transmitters

telegraph because of the uncomplicated design. It was a battery powered system with simple

transmitters by wire. The main failure mode was a broken wire or insufficient voltage. After WWI,

greatly improved transmitters based on vacuum tubes became available (Figure 2.16).

2.3.2. In the World War II - new electronics failure in military

Before World War II, many concepts in reliability engineering still did not exist. However, many new

electronic products such as electronic switches, vacuum tube portable radios, radar and electronic

detonators are introduced into the military during the WWII (Figure 2.17).
Fig.2.17.ReliabilityMetric Tailored to the Leading Electronic Technology of the World War II and 1950’s

– Vacuum Tube and its assembly, Discreet Transistor & Diodes.

As the war began, it was discovered that half of the airborne electronics equipment in storage was

down in lifetime and unable to meet the military requirements (the Air Core and Navy). Reliability work

for this period had to do with new metal materials testing. Study for failure mechanism was only its

fatigue or fracture. For instance, M.A. Miner published the seminal paper titled “Cumulative Damage

in Fatigue” in 1945 in an ASME Journal. B [9]. Epstein published “Statistical Aspects of Fracture

Problems” in the Journal of Applied Physics in February 1948 [10] (Figure 2.18).

Fig.2.18. Palmgren-Miner linear damage hypothesis

 Fig.2.19. V-1 &V-2 missiles and Lusser's law

For some more interesting facts, Germany during World War II applied the basic reliability concepts to

improve reliability of their V1 and V2 rockets that consist of multi-modules (Figure 2.19). To complete

the mission of V1 and V2 rockets, Germany engineer had to improve their reliability. One of reliability

theory was Robert Lusser's law. By his law, the reliability of a series system is equal to the product of

the reliability of its component subsystems. It means that a series system is "weaker than its weakest

link", as the product reliability of a series of components can be less than the lowest-value

component. After World War II, the United States Department of Defense seriously recognized the

necessity for reliability improvement of its military equipment. This law became theoretical basis of

MIL-HDBK-217 and MIL-STD-756.

2.3.3. In the end of World War II and 1950’s –starting the reliability engineering

In the start of the 1950s the main military applications for reliability were the vacuum tube in radar

systems or other electronics because these systems proved problematic and costly during the World

War II (Figure 2.20). The vacuum tube computers that had a 1024 bit memory were invented to fill a

large room and consume kilowatts of power, , though grossly inefficient in modern..

During the war vacuum tubes mounted in these airplanes had been proved as the problematic parts,

though the component price was cheap. After the war, half of the electronic equipment for shipboard
was failed in lifetime. Failure modes of vacuum tubes in sockets were intermittent working problems.

The action plans for a failed electronic system were to exploding the system, removing the tubes and

re-installing them at proper time. In the renovation process, because the military had to consider the

cost issues, the operation and logistics costs for the vacuum tubes would become huge. To solve the

problem of vacuum tube, Institute of Electrical and Electronic Engineers (IEEE) in 1948 formed the

Reliability Society. Z.W. Birnbaum in 1948 had founded the Laboratory of Statistical Research at the

University of Washington, which served to use the concept of statistics. In 1951, to study reliability

problems with the Air Force Rome, Air Development Center (RADC) was established in Rome and

New York.

Fig.2.20.Typical vacuum tube failure - air leakage into the tube due to the crack tube

In 1950 a study group in military was initiated, which was thereafter called the Advisory Group on the

Reliability of Electronic Equipment (AGREE). By 1959, reports of this group suggested the following

three recommendations for the reliable systems such as vacuum tube: 1) there was a need to develop

reliable components for supplier, 2) the military should establish quality and reliability requirements (or

specifications) for component suppliers, and 3) actual field data should be collected on components to

search out the root causes of problems.

 Fig. 2.21. Bathtub curve for vacuum tube radio systems

A definition of product lifetime originally came from 1957 AGREE Commission Report. Task Group 1 in

AGREE has developed minimum-acceptable measures for the reliability of various types of military

electronic equipment, expressed in terms of Mean Time Between Failures (MTBF), though defining

lifetime for electronic components were currently inadequate. The final report of AGREE committee

suggested the reliability of product such as most vacuum tube followed the bathtub curve.

Consequently, reliability of components is often defined as “the bathtub curve” that has early failure,

useful life, and wear-out failure (Figure 2.21). Today the cumulative distribution function corresponding

to a bathtub curve was replaced with a Weibull chart in reliability engineering.

In the early of 1950s, a conference on electrical contacts and connectors was initiated to study the

reliability physics - failure mechanisms and reliability topics. In 1955, RADC issued “Reliability Factors

for Ground Electronic Equipment.” by Joseph Naresky [11]. The conference was publishing

proceedings, entitled as “Transaction on Reliability and Quality Control in Electronics”, merged with an

IEEE Reliability conference and became the Reliability and Maintainability Symposium.

As television in the 1950s was introduced, more vacuum tubes were utilized in America house. Repair

problems were often due to the failure of one or more vacuum tubes. V acuum tube was a critical
switching device that controls electric current through a vacuum in a sealed container - cathode ray

tube. Typical reliability problems of the tubes with oxide cathodes evolved as 1) reduce its ability to

emit electrons, 2) a stress-related fracture of the tungsten wire, 3) air leakage into the tube, and 4)

glowing plate - a sign of an overloaded tube. Most vacuum tube in radio systems followed a bathtub-

type curve was easy to develop replaceable electronic modules - Standard Electronic Modules

(SEMs), and then restore a failed system.

AGREE committee also recommended to formally testing products with statistical confidence. And it

would carry out the environmental tests that have ultimate temperature and vibration conditions ,

which became Military Standard 781. The AGREE report originally stated that the definition for

reliability is “the probability of a product performing without failure a specified function under given

conditions for a specified period of time”.

Robert Lusser, Redstone Arsenal, pointed out that 60% of the failures of one Army missile system

were due to components that reported on “Predicting Reliability” in 1957. He also stressed that current

quality methods for electronic components were inadequate and that new concepts for electric

components was implemented. ARINC set up an improvement process with vacuum tube suppliers

and reduced infant mortality removals by a factor of four. This decade ended with Radio Corporation

of America (RCA) publishing information in TR1100 on the failure rates of some military components.

RADC used these concepts, which became the basis for Military Handbook 217. Over the next

several decades, Birnbaum made suggestions on Chebychev’s inequalities, non-parametric statistics,

reliability of complex systems, cumulative damage models, competing risk, survival distributions and

mortality rates.

Walodie Weibull was working in Sweden and investigated the fatigue of materials. He created a

Weibull distribution. In 1939 Wallodie Weibull suggested a simple mathematical distribution, which

could represent a wide range of failure characteristics by changing two parameters. The Weibull

failure distribution does not apply to every failure mechanism but it is useful tool to analyze many of

the reliability problems. In 1951 he presented his most famous papers to the American Society of

Mechanical Engineers (ASME) on Weibull distribution with seven case studies. Between 1955 and

1963, he investigated the fatigue and creep mechanisms of materials. He derived the Weibull

distribution on the basis of the weakest link model of failures in materials. By 1959, he produced
“Statistical Evaluation of Data from Fatigue and Creep Rupture Tests: Fundamental Concepts and

General Methods” as a Wright Air Development Center Report 59-400 for the US military [12].

Fig.2.22. Sweden engineer Wallodie Weibull and his Weibull distribution

In 1961, Weibull published a book on materials and fatigue testing while working as a consultant for

the US Air Force Materials Laboratory [13]. The American Society of Mechanical Engineers awarded

Weibull their gold medal in 1972. The Great Gold Medal from the Royal Swedish Academy of

Engineering Sciences was personally presented to him by King Carl XVI Gustaf of Sweden in 1978
(Figure 2.22).

As Weibull analysis methods and applications were propagating, a number of people began to use the

Weibull chart. Dorian Shain in wrote an early booklet on Weibull in the late of 1950s, while Leonard

Johnson at General Motors helped improve the plotting methods by suggesting median ranks and

beta Binomial confidence bounds in 1964. Dr. Robert Abernethy developed a number of applications,

analysis methods and corrections for the Weibull function. Professor Gumbel demonstrated that the

Weibull distribution is a Type III Smallest Extreme Value distribution such as Eqs. (2-6) and (2-7)[14].

Dr. Robert Abernethy developed a number of applications, analysis methods and corrections for the

Weibull function

c
a−x
F ( x )=exp −
( ( ))
b (2-6)

where x  a, b > 0, c > 0

Ea
K= A exp − ( ) RT (2-7)

where k is the rate constant of a chemical reaction, T is the absolute temperature, A is the pre-factor,

Ea is the activation energy, and R is the universal gas constant.

In 1963, Weibull was a visiting professor at Columbia and there worked with professors Gumbel and

Freudenthal in the Institute for the Study of Fatigue and Reliability. While he was a consultant for the

US Air Force Materials Laboratory, he published a book on materials and fatigue testing and the

related reports till 1970.

Aeronautical Radio, Incorporated (ARINC) set up an improvement process with vacuum tube

suppliers to reduce its infant mortality removals. As publishing information in TR1100 on the failure
rates of some military components, it became the basis for Military Handbook 217 - "Reliability

Prediction of Electronic Equipment." Navy Military published Handbook 217 in 1962. Papers for

electronic components were being published at conferences: “Reliability Handbook for Design

Engineers” published in Electronic Engineers, in 1958 by F.E. Dreste and “A Systems Approach to

Electronic Reliability” by W.F. Leubbert in the Proceedings of the IRE (1956) [15]. C.M. Ryerson

produced a history of reliability to 1959 in the proceedings of the IRE entitled as Proceedings of the

IEEE [16].

2.3.4. In the 1960’s and present: mature of reliability methodology – Physics of Failure (PoF)

Fig.2.23, A transistor radio with multiple parts from Wikipedia

Physics of Failure (PoF) for electronic components in 1960s started with several significant events –

invention of the transistor in 1947 and transistor radio in 1954, which became the most popular

electronic communication device during the 1960s and 1970s. People with pocket size listened to

music everywhere (Figure 2.23). These devices had some problems – electromechanical faults,

transistor failure, and capacitor problems. POF is a kind of systematic approach to the design and

development of reliable product to prevent failure. Based on the knowledge for the root cause of
failure mechanisms, system can improve its preformance.

RADC worked in earnest the Physics of Failure in Electronics Conference sponsored by Illinois

Institute of Technology (IIT). In the 1960s America strong commitment to space exploration would turn

into National Aeronautics and Space Administration (NASA), a key efforts to improve the reliability of

components and systems that could work properly to complete the space missions. RADC produced

the document “Quality and Reliability Assurance Procedures for Monolithic Microcircuits.”

Semiconductors were a popular use in small portable transistor radios. Next, low cost germanium and

silicon diodes were able to meet the requirements. Dr. Frank M Gryna published a Reliability Training

Text through the Institute of Radio Engineers (IRE).

Fig.2.24. Andy Grove, Bruce Deal, and Ed Snow at the Fairchild Palo Alto R & D laboratory and first

commercial Metal Oxide Semiconductor (MOS) IC in 1964 from Wikipedia

In this period the nuclear power industry and the military – missiles, airplanes, helicopters and

submarine applications enabled the reliability problems of a variety of technologies to initiate POF.

The study of Electro-Magnetic Compatibility (EMC) system effects was initiated at RADC in the 1960s

(Figure 2.24).

One of the milestones in the Proceedings of the 7 th National Symposium of Reliability and Quality

Control was the proof of the effectiveness of the Arrhenius model for semiconductors in 1962. G.A.
Dodson and B.T. Howard of Bell Labs published the papers, entitled as “High Stress Aging to Failure

of Semiconductor Devices.” [17] This conference also issued lots of other papers. It could look at the

technical improvement of at the technical improvement of other electronic components, and renamed

as the Reliability Physics Symposium (RPS) in 1967. Shurtleff and Workman in the late of sixty issued

the original paper on step stress testing that establishes limits when applied to Integrated Circuits.

Electro-migration in electronic system is one of failure mechanism which applied to the transport of

mass in metals when the metals are stressed at high current densities. J.R. Black published his work

on the physics of electro-migration in 1967. Since the number of free charge carriers increases with

temperature, silicon in semiconductor began to dominate reliability activities for a variety of industries.

The U.S. Army Material Command issued a Reliability Handbook (AMCP 702-3) in 1968. On the other

hands Shooman’s Probabilistic Reliability also was issued to explain statistical methods.

To investigate the failure mode of electronic components, automotive industry was published a FMEA

handbook for technical improvement of suppliers, not yet published as a Military standard. As a series

of commercial satellites were launched, the reliability study for communications was strengthened by

International Telecommunications Satellite Organization (INTELSAT) that was providing international

broadcast services between the U.S. and Europe in 1965. Professionals around the world took part in

reliability conferences. As Apollo was landing a moon, people recognized how far reliability had

progressed in the recent decade.

1 t 2−μ2 1 t μ
f ( t )= 2 2
(√ √ )
2μ γ √ π t u
−
exp − 2 + −2
γ μ t [ ( )]
u t (2-8)

where γ is a shape parameter, μ is a scale parameter

As seen in Eq. (2-8), in 1969, Birnbaum and Saunders suggested a life distribution model that could

be derived from a physical fatigue process where crack growth causes failure. Since one of the best
ways to choose a life distribution model is to derive it from a physical/statistical argument that is

consistent with the failure mechanism, the Birnbaum-Saunders fatigue life distribution is worth

considering.

As the microcomputer had been invented in the 1970s, RAM memory size was growing at a rapid

rate. Vacuum tube was replaced with Integrated Circuit (IC). The variety of ICs – Bipolar, NMOS and

CMOS increased very rapidly. In the middle of the 1970s, Electrostatic discharge (ESD) and Electrical

Over Stress (EOS) were discussed by some papers and eventually became the hot issues of a

conference in the decade end.

In the same manner, studies for passive components – resistor, inductor, and capacitor in

International Reliability Physics Symposium (IRPS) moved to a Capacitor and Resistor Technology

Symposium (CARTS). The progressive papers on gold aluminum inter-metalics, accelerated testing,

and the use of Scanning Electron Microscopes (SEM) were in a few highlights of the decade.

In middle of 1970s, Hakim and Reich published a paper on the evaluation of plastic encapsulated

transistors and ICs on field data. And two most memorable reliability papers were one on soft-errors

from alpha particles first reported by Woods and May and on accelerated testing of ICs with activation

energies calculated for a variety of failure mechanisms by D.S. Peck. In the end of the decade,

Bellcore collected commercial field data and became the basis of the Bellcore reliability prediction

methodology used widely with MIL-STD-217F.

During the Apollo space program, the spacecraft and its components worked reliably all the way to the

moon and back. In coming to the Navy, all contracts should contain specifications for reliability and

maintainability instead of just performance requirements. Military Standard 1629 on FMEA was issued

in 1974, NASA made great strides at designing and developing spacecraft such as the space shuttle.

Their emphasis was on risk management through the use of statistics, reliability, maintainability,

system safety, quality assurance, human factors and software assurance. Reliability had expanded

into a number of new areas as technology rapidly advanced. Emphasizing temperature cycling and

random vibration became ESS testing, eventually issued as a Navy document P-9492 in 1979 and

make a book on Random Vibration with Tustin in 1984. The older quality procedures were replaced

with the Navy Best Manufacturing Practice program.

In the 1980s, televisions had become product that used all semiconductors. Automobiles rapidly

increased their use of semiconductors with a variety of microcomputers. Large air conditioning

systems, microwave ovens, and a variety of other appliances developed one chip electronic

controllers. Communications systems began to adopt electronics to replace older mechanical

switching systems. Bellcore issued the first consumer prediction methodology for telecommunications

and SAE developed a similar document SAE870050 for automotive applications.

Fig.2.25 Bathtub curve and straight line with slope .

As seen in Figure 2.25, during this decade, as the failure rate of many electronic components

including mechanical components were dropped by a factor of 10, engineer questioned on the bathtub
curve. For such a situation, the traditional failure rate typified by the bathtub curve can be reduced to

resemble the failure rate represented by a flat, straight line with the shape parameter .

Software became important to the systems improvement by advancing with work at RADC. Software

reliability developed models such as Musa Basic to predict the number of missed software faults that

might remain in code. The Naval Surface Warfare Center issued Statistical Modeling and Estimation

of Reliability Functions for Software in 1983. Contributions by William Meeker, Gerald Hahn, Richard

Barlow and Frank Proschan developed models for wear, degradation and system reliability.

The PC came into dominance as a tool for measurement & control and enhanced the possibility of

canned programs for evaluating reliability. In the end of the decade, FMEAs, FTAs, reliability

predictions, block diagrams and Weibull Analysis programs were performed in the commercial use.

The challenger disaster caused people to recognize the assessment significance of system reliability.

Many of the military specifications–Military Handbook 217 became obsolete and best commercial

practices were often adopted. Most industries developed their own reliability standards like the

JEDEC Standards for semiconductors and the Automotive Standard Q100 and Q101. Afterward the

last century the rise of the internet created a variety of new challenges - micro-electro mechanical

systems (MEMS), hand-held GPS, and handheld devices - for reliability. Consumers have become

more aware of reliability disasters. In many ways, reliability became part of everyday life and

consumer expectations. The developed methodology in reliability engineering has widely been

developing until now. However, new methodology for reliability is still required to find the problematic

parts that is the main cause of recall before production.

REFERENCES

[1] Saleh JH and Marais K (2006) Highlights from the Early (and pre-) History of Reliability.

Engineering Reliability Engineering and System Safety 91 (2):249-256

[2] Wöhler A (1855) Theorie rechteckiger eiserner Brückenbalken mit Gitterwänden und mit

Blechwänden. Zeitschrift für Bauwesen 5:121-166

[3] Wöhler A (1870) Über die Festigkeitsversuche mit Eisen und Stahl. Zeitschrift für Bauwesen 20:73-
106

[4] Griffith AA (1921) The phenomena of rupture and flow in solids. Philosophical Transactions of the

Royal Society of London A 221: 163–198

[5] J.A. Fleming, U.S. Patent 803,684, 17 Nov. 1815

[6] Ford (1929) 1930 model brochure – Beauty of Line – Mechanical excellence. Retrieved 24 May

2012.

[7] Shewhart WA (1931) Economic control of quality of manufactured product. D. Van Nostrand

Company, New York

[8] Deming WE and Stephan F (1940) On a least squares adjustment of a sampled frequency table

when the expected marginal totals are known. Annals of Mathematical Statistics. 11 (4): 427–444

[9] Miner MA (1945) Cumulative Damage in Fatigue. Journal of Applied Mechanics 12(3):59-64

[10] Epstein B (1948) Statistical Aspects of Fracture Problems Journal of Applied Physics 19.

[11] Naresky JJ (1962) Foreword. Proceedings of First Annual Symposium on the Physics of Failure in

Electronics, September 26-27

[12] Weibull W (1959) Statistical Evaluation of Data from Fatigue and Creep Rupture Tests, Part I:

Fundamental Concepts and General Methods. Wright Air Development Center Technical Report 59-

400, Sweden, September.

[13] Weibull W (1961) Fatigue Testing and Analysis of Results. Pergamon Press: London.

[14] Abernethy R (2002) The New Weibull Handbook. 4th edition self published ISBN 0-9653062-1-6

[15] Lloyd D and Lipow M (1962) Reliability: Management, Methods and Mathematics. Prentice Hall:

Englewood Cliffs.

[16] Knight R (1991) Four Decades of Reliability Progress. Proceedings of Annual RAMS pp156-160

[17] George E (1998) Reliability Physics in Electronics: A Historical View. IEEE Transactions on

Reliability 47(3): 379-389

Chapter 3

Modern definitions in reliability engineering

Seong-woo Woo

Email: [email protected]

Reliability Association of Korea, Seoul, Korea (Republic of)

Abstracts: This chapter will briefly review the modern definitions in reliability engineering that can be

used widely – bathtub, MTBF (Mean Time Between Failure), fundamentals in statistics and probability

theory, statistical distributions like Weibull, and time-to-failure model. From customer's standpoint,

when a product is delivered to the end-user, reliability explained as lifetime and failure rate can be

assessed through product specification. Many reliability concepts used in the product predict the

failure rate/lifetime of components subjected to random loads. So the product reliability is related to

the robust design of mechanical system without design defects in lifetime. For mechanical engineer,

reliability theory may feel complex because of concepts of probability and statistics. However, the

reliability concepts are required to develop new methodology of reliability assessment in product. It

will help to establish the testing method of reliability that points out the design failure or reliability

disasters in the reliability-embedded developing process. As a design parameter method, Taguchi

method has developed many concepts but it still has the weak points. Based on Taguchi concepts,

new reliability methodology is still required to discover the design problem of parts.

Keywords: Reliability concepts, Reliability engineering, Probability theory, Robust design.

3.1 Introduction

Reliability is the ability of an item to work properly the intended functions during its lifetime. The

modern concepts in reliability engineering started through the reliability study of vacuum tube in the
WW2. The reliability concepts except the quality control in product manufacture can focus on the

study of quality itself in design.

3.1.1 Bathtub curve

Fig. 3.1.Bathtub curve

To describe three types of the failure rate (or hazard function in product, the bathtub curve in reliability

engineering comprises as (Figure 3.1):

 Early failures in the first part are a decreasing failure rate (β<1).

 Random failures in the second part are a constant failure rate (β=1).

 Wear-out failures in the third part are an increasing failure rate (β>1).
As the failure of the vacuum tube was studied in WWII, the bathtub curve was created by mapping the

rate of early "infant mortality" failures, the rate of random failures with constant failure rate during its

"useful life", and finally the rate of "wear out" failures as the product exceeds its design lifetime. In the

early life of a product following to the bathtub curve, the failure rate is high but rapidly decreasing as

defective products are removed, and early sources of potential failure such as storage, handling and

installation error are dominated. In the mid-life of a product—generally, once it reaches consumers—

the failure rate is low and constant. In this period product may experience the catastrophic disaster if

design problem in product exist. Before production, design problem should be found by proper

method - accelerated life testing. In the late life of the product, the failure rate increases. Thus, there

are three types of reliability testing in accordance with the failure rate (Figure 3.2).

 Early failures: Because it requires the short test time, it easily improve prior to shipment

 Random failures: Specific environmental tests related to shipping/usage/disposal

- Shipping testing: storage, transportation by vehicle/ship/rail

- Usage Testing: High voltage, lightning, shock, temperature & humidity, EMC (EMI);

- Disposal (Low frequency, Pass/Fail testing)

 Catastrophic disasters: Catastrophic disasters often happen in 1-2 years. It comes from the

design missing. If it is reproduced by accelerated life testing described in Chapter 7, it can be

eliminated.

The term "Military Specification" is also used to describe systems in which the infant mortality section

of the bathtub curve has been burned out or removed. This is done mainly for life critical or system

critical applications as it greatly reduces the possibility of the system failing early in its life.

Manufacturers will do this at some cost generally by means similar to environmental stress screening.

In reliability engineering, the cumulative distribution function corresponding to a bathtub curve may be

transformed and analyzed by using a Weibull chart.

Fig. 3.2. Three types of reliability tests in accordance with the failure rate
3.2 Fundamentals in Probability Theory

The failure behavior of product and components can be expressed as the statistical and probability

theory because of these random events in field. In market data it is proper to know the failure

characteristics of sample data from a population of items. For example, if hundred televisions put on

test and 12 among them fail, analyze the times to failure. If thousand aircraft engine controllers are

operating in service, collect all the times to failure data and analyze them. Here test data may be not

only times but distance or cycles etc.

3.2.1 Probability

The probability was originally established by gamblers who were interested in high stakes. To answer

the question “how probable”, it is that a certain event A occurs in a game of gambling. An early

mathematician, Laplace and Pascal, invented the probability. That is, when N is the number of times

that X occurs in the n repeated experiments, the probability of occurrence of event X, P(X), can be

defined as:

P ( X )= lim ( number of cases favorable ¿ X ¿ ¿ number of all possible cases ) =lim ( N /n )

n→∞ n→ ∞

P ( X )= lim ( number of cases favorable ¿ X ¿ ¿ number of all possible cases ) =lim ( N /n )

n→∞ n→ ∞

(3.1)

where X is a random variable

For example, if trials n approaches∞ , the probability of rolling a 1 with a die is:

1 1
P ( X=3 )= =0.167P ( X=3 )= =0.167
6 6

(3.2)
In general, this is adequate for gambling. On the other hands, in technical reality, the failure

probabilities happen to vary amounts. In modern theory, probability is seen as a basic principle and

has the assumptions:

 Each random variable X has 0 ≤ P ( X ) ≤ 1.

 The area under the curve is equal to 1: ∫ P ( X ) dX =1, where 0 ≤ X ≤ ∞

 If X 1 , X 2 , X 3 ,… are random variables, then P ( X 1 ∪ X 2 ∪ … )=P ( X 1) + P ( X 2 ) +…

a
P ( a≤ X≤b )=∫b f ( X ) dX
 Area under the curve between two values is the probability:

If not all data is normally distributed, other distributions - Weibull analysis is especially suited to failure

rates. Select the failure data and draw histogram. We can find the skewed right (or left) histogram like

Weibull distribution. When failure behavior is represented graphically, basic probability concepts are

mean, median, mode, standard deviation (See Figure 3.3).

Fig. 3.3.Mean, median and mode for skewed left/right, and symmetric distribution
3.2.1.1 Mean

For a data set, mean refers to one measure of the central tendency either of a probability distribution

or of the random variable characterized by that distribution. If we have a data set containing the failure

times t 1 , t 2 , … , t nt 1 , t 2 , … , t n, the mean is defined by the formula:

n n
ti ti
t 1 +t 2+ …+t n ∑i=1 t 1 +t 2+ …+t n ∑i=1
t m= = t = =
n n m n n

(3.3)

The mean describes the parameter where the middle of the failure times approximately locates. The

mathematical mean is affected to the lowest or highest failure times.

3.2.1.2 Median

Median is the number separating the higher half of a data sample. In reliability testing, the median is

the time in the middle of failure data. The median may be determined by the cumulative distribution

function F(t).

F ( t median ) =0 . 5

(3.4)

The mathematical median is not affected to the lowest or highest failure times.
3.2.1.3 Mode

The mode is the value that appears most often in a set of data. In reliability testing, the mode is the

most frequent failure time. The mode is the maximum value of the density function f(t). So it can be

expressed as:

'
f ( t mod e )=0 (3.5)

The mathematical median is not affected to the lowest or highest failure times.

3.2.1.4 Standard Deviation

In statistics, the standard deviation (SD) is used to quantify the variation amount of a set of data

values. In reliability testing, the standard deviation is the square root of the variance. This is

expressed by

n 1/ 2 n 1/ 2

σ= [ ∑ ( t i−t m )
i=1
n
2

] [σ=
∑ ( t i−t m )
i=1
n
2

]
(3.6)

The standard deviation has the same dimension as the failure times ti
3.2.1.5 Expected value

In probability theory, the expected value of a random variable is intuitively the average value of the

long-run repetition experiment. The expected value, E(t), of a continuous random variable is

expressed by

∞ ∞
E(t )=M = ∫ tf ( t ) dt E(t )=M = ∫ tf ( t ) dt
−∞ −∞

(3.7)

3.2.2 Probability distributions

Probability distributions are typically defined in terms of the probability density function. However,

there are a number of probability functions used in applications.

3.2.2.1 Reliability function

The common used function in reliability engineering is the reliability function. This function is the

probability of an item operating for a period of time without failure. The reliability function can be

expressed as:

∞
R ( t ) =P ( T >t )=1−F ( t )=∫t f ( x ) dx
(3.8)

R(t) is the probability that the item will not fail in the interval (0, t]. R(t) is the probability that it will

survive at least until time t – it is sometimes called the survival function (See Figure 3.4).
 Fig. 3.4.Cumulative distribution function F(t) and Reliability function f(t)

3.2.2.2 Cumulative distribution function

The cumulative distribution function (CDF) is the probability that the variable t takes a value less than

or equal to T. CDF associated with the time to failure T is expressed as:

F ( t )=P ( T≤t )
(3.9)

which is the probability that the system fails within the time interval (0; t]. If T is a continuous random

variable, the probability function is related to its probability density function f (t) by

t
F ( t )=∫ f ( x ) dx
0 (3-10)

3.2.2.3 Probability Density function (PDF)

 Fig. 3.5 Density function f(t) and cumulative distribution function F(t)

In probability theory, a probability density function (PDF) is a function that describes the relative

likelihood for this random variable to take on a given value (Figure 3.5). In reliability testing, density

function f(t) is defined by:

t
d ∫ f ( x ) dx
dF ( t ) 0
= =f (t )
dt dt (3-11)

3.2.2.4 Failure rate

Failure rate (or Hazard rate function) is the frequency with which an engineered system or component

fails. Consider the conditional probability:

 P ( t <T ≤t+ Δt ) F ( t +Δt )−F ( t )
P ( t<T <t+ Δt|T >t )= =
R (t ) R (t ) (3-12)

In reliability engineering, failure rate (or hazard rate function) (t) is defined by:

P ( t <T <t +Δt|T >t ) f ( t )

λ ( t )= lim =
Δt→ 0 Δt R (t )
(3-13)

(t) dt is the probability that the system will fail during the period (t; t + dt], given that it has survived

until time t.

3.2.2.5 Cumulative hazard rate function

A survival and hazard function is to analyze the expected duration of time until one or more events

happen, such as failure in mechanical systems. Cumulative hazard rate function (t) is defined by:

t
Λ ( t ) =∫ λ ( x ) dx
0

(3-14)

Suppose the failure rate (t) is known. Then it is possible to obtain f (t), F(t), and R(t).

dF ( t ) dR (t ) dR/ dt
f (t )= =− ⇒ λ ( t ) =−
dt dt R (3-15)
 dR
=−λ ( t ) dt
R
(3-16)

If Eq. (3-16) is integrated, then reliability function becomes

[
R ( t ) =exp −∫ λ ( τ ) dτ
0
] (3-17)

So the density function and cumulative distribution function are defined as:

[
f ( t )= λ ( t ) exp −∫ λ ( τ ) dτ
0
] (3-18)

[
F ( t )=1−exp −∫ λ ( τ ) dτ
0
] (3-19)

Relationship between reliability function R(t) and cumulative distribution function F(t) can be

summarized in Figure 3.6.

Fig. 3.6 Relationship between Reliability function R(t) and Cumulative distribution function F(t)
3.3 Reliability lifetime metrics

An important goal for reliability designers is to assess lifetime against product failures. Reliability

lifetime metrics are used to quantify a failure rate and the resulting time of expected performance.

MTTF, MTBF, MTTR, FIT and BX% life are reliability lifetime metrics as follows:

 MTTF (Mean Time To Failure)

 MTBF (Mean Time Between Failure),

 MTTR (Mean Time To Repair),

 BX% life

3.3.1 Mean Time To Failure (MTTF)

MTTF is a basic lifetime metric of reliability to specify the lifetime of non-repairable systems – “one-

shot” devices like light bulbs. It is the mean time until a piece of equipment fails at first statistically.

MTTF is the mean over a long period of time with a large unit (Figure 3.7).

Fig. 3.7. Concept of Mean Time To Failure

 t 1 +t 2 +.. . .. .+t n
MTTF =
n (3-20)

As seen in Figure 3.7., we know that MTTF is 23,000 km if using Eq. (3.20). And the MTTF can be

described with other mathematical terms:

∞ ∞∞
dR ( t )
MTTF =E ( T )=∫ t⋅f ( t ) dt=−∫ t dt=∫ R (t ) dt
0 0 dt 0 (3-21)

d d
f ( t )= F ( t )=− R ( t )
where dt dt

Example 3.1. Consider a system with reliability function

1
R (t)= , for t>0
( 0 . 2t +1 )2 (3-22)

Find the probability density function, failure rate, and MTTF

d 0. 4
f ( t )=− R (t)=
Probability density
dt ( 0 . 2t +1 )3

f (t) 0.4
λ ( t )= =
Failure rate R ( t ) ( 0. 2 t+1 )

∞
MTTF =∫ R ( t ) dt= 5 months
Mean time to failure 0
3.3.2 Mean Time Between Failure (MTBF)

Mean Time Between Failure (MTBF) is a reliability metric used to describe the mean lifetime of

repairable components - computers, automobiles, and airplanes. MTBF remains a basic measure of a

systems’ reliability for most products, though it still is debated and changed. MTBF still is more

important for industries and integrators than for consumers.

Fig. 3.8. Concept of Mean Time Between Failure

T
MTBF =
n (3.23)

MTBF value is equivalent to the expected number of operating hours (service life) before a product

fails. There are several variables that can impact failures. Aside from component failures, customer

use/installation can also result in failure. MTBF is often calculated based on an algorithm that factors

in all of a product’s components to reach the sum life cycle in hours. MTBF is considered a system

failure. It is still regarded as a useful tool when considering the purchase and installation of a product.

For repairable complex systems, failures are considered to be those out of design conditions which

place the system out of service and into a state for repair. Technically, MTBF is used only in reference
to a repairable item and non-repairable items, while MTTF is used for non-repairable items like

electric components.

3.3.3 Mean Time To Repair (MTTR)

Mean Time To Repair (MTTR) is the average lifetime needed to fix a problem. In an operational

system, repair generally means replacing a failed hardware part. Thus, hardware MTTR could be

viewed as mean time to replace a failed hardware module. Taking too long to repair a product drives

up the cost of the installation in the long run, due to down time until the new part arrives and the

possible window of time required scheduling the installation. To avoid MTTR, many companies

purchase spare products so that a replacement can be installed quickly. Generally, however,

customers will inquire about the turn-around time of repairing a product, and indirectly, that can fall

into the MTTR category. And relationship among MTTF, MTBF and MTTR can be described in Fig.

3.9.

Fig. 3.9. A schematic diagram of MTTF, MTTR, and MTBF

3.3.4 BX% life

The BX life metric originated in the ball and roller bearing industry, but has become a a product

lifetime metric used across a variety of industries today. It’s particularly useful in establishing warranty
periods for a product. The BX% life is the lifetime metric which takes to fail X% of the units in a

population. For example, if an item has a B10 life of 1000 km, then 10% of the population will have

failed by 1000 km of operation.

Fig. 3.10. Concept of BX% life

Alternatively, the B10% life has the 90% reliability of a population at a specific point in product lifetime.

The “BX” or “Bearing Life” nomenclature refers to the time at which X% of items in a population will

fail. The B10 life metric became popular among product industries due to the industry’s strict

requirement. Now B1, B10 and B50 lifetime values serve as a measurement for the reliability of a

product (Fig. 3.10).

3.3.5 The inadequacy of the MTTF (or MTBF) and the alternative metric BX life

Two representative metrics of reliability may describe product lifetime and the failure rate. The failure

rate is adequate for understanding situations that include unit periods, such the annual failure rate.

But the lifetime is frequently indexed using the mean time to failure.

The MTTF are misinterpreted. For instance, assume that the MTTF of a printed circuit assembly for

television is 40,000 hours. Annual usage reaches 40,000 divided by 2,000 and become 20 years,

which is regarded as the lifetime of the unit. The average lifetime of the television PCA is assumed to

be 20 years. But because actual customer experience is that the lifetime of a television is a 10 years,

this can lead to misjudgments or overdesign that wastes material.

Table 3.1 Results of 1987 Army SINCGARS Study*

MIL - HDBK-217 Actual Test

Vendor
MTBF(hour) MTBF(hour)
A 2840 1160

B 1269 74

C 2000 624

D 1845 2174

E 2000 51

F 2304 6903

G 3080 3612

H 811 98

I 2450 472

* The transition from statistical field failure based models to physics-of-failure based models for

reliability assessment of electronic packages, EEP-Vol. 10-2, 619-625, Advances in Electronic

Packaging ASME 1995, - T.J. Stadterman et al

MTTF is often assumed to be the same as lifetime because customers understand the MTTF as,

literally, the average lifetime because customers understand the average lifetime of their appliances,

so they suppose products will operate well with until they reach the MTTF. In reality, this does not

happen. By definition, the MTTF is an arithmetic mean; specifically, it equals the period from the start

of usage to the time that the 63 rd item fails among 100 sets of one production lot when arranged in the

sequence of failure times.

Under this definition, the number of failed televisions before the MTTF is reached would be so high

that customers would never accept the MTTF as a lifetime index in the current competitive market.

The products of first-class companies have fewer failures in a lifetime than would occur at the MTTF.

In the case of home appliances, customers expect no failure for the 10 years. The failure of the TV is

accepted from the customer’s perspective in the later time. Customers would expect the failure of all

televisions once the expected use time is exceeded – 12 years in the case of a television set – but

they will not accept major problems within the first 10 years,

The MTTF is inappropriate as a lifetime index. Alternatively, it is reasonable to define the lifetime as

the point in time when the accumulated failure rate has reached X %. This is called the BX life. The

value X may vary from product to product, but for home appliance, the time to achieve a 10 to 30%

cumulative failure rate failure rate, B20~30 life, exceeds 10 years. Thus, an average annual failure

rate equals to 1 ~ 3%.

Now let’s calculate the B10 life from the MTTF of 40,000 hours. Since the annual usage is 2,000

hours, the B10 life is 2 years, which means that the yearly failure rate would be 5%. The reliability

level of this television, then, would not be acceptable in light of the current annual failure rate of

1~3%. The misinterpretation of reliability using an MTTF of 20 years would lead to higher service

expenses if the product were released into the market without further improvement. The lifetime of a

television is 12 to 14 years, not 20 years. Since random failure cannot account for the sharply

increasing failure rate, the MTTF based on random failure or on an exponential distribution is

obviously not the same as the design lifetime of product (Table 3.1).
3.4 Statistical Distributions

3.4.1 Poisson Distributions

Fig. 3.11. Poisson Distributions

The Poisson distribution is named after Simeon Poisson (1781-1840), a French mathematician, and

used in situations where big declines in a time period occurs with a specific average rate, regardless

of the time that has elapsed. More specifically, this distribution is used when the number of possible

events is large, but the occurrence probability over a specified time period is small. Two examples of

such a situation are as follows:

 A store that rents books has an average rental of 200 books every Saturday night. Using this

data, you can predict the probability that more books will sell (perhaps 300 or 400) on the
 following Saturday nights.

 Another example is the number of diners in a certain restaurant every day. If the average

number of diners for seven days is 500, you can predict the probability of a certain day having

more customers.

A Poisson distribution has the following properties:

 The experiment results in outcomes that can be classified as successes or failures.

 The average number of successes (μ) that occurs in a specified region is known.

 The probability that a success will occur is proportional to the size of the region.

 The probability that a success will occur in an extremely small region is virtually zero.

This distribution also has applications in many reliability areas when one is interested in the

occurrence of a number of events that are of the same type. Each event’s occurrence is denoted as a

time scale and each event represents a failure. The Poisson density function is expressed by

( λt ) x e− λt
f ( x )=
x! for x=0, 1, 2,…..,n (3-24)

where  is the constant failure rate, t is the time

If exponential distribution follows, we can let m = t.

 ( m) x e−m
P ( x , m )=
x! (3-25)

The cumulative distribution function is given by

y
F ( y )=∑ [ ( λt )i e− λt ] /i!
i=0 (3-

26)

In a certain region, the number of traffic accidents averages one per two days happens. Find the

probability that x= 0, 1, 2 accidents will occur in a given day.

So the number of traffic accidents averages one per two days, m = t = 0.5,

( 0. 5 )0 e−0 . 5
f ( 0 )= =0 . 606
X=0, 0! ,

(3-27)

Accident days = 365 day  0.606 = 221 day

( 0 .5 )1 e−0. 5
f ( 1 )= =0. 303
X=1, 1! , (3-28)

Accident days = 365 day  0.303 = 110 day

 ( 0 . 5 )2 e−0. 5
f ( 2 )= =0. 076
X=2, 2! ,

(3-29)

Accident days = 365 day  0.076 = 27 day

TV is selling in a certain area and average failure rate is 1%/2000hr. If 100 TV units are sampling and

testing for 2,000 hours, find the probability that no accidents, x= 0, will occur.

m=n⋅λ⋅t=100×0 . 01/2000×2000=1
(3-30)

Because no accident, the probability is

( 1 )0 e−1
X =0 . f ( 0 )= =0 . 36
0! (3-31)

We can estimate the confidence level is 63% for 100 TV units. If no accidents, x= 0, keep and the

confidence level would like to increase to 90%, how many TV units will it requires?

( m )0 e−m
X =0 , f ( 0 ) = =0 . 1
0! (3-32)
So if m = 2.3, the required sample size n = 230 will be obtained as

m=n⋅λ⋅t=n×0 . 01/2000×2000=2. 3
(3-33)

3.4.2 Exponential Distributions

The exponential distribution has a widely used application in reliability engineering because many

engineering modules exhibit constant failure rate during the product lifetime. Also, it is relatively easy

to handle in performing reliability analysis.

From Eq. (3-25), let x = 0. We can also obtain the reliability function

( m )0 e−m −m −λt
R(t )=P ( 0 , m )= =e =e
0! (3-34)

So the cumulative distribution function also is obtained as:

F ( t )=1−e−λt (3-35)

If the cumulative distribution function is differentiated, the probability density function is obtained as:

f ( t )=λe− λt t  0,  > 0 (3-36)

Hazard rate function (t)is defined by:

λ ( t )=f ( t ) / R (t )= λe− λt /e− λt =λ (3-37)

In general, if product follows the exponential distribution, mean time to failure (MTTF) is 0.63 at 1/λ.

Fig.3.12. Cumulative distribution function F(t) of exponential distribution

3.5 Weibull distributions and its applications

3.5.1 Introduction

In characterizing the failure times of certain components one often employs the Weibull distribution.

As it was developed by Weibull in the early 1950s, this distribution can be used to represent many

different failure behaviors. Many other extensions of the Weibull distribution have been proposed to
enhance its capability to fit diverse lifetime data since 1970s.

This is mainly due to its weakest link properties, but other reasons have its increasing failure rate with

component age and the variety of distribution shapes. The increasing failure rate accounts to some

extent for fatigue failures. The density function depend upon the shape parameter . For low  values

( < 1), the failure behavior can be similar to the exponential distribution. For  > 1, the density

function always begins at f(t) = 0, reaches a maximum with increasing lifetime and decreasing slowly

again.

As seen in Figure 3.13, the probability density function for two parameter distribution is defined as:

t β
βt β−1 −( )
f ( t )= β e η
η , t  0, >0, >0 (3-38)

where  and  are characteristic life and shape parameters, respectively.

When Eq. (3.38) is integrated, the cumulative distribution function is obtained as:

t β
∞
F ( t )=∫ f ( t ) dt=1−e
−( η)
, t >0
0 (3-39)
 (a) Probability density function (b) Cumulative distribution function

Fig. 3.13 Probability density and cumulative distribution function on the Weibull distributions

Reliability function R(t) is defined as:

R ( t ) =1−F ( t )=e
− ( ηt ) , t >0
(3-40)

Hazard (or failure) rate function (t) can be described by:

f (t) β t β
λ ( t )= =
R (t ) η η(), t>0
(3-41)

For  =1 and 2, the exponential and Rayleigh distributions are especially called in Weibull distribution,
respectively. The various failure rates of the Weibull distribution specified in bathtub curve can be

divided into three regions.

  < 1.0: Failure rates decrease with increasing lifetime (early failure)

  = 1.0: Failure rates are constant.

  > 1.0: Failure rates increase with increasing lifetime

For the time  = 1.0 and t = , the cumulative distribution function F(t) from Eq.(3.39) can be

calculated:

−1
F ( t )=1−e =0.632 (3-42)

Therefore, the characteristic lifetime  is assigned to the cumulative distribution function F(t) = 63.2%

for exponential distribution.

3.5.2 Shape parameters 

A shape parameter estimated from the data affects the shape of a Weibull distribution, but does not

affect the location or scale of its distribution. The spread of the shape parameters represents the

confidence intervals and a dependency of the stress level. A summary of the determined shape

parameters is approximately described as:

 High temperature, high pressure, high stress: 2.5 < <10

- Low cycle fatigue : depend on cycle times
ex) disk, shaft, turbine
  Low temperature, low pressure, low stress: 0.7 <  < 2
- Degradation : depend on use time
ex) electrical appliance, pump, fuel control value

Shape parameter  of a certain component would be invariable, but its characteristic life  varies

according to use condition and material status. Thus, shape parameter ( ) would be estimable and

then will be confirmed after test. The density function and hazard rate function for the Weibull

distribution range from shape parameters   1.0 ~ 5.0.

3.5.3 Confidence Interval

In statistics, a confidence interval (CI) is characterized as the probability that a random value lies

within a certain range. CI is represented by a percentage. For example, a 90% confidence interval

implies that in 90 out of 100 cases, the observed value falls within this certain interval. After any

particular sample is taken, the population parameter is either in the interval realized or not. The

desired level of confidence is set by the researcher. A 90% confidence interval reflects a significance

level of 0.1.

The average of failure times can often deviate within a certain range. The Weibull line may describe

experimental results. If the median is used to determine F(ti), 50% of the experimental results lie

below the Weibull line. To know the truth of the Weibull line, it is necessary to determine its confidence

interval

Over an observation of several test samples, the Weibull line drawn in Figure 3.14 is the most

probable in the middle – median values and its confidence intervals. The line in the middle represents

the population mean – observed over several test specimens – thus 50% of the cases lie above and

50% lie below this line.

 Fig.3.14. Weibull plot of five failures with 90% confidence interval

3.5.4 A plotting method on Weibull probability paper

Weibull plotting is a graphical method for informally checking on the assumption of Weibull distribution

model and also for estimating the two Weibull parameters – shape parameter and characteristic life.

The method of Weibull plotting is illustrated both for complete samples of failure times (type I) or for

censored samples (type II).

The cumulative distribution function F(t) has an S-like shaped curve (Figure 3.15(a)). With a Weibull

Probability Paper, If plotted the function F(t) in Weibull Probability Paper, it is useful to evaluate the

lifetime of mechanical component in reliability testing.

After taking inverse number and logarithmic transformation from Eq. (3-40), it can be expressed as:

β
t
−1
()
ln ( 1−F ( t ) ) =
η (3-43)
After taking logarithmic transformation one more time, it can be expressed as:

1
(
ln ln
)
1−F ( t )
=β⋅ln t−β ln η

(3-44)

If F is sufficiently small, then Eq.(3.44) can be modified as:

1
(
ln ln
)
1−F ( t )
≃ln F ( t )=β⋅ln t−β ln η

(3-45)

That is, two parametric Weibull distribution can be expressed as a straight line on the Weibull

Probability Paper. The slope of its straight line becomes the shape parameter  (See Figure 3.15 (b)).

β
1
−
R ( t ) =1−F ( t )=e
( ηt ) (
ln ln
)
1−F ( t )
≃ln F ( t )=β⋅ln t−β ln η
(a) (b)
 Fig. 3.15 A plotting of Weibull probability paper

3.4.5 Probability Plotting for the Weibull Distribution

One method of calculating the parameters of the Weibull distribution is by using probability plotting.

These procedures that are more illustrated are as following:

1) Rank the times-to-failure in according to ascending order t1<t2 …. <tn

i 1 2 3 ………. r-1 r

ti t1 t2 t3 t r-1 tr

By ordering the failure times, an overview is won over the timely progression of the failure times. In

addition, the ordered failure times are required in the next analysis step and are referred to as order

statistics. Their index corresponds to their rank.

Step 2) Determine the failure probability F(ti) of the individual order statistics

i−0.3
F ( t i )≈ ×100
n+0. 4 (3-46)

Step 3) Enter the coordinate (ti, F(ti)) in the Weibull probability paper.

Step 4) Approximate sketch the best fit straight line through the entered points and determine the

Weibull parameters β^ . At the F(t) = 63.2% ordinate point, draw a straight horizontal line until this

line intersects the fitted straight line. Draw a vertical line through this intersection until it crosses the
abscissa. The value at the intersection of the abscissa is the estimate of η^ (See Figure 3.16)

Fig.3.16. A typical characteristics of Weibull plot with a large sample size

Example 3.2. Assume that six automobile units are tested. All of these units fail during the test after

operating the following number of hours

ti : 93, 34, 16, 120, 53 and 75. Estimate the values of the

parameters for a two-parameter Weibull distribution and determine the reliability of the units at a time

of 15 hours.

Solution)

First, rank the times-to-failure in ascending order as shown next.

 Failure order ( i ) Time-to-failure, hours ( ti ) F(ti), %

1 16 10.94

2 34 26.56

3 53 42.19

4 75 57.81

5 93 73.44
6 120 89.06

Second, by using Excel, approximate sketch the best fit straight line through the entered points (ln(t i),

ln[-ln(1-F(t))]) from Eq. (3-45)

Failure order ( i ) ln(ti) ln[-ln(1-F(t))]

1 2.77 -2.16

2 3.53 -1.18

3 3.97 -0.60

4 4.32 -0.15

5 4.53 0.28
6 4.79 0.79

We can obtain the estimated shape parameter β^ = slope = 1.427, estimated characteristic life

6 . 187
η^ (Q(t) is 63.2% ordinate point) = e 1 . 427
=76 . 3226 hours

where ln[-ln(1-0.63)] = -0.00576, -0.00576 = 1.427x - 6.186

 1.00

0.50 f(x) = 1.43 x − 6.19

R² = 0.99

0.00
2.50 3.00 3.50 4.00 4.50 5.00

-0.50

-1.00

-1.50

-2.00

-2.50

A Weibull distribution with the shape parameter  = 1.427 and  = 76.32 hour is drawn on the Weibull

Probability Paper. The cumulative distribution function is described as

t 1 .43

F ( t )=1−e
− ( )
76. 32
(3-47)

In result a straight line is sketched with slope =1.4 on the Weibull Probability Paper. The

characteristic lifetime is 76.0 hour when the cumulative distribution function, iF(t) is 63% (See Figure

3.17).
Fig. 3.17 How to use Weibull CDF
Chapter 4

Failure Mechanics, Design and Reliability Testing

Seong-woo Woo

Email: [email protected]

Reliability Association of Korea, Seoul, Korea (Republic of)

Abstracts: This chapter will be discussed as (qualitative) established method like FMEA, FEA, and

Taguchi method that will improve the product design or quality. On the other hand, as quantitative

method like Weibull analysis, reliability testing and the others also will be discussed further as

methodology of reliability assessment. As time goes, product becomes failure - the state of not

meeting an intended function of the customer’s satisfaction. Product failures in field happen when the

parts cannot withstand the repetitive stresses due to loads over the product lifetime. The failure

mechanics of product can be characterized by the stress (or loads) on the structure and materials

used in the structure. If there is a void (design weak point) in the material where the loads are applied,

engineer would want to move the void in the structure to location away from where the stress is

applied. These activities are called design. Quantitatively, the final goal of these quality activities is to

discover the design problems by reliability testing. Engineer judge whether the product achieve the

reliability target.

Keywords: Reliability testing, Failure mechanics, Design, Quantitative method,.

4.1 Introduction

Reliability can be defined as the probability that a component or product will fulfill its intended function

over lifetime. It is necessary to clearly be understood what a product’s intended function and its failure

are. Intended functions are the product functionalities that perform the voices of the customer. They
are represented as product specifications in company. As time goes, product becomes the state of not

meeting an intended function of the customer’s satisfaction. Consequently, we call it failure (Figure

4.1).

Fig.4.1.Train wreck or train crash from Wikipedia

The most common failures are those caused by specification deficiencies of the product function. In

such cases design might result in the product failure to customer satisfaction (or specification). When

field failure occurs, we also determine whether the company specifications are inappropriate or

whether verifiers are incorrectly conforming to the specifications instantaneously. To find out the

design faults, verification specifications suitable for a newly developed product should be developed

like reliability quantitative specifications that are described in Chapter 7.

If functions break during product usage unexpectedly, we can say it is failure. The definition of a

failure may not precise if a gradual or intermittent loss of performance over time is observed. For an

example, seals experience a degradation of material properties and no longer satisfy the
specifications. In this case we can replace old part with new one. The most critical failures are no

longer satisfied with the customer requirements (or specifications) due to unidentified factors before

the product releases. In such cases design might does not achieve the customer requirements and

experience the recall. Thus, design is a critical process to determine whether the intended function of

product or modules in lifetime fulfills through exact specifications.

A disaster due to product failure is always an undesirable event for several reasons: putting human

lives in jeopardy, causing economic losses and interfering with the availability of products and

services. The failure causes come from improper materials selection, inadequate design of the parts

and its misuse. So it is the engineer responsibility to be prepared when failure is expected to occur.

Engineer needs to assess its cause and take action appropriate preventive measures against future

incidents - the parametric ALT will be discussed in Chapter 7.

4.2 Failure mechanics and designs

The failure mechanics of components that might no longer be functioned can be characterized by two

factors: 1) the stress (or loads) on the structure, 2) The type of materials used in the structure. To

prevent the failure, engineer should know either loads (or stress) or structural (or materials) related

(Figure 4.2).

Fig.4.2. Failure mechanics created by a load on a component made from a specific material
If there is a void (design weak point) in the material where the loads are applied, the structure can

fracture at that location. The engineer would want to move the void in the structure to location away

from where the stress is applied.

Product failure in mechanical system is a physical problem that is created when stress due to loads

causes a fracture. Failure mechanics seeks to understand the process how stress and materials

impact the failure. The applied loads cause stresses on the module structure. The failure site of the

module structure might be observable when the failed products are taken apart in the field or could be

from the results of parametric ALT.

Fig. 4.3. Design process and reliability testing

If the structure is ideally designed and has well-dispersed stresses, there should be no problems with

the failure of the module. As the mechanical design is developed with an optimal design process -

Finite Element Analysis (FEA), it may have design flaws that will show up in the field. Figure 4.3

illustrates an ideal design process that includes design, reliability testing, and field conditions feeding
back into the design and reliability testing. Product failures in the field happen when the parts cannot

withstand the repetitive stresses due to loads over the life time of the module. Most products with

electrical or mechanical components are composed of multi-module structures. If one of modules has

a problem due to an improper design, then that module will determine the lifetime of the failed

product. Consequently, by experiments such as reliability testing, the design problems might be

revealed before launching product. The product design should be effectively connected with Design

Engineering and Test Engineering to achieve the reliability target of modules.

4.2.1 Product design - intended functions

Fig.4.4. Intended function in parameter diagram

We can say to define the intended functions that shall implement the customer’s needs in product. For

television, the intended function is to watch the program that consists of moving images. It is required

as the fundamental advantage like the superiority of picture and sound. A product like TV consists of

multiple modules that can be put together as a subassembly. When product has an input, it has output

as response (intended functions) that want to be implemented. Intended functions are embodied in

the product design process and their performance will be measured by specifications (See Figure

4.4).
Based on design idea, intended functions of product can be implemented through the design

developing process. In the same manner designers like previous market failure use the engineering

design process to satisfy the customer’s needs. Product design would learn from the experiences of

others. Companies often specify the past mistakes by the documents that describe the design

requirements. Performances are to note each of the key features that consist of a variety of

specification. Reliability is to identify the failure mode through testing over time or lifetime.

Manufacture determines whether product conforms to specification in the production process. After

production, the new concept of product determines the feedback of customer. in field (Figure 4.5).

Fig.4.5. Implementing Intended Functions in the design process

There are many good possibilities for solving design problems and choosing the best solution that

product specifications could be embodied in the design process. Product development refines and

improves a solution in the design phase, before a product launches to customers. Basically, the

engineering design process starts to define the problems to be implemented: 1) What is the problem?

2) How have others approached it? 3) What are the design constraints?

A prototype is a first operating version of a solution (or intended function) that satisfies the customer

requirements. The product design process involves multiple iterations like redesigns of your solution
before settling on a final design. Final product can define the performance criteria in evaluating the

specifications of product quality – a) intended functions (or fundamental advantage), b) specified

product life, and c) the failure rate of product under operating and environmental conditions.

In the current global competitive marketplace, product quality of intended functions is an important

requisite to ensure continued success in the marketplace. On the other hand, the failure of product

quality expels from the customer satisfaction and loyalty. Thus, it is important for the product design

team to understand customer expectation or voice and usage conditions.

Intended functions are the functionalities that product is to perform the voices of the customer and

explain the company specifications. The intended functions of a product must be recognized in the

design to ensure whether critical customer requirements (or specifications) work properly in product.

Thus, the fulfillment of each intended function should be understood from a standpoint of the

customer’s expectations. As time goes by, product fails to perform its intended functions because they

don't meet. For example,in the first steps of failure mode effects analysis (FMEA), a product starts to

assess whether a product might fail to perform its intended function.

The engineering design team may often neglect the potential customer uses, though customer’s right

uses. Sometimes the failure of a returned product in field may be perceived to be customer abuses of

product, generally not failure. From a standpoint of robust design, the intended functions might be

designed to withstand the customer misusages (or overloads) for customer's proper uses. In this case

the robust concept is effective. To robustly keep the intended functions, product withstands noise

factors like customer usage (or loads). As seen in Fig. 4.4, Taguchi’s robust design schematic of

product employs two experimental arrays: one for the control array (design) and the other for the

noise array (loads). Optimizing over the control factors, product can be reduced to a signal (output)-to-

noise (load) ratio, and intended function will be designed robustly.

However, a large number of experimental trials in the Taguchi product array may be required because

the noise array is repeated for every row in the control array. As a result it is hard to discover the

optimal design parameters. Alternative approach for a robust design of intended function will explain

the parametric ALT in Chapter 7.

4.2.2 Specified design lifetime

The design life of product is the time period which the product works properly within the life

expectancy. The design life of products might differ from the reliability lifetime metrics – MTBF. For

example, the MTBF of product may be 100,000 hours and the design life is 10,000 hours. It means

that one failure occur every 100,000 population operating hours. Because product cannot reach

100,000 operating hours, most of these units will be replaced by a new unit. Aluminum electrolytic

capacitors, fans, and batteries will fail due to wear-out before they could achieve the operating time -

MTBF. As design life time, BX life can be useful. It means the life time at which X% of the units in a

population will have failed. For example, if unit has a B10 life of 10,000 hours, 10% of the population

will have failed by 10,000 hours of operation.

The specified design lifetime provides a usage or timeframe for reliability analysis or testing. Some

organization might simply choose to design a product to be reliable over the stated warranty

(qualification) period. Because it is commensurate with how long the product is expected to be used in

the field, an enlightened organization might choose a design life. Depending on the designer’s

perspective, the life specification might be based on any of the design life time – BX life.
 Fig.4.6. Qualification Life & Design Life

When product is selling to customer, product is designed to have safe margin between environmental

stress and strength. As time goes on, product failure initially appears at qualification lifetime. Design

life has maximum failure rate at acceptance criteria of design quality. Generally, design lifetime is 2 ~

3 times greater than qualification lifetime. Design life is based on information coming from the

customer and competitive benchmarks. Engineers often define a design life time to represent an

engineer’s specifications of product usage under which the reliability must be verified. It is important

that careful thought go into the synthesis of this specification to ensure that product quality do not

result in customer’s dissatisfaction due to the end of the useful life (Figure 4.6).

4.2.3 Dimensional differences between quality defects and failures

As seen in Table 4.1, quality defects appear for one of three reasons: 1) Incompleteness of design

specification, 2) Nonconformance to specifications during manufacture, 3) Customer misuse of the

product. The established specifications in company can perceive them whether product quality meet.

They also will follow the normal distribution.

Table 4.1. Quality defects and failures

Quality Defects Failure

Concept Out of established specifications Physical trouble

Index Defect rate, ppm Failure rate, lifetime

Unit Percent, ppm Percent/year, year

Area Manufacturing Design

Probability Normal distribution Exponential/Weibull

( x−μ )2 −λt
1
− F ( t )=1−R ( t )=1−e
2 σ2
f ( x )= e
σ √2 π
The quality defects can refer to a variety of problems that appear before customer use, such as flaws

causing poor performance or the failure to work. Similarly, the good quality means good aesthetic

design with fundamental advantage as well as good performance. For some aspect of the product,

quality defect is out of tolerance in manufacturing. If something can be found to be out of

specifications, it is considered to have a quality defect that follows the normal distribution.

On the other hands, failure clearly indicates a physical performance disorder related to the product as

time goes on. The number of failure per year for a given production lot results in the annual failure

rate. Any mistakes or omissions of design will induce results in failure that follows the exponential or

Weibull distributions. The absence of failure means good reliability.

4.2.4 Classification of failures

The definition of failure is obvious when there is a total loss of product (intended) functions that can be

differentially perceived from the viewpoints of the customers or by specifications. If something breaks

during product usage unintentionally, it may fail. However, if only a partial loss of (intended) function is

involved, it will be complicated to define the product failure. In such instances the definition of the

failure may not be precise when one observes a gradual or intermittent loss of performance over time.

Although the activity is completed successfully, a person may still feel dissatisfied if the underlying

process is perceived to be below expected standard (or specification).

For example, a variety of automobile failures might be classified as:

A Class is said that failure will damage the body of passenger or the loss of car control, failure of
brake equipment, and fire risk. Examples are too many to be expected: 1) Accident due to loss of

acceleration control, 2) Differential gear fixation, 3) air exposure of pump, 4) Overheat or

disconnection of cable, 5) Damaged flywheel, 6) Malfunctioned clutch, and 7) Malfunctioned injection

pump (See Figure 4.7).

B1 Class is said that failure will stop car. Examples are too many to be expected: 1) Engine stop or no

starting from computing engine, injection, pump, common rail, ignition coil, car starting, engine control,

engine fixation, distribution chain, 2) Transmission stop, and 3) Stop of gear box, no reverse operation

B2 Class is said that failure may stop car. There are a lot of examples: 1) Abnormal noise of engine or

gear box, 2) Overheat Engine, and 3) Vibration of engine or gearbox

Fig.4.7.Definition for Failure Class (example: automobile)

C Class is said that car can be drivable, but it requires the high cost to recover it. Examples also are

too many to be expected: 1) It make car inoperative, 2) It affects visual, hearing, and smell, 3) Critical

motor surges and power loss, 4) Abnormal noise, oil spill, cooling water spill, smell, over oil leakage,

5) abnormal smell, and 6) Clutch malfunction, inoperative gear transmission. C1 Class also is
inconsistent to the standard of discharge gas. Example is not to meet for standard of emission gas.

D Class is said that using car is no effect but minor operational failure. Examples are too many to be

expected: 1) Driving car is inconvenient, 2) Affect visual and hearing, 3) Over fuel consumption, 4)

Slowly acceleration, 5) Idle speed is instable, 6) Engine is not starting, 7) Vibration, and 8) A few of

noise (discharge noise, cracking noise, vibration noise, cooling pump noise, starting noise, noise in

gear transmission, erosion of engine part).

E Class is wearable parts such as filter, spark plug, and timing belt need to be replaced periodically. In

a result, we can say failure is defined as A, B, C, and D class.

4.3 Failure mode and effect analysis (FMEA)

4.3.1 Introduction

Failure mode and effect analysis (FMEA) is a widely used method to study system problems in the

reliability engineering. The history of FMEA goes back to the early 1950s with the development of

flight control systems when the U.S. Navy’s Beau of Aeronautics developed a requirement called

“Failure Analysis”. In the mid-sixties FMEA was set to work in earnest by NASA for the Apollo project.

In the 1970s, the U.S. Department of Defense developed military standards entitled “Procedures for

Performing a Failure Mode, Effect, and Critically Analysis”. For use in aerospace, defense, and

nuclear power generation, FMEA/FMECA methods are widely used to conduct analysis of systems.

The Ford Company integrated this method into its quality assurance concept.

 In the early of 1950s: Propeller airplane  Reliability design of Jet airplane with flight control

systems

 In the middle of 1960s: Apollo program used in real earnest

 In the 1970s: US NAVY (MIL-STD-1629) was adopted

 In the late of 1970s: widely used in the industry due to the introduction of product liability law
In the first step of a system reliability study, FMEA involves reviewing the design of many components,

assemblies, and subsystems to identify failure modes, and their causes and effects. To determine

whether an optimum criterion of reliability assessment is achieved, FMEA is to analyze and modify

many components in system. FMEA uses the risk priority number (RPN). A qualitative analysis is

mainly used in the light of FMEA team experience.

It may be described as an approach used to perform analysis of each potential failure mode in the

systems under consideration to examine the effects of such failure modes on that system. When

FMEA is extended to classify each potential failure effect according to its severity, the method is

known as failure mode effects and criticality (FMECA).

Fig.4.8.Members of FMEA team

As seen in Figure 4.8, the FMEA is carried out in interdisciplinary groups – members in the planning,

R&D and QA. It is reasonable to execute an FMEA in teams, since it is only then possible to

incorporate all operational areas affected by the analysis. In practice it is beneficial to execute an

FMEA under the direction of an FMEA moderator, who is familiar with the methodical procedure. In

this way, time consuming discussions concerning the method can be avoided.
In general, the FMEA team consists of a moderator, who offers methodical knowledge. They can offer

technical knowledge concerning the product or process to be analyzed. The moderator, who also may

possess a marginal know-how concerning the product or process, certifies that the team members

acquire a basic knowledge of the FMEA methodology. A brief training at the beginning of an FMEA

assignment is useful.

FMEA is a systematical method that the fundamental idea is the determination of all possible failure

modes for arbitrary systems or modules and the possible failure effects and failure causes are

presented. The aim of the method is to recognize the risks and weak points of product design as early

as possible in order to enable execution improvements in timely manner. There are many terms used

in performing FMEA/FMECA and some of them are as follows:

 Failure cause. The factors such as design defects, quality defects, physical or chemical

processes, or part misapplications are the primary reason for failure or they start the physical

process which deterioration progresses to failure.

 Failure mode. The notion or manner through which a failure is perceived.

 Failure effect. The consequence a failure mode has on item’s function, operation, or status.

 Single failure point. An item’s malfunction that would lead to system failure and is not

compensated through redundancy or through other operational mechanisms

 Criticality. A relative measure of a failure mode’s consequences and its occurrence frequency.

 Severity. A failure mode’s consequences, taking into consideration the worst case scenario of a

failure, determined by factors such as damage to properly, the degree of injury, or ultimate

system damage.

 Criticality analysis. An approach through which each possible failure mode is ranked with

respect to the combined influence of occurrence probability and severity.

 Undetectable failure. A postulated failure mode in the FMEA for which no failure detection

approach is available through which the concerned operator can be alerted of the failure.
 Local effect. The consequences a failure mode has on the function, operation, or status of the

item currently being analyzed.

4.3.2 Types of FMEA

The types of FMEA are classified as 1) System-level FMEA, 2) Design-level FMEA, and 3) Process-

level FMEA (See Figure 4.9).

4.3.3 System-level FMEA

Failure functions as well as failure modes for product are analyzed in the system-level FMEA. The

analysis is carried out in various hierarchical system levels all the way to the failure on the module

level. This is the highest-level FMEA that can be performed and its purpose is to identify and prevent

failures related to system/subsystems during the early conceptual design. System-level FMEA is

carried out to validate that the system design specifications reduce the risk of functional failure to the

lowest level during the operational period.

 Fig.4.9. Types of Failure mode and effect analysis (FMEA)

Some benefits of the system-level FMEA are identification of potential systemic failure modes due to

system interaction with other systems and/or by subsystem interactions, selection of the optimum

system design alternative, identification of potential system design parameters that may incorporate

deficiencies prior to releasing hardware/software to production, a systematic approach to identify all

potential effects of subsystem/assembly part failure mode for incorporation into design-level FMEA,

and a useful data bank of historical records of the thought processes as well as of action taken during

product development efforts.

4.3.4 Design-level FMEA

The purpose of performing design-level FMEA is to help identify and stop product failures related to

design. This type of FMEA can be carried out upon component-level/subsystem-level design proposal

and its intention is to validate the design chosen for a specified functional performance requirement.

The advantages of performing design-level FMEA include identification of potential design-related

failure modes at system/subsystem/component level, identification of important characteristics of a

given design, documentation of the rationale for design changes to guide the development of future

product design, help in the design requirement objective evaluation of design alternatives, systematic

approach to reduce criticality and risk, accumulated data serve as a useful historical record of the

thought processes and the actions taken during the product development effort, and useful tool to

establish priority for design improvement actions.

4.3.5 Process-level FMEA

This identifies and prevents failures related to the manufacturing/assembly process for a certain

product. The benefits of the process-level FMEA include identification of important characteristics

associated with the process, identification of potential process shortcomings early in the process
planning cycle, development of priorities for process improvement actions, and documentation of

rationale for process changes to help guide the establishment of potential manufacturing processes.

4.3.6 Steps for performing FMEA

The distinction between technical knowledge in various fields and the methodology of an FMEA

execution offers the advantage that the experts from the respective fields only offer their technical

knowledge free of any methodical considerations. Thus, merely a basic knowledge of FMEA is

adequate of the team of experts. The team size ranges ideally between 4-6 members – supervisors,

moderator, and a small product team.

The fundamental step of an FMEA searches for all conceivable failure modes. Thus, this step should

be executed most carefully. Each failure mode not found can lead to dangerous failure effects.

Options available to discover failure modes are damage statistics, experience of the FMEA

participants, checklists, brainstorming, and systematic analysis over failure functions. An imperative

principle is the observation of former arisen failures in similar cases. All further failure modes can be

derived with the help of the experience of the FMEA participants.

The first sections of the FMEA form sheet are reserved for the description of the system, product or

process and their function. The next section of the form sheet deals with the risk analysis. This is

followed by a risk assessment in order to rank the numerous failure causes. The last step is a concept

optimization derived from the analysis of the risk assessment (See Figure 4.10).

The completed form sheet represents a tree structure. A certain component has one or more functions

and normally several failure modes. Each failure mode has again various failure effects and different

failure causes (See Figure 4.11).

FMEA can be performed in the following six steps:

4.3.6.1 Defines system and its associated requirements (Step1)

This is a first step of FMEA. Define the system under consideration how complex the system is. The

analyst must develop the system definition using documents such as reports, drawings, development

plans (or specifications). The system structure arbitrarily orders the individual system elements into

various hierarchical levels.

 The definition of design (or process) interfaces

 Dividing the system into its individual system elements – module and components

 Arranging system elements hierarchically in product structure

Fig.4.10. FMEA form sheet that consists of four sections
Fig.4.11 FMEA example for mechanical/civil parts with tree structure
4.3.6.2 Describe the System and Its Associated Functional Blocks (Step 2)

 The arrangement of the system structure is the basis for determining the preparation of the

description of the system under consideration. Such description may be grouped into two

parts:

 Narrative functional statement (Top down). The functions are created by preparing for

each module and component as well as for the total system (See Figure 3). It provides

narrative description of each item’s operation for each mode/mission phase. The degree of

the description detail depends on factors such as an item’s application and the uniqueness of

the functions performed.

 System block diagram. The purpose of this block diagram is to determine the

success/failure relationships among all the system components

4.3.6.3 Identify failure modes and their associated effects (Failure analysis, Step 3)

A failure analysis performs the analysis and determination of the failure modes and their effects. The

failure leads to the dissatisfaction of a module. Compensating provisions and Criticality classification

are described below.

 Compensating provisions. Design provisions or operator actions that is circumventing or

mitigating the failure effect.

 Criticality classification. This is concerned with the categorization of potential effect of

failure.

- People may lose their lives due to failure

- Failure may cause mission loss

- Failure may cause delay in activation

- Failure has no effect

4.3.6.4 RISK ASSESSMENT (Step 4)

The objective of the risk assessment is to prioritize the failure modes discovered during the system

analysis on the basis of their effects and occurrence likelihood. Thus, for making an assessment of

the severity of an item failure, two commonly used methods are Risk Priority Number (RPN)

Technique that is widely used in the automotive industrial (Table 4.2~4.4).

4.3.6.5 RPN (Risk Priority Number)

This method calculates the risk priority number for apart failure mode using three factors: (1) failure

severity ranking (SR), (2) failure mode occurrence ranking (OR), and (3) failure detection probability

(DR). For example, if people are put into danger, the severity is evaluated higher, whereas a minimum

limitation of comfort would receive a respectively lower value. With assessment value DR it is

determined how successful the detection of the failure cause is before delivery to the customer. More

specifically, the risk priority number is computed by multiplying the ranking (i.e., 1-10) assigned to

each of these three factors. Thus, the risk priority is expressed by

Risk Priority Number (RPN) = (Severity) (Occurrence) (Severity) (4-1)

With the RPN a ranking of the identified failure causes and their failure connection to the failure effect

can be done.

Since the above three factors are assigned rankings from 1 to 10, the value of the RPN will vary from

1 to 1000. The average RPN is normally 125 (5*5*5). Failure modes with a high RPN are considered

to be more critical; thus, they are given a higher priority in comparison to the ones with lower RPN.

Nonetheless, ranking and their interpretation may vary from one organization to another. Table 3

through 3 present rankings for failure detection, failure mode occurrence probability, and failure effect

severity used in one

4.3.6.6 Optimization (Step5)

The last phase of the FMEA is the optimization phase. First, the calculated RPN are ordered

according to their values. According to the Pareto principle, 20~30% of the RPN has been optimized.

 Ranking of failure causes according to their RPN value

 Concept optimization beginning with the failure causes with the highest RPN (Pareto

principle)

 Failure causes with OR>8, DR>8, DR>8 separately

The new optimization actions are entered on the right side of the form sheet for the optimized failure

causes and the responsibility is recorded. An improved RPN is calculated for the improved state the

new assessment values assigned to RPN.

Example 4.1 Develop a FMEA for pressure cooker like Figure 4.12. The safety features of pressure

cooker have as following:

1) Safety valve relieves pressure before it reaches dangerous levels.

2) Thermostat opens circuit through heating coil when the temperature rises above 250° C.

3) Pressure gauge is divided into green and red sections. That is, "Danger" is indicated when the

pointer is in the red section.

First of all, problem should define scope:

1) Resolution: The analysis will be restricted to the four major subsystems (electrical system, safety

valve, thermostat, and pressure gauge)

 Table 4.2. Failure Detection Ranking

Item Likelihood of
Rank meaning Rank
No. detection
1 Very High Potential design weakness almost certainly detected 1, 2
2 High There is a good chance of detecting 3, 4
3 Moderate There is a possibility of detecting potential design weakness 5, 6
4 Low Potential design weakness is unlikely to be detected 7, 8
5 Very low Potential design weakness probably will not be detected 9
6 Delectability Potential design weakness cannot be detected 10
Absolutely
uncertain

Table 4.3. Failure Mode Occurrence Probability

Item Occurrence
Ranking term Rank meaning Rank
No. probability
< 1 in 106
1 in
20,00
0
1 in
1 Remote Occurrence of failure is quite unlikely
4,000 1
2 Low Relatively few failures are expected
1 in 2
1,000 3
3 Moderate Occasional failures are expected
1 in 4
400 5
1 in 6
80 7
4 High Repeated failures will occur
1 in 8
40 9
5 Very high Occurrence of failure is almost inevitable
1 in 10
20
1 in
8
1 in
2

Table 4.4. Severity of the Failure-Mode Effect

Failure effect
Item
severity Severity category description Rank
No.
category
1 Minor No real effect on system performance and the customer
1
may not even notice the failure
2 Low The occurrence of failure will only cause a slight customer
2, 3
annoyance
3 Moderate Some customer dissatisfaction will be caused by failure
4, 5, 6
4 High High degree of customer dissatisfaction will be caused by
7, 8
failure but the failure itself does not involve safety or
noncompliance with government rules and regulations
The failure affects safe item operation, involves
9, 10
5 Very high noncompliance with government rules and regulations
2) Focus - Safety

(a). A schematic diagram of pressure cooker

(b). System block diagram of pressure cooker

Figure 4.12 FMEA for pressure cooker

Based on a focus of safety of pressure cooker, perform Failure Modes, Effects and (Criticality)

Analysis for a Pressure Cooker with

 Table 4.5 Failure Modes, Effects and (Criticality) Analysis for a Pressure Cooker

Item Failure Mode Failure Causes Failure Effects S O C Control Measures/Remarks

Defective cord
Cooking interruption Use high-quality components.
No current Defective plug 1 2 2
(mission failure) Periodically inspect cord and plug.
Electrical Defective heating coil
System
Current flows to Use a grounded (3-prong) plug.
Faulty Insulation Shock 2 1 2
ground by an Only plug into outlets controlled by
Cooking interruption
alternate route ground-fault circuit interrupters.

Steam could burn Design spring to handle the fatigue

Open Broken valve spring operator 2 2 4 and corrosion that it will be
Safety
Increased cooking time subjected to.
Valve
Corrosion Potential over Use corrosion-resistant materials.
Closed 1 2 2
Faulty manufacture pressurization Test the safety valve.

Defective
Open Cooking interruption 1 2 2 Use a high-quality thermostat.
thermostat
Thermosta
t
Closed Over pressurization 1 2 2 Use a high-quality thermostat.
Defective thermostat
eventually opens valve

.
4.4 Fault Tree Analysis (FTA)

4.4.1 Concept of FTA

(a) Case I (b) Case II

Fig.4.13. A typical example of Fault Tree Analysis

Fault tree analysis (FTA) is one of the most widely used methods in the industrial area to identify the

internal (or external) causes of failures (Figure 4.13). Thus, the FTA defines the system behavior in

regard to fault. FTA was developed in the early 1950s at Bell Telephone Laboratories and started to

use the FTA for the development of commercial aircrafts (1966). In the 1970s this method was used in

the area of nuclear power. Now it is spread in many different areas – automobile, communication, and

robotics.

FTA is used to show the system functions and their reliability. In the early design stage it may be

applied as a diagnosis and development tool. The potential system faults can be identified and the

design action plans can be setup. One of the major advantages of FTA is that the method provides

both qualitative and quantitative results.FTA also with Boolean algebra and probability theory is

beneficial to the preventative quality assurance. (See Figure 4.14).

 Fig.4.14 Commonly used fault tree symbols: (i) AND gate, (ii) OR gate, (iii) rectangular, (iv) OR gate

Although many symbols are used in performing FTA, the four commonly used symbols are described

as:

 AND gate. This denotes that an output fault event occurs only if all of the input fault events

occur

 Or gate. This denotes that an output fault event occurs only if one or more of the input fault
events occur

 Rectangle. This denotes a fault event that results from the logical combination of fault

events through the input of a logic gate

 Circle. This represents a basic fault event or the failure of an elementary component. The

event’s probability of occurrence, failure, and repair rates are normally obtained from field

failure data.

The objectives of FTA are 1) systematic identification of all possible failures and their causes, 2)
illustration of critical failures, 3) evaluation of system concepts, and 4) documentation of the failure

mechanism and their functional relations. It begins by identifying an undesirable system event (Top

Event). Top event are generated and connected by logic gates such as OR and AND. The fault tree

constructions are repeated successively until the lowest events are developed.

Example 4.2

Assume that electric circuit system contains motor system, two switches, and electric power source.

Develop a fault tree for the top event “no operating motor”, if the interruption of motor power can only

be caused either by current failure or motor failure (See Figure 4.15).

Switch
Power C Motor E
(Motor E consists of brush
E1 and Coil E2)

Fig. 4.15. A circuit diagram for Example 4.2

By using the Figure 4.14 symbols, a fault tree for motor system can be developed as following:
 No operating
motor X0

Current X1 Motor X2
failure failure

Switch Brus Coil

Power
X3 X4 h failur
failure failure X5
failur e X6
e

failu failu
re re
X7 X8

Fig. 4.16. A fault tree for Example 4.1

Each fault event in the figure is labeled as X 0, X1, X2, X3, X4, X5, X6, X7, and X8. For independent fault

events, the probability of occurrence of top events of fault trees can easily be evaluated by applying

the basic rules of probability to the output fault events of logic gates. For example, we have

P ¿¿ (4-2)

P ¿¿ (4-3)

P ( x 2 )=P ( x 5 ) +P ( x 6 ) −P ( x 5 ) P ( x 6 )
(4-4)
 P ( x 0 ) =1−[ 1−P ( x1 ) ][ 1−P ( x 2 ) ]

(4-5)

where P(Xi) is the probability of occurrence of fault event Xi, for i = 1, 2, 3,….,8.

Example 4.3

In Figure 4.16, assume that the probability of occurrence of fault events X 3, X5, X6, X7, and X8 are 0.02,

0.02, 0.03, 0.01, and 0.02, respectively. Calculate the probability of occurrence of the top event “no

operating motor” by using Eqs. (4.2) – (4.5).

Thus, by substituting the given data values into Eqs. (4.2) – (4.5), we can get

P ( x 4 )=P ( x 7 ) +P ( x8 ) −P ( x 7 ) P ( x 8 ) =( 0 .01 ) + ( 0 . 02 )−( 0 .01 ) ( 0. 02 )=0 .0298

(4-6)

P ¿ ¿ (4-7)

P ( x 2 )=P ( x 5 ) +P ( x 6 ) −P ( x 5 ) P ( x 6 ) =( 0 . 02 )+ ( 0 .03 )−( 0 . 02 )( 0 . 03 )=0 . 0494

(4-8)

P ( x 0 ) =1−[ 1−P ( x1 ) ][ 1−P ( x 2 ) ]=1−( 1−0 . 0492 )( 1−0. 0494 )=0 . 0962
(4-9)

Thus, the probability of occurrence of the top event “no operating motor” is 0.0962
4.4.2 Reliability Evaluation of Standard Configuration

Engineering systems can form various types of configurations in performing reliability analysis. A

system is said to be a serial system if failure of one or more components within system results in

failure of the entire system. On the other hands, parallel system is that the failure of all components

within the system results in the failure of the entire system. For example, the lighting system that

consists of four bubs in a room is a parallel system, because room blackout happens only when all

four bulbs break. The reliabilities the serial or parallel systems are summarized in Table 4.5.

Example 4.4

Assume that an aircraft has four independent and identical engines and all must work normally for the

aircraft to fly successfully. Calculate the reliability of the aircraft flying successfully, if each engine’s

reliability is 0.99.

By substituting the given data values of equation system reliability in Table 4.6, we can get

4
Rs =( 0 . 99 ) =0. 9606 (4-10)

Thus, the reliability of the aircraft flying successfully is 0.9606

 Table 4.6 Standard Configuration with m units

System structure Serial systems Parallel systems

Block Diagram

Functional Tree

m m
System reliability Rs ( t )= ∏ Ri ( t ) Rs ( t )=1−∏ ( 1−Ri ( t ))
i=1 i=1
m
−∫ λ i dt −λ i t
ithunit constant failure rate Ri ( t ) =e =e R ps=1−∏ ( 1−e λi )
i=1
∞ n ∞ n
MTTF ∫ Ri ( t ) dt=1 /∑ λi ∫ Ri ( t ) dt= 1λ ∑ ii
0 i=1 0 i=1

Example 4.5

A system is composed of two independent and identical active units and at least one unit must

operate normally for the system success. Each unit’s constant failure rate is 0.0008 failures per hour.

Calculate the system mean time to failure and reliability for a 150-hour mission.

Substituting the given data values of parallel system MTTF Eq. in Table 4.5 yields
 ∞ m
m 1 1 1 1
MTTF ps=∫ [ 1−( 1−e− λt ) ] dt=
0
∑ = ( )
λ i=1 i ( 0 .0008 )
1+ =1875
2 hours (4-11)

Using the specified data values of parallel system reliability equation in Table 4.6 yields

−( 0 .0008 ) (150 ) 2
[
R ps (150 )= 1−{1−e } ]=0. 9872
(4-12)

Thus, the system mean time to failure and reliability are 1875 hours and 0.9872, respectively.

4.5 Robust Design (or Taguchi methods)

Fig. 4.17 Robust design - inputs that make the outputs less sensitive

Robust design first developed by Taguchi is a powerful technique for improving reliability at low cost in

a short time. Robust design is a statistical engineering methodology for optimizing product conditions
so that product performance is minimally sensitive to various noise sources of variation (See Figure

4.17). Since 1980s, it has been applied extensively to improve the quality of countless products and

processes.

Robustness is defined as the ability of a product to perform its intended function consistently at the

presence of noise factors such as environmental loads. Here, the noise factors are the variables that

have adverse effects on the intended function and are impractical to control. Environmental stresses

(or loads) are the typical noise factors. This definition is widely applied in the field of quality

engineering to address initial robustness when the product service time is zero. If customer

satisfaction over time is concerned, the effect of time should be taken into account.

Reliability of mechanical/civil system can be perceived as robustness over time or environmental

conditions. A reliable product has a high robustness value under different use conditions. To achieve

high robustness, Taguchi methods recommend the optimal design parameters insensitive to noise

parameters.

Taguchi methods are originally a kind of method to improve the product quality and recently applied to

engineering as robust design method. Professional statisticians have welcomed the goals and

improvements brought about by Taguchi methods, particularly by Taguchi's development of designs

for studying variation, but have criticized the inefficiency of some of Taguchi's proposals. As

alternative methods, parametric Accelerated Life Testing in Chapter 7 will be studied.

4.5.1 A specific loss function

To estimate these hidden quality costs, Taguchi’s quality loss function (QLF) has been proposed.

Taguchi’s approach is different than the traditional approach of quality costs. In the traditional

approach, if you have two products that one is within the specified limits and the other is just outside

of the specified limits, the difference is small. Although the difference is small, the product within the

limits is considered a good product. On the other hands, the outside one is considered a bad product.

Taguchi disagrees with this approach. Taguchi believes that when a product moves from its target

value, that move causes a loss. It doesn’t matter if the move falls inside or outside the specified limits.
For this reason, Taguchi developed the QLF to measure the loss that is associated with hidden quality

costs. This loss happens when a variation causes the product to move away from its target value.

As seen in Figure 4.18, QLF is a “U” shaped parabola. The horizontal axis is tangent with the

parabola at the target value. This is a quadratic loss function because it assumes that when a product

is at its target value (T), the loss is zero.

Fig. 4.18 Taguchi quality loss function for a nominal-the-best (on-target) characteristics

Quality characteristics can be categorized into three situations: (1) On-target, minimum-variation, (2)

Smaller the better, and (3) Larger the better.

4.5.1.1 On-target, minimum-variation (for example, a mating part in an assembly).

In engineering design we frequently encounter the on-target characteristics. Due to production

process variation, the characteristics are allowed to vary within a range, say , where d is called

the tolerance. Eq. (4-13) describes the quality loss of this type of characteristics.

The unit loss is determined by the formula:

 2
L ( t )=k ( y−T )
(4-13)

where k = a proportionality constant dependent upon the organization’s failure cost structure, y =

actual value of quality characteristic, T = target value of quality characteristic.

The value of k must first be determined before the loss can be estimated. To determine the value of k:

k=c/ Δ2 (4-14)

where c = loss associated with the specification limit, and  = deviation of the specification from the

target value.

The target function is described as

2
f ( y )=1 / ( y −t ) (4-15)

4.5.1.2 Smaller the better – variance (for example, carbon dioxide emissions)

If y is a smaller-the-better characteristic, its range can be written as [ 0 , d ], where 0 is the target value
and d is the upper specification limit. The quality loss function is obtained by substituting T = 0 into Eq.

(4-13) and can be written as

 2 (4-16)
L ( t )=ky

The target function is described as

2 (4-17)
f ( y )=1/ y

4.5.1.3 Larger the better – performance (for example, agricultural yield)

If y is a larger-the-better characteristic, its range is [ d , ∞ ], where d is the lower limit. Because the

reciprocal of a larger-the-better characteristic has the same quantitative behavior as a smaller-the-

better one, the quality loss function can be obtained by substituting 1/y for y in Eq. (4-13). Then we

have

2
1
L ( t )=k
y () (4-18)

The target function is described as

2 (4-19)
f ( y )= y

Example 4.6

A product with on-target and minimum-variation has 100 (target value). The unit loss is determined by

the formula:

2
L ( t )=40 ( y−100 )
(4-20)
Find out the expectation of quality loss function of process line1 and line2

So expectation of quality loss function can be expressed as:

2
E [ L ] =E [ k ( y−m )2 ]=kVar [ y ] +k ( E [ y ] −m )
(4-21)

If process line1 has mean 96 and standard deviation 3, the expectation of quality loss function of

process line1 is

2 2
E [ L ] =40 ( 3 ) +40 ( 96−100 ) =1000 $
(4-22)

If process line 2 has mean 98 and standard deviation 5, the expectation of quality loss function of

process line2 is

2 2
E [ L ] =40 ( 5 ) +40 ( 98−100 ) =1160 $
(4-23)

If the standard variation of process line 2 decreases from 5 to 3, the cost reduction is

2 2
ΔE [ L ] =40 (5 ) −40 ( 3 ) =640 $

(4-24)

4.5.2 Robust Design Process

 Fig. 4.19 Steps of robust design

As seen in Figure 4.19, robust design is a statistical engineering methodology for minimizing the

performance variation of a product by choosing the optimal design conditions of the product to make

the performance insensitive to noise factors. Taguchi realized that the best opportunity to eliminate

variation is during the design of a product and its manufacturing process. Consequently, he developed

a strategy for quality engineering that the process consists of three stages – system design,

parameter design, and tolerance design.

4.5.2.1 System design

System design involves selection of technology and components for use, design of system

architecture, development of a prototype that meets customer requirement, and determination of

manufacturing process. System design has significant impacts on cost, yield, reliability,

maintainability, and many other performances of a product. It also plays a critical role in reducing

product sensitivity to noise factors. If a system design is defective, the subsequent parameter design

and tolerance design aimed at robustness improvement are ineffective. In recent years, some system

design methodologies have emerged and shown effective, such as TRIZ (a problem-solving, analysis

and forecasting tool derived from the study of patterns of invention in the global patent literature).

This step is indeed the conceptual design level, involving creativity and innovation.
  Getting into the ‘design space’

 Creating a feasible design

 Involves innovation

4.5.2 Parameter (measure) design

Parameter design aims at minimizing the sensitivity of the product performance to noise factors by

setting its design parameters at the optimal levels. In this step, designed experiments are usually

conducted to investigate the relationships between the design parameters and performance

characteristics of the product. Using such relationships, one can determine the optimal setting of the

design parameters.

Once the concept is established, the nominal values of the various dimensions and design

parameters in the product need to be set, the detail design phase of conventional engineering. In

many circumstances, this allows the parameters to be chosen so as to minimize the effects on

performance arising from variation in environmental noise – loads. Strictly speaking, parameter design

might signify the robust design

 Optimizing within the ‘design space’ (not changing anything fundamentally)

 Settings for the factors identified in systems design

4.5.3 Tolerance design

With a successfully completed parameter design, tolerance design is to choose the tolerance of

important parts to reduce the performance sensitivity to noise factors under cost constraints.

Tolerance design may be conducted after the parameter design is completed. If the parameter design

cannot achieve sufficient robustness, tolerance design is completed. In this step, the important parts

whose variability has the largest effects on the product sensitivity are identified through
experimentation. Then the tolerance of these parts is tightened by using higher-grade parts based on

the trade-off between the increased cost and the reduction in performance variability.

 Tightening tolerances on important factors

 Better materials/new equipment may be needed

 Has significant cost implications

4.5.4 A Parameter Diagram (P-Diagram)

A P-diagram in the mechanical/civil system illustrates the input (signals), outputs (intended functions

or responses), control factors, and noise factors. Figure 4.20 shows a generic P- diagram

Fig.4.20 Generic parameter diagram

Signals are inputs from subsystems or modules to the system. The system transforms the signals into

functional responses. Signals are essential to fulfilling the function of a system. Noise factors are

variables that have adverse effects on robustness. Typical examples in the mechanical/civil system

are a variety of random loads.

Control factors are the design parameters whose levels are specified by designers. The purpose of a

robust design is to choose optimal levels of the parameters. In practice, mechanical/civil systems

have a large of design parameters, which are not important in terms of the robustness. Thus, the key

design parameters are included in a robust design. These key design parameters are identified by

using engineering judgment, analytical study, a preliminary test, or field data (See Figure 4.21).

(a) Compressor system in a vapor-compression refrigeration cycle

 (b) Parameter diagram of compressor system

Fig.4.21 Schematic diagram for Mechanical System (Example: compressor system)

4.5.5 Taguchi’s Design of Experiment (DOE)

Design of experiment is a statistical technique for studying the effects of multiple factors on the

experimental response. The factors are laid out in a structured array in which each row combination

are conducted and response data are collected. Through experimental data analysis, we can choose

the optimal levels of controls factors that minimize the sensitivity of the response to noise factors.

4.5.5.1 Orthogonal Arrays

An orthogonal array is a balanced fractional matrix in which each row represents the levels of factors

of each run and each column represents the levels of specific factor that can be changed from each

run. In a balanced matrix,

 All possible combinations of any two columns of the matrix occur an equal number of times

within the two columns. The two columns are also said to be orthogonal.

 Each level of specific factor within a column has an equal number of occurrences within the

column.

Example 4.7 If product on-target and minimum-variation have the following experimental data, find

out the optimal factor combinations.

Table 4.7 Experimental data

A B AB C D E E Experiment S/Ni

1 2 3 4 5 6 7 1 2 3 4 5

1 0 0 0 0 0 0 0 34 22 29 14 25 -28.20
 2 0 0 0 1 1 1 1 32 24 26 16 28 -28.22

3 0 1 1 0 0 1 1 24 18 25 27 22 -27.38

4 0 1 1 1 1 0 0 27 22 26 23 25 -27.84

5 0 0 1 0 1 0 1 30 25 27 29 20 -28.44

6 0 0 1 1 0 1 0 19 16 33 34 19 -28.04

7 0 1 0 0 1 1 0 25 33 24 25 21 -28.27

8 0 1 0 1 0 0 1 26 27 27 28 26 -28.57

First of all, make the simplified analysis for level 1 and 2 per A, B. C, D like Table 4.8.

Table 4.8 Simplified analysis for experimental data

Factor A B AB C D Total

Level 1 -111.64 -112.95 -113.26 -112.29 -112.24

S/N Ratio -225.01
Level 2 -113.37 -112.06 -113.26 -112.29 -112.24

Level Range 1.73 0.89 1.51 0.43 0.53 5.09

Contribution (%) 34.0 17.5 29.7 8.4 10.4 100.0

Draw the Pareto chart and find the factors that consists of 80~90%, based on the accumulated total

sum.
 (a) Pareto Charts

(b) Interaction effects between A, B, C, D factors

Fig.4.22 Simplified analysis for A, B, C, D factors

Then we know that A, B, and AB occupy approximately 80% from Pareto charts. Find out the proper

levels of A, B factors by Table 4.8.

Table 4.8 Simplified analysis for A, B factors

A0 A1

B0 S/N00=-56.42 S/N10=-56.53
 B1 S/N01=-55.22 S/N11=-56.84

Consequentially, we know that S/N ratio for A0 and B1 is the smallest among the levels of A, B, and

AB

4.5.6 Inefficiencies of Taguchi's designs

Taguchi’s robust design method uses parameter design to place the design in a position where

random ‘‘noise” does not cause failure and to determine the proper design parameters and their

levels. However, for a simple mechanical structure, a lot of design parameters should be considered

in the Taguchi method’s robust design process. Those mechanical/civil products with the missing or

improper minor design parameters may result in recalls and loss of brand name value.

Interactions also are part of the real world. In Taguchi's arrays, interactions are confounded and

difficult to resolve. Response surface methodology (RSM) is a "follow-up design" that resolves only

the confounded interactions. RSM design may be used to explore possible high-order univariate

effects of the remaining variables and have great statistical efficiency. The sequential designs of

response surface methodology decreases experimental runs than would a sequence of Taguchi's

designs. But Taguchi's designs still require a lot of experimental runs.

4.6. Reliability Testing

4.6.1 Introduction

Product lifetime can be estimated by using quantitative reliability testing methods. Reliability testing is

performed to see whether intended function of product is feasible. Regardless of demonstrating

successful requirement achievement, it causes product failures to use concepts of load and strength

under severe test conditions. It will be used to determine whether the product is adequate to meet the

requirements of performance and reliability. Reliability testing during design and development
therefore is mandatory to prove whether the lifetime of product is sufficient for customer requirements.

Product is the operational certainly for a stated time interval (or lifetime). The goal of product reliability

is to develop product with a longer lifetime, based on the reliability target of product (or module).

Today, the reliability of an element or of a system is defined as the probability that an item will perform

its required function under given conditions for a stated time interval. The definition terms have to be

explained as:

 Perform means that the item does not fail

 The given conditions include total physical environment, i.e., mechanical, electrical, and

thermal conditions.

 The stated time interval can be very long (twenty years, for telecommunication

equipment), long (a few years) or short (a few hours or weeks, for space research

equipment). But the time could be replaced by other parameters, such as: the mileage

(of an automobile) or the number of cycles (i.e., for a relay unit).

Based on the observed data of reliability testing, maximum-likelihood estimation (MLE) is a popular

method to predict the product lifetime. MLE is the statistical method of estimating the parameters of a

statistical model - some unknown mean and variance that are given to a data set. Maximum likelihood

selects the set of values of the model parameters that maximizes the likelihood function.

For example, one may be interested in the lifetime of product (or module), but be unable to measure

the lifetime of every single product in a population due to cost or time constraints. Assuming that the

lifetime are normally distributed with some unknown mean and variance, the mean and variance can

be estimated with MLE while only knowing the lifetime of some sample of the overall population. MLE

would accomplish this by taking the mean and variance as parameters and finding particular

parametric values that make the observed results the most probable given the model.
4.6.2 Maximum Likelihood Estimation

For a fixed set of data and underlying statistical model, the method of maximum likelihood estimates

the set of values of the model parameters that maximizes the likelihood function. Intuitively, this

maximizes the "agreement" of the selected model with the observed data, and for discrete random

variables it indeed maximizes the probability of the observed data under the resulting distribution.

MLE would give a unified approach to estimation, if the case of the normal distribution or many other

problems is well-defined.

One very good statistical method for the determination of unknown parameters of a distraction is the

Maximum Likelihood Method. It assumes that the histogram of the failure frequency depicts the

number of failure per interval. For larger test sample sizes n it is possible to derive function out of the

histogram and thus to exchange the frequencies to the probabilities.

In this way it is possible to state, for example that during the first interval probably 3% of all failures

will occur. In the second interval it is most likely that 45% of the failures occur, etc. According to

theory, the probability L of test sample can be obtained by the product of the probability of the

individual intervals.

Suppose there is a sample t1, t2, …,tn of n independent and identically distributed failure times, coming

from a distribution with an unknown probability density function f0 (·). On the other hands, supposed

that the function f0 belongs to a certain family of distributions {f(·| θ), θ∈ Θ} (where θ is a vector of

parameters), called the parametric model, so that f0 = f(·| θ0). The unknown value θ0 is expected to as

the true value of the parameter vector. An estimator θ^ would be fairly close to the true value θ0. The
observed variables ti and the parameter θ are vector components.

n
L ( Θ;t 1 , t2 ,⋯, t n )=f ( t 1 , t 2 ,⋯, t n|Θ ) =∏ f ( t i|Θ )
i=1 (4-25)

This function is called the likelihood. The idea of this procedure is to find a function f, for which the
product L is maximized. Here, the function must possess high values of the density function f in the

corresponding region with several failure times ti. At the same time only low value of f in regions with

few failures are found. Thus, the actual failure behavior is accurately represented. If determined in this

way, the function f gives the best probability to describe the test samples.

It is often more effective to use the log-likelihood function. Thus the product equation becomes an

addition equation, which greatly simplifies the differentiation. Since the natural log is a monotonic

function, this step is mathematically logical.

n
ln L ( Θ;t 1 , t 2 ,⋯, t n ) =∑ ln f ( t i|Θ )
i=1 (4-26)

By differentiate Eq. (4-26), the maximum of the log-likelihood function and thus the statistically optimal

parameters Θ l can be obtained as:

∂ln ( L ) n 1 ∂ f ( t i ;Θ )
=∑ ⋅ =0
∂ θl i=1 f ( t i ;Θ ) ∂ θl
(4-27)

These equations can be nonlinear in the parameters; therefore it is often useful to apply approximate

numerical procedures. By the Likelihood function value the opportunity is given to estimate the quality

of the adaptation of a distribution to the failure data. The greater the likelihood function value is, the

better the conclusive distribution function represents the actual failure behavior. However, based on

MLE, the characteristic life MLE from the reliability testing (or lifetime testing) can be estimated on the

Weibull chart (Refer to Section 3.4).

4.6.3 Time-to-Failure Models

The failure time T of a product is a random variable. Time can take on different meanings depending

on operational time, distance driven by a vehicle, and number of cycles for a periodically operated

system. Time-to-Failure model usually provides all the tools for reliability testing, especially

accelerated life testing data analysis. It is designed for use with complete (time-to-failure), right

censored (suspended), interval or left censored data. Data can be entered individually or in groups.

Time-to-Failure Model has the following types:

 Arrhenius: a single stress model typically used when temperature is the accelerated

stress.

 Inverse Power Law (IPL): a single stress model typically used with a non-thermal stress,

such as vibration, voltage or temperature cycling.

 Eyring: a single stress model typically used when temperature or humidity is the

accelerated stress.

 Temperature-Humidity: a double-Arrhenius model that is typically used when

temperature and humidity are the acceleration variables.

 Temperature-Nonthermal: a combination of the Arrhenius and IPL relationships that is

typically used when one stress is temperature and the other is non-thermal (e.g.,

voltage).

4.6.3.1 Arrhenius Equation

The Arrhenius equation proposed by Arrhenius in 1889 is a formula for the temperature dependence

of reaction rates.
 Fig. 4.23 Arrhenius Equation

As seen in Figure 4.23, reactivity modeling consists of computing the energy of the products, the

reactants, and the transition state (TS) connecting them. These three points are the critical features

on a reaction pathway. The difference between the energies of the transition state and reactants ( Ea

= ETS – Er) is the activation energy Ea. The activation energy is important in understanding the rate of

chemical reactions as expressed in the Arrhenius Equation which relates the rate constant K of a

chemical reaction to its activation energy:

One of the earliest and most successful acceleration models predicts how time-to-fail varies with

temperature. The Arrhenius equation empirically based model is known as:

Ea
TF=A exp ( )
kT (4-28)

where T denoting temperature measured in degrees Kelvin at the point when the failure process takes

place, k is Boltzmann's constant (8.617Ⅹ10-5 in eV/K), and constant A is a scaling factor that drops
out when calculating acceleration factors, with Ea denoting the activation energy, which is the critical

parameter in the model. If Eq. (4-28) takes logarithm, the simple straight line can be obtained (Figure

4.24):

Fig. 4.24 Arrhenius model

The acceleration factor between a higher temperature T2 and a lower temperature T1 is given by

Ea 1 1
AF=exp
( ( −
k T1 T2 )) (4-29)

The value of Ea depends on the failure mechanism and the materials involved. It typically ranges from

0.3 or 0.4 up to 1.5, or even higher.

4.6.3.2 Inverse Power Law

In statistics, a power law is a functional relationship between two quantities, where a relative change

in one quantity results in a proportional relative change in the other quantity, independent of the initial

size of those quantities: one quantity varies as a power of another.

For example, the "life" of a product will go down as stress goes up. While this is not a hard and fast

rule, very few systems do not behave in this intuitive fashion. This allows for shorter test times at

higher levels of stress. With solid knowledge of the life-stress relationship, effective predictions of life

at normal or usage conditions can be made.

The most important and widely used model for mechanical systems is the inverse power law (IPL). It

has forms:

−n (4-30)
TF=AS

n (4-31)
K=BS

The most critical factor is n, the life-stressor slope with s being stress applied to the system. A is a

constant; in reality it relates the basic mechanical strength of the design to resist the stress applied to

it. If Eq. (4-30) takes logarithm, the simple straight line can be obtained (Figure 4.25):

Fig. 4.25 Inverse power Law

4.6.3.3 Eyring Equation

For chemical reaction rate theory, Eyring acceleration model has led to a very general and powerful

one. This model has several key features:

 It has a theoretical basis from chemistry and quantum mechanics.

 If a chemical process (chemical reaction, diffusion, corrosion, migration, etc.) is causing

degradation leading to failure, the Eyring model describes how the rate of degradation varies with

stress or, equivalently, how time to failure varies with stress.

 The model includes temperature and can be expanded to include other relevant stresses.

 The temperature term by itself is very similar to the Arrhenius empirical model, explaining why

that model has been so successful in establishing the connection between the ΔE parameter and

the quantum theory concept of "activation energy needed to cross an energy barrier and initiate a

reaction".

The model for temperature and one additional stress takes the generic form:

Ea
TF=B exp ( ) kT
×S−n

(4-32)

where k is the Boltzmann constant, T is the thermodynamic temperature, and Ea.is the activation

energy.

The acceleration factor between a higher temperature T2 and a lower temperature T1 is given by
 S2 n Ea 1 1
AF =
( ) ( (
S1
exp −
k T 1 T2 )) (4-33)

We know that the acceleration factor in Eyring equation Eq. (4.32) is similar to Eq. (7.17) that is

derived from the generalized stress model Eq. (7.16).

4.6.4 Reliability Testing

The time and effort in testing can be significantly reduced by censored tests, and they can estimate

the product lifetime. If a test trial is interrupted before all n test units have failed, a censored test may

produce. If the interruption occurs after a given time, one is dealing with censoring of type I.

On the other hands, if a trial is interrupted after a given amount of test units r has failed, one is dealing

with censoring of type II. The trials stop after 4 failures. The point in time at which the failure r occurs

is a random variable. Thus, leaving the entire trial time length opens until the end of the trial.

The fact that n-r test units have not failed is taken into account by substituting r for n in the

denominator of the approximation equation. With type I or II censoring it is necessary to estimate the

characteristic lifetime  in the Weibull chart by extrapolating the best fit line beyond the last failure

time. This is generally problematic as long as further failure mechanisms cannot be neglected. A

statistical statement concerning the failure behavior can be obtained on the observed lifetime.

The procedures and methods for the assessment of complete data or censored data can be found in

Table 4.9 and Figure 4.26.

Table 4.7 Overview of procedures for the assessment of censored data

Data Type Type of Censor Description Procedure

Complete
No censoring All samples have failed Median Procedure
Data
 i−0.3
F ( t i )≈
r=n n+0.4
For i=1,2,…,n
Lifetime characteristics of all Median Procedure
Censored
Censoring Type I intact units are larger than the i−0.3
Data F ( t i )≈
or Type II lifetime characteristics of the n+0.4
r<n
units r which failed last For i =1,2,…,r

Example 4.8

Select the ten samples from a Integrated Circuit chip manufactured in august, 2016 and perform the

reliability testing at 120℃, 135℃, and 150℃. Under 30℃normal conditions, search B10 life for

30℃on Weibull or excel program.

(a) Type I censoring (b) Type II censoring

Fig.4.26.Schematic of type I (a) and type II censoring (b)

 Temperature, ℃

120 130 150

1 3450 3300 2650

2 4340 3720 3100

3 4760 4180 3400

4 5320 4560 3800

5 5740 4920 4100

6 6160 5280 4400

7 6580 5640 4700

8 7140 6230 5100

9 8100 6840 5700

10 8960 7380 6400

Case I) for 120℃, because temperature data have complete with no censoring, plot them on Weibull

chart. We can approximate sketch the best fit straight line through the entered points and determine

the Weibull parameters β^ = 3.812. At the Q(t) = 63.2% ordinate point, draw a straight horizontal

line until this line intersects the fitted straight line. Draw a vertical line through this intersection until it

crosses the abscissa. The value at the intersection of the abscissa is the estimate of η^ = 6692.
 i
120oC F(t)*100

1 3450 6.73077

2 4340 16.3462

3 4760 25.9615

4 5320 35.5769

5 5740 45.1923

6 6160 54.8077

7 6580 64.4231

8 7140 74.0385

9 8100 83.6538

10 8960 93.2692

Case II) Using Excel program, for 120 oC, we obtain the estimated shape parameter β^ = slope =

33 . 576

3.812, estimated characteristic life η^ (F(t) is 63.2% ordinate point) = e 3 .812

=6692 hours

where ln[-ln(1-0.63)] = -0.00576, -0.00576 = 3.811x – 33.57

X: LN(t) Y: LN(LN(1/(1-F(t)))) 1.50

8.15 -2.66384
1.00
f(x) = 3.81 x − 33.58
8.38 -1.72326
0.50 R² = 0.99
8.47 -1.20202
8.58 -0.82167 0.00
8.66 -0.50860 8.00 8.20 8.40 8.60 8.80 9.00 9.20
-0.50
8.73 -0.23037
-1.00
8.79 0.03293
8.87 0.29903 -1.50

9.00 0.59398 -2.00

9.10 0.99269 -2.50

-3.00
In same way, we can obtain the analysis result of data for 130 oC and 150 oC. Plot them on the

Weibull chart.
 1
B10 life for 30℃ can be obtained from
(
LβBX = ln
1−x )
⋅η β

Temp, ℃ B10 Life 1/T Ln(B10)

120 3706 0.002544 8.217708

130 3355 0.00248 8.118207

150 2728 0.002363 7.911324

Ea 1
ln ( L )=ln ( A )+ ×
k T y=3 . 8782+1707 . 4 x
 E
ln( A )=3 . 878204, =1707 . 392
k

A=exp(a)=48.3373, E=1707.392x(8.617x10-5) = 0.147126 eV

So the estimated life-stress model from Eq. (4-28) is

TF=A exp ( kTE )=48 .3373 exp (1707T.392 )

a

So B10 life for 30℃is obtained from

TF=48 . 3373 exp (1707 .392

30+273 . 16 )
=13 , 506
Chapter 5

Load Analysis

Seong-woo Woo

Email: [email protected]

Reliability Association of Korea, Seoul, Korea (Republic of)

Abstracts: This chapter will explain how to model the mechanical/civil systems – automobiles, aircraft,

satellites, rockets, space stations, ships, bridge, and building subjected to the random loading.

Product have their own particular structural loads in field. A typical pattern of repeated load or

overloading may cause structural failure in product lifetime. Such possibility should be assessed in the

design phase whether structure subjected to loads endures in its lifetime. A modeling is a

mathematical representation of the dynamics system to describe the real world used by traditional

system modeling method like Newtonian. Here, as alternative method, the bond-graph will be

introduced because it is easily applicable to the mechanical/civil systems. If products are modeled, the

time response of system simulation for (random) dynamic loads will obtain. As the time response is

simplified and counted as a sinusoidal input, the rain-flow counting method and miner’s rule can

assess the system damage. Because there are a lot of assumptions, this analytic methodology is

exact but complex to reproduce the reliability disasters due to the design failures. So we should

develop the final solutions – experimental method like parametric ALT that will be discussed in

Chapter 7. Load analysis will be helpful to figure out the failure of problematic parts and finally

discover them in the reliability-embedded design process..

Keywords: Load Analysis, Mathematical modeling, Bond-graph, Miner’s rule, Rain-flow counting,
5.1 Introduction

Loads cause stresses, deformations, and displacements in the structures of product. Assessment of

their effects can be implemented by the structural modeling and its analysis using finite element. In a

result repeated load or overloading may cause structural failure.

Two generic types of mechanical static or dynamic loading exist. A static load – tension or

compression can exhibit motion or permenent change like dislocation if repeated in a lifetime.

Eventually, they will be a permanent deformation. The examples of static loading are as following:

 Structural load and deflection versus material stress and strain

 Tension and compression loads

 Torsion and bending loads

A dynamic load, sometimes also referred to as probabilistic loads, is a force exerted by a moving body

on a resisting member, usually in a relatively short period of time. Because such loads are usually

unstable, we can say the dynamic load. Dynamic loads involve motion and therefore are time varying

load conditions. The examples of dynamic loading are as following:

 Impact, vibration and shock loads

 Unbalanced inertia loads

An impact load is one whose time of application on a material is less than one-third of the natural

period of vibration of that material. A variety of cyclic loads on a structure can lead to fatigue damage,

cumulative damage, or fracture. These loads come from repeated loadings on a structure or can be

due to vibration.
5.2 Modeling of Mechanical System

5.2.1 Introduction

The modeling of mechanical product is a mathematical representation of the dynamics structural

systems to figure out their characteristics. Typical modeling methods – Newton, Lagrange,

Hamiltonian mechanics, and D'Alembert's Law are commonly used in dynamic system. As an output,

models might describe the system behavior that can be represented in random variables (or state

space). In a result the state space are expressed as vectors and provides a convenient and compact

way to analyze systems with multiple inputs and outputs.

Fig. 5.1 Operational loads due to the random vibration on road

When observed in most mechanical/civil components, loads in field follows a more or less random
curve that constant load amplitudes are quite seldom. For example, automobiles possess completely

random stochastic load curves due to the street roughness, car speed, and environmental conditions.

And for airplane, a mean load change repetitively occurs on the wing of a transportation airplane

when it takes off or lands (See Figure 5.1).

On the other hands, the load of the gas turbine blade in an airplane is to a large extent deterministic

that there is no randomness in the system states, though the load sequence is still variable. With

simple algorithms and fast processors an on-line load measurement for parts can be directly

measured during operation.

However, a measurement during operation is quite time-consuming and actually impossible to figure

out the whole transmitted loads in product lifetime. To do that, the engineer depends on the

mathematical modeling, analysis and response such as the Newtonian model that was develeoped for

long time ago (See Figure 5.2).

(a) Base random vibrations

 (b) A simplified modeling of the automobiles

Fig. 5.2. Random loads and modeling of the automobiles by Newtonian modeling

5.2.2 D'Alembert's modeling for automobile

Engineer uses D'Alembert's principle and free body diagram to model mechanical system. If there is

an automobile that is used for transportation, we can model a simple system with a mass that is

separated from a wall by a spring and a dashpot. The mass could represent an automobile, with the

spring and dashpot representing the automobile's bumper. If only horizontal motion and forces are

considered, it is represented in Figure 5.3.

(a) Typical automobile subjected to wind flow

 (b) Mass-spring-dashpot system

Fig. 5.3 Typical Mechanical Automobile Modeling

The free body diagram is a drawing method showing all external forces acting on a body. There is

only one position in this system defined by the variable "x" that is positive to the right. It is assumed

that x=0 when the spring is in its relaxed state. As seen in Figure 5.4, there are four forces to develop

a model from the free body diagram: 1) An external force (Fe) such as friction force and air resistance

force, 2) A spring force that will be a force from the spring, k·x, to the left, 3) A dashpot force that will

be a force from the dashpot, b·v, to the left, 4) Finally, there is the inertial force which is defined to be

opposed to the defined direction of motion. This is represented by m·a to the left.

Fig. 5.4.Completed free body diagram for automobile modeling

Newton's second law states that an object accelerates in the direction of an applied force, and that

this acceleration is inversely proportional to the force, or

 ∑ F=m⋅a
all lexternal

(5-1)

Subtracting the right-hand side results in D'Alembert's principle,

∑ F−m⋅a=0
all lexternal (5-2)

If we consider the m·a term to be inertia force (or D'Alembert's force), D'Alembert's law will be left

∑ F⋅δr=0
all (5-3)

To visualize this consider pushing against a mass (in the absence of friction) with your hand in the

positive direction. Your hand experiences a force in the direction opposite to that of the direction of the

force (this is the -m·a term). The inertial force is always in a direction opposite to the defined positive

direction. We sum all of these forces to zero and get

F e ( t )−ma ( t ) −bv ( t )−k⋅x ( t )=0 (5-4)

In other words, we can change

 2
d x dx
m 2
+b + kx ( t )= F e ( t )
dt dt (5-

5)

5.3 Bond graph modeling

5.3.1 Introduction

Bond Graph is an explicit graphical tool for modeling multidisciplinary dynamic systems including

components from different engineering areas – the mechanical/civil, the electrical, the thermal, and

the hydraulic system. When designing a new dynamic system, it is a good method to utilize a

graphical representation for communicating other engineers to express the dynamic modeling. In

engineering disciplines, linear graphs have long traditions among several graphical representation

means.

In 1959 Bond Graph method was developed by Professor Henry Payner and his former students at

MIT, who gave the revolutionary idea of portraying systems in terms of power bonds, connecting the

elements of the physical system to the so called junction structures which were manifestations of the

constraints. This power exchange portray of a system is called bond graph.

In 1961 the Paynter’s books were published as entitled “Analysis and Simulation of Simulation of

Multiport Systems.” In 2006, the three authors have published the fourth edition entitled as “System

Dynamics – Modeling and Simulation of Mechatronic Systems”. Now several disciplines of Bond

Graph have been widely accepted in the world as a modeling methodology. There are many

literatures about Bond Graph method and its applications to analyze dynamic systems.

In a result this method will give a brief description for analyzing loads applied to structure and

understanding its work.

 Fig. 5.5.A typical modeling of the automobiles subjected to repetitive random vibrations

Fig. 5.6.Typical hydrostatic transmission modeling

5.3.2 Basic elements, Energy relations, and Causality of Bond Graph

A Bond Graph is a graphical representation of a physical dynamic system. It is similar to the better

known block diagram and signal-flow graph. While the symbols in Bond Graph represent bi-directional

exchange of physical energy, those in block diagrams and signal-flow graphs represent uni-directional

flow of information. Bond Graph also can be applicable in multi-energy domain - mechanical/civil,

electrical, and hydraulic system.

The dynamic systems analysis is relatively simple when the steady state behavior or the few degrees

of freedom has. In most of the cases, the main concern of engineers is to establish the mathematical

model that represents the dynamic behavior of the system and how the different parameters influence
the system behavior, because the system dynamic equations are usually partial differential equations,

whose solutions require deep mathematical knowledge.

As the fundamental bases of the Bond Graph theory, energy flow is a basic element in a system. It

flows in from one or more sources, is temporarily stored in system components or partially dissipated

in resistances as heat, and finally arrives at “loads” where it produces some desired effects. Power is

the rate of energy flow without direction.

Fig. 5.7. Power flow in Bond Graph for electric-hydraulic system

Bond Graph represents this power flow between two systems. This flow is symbolized through an

arrow (Bond) as Figure 5.7 illustrated. Each bond represents the instantaneous energy flow or power.

The flow in each bond is denoted by a pair of variables called 'power variables' whose product is the

instantaneous power of the bond. Because power is not easy to measure directly, engineers can be

represented as two temporary variables - flow and effort. Every domain has a pair of effort and flow

variable. For example in mechanical system, flow represents the “velocity” and effort the “force”, in

electrical system, flow represents the “current” and effort the “voltage”. The product of both temporary

variables – power is represented as:

P=e ( t )⋅f ( t )
(5-6)
The method makes possible the simulation of multiple physical domains, such as mechanical,

electrical, thermal, hydraulic, etc. Flows and efforts should be identified with a particular variable for

each specific physical domain which is working. Table 5.1 also shows the physical meanings of the

variables in different domains.

Table 5.1 Energy flow in the multi-port physical system

Modules Effort, e(t) Flow, f(t)

Mechanical translation Force, F(t) Velocity, V(t)

Mechanical rotation Torque, (t) Angular velocity, (t)

Compressor, Pump Pressure difference, P(t) Volume flow rate, Q(t)

Electric Voltage, V(t) Current, i(t)

Thermal Temperature, T Entropy change rate, ds/dt

Chemical Chemical potential,µ Mole flow rate, dN/dt

Magnetic Magneto-motive force, em Magnetic flux, φ

The Bond Graph is composed of the "bonds" which link together "1-port", "2-port" and "3-port"

elements. Whether power in bond graph is continuous or not, every element is represented by a multi-

port. Ports are connected by bonds. The basic blocs of standard bond graph theory are listed in Table

5.2.

For 1-ports there are effort sources, flow sources, C-type elements, I-type Elements, and R-type

Elements that can connect power discontinuously. For 2-ports, there are Transformer and Gyrator that

can connect power continuously. For 3-ports, there are 0-junction and 1-junction that can make up the

network.
Power bonds may join at one of two kinds of junctions: a “0” junction and a “1” junction. In a “0”

junction, the flow and the efforts satisfy Eqs (5.7) - (5.8):

∑ flowiinput =∑ flowoutput (5-7)

effort 1 =effort 2 =. .. . ..=effort n (5-8)

This corresponds to a node in an electrical circuit (where Kirchhoff's current law applies). In a “1”

junction, the flow and the efforts satisfy Eqs. (5.9) - (5.10):

∑ effort iinput =∑ effort output (5-9)

flow 1=flow 2 =.. . .. .=flow n

(5-10)
 Table 5.2 Basic Elements of Bond Graph

Elements Symbol Relation equations

Effort Se Se =e ( t )
1-Port
Elements
(Sources)
Flow Sf Sf =f (t )

de ( t ) 1
C-type Elements C = f (t)
dt C

1-Port df ( t ) 1
I-type Elements I = e (t)
Elements dt I

R-type Elements R e (t )=R ∙ f (t )

10 20
00 00 TF
e 2 ( t )=TF ∙ e 1 ( t )
Transformer 1
f 2 ( t )= ∙ f (t )
2-Port TF 1
Elements 10 20
00 00 GY e 2 ( t )=GY ∙ f 1 ( t )
Gyrator 1
f 2 ( t )= ∙e ( t )
10 20 GY 1
00 0 00
0-junction e 2 ( t )=e1 ( t )
3-Port 10 20
junction 1 00
00
elements
1-junction f 2 ( t )=f 1 ( t )
This corresponds to force balance at a mass in a system. An example of a “1” junction is a resistor in

series. In junction, the premise of energy conservation is assumed, no lost is allowed. There are two

additional variables, important in the description of dynamic systems.

For any element with a bond with power variables – effort and flow, the energy variation from t0 to t

can be expressed by:

t
H (t )−H ( t 0 )=∫t e ( τ ) f ( τ ) dτ
0

(5-11)

For C-type elements, e (effort) is a function of q (displacement). If displacement is differentiated, flow

is obtained as

dq
q ( t )=∫ f ( t ) dt ⇒ =f ( t )
dt (5-12)

If Eq. (5.11) is changing variables from t to q, the linear case can be expressed as:

1 2 2
H ( q )−H ( q0 )= q −q 0 )
2C (
(5-13)

For I-type elements, f (flow) is a function of p (momentum). If momentum is differentiated, effort is

obtained as
 dp
p (t )=∫ e ( t ) dt ⇒ =e ( t )
dt (5-14)

If Eq. (5.11) is changing variables from t to p, the linear case can be expressed as:

1
H ( p )−H ( p 0 ) = ( p 2− p20 )
2I (5-15)

Resistor elements represent situations where energy dissipates - electrical resistor, mechanical

damper, and coulomb frictions. In these sorts of elements there is a relationship between flow and

effort as the Eq. (5.16) shows. The value of “R” can be constant or function of any system parameter

including time.

e ( t )=R⋅f ( t ) (5-16)

Compliance elements represent the situations where energy stores - electrical capacitors, mechanical

springs, etc. In these sorts of elements there is a relationship between effort and displacement

variable as the Eq. (5.17) shows. The value of “K” can be constant or function of any system

parameter including time.

e ( t )=K⋅q ( t )
(5-17)
Inertia elements represent the relationship between the “flow” and Momentum (electrical coil, mass,

moment of inertia, etc.) as the Eq. (5.18) shows. The value of “I” tends to be constant

p (t )=I⋅f ( t ) (5-18)

A transformer adds no power but transforms it, such as an electrical transformer or a lever.

Transformers represent those physical phenomena that are variation of the values of output flow and

effort on the values of input flow and effort. If the transformation ratio is given by the “TF” value, then

the relationship between input and output is shown in Eqs. (5.19) - (5.20).

e output ( t )=TF⋅e input ( t ) (5-19)

1
f output ( t )= ⋅f (t )
TF input
(5-20)

Fig. 5.8."half-arrow" sign convention and meaning of the causal stroke

One is the "half-arrow" sign convention. This defines the assumed direction of positive energy flow. As

with electrical circuit diagrams and free-body diagrams, the choice of positive direction is arbitrary,

with the caveat that the analyst must be consistent throughout with the chosen definition. The other

feature is the "causal stroke". This is a vertical bar placed on only one end of the bond. It is not

arbitrary (Figure 5.8).

On each Bond, one of the variables must be the cause and the other one the effect. This can be

deduced by the relationship indicated by the arrow direction. Effort and flow causalities always act in

opposite directions in a Bond. The causality assignment procedure chooses who sets what for each

bond. Causality assignment is necessary to transform the bond graph into computable code.

Any port (single, double or multi) attached to the bond shall specify either "effort" or "flow" by its

causal stroke, but not both. The port attached to the end of the bond with the "causal stroke" specifies

the "flow" of the bond. And the bond imposes "effort" upon that port. Equivalently, the port on the end

without the "causal stroke" imposes "effort" to the bond, while the bond imposes "flow" to that port.

Once the system is represented in the form of Bond-graph, the state equations that govern its

behavior can be obtained directly as a first order differential equations in terms of generalized

variables defined above, using simple and standardized procedures, regardless of the physical

domain to which it belongs, even when interrelated across domains.

5.3.2 Case Study: Hydrostatic Transmission (HST) in sea-borne winch

The winch structure is designed for launching, owing, and handling the cable and array in ship. The

operation conditions of sea-borne winch can be varied such as operation conditions – sea state, ship

speed, and towing cable length. Because its operation requires high tension, sea-borne winch is

commonly used by the hydrostatic transmission (HST). It consists of electric motor, pump, piping,

hydraulic motor, and loads. Tension and the response characteristics under the states of launching,

towing, and hauling should be known before the design of HST. Tension data can be obtained from

tension experiment. However, as an experiment, obtained the exact time response characteristics has
many difficulties. And many previous design methods for HST involve extensive calculations because

energy type of HST changes from mechanical to hydraulic, and then mechanical system. Bond Graph

can easily model HST system and the dynamic response (Figure 5.8).
Fig. 5.8.Hydraulic-driven winch system in ship-borne
Fig. 5.19.Bond Graph and derivation of the state equations of the Hydrostatic transmission in sea-

borne winch
HST as shown in Figure 5.9 is commonly divided into electric motor, hydraulic pump, piping system,

safety switches, and hydraulic motor. A rotating electric motor operates a hydraulic pump, which

supplies oil to pipe system.

As cylinders in a hydraulic motor are filled with oil, shat rotates load. Therefore, HST is a kind of the

closed-loop power transmission. The effort and flow in the rotating mechanical/civil system are torque

and angular velocity, respectively. If two elements are integrated, they became momentum and

volume. No matter what systems in HST may be, power does not change.

Bond Graph of electric motor and hydraulic pump is shown in Figure 5.11. Source flow SF 11 indicates

an electric motor with constant angular velocity. It is assumed that a 10% among total torque perishes

out by resistance element R12. Transducer element MTF11 represents the capacity of a variable piston

pump which can control capacity with swivel angle o.

Fig. 5.10.Electric motor and hydraulic pump modeling

A bulk modulus B with implies oil compressibility chooses 10,000 bar among 6,000 ~ 12,000 bar. Fluid

condensers C23 = C21 are described as V/B. Fluid inertia I24 represents oil mass. Using the least

square method, resistance R22 and R26 are calculated from the pump and motor leakage. Because
pipe flow is laminar, fluid resistance R 25 can be calculated. Motor capacity TF 3128 is determined from

the number of filling cylinders. Moment of inertia of drum and flange I 33 can be calculated. It is

assumed that torque loss of flange R32 is about 10%. When Bond Graph is drawn from top and

bottom – starting with the electric motor and ending with the load, a total Bond Graph and derivation

of the state equations of a HST is represented as:

To obtain non-dimensional state equations, non-dimensional variables are introduced as

~ P Q ~ t 1
p= ,~
q= , t = −1 =
I Q̄ or ω C P̄ ωn √ IC (5-21)

dQ d ( C P̄ ~ q ) d ~t
q ) d ( C P̄ ~
= = ~ =C P̄ ωn ~
q̇
dt dt dt dt
(5-22)

dP d ( I Q̄ ~p ) d ( I Q̄ ~p ) d ~t
= = ~ =I Q̄ ω n ~ṗ
dt dt d t dt (5-23)

where P,Q and t are dimensional integral of pressure, volume and time ~ q and~t are non-
p, ~
dimensional integral of pressure, volume and time, respectively. Therefore non-dimensional state

equations are derived as:

C 23 ωn P̄ ~ P̄ ~ Q̄ ~
q̇ 23 + q 23 + p =1
SF 11 MTF 2113 R 23 SF 11 MTF 2113 SF 11 MTF 2113 24 (5-24)
 I 24 ωn ~ P̄ ~ ~ P̄ ~
ṗ 24− q 23 + p24 + q =0
R25 R25 Q̄ R25 Q̄ 27 (5-25)

~ Q̄ R26 ~ ~ ωTF 3128 R 26 ~

C27 ω n R26 q̇ 27 − p +q + p 33=0
P̄ 24 27 P̄ (5-26)

I 33 ωωn ~ P̄ TF 3128 ~ R 32 ω ~
ṗ − q 27 + p =0
SE 34 33 SE 34 SE 34 33

(5-27)

To investigate the dynamic stability of the system, simple asymptotic approach can be used and

perturbations around stable points are expressed as:

~ q 230 +ε 1 ~
q 23=~ q 231 +O ( ε 2 ) (5-28)

~ p 240 +ε 1 ~
p 24=~ p241 +O ( ε 2 ) (5-29)

~ q 270 +ε 1 ~
q 27=~ q 271 +O ( ε 2 ) (5-30)

~ p330 + ε 1 ~
p 33=~ p 331 +O ( ε 2 ) (5-31)

where1 is very small value.

Substitute Eq. (5.24) to (5.27) into (5.28) to (5.31), then the terms of 0 is yield

P̄ ~ Q̄ ~
( q 230 ) + p =1
R 22 SF 11 MTF 2113 SF11 MTF 2113 240 (5-32)

P̄ ~ ~ P̄ ~
− q 230 + p 240 + q =0
R 25 Q̄ R25 Q̄ 270 (5-33)

Q̄ R26 ωTF 3128 R26

~p +~ ~p =0
− 240 q 270 + 330
P̄ P̄
(5-34)

P̄ TF 3128 ~ R ω
− q 270 + 32 ~p =−1
SE 34 SE 34 330

(5-35)

And then the terms of 1 is yield

C 23 ωn P̄ ~ P̄ ~ Q̄ ~
q̇ 231 + q 231 + p =0
SF 11 MTF 2113 R23 SF 11 MTF 2113 SF 11 MTF 2113 241 (5-36)

I 24 ωn ~ P̄ ~ ~ P̄ ~
ṗ 241− q 231+ p241 + q =0
R25 R25 Q̄ R25 Q̄ 271 (5-37)
 ~ Q̄ R26 ωTF 3128 R26
C27 ω n R26 q̇ 271 − ~
p 241 +~
q 271 + ~
p 331=0
P̄ P̄ (5-38)

I 33 ωωn ~ P̄TF 3128 ~ R ω

ṗ 331− q 271 + 32 ~p =0
SE 34 SE 34 SE 34 331 (5-39)

If the perturbed Eqs. (5.36) to (5.39) are expressed as state space form [ dx /dt ] =[ A ][ X ], then

1 Q̄
0 0

[ ]
− −
R 22 C23 ω n C 23 ωn P̄
~
q̇ 231 R 25 ~
− P̄ P̄ q 231

[]
~
ṗ241
~ =
q̇ 271
~
ṗ 331
Q̄ I 24 ω n

0
−
I 24 ω n
1
P̄ C 27 ω n

0
−

−
Q̄ ω n I 24
1
C 27 ωn R 26
− P̄TF 3128
−

−
0

ω TF 3128
P̄ C 27 ω n
R 32
[]
~
p 241
~
q 271
~
p 331

I 33 ωωn I 33 ω n
(5-40)

To investigate the dynamic stability of the non-dimensional state Eq. (5.40), eigen-value of The Bond

Graph can be represented as a state equation form | A−λI |[ X ] =0. The system is unstable if eigen-
value are  > 0 and the system is stable  < 0. When the state equations are represented as state

space form of
 dQ 23 1 1
0 0

[ ][ ]
− −
dt C 23 R22 I 24
dP 24 1 R25 1 Q 23 MTF 2113 0

[ ][ ] [ ]
− − 0
dt = C 23 I 24 C 27 P 24 0 0 SE
dQ 27 TF 3128 Q 27
+
0
[ SF 11 ] + 0 [ 34 ]
1 1
0 − −
dt I 24 C 27 R 26 I 33 P33 0 −1
dP 33 TF 3128 R32
dt 0 0 −
C 27 I 33
(5-

41)

When Eq. (5.41) is integrated, the pump pressure and motor pressure are obtained as

P̄ pump 1/C 23 0 Q 23
[ ][
P̄motor
=
0 1/C 27 Q27 ][ ]
(5-42)

HST simulations are classified as models of low speed, high, and maximum tension. The tension

values might be obtained by the drag force analysis of cable. A steady solution of 0 equation and

eigen-values from high speed, low speed, and maximum tension are calculated as stable. The values

of (a) perturbed state Q23 (b) perturbed state P24 (c) perturbed state Q27 (d) perturbed state P33 from

high speed mode are shown in Figure 5.12.State variables are converged after they perturbed around

steady state value 0. It can figure out that simulations results with a big overshoot reach a stead state

value (Figure 5.13).

(a) Perturbed state~
q 23
(b) Perturbed state~
p 24

(c) Perturbed state~

q 27

(d) Perturbed state~

p 33
 Fig. 5.12.Perturbed state ~
q 23, ~
p 24, ~
q 27 and ~
p 33

Fig. 5.13. Simulation results for hydrostatic transmission

5.3.2 Case Study: Failure analysis and redesign of a helix upper dispenser

The mechanical icemaker system in a side-by-side (SBS) refrigerator with a dispenser system

consists of many structural parts. Depending on the customer usage conditions, these parts receive a

variety of mechanical loads in the ice making process. Ice making involves several mechanical

processes: (1) the filtered water is pumped through a tap line supplying the tray; (2) the cold air in the

heat exchanger chills the water tray; and (3) after ice is made, the cubes are harvested, stocking the

bucket until it is full. When the customer pushes the lever by force, cubed or crushed ice is dispensed.

In the United States, the customer typically requires an SBS refrigerator to produce 10 cubes per use

and up to 200 cubes a day. Ice production may be influenced by uncontrollable customer usage

conditions such as water pressure, ice consumption, refrigerator notch settings, and the number of

times the door is opened. When the refrigerator is plugged in, the cubed ice mode is automatically

selected. A crusher breaks the cubed ice in the crushed mode. Normally, the mechanical load of the

icemaker is low because it is operated without fused or webbed ice.

However, for Asian customers, fused or webbed ice will frequently form in the tray because they

dispense ice in cubed mode infrequently. When ice is dispensed under these conditions, a serious

mechanical overload occurs in the ice crusher. However, in the United States or Europe, the icemaker

system operates continuously as it is repetitively used in both cubed and crushed ice modes. This can

produce a mechanical/civil overload.

Figure 5.14 overviews the schematic of the ice maker. Figure5.15 and Figure5.16 show a schematic

diagram of the mechanical/civil load transfer in the ice bucket assembly and its bond-graphs. An AC

auger motor generates enough torque to crush the ice. Motor power is transferred through the gear

system to the ice bucket assembly – that is, to the helix upper dispenser, the blade dispenser and the

ice crusher.

Fig.5.14. Robust design schematic of ice maker

 Fig.5.15. Schematic diagram for mechanical ice bucket assembly

Fig.5.16.Bond Graph of ice bucket assembly

The Bond Graph can be represented as a state equation form, that is,

dfE 2 /dt=1/ La ×eE 2 (5-43)

 dfM 2 /dt=1/J ×eM 2 (5-44)

The junction from Eq. (5-43)

eE 2 =e a −eE 3

(5-45a)

eE 3 =Ra ×fE 3

(5-45b)

The junction from Eq. (5-44)

eM 2 =eM 1 −eM 3 (5-46a)

eM 1 =( K a×i ) −T Pulse

(5-46b)

eM 3=B×fM 3 (5-46c)

Because
fM 1=fM 2 =fM 3 =ω and
i=fE1 =fE 2 =fE 3 =i a ,

From Equation (5-45)

eE 2 =e a −Ra ×fE 3 (5-47)

fE 2 =fE 3=i a

(5-48)

If substituting Eqs. (5-47) and (5-48) into (5-43), then

di a / dt =1/ L a× ( e a −Ra ×i a )
(5-49)

And from Eq. (5.46) we can obtain

eM 2 =[ ( K a ×i ) −T Pulse ]−B×fM 3
(5-50a)
 i=i a (5-50b)

fM 3 =fM 2 =ω (5-50c)

If substituting Eq. (5-50) into (5-44), then

dω / dt =1/ J ×[ ( K a ×i ) −T Pulse ] −B×ω

(5-51)

So the state equation can be obtained from Eq. (5.49) and (5.51) as following

di a / dt −R a / La 0 i a 1/ La
e a + 1 T Pulse
[ ][
dω / dt
=
mk a −B/ J ω][ ] [ ] [ ]
+
0 −1/ J
(5-52)

When Eq.(5-52) is integrated, the angular velocity of the ice bucket mechanical assembly is obtained

as

ia
y p =[ 0 1 ]
[]
ω
(5-53)

5.4 Load Spectrum and Rain-flow Counting

5.4.1 Introduction

As seen in previous sections, we know that product subjected to a variety of loads can be simulated

through dynamics modeling like bond-graph. On the other hands, to experimentally measure the load

over time, strain gage type transducers are attached to the critical areas of the component. The

acquired data from the transducers are usually recorded and stored by a computer or by other

devices. After the recorded data is filtered to isolate the primary loads from noise, the recorded data
converted from the strain values to torque are counted by rain-flow counting methods. After simplifying

the fatigue damage computations, we can apply the Miner's rule (See Figure 5.17).

With the measurement data over time, we can perform a peak and trough detection to find the turning

points in the data. This is known as rain-flow counting. The output of this calculation is called the

torque count statistics. Some engineers stop at this point and define the rain-flow data as the load

spectrum, however it is not. Using the rain-flow data, it is then possible to calculate the histogram.

This histogram is the load spectrum. This load spectrum is very import during the design phase or a

refinement phase. The information from the load spectrum can be used with test rigs or simulation

software to reduce, but not remove, the need for field tests.

Realistic representation of loads is a key ingredient to successful fatigue analysis & design. It will

accurately measure the applied loads on an existing product and predict loads on a component or

structure that does not yet exist. Historically, complex load histories are often replaced by more

simplified loadings. The rain-flow cycle counting is a method for counting fatigue cycles from a time

history. The fatigue cycles are stress-reversals. The rain-flow method allows the application of Miner's

rule in order to assess the fatigue life of a structure subject to complex loading. And rain-flow counting

method may enable cumulative damage or the fatigue effects of loading events. The term "spectrum"

in fatigue often means a series of fatigue loading events other than uniformly repeated cycles.

Sometimes spectrum means a listing, ordered by size, of components of irregular sequences.

Maximum and minimum loads are also used to define the classifications in which the counts of cycles

are listed.
Fig.5.17 Classification and counting of the dynamic load
5.4.2 Rain-flow Counting

With the load-time, stress-time, or strain-time history, rain flowing down a roof can be represented by

the history of peaks and valleys. Rain-flow counting is a concept developed in Japan by Tatsuo Endo

and M. Matsuishi in 1968 [1] and in the USA for the segmentation of any arbitrary stress curve into

complete oscillation cycles. Rain-flow counting counts closed hysteresis loops in a load-time-function,

which are decisive for the damage of metal materials.

The following assumptions are valid for rain flow counting

 Cyclic stable material behavior, that means that the cyclic stress-strain curve remains

constant, thus no hardening or softening of the material takes place.

 Validity of the masing hypothesis, which means that the form of the hysteresis loop branches

correspond to the double of the initial load curve.

 Memory behavior of the material which means that after a closed hysteresis loop, a

previously not yet completely closed hysteresis loop follows the same ,  path.

As seen in Figure 5.18, the tips of the largest hysteresis loop are at the largest tensile and

compressive loads in the load history (points 1 and 4). The notch strain-time history (Figure 5.18(c)) is

quite different from the corresponding notch stress-time history (Figure 5.18(e)). During each segment

of the loading the material “remembers” its prior deformation (called material memory). The damage

from each counted cycle can be computed from the strain amplitude and mean stress for that cycle as

soon as it has been identified in the counting procedure. The corresponding reversal points can then

be discarded.

That sequence clearly has 10 cycles of amplitude 10 MPa and a structure's life can be estimated from

a simple application of the relevant S-N curve.

.
Fig.5.18. Rain-flow counting method
An advantage of rain-flow counting is when it is used with notch strain analysis. The rain-flow counting

results in closed hysteresis loops, which representing a counted cycle. Therefore, the closed

hysteresis loops can also be used to obtain the cycle counting. If the dynamic load-time behaviors are

acquired in Figure 5.19., they can be summarized by rain-flow counting as Table 5.3.

Fig.5.19.Acquisition of the dynamic load-time behavior with rain-flow counting algorithm

 Half-cycle starts at (A) and terminates opposite a greater tensile stress, peak (B); its range is

100MPa.

 Half-cycle starts at tensile peak (B), flow through (C), and terminates a greater tensile stress,

peak (E); its range is 200MPa.

Consequently, as seen in Table 5.3, we can count two cycles for 50 MPa range, two cycles for 100

MPa range, one cycle for 150 MPa range, and one cycle for 200 MPa range. Since calculated lifetime

estimations are afflicted with large uncertainties, it is desired to reconstruct the stochastic load-time

functions out of the load spectrums, in order to carry out experimental lifetime proofs with servo-

hydraulic facilities.

However, the reconstruction of a representative load-time function is not possible with the load

spectra alone. Two parametric rain-flow counting method is the most suitable method for the

acquisition of the local stress-strain hysteresis curves and influences the result of lifetime estimation.

Table 5.3 Summary of the dynamic loads by using rain-flow counting

Path From [MPa] To [MPa] Range [MPa] Cycles

A-B 0 100 100 0.5

B-E 100 -100 200 0.5

C-D -50 50 100 0.5

D-C 50 -50 100 0.5

E-F -100 100 200 0.5

F-I 75 -75 150 0.5

G-H 0 50 50 0.5

H-G 50 0 50 0.5

K-J -50 0 50 0.5

J-K 0 50 50 0.5

I-F -75 75 150 0.5

L-M 100 0 100 0.5

5.4.3 Goodman Relation

In the presence of a steady stress superimposed on the cyclic loading, the Goodman relation [2] can

be used to estimate a failure condition. It plots stress amplitude against mean stress with the fatigue
limit and the ultimate tensile strength of the material as the two extremes.

σm
(
σ a =σ 'e× 1−
σ 'u ) (5-54)

'
where σ e effective alternating stress at failure for a lifetime of Nf cycles, σ ' u is ultimate stress.

A very substantial amount of testing is required to obtain as S-N curve for the simple case of fully

reversed loading, and it will usually be impractical to determine whole families of curves for every

combination of mean and alternating stress. There are a number of strategems for finessing this

difficulty, one common one being the “Augmented” Modified-Goodman diagram, shown in Figure 5.20.

Fig.5.20. “Augmented” Modified-Goodman diagram

Here a graph is constructed with mean stress as the abscissa and alternating stress as the ordinate,

and a straight “lifeline” is drawn from σ e on the σ a axis to the ultimate tensile stress σ f on the σ m axis.

Then for any given stress, the endurance limit (or fatigue limit) – the value of alternating stress at

which fatigue facture never occurs - can be read directly as the ordinate of the lifeline at line is drawn
from the origin with a slope equal to that ratio. Its intersection with the lifeline then gives the effective

endurance limit for that combination of σ f and σ m

5.4.4 Palmgren-Miner’s law for cumulative damage

Fatigue properties of a material (S-N curves) are tested in rotating-bending tests in fatigue testing

apparatus. The S-N curve is required as a description of the material behavior for the calculation of

fatigue strength and operational fatigue strength. Well before a micro-structural understanding of

fatigue processes was developed, engineers had developed empirical means of quantifying the

fatigue process and designing against it. Perhaps the most important concept is the S-N diagram,

such as those shown in Figure 5.21, in which a constant cyclic stress amplitude S is applied to a

specimen and the number of loading cycles N until the specimen fails. Millions of cycles might be

required to cause failure at lower loading levels, so the abscissa in usually plotted logarithmically.

Fig.5.21. S - N diagram for aluminum and low-carbon steel

There are three zones to distinguish between in the double logarithmic representation of S – N curve
  Low cycle fatigue: high loads, plastic and elastic deformation, N = 10 ~ 103 cycles (1 stage)

 High cycle fatigue: fatigue strength, the zone of the sloped lines, until the corner load cycles

to failure ND = 106 ~ 107 (2 stage)

 Fatigue limit (endurance limit), zone of the horizontal lines starting from N > ND. However,

several materials such as austenite steels dot possess a distinct endurance strength (3

stage)

In this case, the S—N curve becomes horizontal at large N. The fatigue limit is maximum stress

amplitude below which the material never fails, no matter how large the number of cycles is. In most

alloys, S decreases continuously with N. In this case the fatigue properties are described by fatigue

strength at which fracture occurs after a specified number of cycles (e.g. 10 7). Fatigue life is number

of cycles to fail at a specified stress level.

Fatigue failure has three stages:1) crack initiation in the areas of stress concentration or near stress

raisers, 2) incremental crack propagation, and 3) final rapid crack propagation after crack reaches

critical size. The total number of cycles to failure is the sum of cycles at the first and the second

stages. That:

N f =N i +N p
(5-55)

where Nf number of cycles to failure, Ni Number of cycles for crack initiation, Np Number of cycles for

crack propagation

In the fatigue strength zone, the S – N curve can be described by the following equation if represented

in the double logarithmic form.

 −k
σ
N=N D⋅ a
σD ( ) (5-56)

If possible, the determination of the S – N curve for operational fatigue strength calculation should be

carried out on real parts. Often, however, due to cost and time limitations, the calculations are only

carried out on special test samples.

The resulting load cycles to failure are random variables, which mean that they lie scattered around

the mean value. Today, the transformation of results won from a tension/compression trial onto a real

component is difficult. Thus, the exact determination of a notch over the entire load cycle zone is still

not possible today. Therefore, one is forced to rely on tests and trials.

In some materials, notably ferrous alloys, the S – N curve flattens out eventually, so that below a

certain fatigue limit σ e failure does not occur no matter how long the loads are cycled. Obviously, the

designer will size the structure to keep the stresses below σ e by a suitable safety factor if cyclic loads

are to be withstood. For some other materials such as aluminum, no fatigue limit exists and the

designer will size the structure to keep the stresses below σ eby a suitable safety factor if cyclic loads

are to be withstood. For some other materials such as aluminum, no fatigue limit exists and the

designer must arrange for the planned lifetime of the structure to be less than the fatigue point on the

S - N diagram.

Statistical variability is troublesome in fatigue testing; it is necessary to measure the lifetimes of

perhaps twenty specimens at each of ten or so load levels to define the S-N diagram with statistical

confidence. It is generally impossible to cycle the specimen at more than approximately 10Hz and at

that speed it takes 11.6 days to reach 10 7 cycles of loading. Obtaining a full S-N curve is obviously a

tedious and expensive procedure.

At first glance, the scatter in measured lifetimes seems enormous, especially given the logarithmic

scale of the abscissa. If the coefficient of variability in conventional tensile testing is usually only a few
percent, why do the fatigue lifetimes vary over orders of magnitude? It must be remembered that in

tensile testing, we are measuring the variability in cycles at a given number of cycles, while in fatigue

we are measuring the variability in cycles at a given stress. State differently, in tensile testing we are

generating vertical scatters bars, but in fatigue they are horizontal. Note that we must expect more

variability in the lifetimes as the S-N curve becomes flatter, so that materials that are less prone to

fatigue damage require more specimens to provide a given confidence limit on lifetime.

Numerous different researchers have occupied themselves with the damage accumulation hypothesis

in fatigue failure, so that currently several variations exist. In general, the variations only distinguish

themselves by the fundamental S-N curve used: either fictitiously extrapolated or the real curve itself.

Oscillating loads cause an effect in materials, this is often referred to as “Damage” as soon as this

load surpasses a certain limit. It is assumed that this damage accumulates from the individual load

cycles and leads to a material fatigue. For an exact calculation this damage must be collected and

recorded quantitatively. This, however, has not yet been achieved with success.

Despite this fact, in order to gather information concerning the lifetime L out of the results of Wöhler

trials with irregular load cycle effects, around the year 1920, Palmgren [3] developed the fundamental

idea of linear accumulation, specific for roll bearing calculation. In 1945, Miner published the same

idea in a general form.

Fig.5.22.Schematic representation of Miner’s cumulative damage summation

Miner assumes that a part absorbs work during the fatigue process (Figure 5.22). The ratio of already

absorbed work to the maximal work which can be absorbed is a measurement for the current

damage. Thus, the ratio of the load cycle number n to the load cycles to failure N, which is determined

in the single-stage zone with the corresponding amplitude, is equal to the ratio of absorbed work w to

absorbed work W. This is denoted as the damage portion

w n
=
W N (5-57)

When the cycle load level varies during the fatigue process, a cumulative damage model is often

hypothesized. By definition of the S-N curve, take the lifetime to N 1 cycles at stress level S1 and N 2at

S2. If damage is assumed to accumulate at a constant rate during fatigue and a number of cycles n1 is

applied at stress S1, where n1 < N 1, then the fraction of lifetime consumed will be n1 / N 1.

The Palmgren-Miner hypothesis asserts that the damage fraction at any level Si is linearly

proportional to the ratio of number of cycles of operation to the total number of cycles that would

produce failure at that stress level; that is

nj
D i=
Nj (5-58)

The limiting condition of strength happens when the absorbed work and absorbable work are the

same. That is, the prerequisite that the absorbed fracture work W is the same for all occurring load

sizes, allows the addition of the individual damage portions for load cycles of different sizes
 w1 + w2 +⋯+ wm
=1
W (5-59)

So failure is predicted as follows,

w 1 w2 w n n n
D 1 + D2 + .. .+ D i= + +⋯+ m = 1 + 2 +⋯+ m =1
W W W N1 N2 Nm

(5-60)

The generalization of this approach is called Palmgren-Miner’s Law, and can be written

nj
∑ N ≤1
j (5-61)

wheren j is the number of cycles applied at a load corresponding to a lifetime of Nj

Miner confined the applicability of this equation by the following conditions

 Sinus formed load curve

 No hardening or softening appearances in the material

 The begin of a crack is considered as an incipient damage

 Some loads lie above the endurance strength

Minor’s law should be viewed like many other material laws that might be accurate enough to use in

design. But damage accumulation in fatigue is usually a complicated mixture of several different

mechanisms, and the assumption of linear damage accumulation inherent in Miner’s law should be

viewed skeptically. If portions of the material’s microstructure become unable to bear load as fatigue

progresses, the stress must be carried by the surviving micro-structural elements. The rate of damage

accumulation could drop during some part of the material’s lifetime. Miner’s law ignores such effects,

and often fails to capture the essential physics of the fatigue process

With knowledge of the load spectrum and the tolerable material load in the form of the S – N curve, a

lifetime prediction can be made for a mechanical/civil system with the help of a damage accumulation

hypothesis. Here it should be considered, that this prediction can only be made with a certain

probability, since among other things the load spectrum as well as the load capacity expressed in the

form of S – N curve are random variables. Likewise, the damage accumulation hypotheses known

today have only been proven empirically in material science. Therefore, a practical lifetime prediction

requires balance field tests, test stand trials, calculation and a careful assessment and evaluation of

the data, if the prediction should be able to serve as an effective tool for the designer.

Example 5.1 Stress σ1 has lifetime N1 = 104 cycles, and a more rigorous stress σ 2 has lifetime N2 =103

cycles. If 700 cycles at stress σ2 is operated, when will it stop to operate at stress σ 1?

Solution) From Palmgren-Miner’s Law Eq. (5-61), we can calculate the cycles to fail.

700 x
+ =1
1000 10000

So the expected failure cycle is x = 3000 cycles

Example 5.2 A part is subjected to a fatigue environment where 10% of its life is spent at an

alternating stress level, σ1, 30% is spent at a level σ 2, and 60% at a level σ 3. How many cycles, n, can
the part undergo before failure?

If, from the S-N diagram for this material the number of cycles to failure at I (i=1, 2, 3), then from the

Palmgren-Miner rule failure occurs when:

0.1n 0.3n 0.6n

+ + =1
N1 N 2 N 3

so solving for n gives

1
n=
0.1 0.3 0.6
+ +
N1 N2 N3

If N1, N2, N3 are 103,104, and 105, the time to failure n will be 7353 cycles

REFERENCES

[1]. Matsuishi M & Endo T (1968) Fatigue of metals subjected to varying stress. Japan Soc. Mech.

Engineering.

[2]. Mott & Robert L. (2004) Machine elements in mechanical design (4 th ed.) Upper Saddle River NJ:

Pearson Prentice Hall pp. 190–192

[3]. Palmgren AG (1924) Die Lebensdauer von Kugellagern Zeitschrift des Vereines Deutscher

Ingenieure 68(14): 339–341

Chapter 6

Mechanical System Failures

Seong-woo Woo

Email: [email protected]

Reliability Association of Korea, Seoul, Korea (Republic of)

Abstracts: This chapter will review the concepts of fracture and fatigue that occupy most of part

failures in mechanical system subjected to random stress (or loads). To figure out the mechanical

system failures, it will benefit for engineer to design the product structures – automobile, bridge,

skyscrapers, and the others – in the allowable stress and strain as mechanical properties. However,

as the current reliability methodology, engineer still doesn’t know whether the stress in product lifetime

overcomes the random stress in lifetime. Failure of mechanical components in aircraft wing during a

long flight can occur in short time or tens of thousands of vibration load cycles. Fatigue fracture

catastrophically occurs in product lifetime when there are stress raisers such as holes, notches, or

fillets in design. Mechanical engineer should clearly figure out the failure mechanism like fracture or

fatigue to design the product structure subjected to random loading. Consequently, to discover the

problematic part, we need new reliability methodology like parametric accelerated life testing in the

reliability-embedded design process.

Keywords: Mechanical system failures, Fracture, Fatigue, Design, Failure analysis

6.1 Introduction

Fatigue was coined by France engineer Jean-Victor Poncelet in the middle of the nineteenth century.

It meant to represent that the material got tired due to repeated loading, and eventually disintegrated

[1]. The National Bureau of Standards and Battelle Memorial Institute estimated the costs for failure
due to fracture to be $119 billion per year in 1982 [2]. The required costs are important, but the safety

of many failures in human life and injury is infinitely more so.

Fracture mechanics is the study field concerned with the study how cracks in materials propagate. It

uses methods of analytical solid mechanics to calculate the driving force on a crack and those of

experimental solid mechanics to characterize the material's resistance to fracture. When subjected to

a variety of loading, fractures have occurred in design inadequacies. Design against fracture still has

area of current research and its final goal

As seen in Figure 6.1, stress is a physical quantity that expresses the response of the material on the

unit area (A) acted in the external (or internal) forces (F). And strain is physical deformation response

of a material to stress.

F
σ=
A
(6-1)

Fig.6.1. Definition of stress

The linear portion of the stress-strain curve is the elastic region and the slope is Young's Modulus.

The elastic range ends when the material reaches its yield strength. After the yield point, the curve

typically decreases slightly and deformation continues. Strain hardening and plastic deformation

begins until it reaches the ultimate tensile stress.

Deformation refers to any changes in the shape of an object due to an applied force. Elastic

deformation is that once the forces are no longer applied, the object returns to its original shape. This

type of deformation involves stretching of the atoms bonds. Linear elastic deformation is governed by

Hooke's law, which states:

σ =Eε
(6-2)

where  is the applied stress, E is Young's modulus, and ε is the strain.

Applied force consists of tension, compression, shear, and torsion (Figure 6.2). Tensile means the

material is under tension. The forces acting on it are trying to stretch the material. Compression is

when the forces acting on an object are trying to squash it.

 Axial loading (tension/compression) - The applied forces are collinear with the longitudinal

axis of the member. The forces cause the member to either stretch or shorten.

 Transverse loading (shear) - Forces applied perpendicular to the longitudinal axis of a

member. Transverse loading causes the member to bend and deflect from its original

position, with internal tensile and compressive strains accompanying the change in curvature

of the member. Transverse loading also induces shear forces that cause shear deformation

of the material and increase the transverse deflection of the member.

  Torsional loading - Twisting action caused by a pair of externally applied equal and

oppositely directed force couples acting on parallel planes or by a single external couple

applied to a member that has one end fixed against rotation.

Fig.6.2. A variety of stresses due to four kinds of static loads

Many materials are made up of many grains which may have second phase particles and grain

boundaries. It is therefore easier to study plastic deformation in a single crystal to eliminate the effects

of grain boundaries and second phase particles.

 Fig.6.3. Plastic deformation (metals) for axial force

If a single crystal of a metal is stressed in tension beyond its elastic limit, it elongates slightly that it is

called plastic deformation (Figure 6.3). Plastic deformation involves the breaking and remaking of

atomic bonds. Plastic deformation may take place by slip, twinning or a combination of both methods.

Plastic deformation cannot be restored to its initial state by changes, .i.e. irreversible process. Under

tensile stress, plastic deformation is characterized by a strain hardening region, necking region and

finally, fracture (also called rupture).

During strain hardening the material becomes stronger through the movement of atomic dislocations.

The necking phase is indicated by a reduction in cross-sectional area of the specimen. Necking

begins after the ultimate strength is reached. During necking, the material can no longer withstand the

maximum stress and the strain in the specimen rapidly increases. Plastic deformation ends with the

fracture of the material. Fracture is the separation of a single body into pieces by an applied stress.

6.2 Mechanism of Slip

 Fig.6.4. Components of force on a slip plane

Slip occurs on planes that have highest planer density of atoms and in the direction with highest linear

density of atoms (Figure 6.4). That is, slip occurs in directions in which the atoms are most closely

packed since this requires the least amount of energy. Therefore they can slip past each other with

force. Slip flow depends upon the repetitive structure of the crystal which allows the atoms to shear

away from their original neighbors. It therefore slides along the face and join up with the atom of new

crystals.

Slip takes place as a result of simple shearing stress. Resolution of axial tensile load F gives two

loads. One Fs is shear load along the slip plane and the other FN a normal tensile load perpendicular

to the plane. By analysis and experiment maximum shear stress happens at 45. Above Figure 6.4

right shows the packing of atoms on a slip plane. We know that there are three directions in which the

atoms are close-packed, and these would be the easy slip directions.

Portions of the crystal on either side of a specific slip plane move in opposite directions and come to

rest with the atoms in nearly equilibrium positions, so that there is very little change in the lattice

orientation. Thus the external shape of the crystal is changed without destroying it. Schematically, slip

can be explained in a face centered cubic (FCC) lattice. The (111) plane is the slip plane having
maximum number of atoms (densest plane). It intersects the (001) plane in the line AC, (110) direction

having maximum number of atoms on it. Slip is seen as a movement along the (111) planes in the

close packed (110) direction (Figure 6.5 (a)).

(a) Slip plane in FCC lattice (b) its schematic diagram

Fig.6.5. Schematic diagram of slip plane in FCC lattice

From the schematic diagram of slip in a FCC crystal, one may assume that the atoms slip

consecutively, starting at one place or at a few places in the slip plane, and then move outward over

the rest of the plane. For instance, if one tries to slide the entire rug as one piece, the resistance is too

much. What one can do is to make a wrinkle in the rug and then slide the whole rug a little at a time

by pushing the wrinkle along. A similar analogy to the wrinkle in the rug is the movement of an

earthworm. It advances in a direction by advancing a part of its body at a time.

By application of the shear force, first an extra plane of atoms (called a dislocation) forms above the

slip plane. On application of force, bond between atoms breaks and creates a new bond between

atoms and a dislocation. On continued application of force, this dislocation advances by breaking old

bonds and making new bonds. In the next move, bond between atoms is broken and a new bond is

made between atoms, resulting in a dislocation. Thus, this dislocation moves across the slip plane

and leaves a step when it comes out at the surface of the crystal. Each time the dislocation moves

across the slip plane, the crystal moves one atom spacing (Figure 6.5 (b)).
6.3 Facture failure

Fracture is the separation of a body into pieces subjected to stress. Fracture takes place whenever

the applied loads (or stresses) are more than the resisting strength of the body. It starts with a crack

that breaks without making fully apart. Fracture due to overstress is probably the most prevalent

failure mechanism in mechanical/civil system and might be classified as ductile fracture, brittle

fracture, and fatigue fracture.

As seen in Figure 6.5, brittle fracture is the failure of a material with minimum of plastic deformation.

Brittle fracture propagates rapidly on a crack with minimum energy absorption and plastic

deformation. Brittle fracture occurs along characteristics crystallographic planes called as cleavage

planes. The mechanism of Brittle fracture was initially explained by Griffith theory [3]. Griffith

postulated that in a brittle material there are micro cracks which act to the concentrated stress at their

tips. The crack could come from a number of sources as flow occurred during solidification or a

surface scratch.
 (a) Brittle (cleavage) fracture mechanism (b) Example: grass

Fig.6.5 Brittle (cleavage) fracture mechanism

(a) Necking, (b) Formation of micro-voids, (c) Coalescence of micro-voids to form a crack, (d) Crack propagation by shear deformation, (e) Fracture
Fig.6.6 Ductile fracture failure mechanism
Brittle materials are glasses, ceramics, some polymers,metals. They have the following

characteristics:

 No appreciable plastic deformation

 Crack propagation is very fast

 Crack propagates nearly perpendicular to the direction of the applied stress

 Crack often propagates by cleavage – breaking of atomic bonds along specific

crystallographic planes (cleavage planes).

Ductile fracture occurs after considerable plastic deformation. The crack will not extend unless an

increased stress is applied. The failure of most polycrystalline ductile materials occurs with a cup-and-

cone fracture associated with the formation of a neck in a tensile specimen. In ductile material the

fracture begins by the formation of cavities (micro-voids) in the center of the necked region. In most

commercial metals, these internal cavities probably form at nonmetallic inclusions. Increasing the

load, increasing the permanent elongation and simultaneously decrease the cross sectional area. The

decrease in area leads to the formation of a neck in the specimen.

The neck region has a high dislocation density and the material is subjected to a complex stress. The

dislocations are separated from each other because of the repulsive inter atomic forces. As the

resolved shear stress on the slip plane increase, the dislocation comes closely together. The crack

forms due to high shear stress and the presence of low angle grain boundaries. Once a crack is

formed, it can grow or elongated by means of dislocations. Crack propagates along the slip plane for

this mechanism. Once crack grows at the expense of others, finally cracks growth results in failure.

The final stage leaves a circular lip on one half of the sample and a bevel on the surface of the other

half. Thus one half has the appearance of a shallow cup, and the other half resembles a cone with a

flattened top (See Figure 6.6 & Figure 6.7).

  Brittle Fracture: Separation along crystallographic planes due to breaking of atomic bonds

(V-shaped Chevron, Cleavage, Inter-granular)

 Ductile Fracture: Initiation, growth and coalescence of micro-voids (Cup-and-cone, Dimple)

Fig.6.7. Brittle Vs Ductile Fracture in material

However, commercial material is made up of polycrystalline, whose crystal axes are oriented at

random. When polycrystalline material is subjected to stress, slip starts first in those grains in which

the slip system is most favorably situated with respect to the applied stress. Since contact at the grain

boundaries is maintained, it may be necessary for more than one slip system to operate. The rotation

into the axis of tension brings other grains, originally less favorably oriented, into a position where

they can now deform. As deformation and rotation proceed, the individual grains tend to elongate in

the direction of flow.

When a crystal deforms, there is some distortion of the lattice structure. This deformation is greatest

on the slip planes and grain boundaries and increases with deformation. This is evident by an

increase in resistance to further deformation. The material is undergoing strain hardening or work
hardening. Since dislocations pile up at grain boundaries, metals can be hardened by reducing the

size of the grains.

6.4 Fatigue failure

6.4.1 Introduction

Another deformation mechanism is fatigue failure, which occurs primarily in ductile metals. Fatigue

may occur when a member is subjected to repeated cyclic loadings. The fatigue phenomenon shows

itself in the form of cracks developing at particular locations in the structure. Cracks can appear in

diverse types of structures such as: planes, boats, bridges, frames, cranes, overhead cranes,

machines parts, turbines, reactors vessels, canal lock doors, offshore platforms, transmission towers,

pylons, masts and chimneys.

Design faults such as stress raisers are deformed. After repetitive deformations, cracks will begin to

appear. Depending on the material, shape, and how close to the elastic limit it deforms, failure may

require a lot of deformation cycles. Structures subjected to repeated cyclic loadings can undergo

progressive damage, which is called fatigue (Figure 6.8).

Fig.6.8 Facture of train wreck due to metal fatigue failure of rail from Wikipedia
The fatigue life of a structural detail subjected to repeated cyclic loadings is defined as the number of

stress cycles it can stand before failure. The physical effect of a repeated load on a material is

different from the static load. Failure always may be brittle fracture regardless of whether the material

is brittle or ductile. Mostly fatigue failure occurs at stress well below the static elastic strength of the

material.

Depending upon the member or structural detail geometry, its fabrication or the material used, four

main parameters can influence the fatigue strength – 1) the stress difference, or as most often called

stress range, 2) the structural geometry, 3) the material, 4) the environment

6.4.2 Type of fatigue loading

Cyclic loading due to repeated force and weight of product is a universal loading condition. Essentially

all structural components are subjected to some type of fluctuating loading during product lifetime, so

they develop fatigue inducing varying stresses. It is virtually impossible to think of any structural

component that does not experience some form of alternating loading.

Three different fluctuating stress-time modes are symmetrical about zero stress, asymmetrical about

zero stress, and random stress cycle. For reversed stress cycle, amplitude is symmetric about a mean

zero stress level. It alternates from σ max to σ min of equal magnitude. Repeated stress cycle is

asymmetrical about σ maxand σ min relative to zero stress level. Random stress cycle is that stress level

fluctuates very randomly in amplitude and frequency.

For asymmetrical about zero stress, cyclic stresses that arise fatigue are characterized by mean

stress m, the range of stress , alternating component a, amplitude ration A, and the stress ratio R

(See Figure 6.9), They are represented as the following Eqs. (6.3) – (6.7).
 m

Tensio
ax

+
CompressiStress
a
n 
 r

0
- a

mi Ti
on

n m
e
Fig.6.9 Fatigue: Failure under fluctuating and cyclic stresses asymmetrical about zero stress

( σ max +σ min )
σ m=
2 (6-3)

Δσ=( σ max −σ min )

(6-4)

( σ max −σ min )
σ a=
2 (6-5)

σa
A=
σm (6-6)

R=σ min /σ max

(6-7)

As seen in Figure 6.10, another cyclic stresses that arise fatigue are 1) periodic and symmetrical

about zero stress, 2) random stress fluctuations. In mechanical/civil system such as bridges, aircraft,

machine components, and automobiles, fatigue failure under fluctuating/cyclic stresses are required:
 A maximum tensile stress of sufficiently high value

 A large enough variation or fluctuation in the applied stress, and

 A sufficiently large number of applied stress cycles.

m
ax
Compress s Tensi

+
on
Stres

0
-

m Ti
ion

in m
e
(a) Periodic and symmetrical about zero stress

m
ax
Tensio

+
CompressiStress
n

0
-

m
Ti
on

in m
e
(b) Random stress fluctuations
 Fig.6.10 Fatigue: Failure under fluctuating and cyclic stresses

There are two ways to determine when component is in danger of metal fatigue; either predicts when

failure will occur due to the material/force/shape/iteration combination, and replace the vulnerable

materials before this occurs, or perform inspections to detect the microscopic cracks and perform

replacement once they occur. Selection of materials not likely to suffer from metal fatigue during the

life of the product is the best solution, but not always possible. Avoiding shapes with sharp corners

limits metal fatigue by reducing stress concentrations, but does not eliminate it.

Fatigue is a critical failure mechanism to be considered in product designs. It is a process in which

damage accumulates due to the repetitive loads below the yield point, which is brittle-like even in

normally ductile materials. Fatigue cracks begin very small and initially grow very slowly until the crack

length approaches the critical length. So it is dangerous because it is difficult to initially detect

cumulative fatigue damage with the naked eye until the crack has grown to near critical length. Typical

fracture surface is perpendicular to direction of applied stress. Fatigue failure has three distinct

stages: 1) crack initiation in the areas of stress concentration (near stress raisers), 2) incremental

crack propagation, and 3) final catastrophic failure.

The examples of “fatigue” for a multitude of reasons has been studied as the disaster of Comet

aircraft and Versailles rail accident occurred when they became large enough to propagate

catastrophically (See Chapter 2). Fatigue failure occur in both metallic and non-metallic materials, and

are responsible for about estimated 80-90% of all structural failures - automobile crank-shaft, motor

shaft, bridges, aircraft landing gear machine components, and the others. Thus, designing for

maximum stress will not ensure adequate product lifetime. Most fracture induced belongs to this

category.

Engineering stress is irregular around stress raisers such as holes, notches, or fillets that concentrate

on the stress. For complex drawings, engineer frequently neglects these design flaws that might

cause the reliability disasters. For instance the vibration of aircraft wing during a long flight can result

in tens of thousands of load cycles. If designed improperly, these structures will fracture. It is

important to find the design faults. In Chapter 7 we will discuss how to find the missing design
parameters by using the parametric ALT.

The central difficulty in designing against fracture in high-strength materials is that the presence of

cracks can modify the local stresses to such an extent that the elastic stress analyses done so

carefully by the designers are insufficient. When a crack reaches a certain critical length, it can

propagate catastrophically through the structure, even though the gross stress is much less than

would normally cause yield or failure in a tensile specimen. The term “fracture mechanics” refers to a

vital specialization within solid mechanics in which the presence of a crack is assumed, and we wish

to find quantitative relations between the crack length, the material’s inherent resistance to crack

growth, and the stress at which the crack propagates at high speed to cause structural failure.

Fast fracture can occur within a few loading cycles. For example, fatigue failures in 1200 rpm motor

shafts took less than 12 hours from installation to final fracture, about 830,000 cycles. On the other

hand, crack growth in slowly rotating process equipment shafts has taken many months and more

than 10,000,000 cycles to fail.

6.4.3 Stress concentration at crack tip

Fracture strength of a material is related to the cohesive forces between atoms. One can estimate that

the theoretical cohesive strength of a material should be one-tenth of the elastic modulus (E).

However, the experimental fracture strength for brittle material is normally E/100 - E/10,000 below this

theoretical value. This much lower fracture strength is caused from the stress concentration due to the

presence of microscopic flaws or cracks found either on the surface or within the material. As seen in

Figure 6.11, stress profile along X axis is concentrated at an internal, elliptically-shaped crack.

Stress has a maximum at the crack tip and decreased to the nominal applied stress with increasing

distance away from the crack. Flaws such stress concentrators or stress raisers have the ability to

amplify the stress at a given point. The magnitude of amplification depends on crack geometry and

orientation.
 (a) Geometry of internal cracks (b) Schematic stress profile along X axis in (a)

Fig.6.11. Stress concentration at crack tip positions

If the crack is similar to an elliptical hole through plate and is oriented perpendicular to applied stress,

the maximum stress max occurs at a crack tip and approximated by Eq. (6-8)

[ √]
σ max =σ ∞ 1+2
a
ρ (6-8)

where  = radius of curvature,  = applied stress, σ max = stress at crack tip, a = half-length of internal

crack or the full length for a surface flaw.

The magnitude of the nominal applied tensile stress is ; the radius of the curvature of the crack tip is

; and a represents the length of a surface crack, or half the length of an internal crack. For a

relatively long micro-crack, the factor ( a /❑ )1 /2 may be very large. So Eq. (6-8) can be modified as:
 1/2
a
σ m ≃2σ ∞ ()
ρ (6-9)

The ratio of the maximum stress and the nominal applied tensile stress is denoted as the stress

concentration factor Kt. The stress concentration factor is a simple measure of the degree to which an

external stress is amplified at the tip of a small crack and described as:

σ a 1/2
K t = max ≈2
σo ρt() (6-10)

Because an external stress is amplified at the tip of a crack, Eq. (6.10) can be rearranged as:

1/2
a
σ max =2σ ∞
ρ() =K t σ ∞
(6-11)

Stress amplification not only occurs at small flaws or cracks on a microscopic level of material but can

also occur in sharp corners, holes, fillets, and notches on the macroscopic level. Cracks with sharp

tips propagate easier than cracks having blunt tips. Because of amplifying an applied stress, stress

concentration may occur at microscopic defects, internal discontinuities (voids/inclusions), sharp

corners, scratches and notches that are often called stress raisers. Stress raisers are typically more

destructive in brittle materials. Ductile materials have the ability to plastically deform in the region

surrounding the stress raisers which in turn evenly distributes the stress load around the flaw. The

maximum stress concentration factor results in a value less than that found for the theoretical value.

Since brittle materials cannot plastically deform, the stress raisers will create the theoretical stress

concentration situation.
The magnitude of this amplification depends on micro-crack orientations, geometry and dimensions.

For example, stress concentration at sharp corners depends on fillet radius (See Figure 6.12).

Fig.6.12.Stress concentration at sharp corners in accordance with fillet radius [4]

6.4.4 Crack propagation and fracture toughness

Cracks with sharp tips propagate easier than cracks having blunt tips. In ductile materials, plastic

deformation at a crack tip “blunts” the crack. Elastic strain energy is stored in material as it is

elastically deformed. This energy is released when the crack propagates. And creation of new

surfaces requires energy. Critical stress that is required for crack propagation is described as:

1/2
2Eγ s
σ c= ( )
πa
(6-12)

where s = specific surface energy

When the tensile stress at the tip of crack exceeds the critical stress value, the crack propagates and

results in fracture. Most metals and polymers have plastic deformation. For ductile materials specific

surface energy γs should be replaced with γ s + γ pwhere γ p is plastic deformation energy. So Eq. (6-12)

can be described as:

1/2
2 E ( γ s +γ p )
σ c= ( πa ) (6-13)

For highly ductile materials, γ p ≫ γ s is valid. So Eq. (6-13) can be modified as:

1/2
2Eγ p
σ c= ( )
πa
(6-14)

All brittle materials contain a population of small flaws that have variety of sizes. When the magnitude

of the tensile stress at the tip of crack exceeds the critical stress value, the crack propagates and

results in fracture. Very small and virtually defect-free metallic and ceramic materials have been

produced with facture strength that approaches their theoretical values.

Example 6.1 There is a long plate of glass subjected to a tensile stress of 30 MPa. If the modulus of

elasticity and specific surface energy for this glass are 70 GPa and 0.4 J/m 2, find out the critical length

of a surface flaw that can have no fracture.

From Eq. (6-12), E= 70GPa, s = 0.4 J/m2,  = 40MPa. So the critical length can be obtained as
 2 Eγ s 2⋅70 GPa⋅0 . 4 J /m2
ac =( )(
πσ 2
=
π⋅( 30 MPa ) 2 ) =2 . 0×10−6 m

Fracture toughness Kc is a material’s resistance to fracture when a crack is present. It therefore

means the amount of stress required to propagate a flaw. It can be described as:

K c =σ c √ πa (6-15)

Fracture toughness depends on temperature, strain rate and microstructure. Its magnitude diminishes

with increasing strain rate and decreasing temperature. If yield strength due to alloying and strain

hardening improve, fracture toughness will increase with reduction in grain size.

6.4.5 Crack growth rates

The metal fatigue begins at an internal (or surface) flaw by the concentrated stresses, and progress

initially of shear flow along slip planes. As previously mentioned in section 6.2, slip can happen (111)

plane in a FCC lattice because the atoms are most closely packed. Over a number of random loading

cycles in field, this slip generates intrusions and extrusions that begin to resemble a crack. A true

crack running inward from an intrusion region may propagate initially along one of the original slip

planes, but eventually turns to propagate transversely to the principal normal stress.

After repeated loadings, the slip bands can grow into tiny shear-driven micro-cracks. These Stage I

cracks can be described as a back and forth slip on a series of contiguous crystallographic plane to

form a band. It is within this slip bands that the process of pores nucleation and coalescence. The

process eventually leads to micro cracks formation. Often, extrusion and intrusions may also appear

which, being a very localized discontinuity, results in a much faster micro crack formation.

Micro cracks join to form a macro crack in Stage II of fatigue. Now the crack is already long enough to
escape shearing stress control and be driven by normal stress which produces a continuous growth,

cycle by cycle, on a plane that is no longer crystallographic, but rather normal to external loads.

Ahead of this macro crack two plastic lobes are generated by stress concentration. The cracks grow

perpendicular to the dominant stress and increases dramatically by plastic stresses at the crack tip as

seen in Figure 6.13.

(a) Fatigue due to repeated loads (b) Nucleation and growth of voids

Fig.6.13 A schematic diagram of general slip producing nucleation and growth of voids.

It is vital that engineers be able to predict the rate of crack growth during load cycling in aircraft as

well as in other engineering structures, so that the problematic parts be replaced or repaired before

the crack reaches a critical length. A great deal of experimental evidence supports the view that the

crack growth rate can be corrected with the cycle variation in the stress intensity factor [5]:

da
=A ΔK m
dN (6-16)

where d a/dN is the fatigue crack rate per cycle, ∆ K =K min −K max is the stress intensity factor

range during the cycle, and A and m are parameters that depend the material, environment,
frequency, temperature and stress ratio.

Fatigue crack propagation rate during Stage II depends on stress level, crack size, and materials. This

is sometimes known as the “Paris Law,” and leads to plots similar to that shown in Figure 6.14.

Fig.6.14. Paris law for fatigue crack growth rates

Table 6.1 Numerical parameters in the Paris equation

Alloy m A

Steel 3 10-11

Aluminum 3 10-12

Nickel 3.3 410-12

 Titanium 5 10-11

Some specific values of the constants m and A for various alloys are given in Table 6.1. The exponent

m is often near 4 for metallic systems, which might be rationalized as the damage accumulation being

related to the volume Vp of the plastic zone: since the volume Vp of the zone scales with r 2p and

r p ∝ K 2I , then da /dn∝ ∆ K 4 .

6.4.6 Ductile-brittle transition temperature (DBTT)

Fig.6.15. Ductile-to-Brittle Transition Temperature

The Ductile-to-Brittle Transition Temperature (DBTT) is widely observed in metals that are dependent

on the composition of the metal. For some steels the transition temperature can be around 0°C, and in

winter the temperature in some parts of the world can be below this. As a result, some steel structures

are very likely to fail in winter. The controlling mechanism of this transition still remains unclear despite
of large efforts made in experimental and theoretical investigation. All ferrous materials (except the

austenitic grades) exhibit a transition from ductile to brittle when tested above and below a certain

temperature, called as ductile-to-brittle transition temperature (DBTT). FCC metals such as Cu, Ni

remain ductile down to very low temperatures. For ceramics, this type of transition occurs at much

higher temperatures than for metals (Figure 6.15).

Since the famous weld fractures in some US army ships (Liberty Ships, tankers) during World War II

are investigated, the ductile-to-brittle transition can be measured by impact testing such as Charpy V-

notch testing (Figure 6.16). The impact energy needed for fracture drops suddenly over a relatively

narrow temperature range – temperature of the ductile-to-brittle transition. Primary function of Charpy

V-notch testing is to determine whether a material experiences a ductile to brittle transition with

decreasing temperature. When the results of a number of tests performed in different temperatures

are plotted, ductile-to-brittle transition curves may be obtained.

Steels were used having DBTT’s just below room temperature. Low temperatures can severely

become brittle steels. At higher temperature, the impact energy is large, corresponding to a ductile

mode of fracture. As the temperature is lowered, the impact energy drops suddenly over a relatively

narrow temperature range which corresponding to the mode of brittle fracture. Fatigue cracks

nucleated at the corners of square hatches and propagated rapidly by brittle fracture.
 Fig.6.16. Schematic of a conventional Charpy V-notch testing

6.4.7 Fatigue Analysis

The majority of component designs involve parts subjected to fluctuating or cyclic loads. Such loading

induces fluctuating or cyclic stresses that often result in failure by fatigue. About 95% of all structural

failures occur through a fatigue mechanism. The concept of fatigue that describes structural system

subjected to repeated loadings was originated in the mid-eighteenth century in the railroad industry.

When fatigue failures of railway axles became a widespread problem in the middle of the nineteenth

century, this drew attention to cyclic loading effects. This was the first time that many similar

components had been subjected to millions of cycles at stress levels well below the monotonic tensile

yield stress.
 Fig.6.17 Some of Wöhler's data for rail car axles steel on the S-N diagram [6],

The modern study of fatigue is generally dated from the work of A. Wöhler, a German engineer in the

railroad system in the mid-nineteenth century. Wöhler was chief superintendent of rolling stock on the

Lower Silesia-Brandenberg Railroad. Wöhler was concerned by the causes of fracture in rail car axles

after prolonged use. A railcar axle is essentially a round beam in four-point bending, which produces a

compressive stress along the top surface and a tensile stress along the bottom. After the axle has

rotated a half turn, the bottom becomes the top and vice versa, so the stresses on a particular region

of material at the surface vary repeatedly form tension to compression. Though the metal became

tired, fatigue was named to describe this type of damage. This is now known as fully reversed fatigue

loading. At the same time, other engineers began to concern themselves with problems of failures

associated with fluctuating loads in bridges, marine equipment, and power generation machines (See

Figure 6.17).
Since 1830, it has been recognized that metal under a repetitive or fluctuating load will fail at a stress

level lower than required to cause failure under a single application of the same load. Figure 6.18

shows a bar-shaped component subjected to a uniform sinusoidally varying force. After a period of

time, a crack can be seen to initiate on the circumference of the hole. This crack will then propagate

through the component until the remaining intact section is incapable of sustaining the imposed

stresses and the component fails.

Fig.6.18 Bar-shaped component subjected to a uniform sinusoidally varying force,

The physical development of a crack is generally divided into 2 separate stages. These relate to the

crack initiation phase (Stage I) and the crack growth phase (Stage II). Fatigue cracks initiate through

the release of shear strain energy. The following diagram shows how the shear stresses result in local

plastic deformation along slip planes. As the sinusoidal loading is cycled, the slip planes move back

and forth like a pack of cards, resulting in small extrusions and intrusions on the crystal surface.

These surface disturbances are approximately 1 to 10 microns in height and constitute embryonic

cracks.

A crack initiates in this way until it reaches the grain boundary. The mechanism at this point is

gradually transferred to the adjacent grain. When the crack has grown through approximately 3
grains, it is seen to change its direction of propagation. Stage I growth follows the direction of the

maximum shear plane, or 45° to the direction of loading. The physical mechanism for fatigue changes

during Stage II. The crack is now sufficiently large to form a geometrical stress concentration. A

tensile plastic zone is created at the crack tip as shown in the following diagram. After this stage, the

crack propagates perpendicular to the direction of the applied load.

6.5 Stress–strength analysis

Fig. 6.19. Applied fatigue stress-strength interference model

Stress–strength analysis in reliability engineering is the analysis of the strength of the materials and

the interference of the stresses placed on the materials [7]. A product's probability of failure is equal to

the probability that the stress experienced by that product will exceed its strength. If given one

probability distribution function for a product's stress and its strength, the probability of failure can be

estimated by calculating the area of the overlap between the two distributions. This overlapping region

may be also referred to as stress-strength interference. However, if there is the design failure like

stress raiser in structure, stress–strength interference analysis is not a good expression that can

express the root cause of reliability disasters (Figure 6.19.).

If the distributions for both the stress and the strength both follow a normal distribution, the expected

probability of failure, F, can be calculated as:

∞
F=P [ stress≥strength ] =∫0 f strength ( x )⋅R stress ( x ) dx
(6.17)

The expected reliability, R, is calculated as:

∞
R=P [ stress≤strength ]=∫0 f stress ( x )⋅R strength ( x ) dx
(6.18)

There are two ways to increase reliability: 1) increase the difference (or safety margin) between the

mean stress and strength values, 2) decrease the standard deviations of the distributions of stress

and strength. The estimates of stresses and strengths for all component of a product would be

perfectly accurate, but this is too costly to accomplish. And the stress conditions depend on the way

the product is used– the customer profiles and environmental conditions.

The strength distribution mainly depends on the material used in the product, its dimensions and the

manufacturing process. To improve the product reliability, the product in the design phase should

increase its strength by using the optimal design and reliability testing. One method of reliability will be

discussed with the parametric ALT testing in Chapter 8.

Environmental stresses have a distribution with a mean x and a standard deviation Sx and component

strengths have a distribution with a mean y and a standard deviation Sy. The overlap of these

distributions is the probability of failure Z. This overlap is also referred to stress-strength interference.

If stress and strength are normally distributed random variables and are independent of each other,

the standard normal distribution and Z tables can be used to quantitatively determine the probability of

failure. First, the Z-statistic is calculated as follows:

 μ x−μ y
Z =−
√ S2x +S2y
(6-19)

Using the Z value table for a standard normal distribution, the area above the calculated Z-statistic is

the probability of failure. P(Z) can be determined from a Z table or a statistical software package. For

example, if μx is 2500kPa, μy is 4500kPa, Sx is 500kPa, and Sy is 400kPa, the probability of failure can

be calculated:

μ x−μ y 2500−4500
Z =− =− =2 .34
√ S2x +S2y √500 2+400 2
(6-20)

Using the Z-value table for a standard Normal distribution, the area above a Z value of 2.34 (2.34

standard deviations) is 0.0096. Therefore, the probability of failure is 0.96%. Likewise, the reliability is

1-0.0096 = 0.9904 or 99.04%.

6.6 Failure Analysis

6.6.1 Introduction
 Fig.6.20. Typical failure mechanisms in product

Using microscopy and spectroscopy, failure analysis is to search out the root cause of failed

components in field and to improve product reliability. Failure analysis is designed to identify the

failure modes, the failure site, and the failure mechanism. It determines the root causes of the design

and recommend failure prevention methods.

The process begins with the most non-destructive techniques and then proceeds to the more

destructive techniques, allowing the gathering of unique data from each technique throughout the

process. The sequence of procedures is visual Inspection, electrical testing, non-Destructive

evaluation, and destructive evaluation.

As seen in Figure 6.20, failure mechanism of product might be classified as overstress mechanisms

and wear mechanisms. Some modes of failure mechanisms are excessive deflection, buckling, ductile

fracture, brittle fracture, impact, creep, relaxation, thermal shock, wear, corrosion, stress corrosion

cracking, and various types of fatigue. Over time, as more is understood about a failure, the root

cause evolves from a description of symptoms and outcomes. The more complex the product or
situation, the more necessary a good understanding of its failure cause is to ensuring its proper

operation (or repair).

Materials may be degraded by their environment by corrosion processes, such as rusting in the case

of iron and steel. Such processes can also be affected by load in the mechanisms of stress corrosion

cracking and environmental stress cracking.

6.6.2 Procedure of failure analysis

To improve reliability targeting of product or modules, the design of a part structure often requires the

engineer to minimize the possibility of failure. It therefore is a critical process to understand the failure

mechanics - fracture and fatigue. Reliability engineer is familiar with appropriate design principles that

can be employed to prevent the failures. By design feedback, reliability engineer can modify the

design by correcting the missing design parameters. Manufacturers also need to know “why things

fail” as much as they know “how things work.”

Failure analysis is a systematic examination of failed products to determine the root cause of failure

and to use such information to eventually improve product reliability (See Figure 6.21). Failure

analysis is designed to 1) identify the failure modes (the way the product failed), 2) identify the failure

site (where in the product failure occurred), 3) identify the failure mechanism (the physical phenomena

involved in the failure), 4) determine the root cause (the design, defect, or loads which led to failure),

and 5) recommend failure prevention methods.

It will inspect whether the load applied cyclically or was overload, the direction of the critical load, and

the influence of outside forces such as residual stresses. Then, accurately knowing the physical roots

of the failure leads to pursue the human errors or the latent causes of these physical roots. Failure

analysis might be classified as non-destructive analysis and destructive analysis.

 (a) Structure of conventional package in semi-conduct

(b) Procedure of failure analysis in electrical system (example)

Fig.6.21. Failure analysis in electrical system

The process begins with the most non-destructive techniques and then proceeds to the more

destructive techniques, allowing the gathering of unique data from each technique throughout the

process. When properly analyzed, this data leads to a viable failure mechanism. The use of

destructive techniques in the early process leads to lose the valuable information that might be

required later. The sequence of procedures is:

 Visual Inspection

 Mechanical or Electrical Testing

 Non-Destructive Evaluation

 Destructive Evaluation (using relevant techniques)

To increase the product reliability, the results of failure mechanism must be modeled by the physics-

of-failure (PoF). It allows designers to properly select materials, which minimize the susceptibility of

future designs to failure by degrading it. In addition, it allows the user to select environmental and

operational loads that minimize the susceptibility of the current design to failure during product

lifetime.

The identification of the critical failure mechanisms and failure sites of assemblies in field also permits

the development of a focused accelerated test program. The benefits of accelerated testing are that it

allows the proper test stresses (e.g., temperature, relative humidity, temperature cycling) so as to

cause wear-out failure in the shortest time without changing the failure mechanism or mode.

This is a vast improvement over the old method of choosing a random set of test loads and levels, or

subjecting the assemblies to a set of "one size fits all" standard tests prescribed by decades-old

military and commercial standards. In addition, the failure distribution in the accelerated tests can be

converted to a failure distribution in the intended use environment using the acceleration factors

calculated by the PoF models. Typical equipments of failure analysis in product might be used as

optical microscope, X-ray, SEM, SAM, FTIR and the others (See Figure 6.22).
 Fig.6.22. Typical equipments of failure analysis

Non-Destructive Evaluation (NDE) is designed to provide as much information on the failure site,

failure mechanism, and root cause of failure without causing any damage to the product or removing

valuable information. A significant amount of failure information is available through visual inspection

and the more traditional NDE methods – X ray or SAM.

For mechanical or electronic device, X-ray microscopy assesses the internal damage, defects, and

degradation in micro-electronic devices. Illuminating a sample with X-ray energy provides images

based on material density that allow characterization of solder voiding, wire-bond sweep, and wire-

bond breakage in components. Consequently, X-ray microscopy is a powerful non-destructive tool for

pinpointing failure sites in product (See Figure 6.23).

 (a) X-ray analysis of 2nd Bonding Broken and Inner wire neck broken (example)

(b) X-Ray Microscopy showing a pitting corrosion on the evaporator tube (example)

Fig.6.23. Failure analysis using X-ray micro-graphy in product

As destructive evaluation, Scanning Electron Microscopy (SEM) is a natural extension of optical

microscopy. The use of electrons instead of a light source provides much higher magnification (up to

100,000 times) and much better depth of field, unique imaging, and the opportunity to perform

elemental analysis and phase identification.

6.6.3 Case Study: PAS (Photo Angle Sensor) in Automobile

 Summary: No ignition of automobile due to the frequent failures of the PAS application IC in

field (Figure 6.24 (a))

 Electrical test (by curve tracer): Electrical open of Pin #4 (Figure 6.24 (b))

 Non-Destructive Inspection by Scanning Acoustic Microscope, X-Ray radiography

- SAM : Die paddle(top, bottom), lead-frame de-lamination (Figure 6.24 (c))

- X-ray radiography : Package crack (Figure 6.24 (d))

 Microscopy analysis (by SEM): Wedge bond open of Pin #4 (Figure 6.24 (e))

(a) Photo Angle Sensor (b) Electrical test

(c) SAM: Die paddle (top, bottom), lead-frame de-lamination

 (d) X-ray radiography : Package crack (e) Microscopy analysis (by SEM)

Fig.6.24. Failure analysis: PAS application IC in Automobile

For de-lamination in semi-conductor, when surface mount device is mounted, because the whole

package is exposed to high temperature and humidity, there are problems such as de-lamination of

resin from frame materials or absorbed moisture inside package vapor blasts, and resulting in

package deformation or pop-corning crack (Figure 6.25).

Fig.6.25. Failure Mechanism: De-lamination in semi-conductor

6.6.4 Fracture faces of product subjected to a variety of loads in fields

Fatigue failure can be recognized from the specimen fracture surface with the different growth zones
and the major physical features: 1) region of slow crack growth is usually evident in the form of a

“clamshell” concentric around the location of the initial flaw, 2) clamshell region often contains

concentric “beach marks” at which the crack may become large enough to satisfy the energy or stress

intensity criteria for rapid propagation, 3) final phase produces the granular rough surface before final

brittle fracture.

For example, the suction reed valves open and close to allow refrigerant to flow into the compressor

during the intake cycle of the piston. Due to repetitive stresses, the suction reed valves of domestic

refrigerator compressors used in the field were cracking and fracturing, leading to failure of the valve.

From SEM microscopy, the fracture started in the void of the suction reed valve and propagated to the

end (Figure 6.26).

 Fig.6.26. Fatigue fracture surface of compressor suction reed valve

The fracture face of a fatigue failure shows both the load type (bending, tension, torsion or a

combination) and the magnitude of the load. To understand the type of load, look at the direction of

crack propagation. It is always going to be perpendicular to the plane of maximum stress. The

following examples reflect the fracture paths on accordance with a variety of loads.

Fig.6.27. Fatigue failure of splined shaft due to reversed torsional loads

Figure 6.27 describes the reversed torsional fatigue failure of splined shaft from a differential drive

gear. The mating halves of the fracture reveal how two separate cracks initiated in a circumferential

recess adjacent to the end of the splines and began to propagate into the cross section following

helical paths. Because the cycles of twisting forces acted in opposite directions, each crack follows

opposing helices which progressively reduced the effective cross sectional area and, consequently,

increased the levels of cyclic stresses from the same applied loads. Shortly before the shaft finally

broke bending forces initiated a third crack at the opposite side of the shaft and this had begun to

propagate as a plane fracture at 90° to the shaft axis until the splined end finally broke away.

Torsional fatigue is involved in 10 to 25 percent of rotating equipment failures. Torsion fatigue failures
can identify them as the fracture oriented 45 degrees to the shaft centerline. The fracture face

typically has one or more origins, a fatigue zone with progression lines and an instantaneous zone. A

large fatigue zone and small instantaneous zone mean the fatigue load was small. A small fatigue

zone and large instantaneous zone mean the fatigue load was high.

Torsional fatigue fractures frequently occur in a shaft that is inside a hub or coupling. These fractures

usually start at the bottom of a keyway and progress around the shaft’s circumference. In Figure 6.27,

the fracture travels around the shaft, climbing toward the surface so the outer part of the shaft looks

like it was peeled away. The fracture surface has characteristics of a fatigue fracture: one or more

origins, ratchet marks and a fatigue zone with progression lines. The shaft fragment is usually held in

place by the coupling or hub, so there is typically a very small or no instantaneous zone.

A shaft fracture may have both torsion and bending fatigue forces. When this occurs, the orientation of

the fracture face may vary from 45 degrees to 90 degrees with respect to the shaft centerline. As the

fracture is closer to 90 degrees, the shaft combines dominant bending with torsion. The fracture angle

therefore offers key evidence as following:

 Closer to 90 degrees, it is a dominant bending force.

 Midway between 45 degrees and 90 degrees, it is a combination of torsion and bending

forces.

 Closer to 45 degrees, it is a dominant torsion force.

Evidence of torsional fatigue also may be found on gear and coupling teeth. Most equipment runs in

one direction, so wear is expected on one side of a gear or coupling teeth. Wear on both sides of a

gear or coupling teeth that rotate in one direction is an indication of varying torsional force. When

coupling alignment is good and wear occurs uniformly on both sides of all coupling teeth, it usually

indicates torsional vibration. Alignment quality can be verified from vibration spectra and phase

readings. An absence of 2X running speed spectral peaks and uniform phase across the coupling

occurs when the alignment is good.

REFERENCES

[1] Timoshenko SP (1953) History of Strength of Materials. McGraw-Hill Book Co.: New York

[2] Anderson TL Fracture (1991) Mechanics: Fundamentals and Applications. CRC Press: Boca

Raton.

[3] Griffith AA (1921) The phenomena of rupture and flow in solids. Philosophical Transactions of the

Royal Society of London, A, 221: 163–198

[4] Neugebauer GH (1943) Stress concentration factors and their effect in design. Prod. Eng.: NY, A,

14: 82–87

[5] Paris PC, Gomez MP, and Anderson WE (1961) A rational analytic theory of fatigue. The Trend in

Engineering 13: 9-14.

[6] Wöhler A (1870) Über die Festigkeitsversuche mit Eisen und Stahl, Zeitschrift für Bauwesen.

20:73-106

[7] ASME (1965) Mechanical Reliability Concepts. ASME Design Engineering Conference ASME: New

York
Chapter 7

Parametric Accelerated Life Testing in Mechanical/civil System

Seong-woo Woo

Email: [email protected]

Reliability Association of Korea, Seoul, Korea (Republic of)

Abstracts: In this chapter, as new quantitative methodology in reliability-embedded developing

process, parametric Accelerated Life Testing (ALT) will be discussed. Engineer in the design process

has final goal to find the problematic part of product and achieve the reliability target. However, it has

pending questions - the testing time and sample size. If fewer or limited samples are selected, the

statistical assessment for product reliability becomes more uncertain. If a sufficient quantity of parts

for more accurate result is tested, the cost and time will demand considerably. It therefore is

reasonable to proceed the accelerated life testing. The parametric ALT is shortly carried out until a

certain number of failed components have been reached under accelerated conditions. As a reliability

methodology, parametric ALT has to derive the sample size equation with accelerated factors. The

accelerated factors could be found in analyzing the load conditions of real dynamics system. Typical

failure mechanisms in mechanical system are fatigue and fracture. To achieve the reliability target of

product, parametric ALT should search out the missing design parameters to robustly withstand the

loads in product lifetime.

Keywords: Parametric Accelerated Life Testing (ALT), Loading conditions, Sample size equation,

Accelerated Factor (AF), Loading Conditions.

7.1 Introduction

Reliability describes the ability of a system or module to function under stated conditions for a
specified period of time [1]. Reliability is often illustrated in a diagram called “the bathtub curve” shown

as the top curve in Figure 7.1. The first part of the curve, called the “infant mortality period”,

represents the introduction of the product in the market. In this period, there is a decreasing rate of

failure. It is then followed by what is usually called the “normal” life period with a low but consistent

failure rate. It then ends with a sharp increase in failures as the product reaches the end of its useful

life. If product in the mechanical/civil systems were to exhibit the failure profile in the bathtub curve

with a large number of failures in the early life of product, it would be difficult for the system to be

successful in the marketplace. Improving the reliability of a system through systematic testing should

reduce its failure rate from the traditional failure rate typified by the bathtub curve to the failure rate

represented by a flat, straight line with the shape parameter β in Figure 7.1. With the second curve,

there are low failure rates throughout the lifetime of the system or component until reaching the end of

its useful life that the failure rate begins to increase.

Fig. 7.1. Bathtub curve and straight line with slope β toward the end of the life of the product.

The product reliability function can be quantified from the expected product lifetime LB and failure rate
 in Figure 1 as follows:

−λL B
R ( LB ) =1−F ( LB )= e ≃1- λLBX (7-1)

In a practical sense, this proportionality is applicable below about 20 percent of cumulative failure rate

[2]. Improving the design of a mechanical system to increase its reliability can be achieved by

quantifying the targeted product lifetime LB and failure rate  by finding the appropriate control

parameters affecting reliability and then modifying the design with the results from parametric

accelerated life testing.

7.2 Reliability Design in Mechanical System

As setting an overall parametric accelerated life testing plan, reliability design of the product can be

achieved by getting the targeted reliability of product - lifetime LB and failure rate  after finding the

missing control parameters and modifying the defective configuration of structures.

The product (or module) with the modified design might meet the assigned reliability target. As

product (or module) carries out test for significantly longer time, the parametric ALTs might obtain the

missing design parameters in the design phase of the mechanical system. Under consumer usage

conditions, these new reliability methodologies in the reliability-embedded design process will provide

the reliability quantitative (RQ) test specifications of a mechanical structure that conforms to the

reliability target.
 (a). Breakdown of Automobile with multi-modules

(b). Product lifetime LB and failure rate λ with multi-modules

Fig.7.2. Product lifetime LB and failure rate λs with multi-modules

As you can see in Figure 7.2, a product can consist of several different modules. For example,

automobiles consists of modules, such as the engine, transmission, drive, electrical, and body parts.

The product lifetime LB and failure rate λs with multi-modules should be determined for each module.

For example, suppose that there were no initial failures in a product, the product lifetime could be

represented by the product lifetime for module #3 in Figure 7.2. The cumulative failure rate of the

product over its lifetime would be the sum of the failure rate of each module as seen in Figure 7.2(b).
One core module #3 will seriously damage the reliability of the whole product and determine the

product lifetime. If the product lifetime was given by Y and the total failure rate was X, the yearly

failure rate can be calculated by dividing total failure rate X by product lifetime Y. The product reliability

may be given as reliability (1-X*0.01) with a yearly failure rate of X/Y and LBX Y years.

Based on failure data from the field, the parametric accelerated life testing plan of the product can be

established for a newly designed module and any modified module. Table 7.1 shows the parametric

ALT for several modules. For module D, a modified module, the yearly failure rate was 0.2 %/year and

LBx life was 6 years from the field data. Because this was a modified design, the expected failure rate

was 0.4 %/year and the expected LBx life was 3.0. To increase the targeted product life, the lifetime of

the new design was targeted to be LBx (x=1.2) 12 years with a yearly failure rate of 0.1%. The product

reliability might be determined by summing the failure rates of each module and lifetimes of each

module. The product reliability is targeted to be over a yearly failure rate of 1.1% and LBx (x=13.2) 12

years (Table 7.1).

Table 7.1 Overall parametric ALT plan of product

Reliability Market Data Expected Design Targeted Design

Yearly Yearly
No Failure Bx Life, Yearly Failure Rate, Bx Life, Failure
Modules Bx Life, yr
Rate, yr %/yr yr Rate,
%/yr %/yr
1 Module A 0.34 5.3 New x5 1.70 1.1 0.15 12 (x=1.8)
2 Module B 0.35 5.1 Given x1 0.35 5.1 0.15 12 (x=1.8)
Modified
3 Module C 0.25 4.8 x2 0.50 2.4 0.10 12 (x=1.2)
Motor
4 Module D 0.20 6.0 Modified x2 0.40 3.0 0.10 12 (x=1.2)

5 Module E 0.15 8.0 Given x1 0.15 8.0 0.1 12 (x=1.2)

6 Others 0.50 12.0 Given x1 0.50 12.0 0.5 12 (x=6.0)

Total R-Set 1.79 7.4 - - 3.60 3.7 1.10 12 (x=13.2)

In targeting the reliability of the new module where there was no field reliability data, the data for

similar modules are often used as a reference. If there has been major redesign of the module, the

failure rate in the field may be expected to be higher. Thus, the predicted failure rates will depend on
the following factors:

1. How well the new design maintains a similar structure to the prior design,

2. For each new module, new manufacturers are assumed to supply parts for the product,

3. Magnitude of the loads compared to the prior design, and

4. How much technological change and additional functions are incorporated into the new

design.

So for Module A, the expected failure rate was 1.7 %/yr and its expected lifetime was 1.1 years

because there was no field data on the reliability of the new design. The reliability of the new design

was targeted to be over LBx (x=1.8) 12 years with a yearly failure rate of 0.15%. To meet the expected

product lifetime, the parametric ALT should help identify design parameters that could affect the

product reliability.

7.3 Reliability block diagram and its connection in product

Fig.7.3 Automobile that consists of multi-modules connected serially

As seen in Figure 7.3, the reliability block diagram is a graphical method that describes how system
and main module connected in product. The configurations of complicated system such as automobile

can be generated from the series or parallel connections between modules. In a reliability block

diagram, components are symbolized by rectangular blocks, which are connected by straight lines

according to their logic relationships. Depending on the purpose of system analysis, a block may

represent a lowest-level component, module, subsystem, and system. It is treated as a block box for

which the physical details may not need to be known. The reliability of the object that a block

represents is the only input that connects system reliability evaluation.

In constructing a reliability block diagram, physical configurations in series or parallel do not indicate

the same logic relations from a standpoint of reliability. A system is said to be a series system if the

failure of one or more modules within the system results in failure of the entire system. A variety of

mechanical products are the serial system at two hierarchical levels that consists of multiple modules.

For example, an automobile engine six cylinders are in series because the engine is said to have

failed if one or more cylinders connected in parallel mechanically are failed. In the same manner

automobile is serially connected in power-train, electrical and control system, chassis, and body.

Suppose that a mechanical series system like automobile consists of n mutually independent

modules. Mutual independence implies that the failure of one module does not affect the life of other

modules. By definition, successful operation of a system requires all components to be functional.

From probability theory, the system reliability is

R =Pr ( E )=Pr ( E1⋅E 2⋅¿⋅En ) (7-2)

where Ei is the event that module i is operational, E the event that the system is operational, and R

the system reliability.

Because of the independence assumption, this becomes

 n
R =Pr ( E )=Pr ( E1 )⋅Pr ( E2 )⋅¿⋅Pr ( En )=∏ Ri
i=1 (7-3)

where Ri the reliability of module i.

Let’s consider a simple case where the times to failure of n modules in a system are modeled with the

exponential distribution. The exponents reliability function for module i is The exponential reliability

function for component i is Ri ( t ) =exp (−λt ), where λ i is the failure rate of component i. Then from (7-

3), the system reliability can be written as

( )
R ( t )=exp −t ∑ λ i =exp (−λt )
i =1 (7-4)

n
λ=∑ λi
where λ is the failure rate of the system and i=1

7.4 Reliability Allocation of Product

7.4.1 Introduction

If the system reliability target is setting in the product planning, it will sequentially be allocated to

individual subsystem, module, and components at the stage of the product design. When each

module achieves the allocated reliability, the overall system reliability target can be attained. Reliability

allocation is an important step in the new reliability testing design process. The benefits of reliability

allocation can be summarized as follows.

 Reliability allocation defines a reliability target for each module. The product has a number of

module or subsystem, which are manufactured by suppliers or internal departments. It is

 important that company share all related parties with the reliability target before delivering

the end product to customer.

 Quantitative reliability targets for modules encourage responsible parties to improve current

reliability through use of reliability techniques.

 Mandatory reliability requirements are closely connected with engineering activities aimed at

meeting other customer expectations in the product design process.

 Reliability allocation drives a deep understanding of product hierarchical structure. The

process may leads to identify the part of design weakness and subsequently improve it.

 As a result, reliability allocation can work on input of other reliability tasks. For example,

reliability assigned to a module will be used to design reliability verification.

Reliability allocation is fundamentally a repetitive process. It is conducted in the early design stage to

support concept design when available information is restricted. As the design process proceeds, the

overall reliability target might be reallocated to reduce the cost of achieving the reliability goal. The

allocation process may be invoked by the failure of one or more modules to attain the assigned

reliability due to technological limitations. The process is also repeated whenever a major design

change takes place.

7.4.2 Reliability allocation of the product

Because some parts are assigned to extremely high reliability goals, it may be unachievable at all. On

the other hands, though there are critical components whose failure causes safety, environmental or

legal consequences, it will be allocated to low-reliability targets. It is important to establish some

criteria that should be considered in reliability allocation.

The task of reliability allocation is to select part reliability targets, R¿1 , R ¿2 , … .. , R ¿n which satisfy the

following equality equation:

 R¿S ≤g ( R ¿1 , R ¿2 R2 ,……, R¿n )
(7-5)

Mathematically, there are an infinite number of such sets. Clearly, these sets are not equally good,

and even some of them are unfeasible. The common criteria are described:

1. Failure possibility. Parts that have a high likelihood of failure previously should be given a low-

reliability target because of the intensive effort required to improve the reliability. Conversely, for

reliable parts, it is reasonable to assign a high-reliability goal.

2. Complexity. The number of constituent parts (or modules) within a subsystem reflects the

complexity of the subsystem. A higher complexity leads to a lower reliability. It is similar to the purpose

of failure possibility.

3. Criticality. The failure of some parts may cause severe effects, including, for example, loss of life

and permanent environmental damage. The situation will be severe when such parts have a high

likelihood of failure. Apparently, criticality is a product of severity and failure probability, as defined in

the FMEA technique described in Chapter 4. If a design cannot eliminate severe failure modes, the

parts should have the lowest likelihood of failure. Consequently, high-reliability goals should be

assigned to them.

4. Cost. Cost is an essential criterion that is a target subject to minimization in the commercial

industry. The cost effects for achieving reliability depend on parts. Some parts induce a high cost to

improve reliability a little because of the difficulty in design, verification, and production. So it may be

beneficial to allocate a higher-reliability target to the parts that have less cost effect to enhance

reliability.

Though several methods for reliability allocation have been developed, the simplest method here is

the equal allocation method. This method can only be applied when the system reliability configuration

is in series. The system reliability is calculated by:

 n
¿
R S = ∏ Ri
i=1 (7-6)

The allocated reliability for each subsystem is:

¿ 1/k
R i= ( R S ) (7-7)

7.4.3 Product breakdown

Typical modern products involved in mechanical system can be outlined as automobile, airplane,

domestic appliance, machine tools, agricultural machinery, and heavy construction equipment. They

can break down several modules to the individual parts. Based on the market data, the reliability

target could be assigned to the product modules like Table 7.1. The targeted reliability of each module

can be quantified as the expected product lifetime LB and failure rate  in Eq. (7-1). The reliability

Testing will be centered on the module of product. For example, if the targeted reliability of refrigerator

is allocated as B20 life 5 year, the reliability for engine will be B4 life 5 year.

Table 7.2. Reliability Target for Mechanical System (Part Count Method)

Level Quantity Target Remark

System 1 System B20 life 10 year Refrigerator

Module 5 Units B4 life 10 year Compressor

Component 500 Components B0.04 life 10 year --

And it is reasonable to carry out the reliability testing per module because test cost for system or
component is higher than that of module.

7.4.3.1 Automobile

Figure 7.4 shows the hierarchical configuration of an automobile connected serially from system to

main modules. It consists of engine, body and main parts, electrical and electronics, interior, power-

train and chassis, miscellaneous auto parts - air conditioning system (A/C), bearings, hose and other

miscellaneous parts. An automobile is a four-wheeled, self-powered motor designed to run on roads

for transport of one to eight people. Each subsystem in automobile is broken down further into

multiple lower-level subsystems. From a reliability perspective, the automobile is a series system

which fails if one or more subsystems (or module) in automobile break. The blocks of the automobile

in the reliability block diagram represent the first-level subsystems, the second-level modules, and the

others which their reliabilities are known. The reliability block diagram of a typical automobile contains

over 20,000 blocks which is including the parts.

Fig.7.4 Breakdown of automobile with multi-modules

However, reliability design of automobile might focus on the modules. They can easily calculate the

module reliability because of the connection of serial system.

7.4.3.2 Airplane

The uses for airplanes are constructed with the objectives of recreation, transportation of goods and

people, and military. The design and planning process, including safety tests, can last up to four

years. The design specifications of the aircraft during the design process often is established. When

the design has passed through these processes, the company constructs a limited number of

prototypes for testing on the ground.

Fig.7.5. Breakdown of airplane with multi-modules

Figure 7.5 shows the hierarchical configuration of a passenger airplane that consists of airframe parts,

wings, fuselage, propulsion (engine), aviation controls and instruments, air conditioning system (A/C),

bearings, hose and other miscellaneous airplane parts. From a reliability perspective, the airplane is a

series system which fails if one or more subsystems (or module) in airplane break. The blocks of the
airplane in the reliability block diagram represent the first-level subsystems, the second-level modules,

which their reliabilities are known. The reliability block diagram of a typical airplane contains over

1,000,000 blocks which is including the parts. Reliability design of airplane will focus on the modules

that are serially connected like other mechanical system in Figure 7.5.

7.4.3.3 Domestic Appliance

Domestic appliance is a large machine used for routine housekeeping tasks such as cooking, washing

laundry, or food preservation. Examples include refrigerator, air conditioner, washing machine, and

cleaner. Major appliances that use electricity or fuel are bigger and not portable. They are often

supplied to tenants as part of otherwise unfurnished rental properties. Major appliances may have

special electrical connections, connections to gas supplies, or special plumbing and ventilation

arrangements that may be permanently connected to the appliance. This limits where they can be

placed in a home (Figure 7.6).

 Fig.7.6 Breakdown of a refrigeration system with multi-modules

The hierarchical configuration of an appliance consists of cabinet, door, internal fixture (selves,

draws), controls and instruments, generating parts (motor or compressor), heat exchanger, water

supply device, and other miscellaneous parts. The reliability block diagram of a typical appliance

contains over 1,000 blocks which is including the parts. Reliability design of domestic appliance will

focus on the modules. They can easily calculate the module reliability because of the connection of

serial system.

7.4.3.4 Machine tools

Fig.7.7 Breakdown of Machine tools with multi-modules

A machine tool is a machine for shaping or machining metal or other rigid materials, usually by cutting,

boring, grinding, shearing, or other forms of deformation. Machine tools employ some sort of tool that

does the cutting or shaping. All machine tools have some means of constraining the work piece and
provide a guided movement of the parts of the machine. Thus the relative movement between the

work piece and the cutting tool (which is called the tool path) is controlled or constrained by the

machine to at least some extent, rather than being entirely "offhand" or "freehand" (Figure 7.7).

The hierarchical configuration of machine tools consists of automatic tool or pallet changing device,

spindle unit, drive unit, hydro-power unit, tilting index table, turret head, cooler unit, CNC controller

and other miscellaneous parts. The reliability block diagram of typical machine tools contains over

1,000 blocks which is including the parts. Reliability design of machine tools will focus on the

modules. They can easily calculate the module reliability because of the connection of serial system.

7.4.3.5 Agricultural machinery and heavy construction equipment

Fig.7.8. Breakdown of tractor with multi-modules

Agricultural machinery such as tractor is used in the operation of an agricultural area or farm. The

hierarchical configuration of agricultural machinery such as automobile consists of engine device,

power supply unit, hydraulic unit, electric device, linkage, PTO driving unit, and other miscellaneous
parts (See Figure 7.8). The reliability block diagram of a typical appliance contains over 4,000 blocks

which is including the parts.

The hierarchical configuration of a construction machine such as excavator consists of engine device,

electric device, track system, upper appearance system, driving system, main control valve unit,

hydraulic operation machine system, cooling system, and other miscellaneous parts. The reliability

block diagram of a typical appliance contains over 5,000 blocks which is including the parts (See

Figure 7.9).

Fig.7.9 Breakdown of excavator with multi-modules

Heavy equipment refers to heavy-duty vehicles, specially designed for carrying out construction tasks,

most frequently ones operating earthwork. They are also known as heavy machines, heavy trucks,

construction equipment, heavy vehicles, or heavy hydraulics. They usually comprise five equipment

systems: engine, traction, structure, power train, control and information. Some equipment frequently

uses hydraulic drives as a primary source of motion. Reliability design of agricultural machinery and

heavy construction equipment will focus on the modules in Figure 7.9. They can easily calculate the

module reliability because of the connection of serial system.

7.5 Failure Mechanics, Design and Reliability Testing
The failure mechanics of product can be described as two factors: 1) the stress on the structure, 2)

the structure materials. The failure mechanisms can be characterized by either loads (or stress)

related failure or structural (or materials) related.

Fig.7.10. Failure mechanics and mechanical system design

If there is void in the structure, the structured will facture like Figure 7.10. On the other hands, if

structure has enough safety margins for load, the structure will deteriorate little by little and facture

near product lifetime. In field two cases - stress and material happen complexly. The typical failure

mechanisms of mechanical system are fracture and fatigue. From bathtub curve, this region would be

described as the constant failure rate that often receives repetitive random stress. As the repeated

load is applied at the stress raisers such as shoulder fillet, the structure that damage is accumulated

will crack. After repetitive stresses, the system will fracture suddenly.

Failure of mechanical/civil systems can happen when the system structures yield at the strength of

materials by the applied loads. The load could be higher than the system was designed for. On the

other hands, the material could be insufficient to handle repetitive loads to which it is subjected.

Consequently, failure occurs when the stress is greater than the material strength, or when the
material cannot withstand the loads. The product engineer would want to move the void in the

structure to a location away from where the stress is applied. This is a design concept.

A product engineer should seek to redesign the structure to either: 1) move the loads, or 2) change

the material type and design shape to withstand the load. The failure site of the product structure

could be found when the failed products are taken apart in the field or after the failed samples of a

parametric ALT. The engineer should identify the failure by experiment using the reliability testing (or

parametric ALT) before launching new product. This failure phenomenon, design, and reliability testing

might be applicable to both mechanical and electronic products because the electric products are

typically housed in mechanical systems. So it is a critical process to search out void such as the

stress raisers that has the missing design parameters by FEA or experimentally using the reliability

testing (or parametric ALT).

With the advent of Finite Element Analysis (FEA) tools, design failure such fatigue can now be

assessed in a virtual environment. Though FEA fatigue assessments do not completely replace

fatigue testing, they will find the detailed or optimal design in the structure of new product. However,

as the system goes from preliminary design to an optimized design, product might have defects. If

modules have a problem due to an improper design, the module will determine the lifetime of the

product (Figure 11 on the left side).

Fig. 7.11. Optimal design and reliability assessment

Even if a detailed product design is optimized using tools such as FEA, it may have still design flaws

that don’t become evident until the system is in the field. In the field, the system may be subjected to

loads that could be very different than what was envisioned in the original design. Field data can be

used by both the design engineer and test engineer to develop appropriate parameters for

accelerated life tests that help validate the expected reliability of the design. Figure 7.11 shows how

for any system, design engineering needs to be effectively connected with test engineering to achieve

the reliability target of a system, subsystem, or module.

As described in the previous section 7.4, the mechanical product such as appliance, car, and aircraft

consists of a multiple of modules. These modules can be put together in a system and have an input

and output similar to what is shown in Figure 7.12. They also have their own (intended) functions just

as a vapor compressor cycle that generates the cold air.

Fig.7.12. Typical robust design schematic (example: refrigerator)

In the field, if a mechanical/civil module functions improperly, consumers would request the module

replaced. Reliability engineers often do not have a clear understanding of the way that a consumer
used the product or the usage patterns that could have contributed to the failure of the product. If the

field usage conditions are fully understood, they could be reproduced in the laboratory testing to be

identical to those of the failure in field. In the design of the product, it is important to understand

potential usage patterns and take corrective actions before launching a product.

Under a variety of environmental and operational conditions, reliability engineers search for potential

failure modes by testing in the laboratory. They will determine the failure mechanisms from the failed

samples. They can create an action plan to reduce the chance of failure. However, it may not be easy

to identify all failure modes attributable to the improper design because, in mechanical/civil systems,

the failure modes come from repetitive stresses which may not be captured in initial testing.

Consequently, modules with specific functions need to be robustly designed to withstand a variety of

loads. In determining product lifetime, the robust design in module determines the control factor (or

design parameters) to endure the noise factor (or stress) and properly work the system, which has the

reliability target - part failure rate  and lifetime LB. Such reliability targeting is known to be

conventionally achieved through the Taguchi methods (SDE) and the statistical design of experiment

[3-8].

Taguchi methods, known to robust designs, use the loss function which quantifies the amount of loss

based on deviation from the target performance. It puts a design factor in an optimal location where

using cost function random “noise” factors are less likely to hurt the design and it helps determine the

best control factors (or design parameters). However, for an uncomplicated mechanical/civil structure,

such as a beam, Taguchi methods should take into account a considerable number of design

parameters. In the design process it is not possible to consider the whole range of the physical,

chemical and the mathematical conditions that could affect the design.

Another experimental methodology, new parametric ALT methods for reliability quantitative test

specifications (RQ), should be introduced so that the product can withstand a variety of repetitive

loads and determine the critical design parameters affecting reliability. Parametric ALT can also be

used to predict product reliability - lifetime, LB and failure rate, . The new parametric ALT discussed in

the next section has a sample size formulation that enables an engineer to determine the design

parameters and achieve the targeted product reliability - lifetime LB and failure rate  [9-21].
7.6 Parametric Accelerated Life Testing

Parametric accelerated life testing uses the sample size equation with Acceleration Factor (AF). It also

is a process that helps designers to find the optimal (or missing) design parameters. If the reliability

target of module in product is allocated, module should accomplish Reliability Quantitative (RQ) test

specifications by obtaining the sample size equation. It can help them better estimate expected

lifetime LB, failure rate of module , and finally determine whether the overall product reliability is

achieved.

The shape parameter  in the sample size equation is calculated from a Weibull distribution chart

tested. From the sample size equation, the durability target h* is determined by the targeted lifetime

LB, acceleration factor AF and the actual testing time ha. So it is important to derive the sample size

equation with the whole parameters – lifetime LB, acceleration factor AF, the actual testing time ha,

and the allowed number of failures.

Under the expected physical and chemical conditions that the product is expected to experience, it is

essential to derive the acceleration factor from a life-stress model, and determine the dominant failure

mechanism for the product. A grasp of physical of failure (PoF) also is required to understand the

failure mechanism. For example, fatigue or fracture due to repetitive stresses is the common

mechanism for failure in mechanical/civil system.

Reliability engineers must also determine how the stresses (or loads) act on the system structure,

which help categorize the potential failure mechanisms under the range in environmental and

operational conditions. Engineers need to develop a testing plan with appropriate accelerated load

conditions to determine the dominant failure mechanisms affecting product lifetime. At the same time,

they also must include other failure mechanisms, such as overstress and wearout stress. The failure

mechanisms in the accelerated life testing should be identical to that under normal conditions

experienced in the field. In the Weilbull distribution, the shape parameter for accelerated conditions

should match those under normal field conditions.

Developing a parametric ALT for reliability quantitative test specifications involves three key steps:
 1). By creating a life-stress model, determine the acceleration factor under severe conditions,

2). Assuming an initial shape parameter that is implied by the intensity distribution of wear failure

in the Weilbull distribution, derive the necessary sample size to carry out the lifetime (or

reliability) target and calculate the testing periods (or Reliability Quantitative (RQ) test

specifications) equivalent to the allocated reliability target.

3). With sample size equation, carry out testing for extended test periods to help determine the

actual shape parameter in the Weibull distribution.

Degradation by loads is a fundamental phenomenon to all products that this is described as entropy of

isolated systems will tend to increase with time – the Second Law of Thermodynamics. For instance,

the critical parameters such as strength will degrade with time. In order to understand the useful

lifetime of the part, it is important to be able to model how critically important product parameters

degrade with time.

7.6.1 Acceleration Factor (AF)

Reliability concerns arise when some critically important mechanical/civil strength due to stress

degrades with time. Let S represent a critically important part parameter like strength and let us

assume that S change monotonically and relatively slowly over the lifetime of the part. A Taylor

expansion about t = 0 produces the Maclaurin Series like Equation 7.8:

( ∂∂tS )
S ( t )=S t =0 +
t =0
t+
1 ∂S
( )
2 ∂t t=0
t 2 +…
(7-

8)

It will be assumed that the higher order terms in the expansion can be approximately by simply

introducing a power-law exponent m and writing the above expansion in a shortened form:
 S=S 0 [ 1± A 0 ( t )m ]
(7-9)

where A0 is a part-dependent constant

The power-law model is one of the most widely used forms for time-dependent degradation. For

convenience of illustration, let us assume that the critical parameter S is decreasing with time and A0

= 1. Eq. (7-9) reduces to:

S
1− = (t )m
S0 (7-10)

For m = 1, one will expect the linear degradation. On the other hands, for m > 1, the degradation will

increase strongly with time and eventually lead to a catastrophic condition.

In reliability engineering, the development of the acceleration factor is fundamental importance to the

theory of accelerated life testing. The acceleration factor speeds up the degradation of product and

permits one to take time-to-failure data very rapidly under accelerated stress conditions. And it can

extrapolate the accelerated time-to-failure results into the future for a given set of operational

conditions. The acceleration factor must be modeled using the time-to failure (TF) models. The

acceleration factor is defined as the ratio of the expected time-to-failure under normal operating

conditions to the time-to-failure under some set of accelerated stress conditions.

( TF ) operation
AF=
(TF )stress (7-11)
Since the TF under normal operation may take many years to occur, experimental determination of

the acceleration factor is impractical. However, if one has proper time-to-failure models, one can

develop the acceleration factor from the TF models.

For solid-state diffusion of impurities in silicon, the junction equation J might be expressed as:

q 1
J =[ aC ( x−a ) ]⋅exp −
[ ( kT
w− aξ ⋅v
2 )]
[Density / Area]·[ Jump Probability ]·[ Jump Frequency ]

qaξ ∂C qa ξ
=−[ a2 ve−qw/kT ]⋅cosh + [ 2ave−qw /kT ] C sinh
2kT ∂ x 2 kT

Q
( )
=Φ ( x , t , T ) sinh ( aξ ) exp −
kT

= A sinh ( aξ ) exp −( kTQ ) (7-12)

where Q is energy, a is coefficients, and A is constants

Reaction process that is dependent to speed might be expressed as:

ΔE−aS ΔE+aS
+ kT −− kT kT − kT
K=K −K =a e −a e
h h

ΔE
kT − aS
=a e kT sinh( )
h kT
(7-13)

where k is Boltzmann’s Constant, T is absolute temperature, S is stress, and A constants

So the reaction rate K can be summarized as:

Ea
K=B sinh(aS ) exp − ( )kT (7-14)

where Ea is the Activation Energy, and B constants

If the reaction rate in Eq. (7-14) and the junction Eq. (7-12) take an inverse number, the generalized

stress model can be obtained like McPherson’s derivation [22],

−1
TF=A [ sinh( aS ) ] exp ( kTE )
a

(7-15)

Because this life-stress model equation was derived from a model of micro-depletion (void) in the

failure domain, it should be relevant to general failure prediction regardless of whether it is a

mechanical, civil or electronic system. Thus, the fatigue in a mechanical/civil system, coil degradation

in a motor, bond-pad corrosion in an IC, etc., can be captured by Eq. (7-15).

 Fig.7.13. Properties of the hyperbolic sine stress term [ sinh(aS)]

The range of the hyperbolic sine stress term [ sinh(aS)] in Eq. (7-12) is increasing the stress as

( S )n
n
following: 1) initially (S ) in low effect, 2) in medium effect, and 3) ( e aS ) in high effect

initially linearly increasing. Accelerated testing usually happens in the medium stress range (See

Figure 7.13).

Thus, time to failure in the level of medium stress can then be described as

Ea
−n
TF=A ( S ) exp ( ) kT (7-16)

The internal (or external) stress in a product is difficult to quantify and use in accelerated testing. It is

necessary to modify Eq. (7-16) into a more applicable form. The power (or energy flow) in a physical
system can generally be expressed as efforts and flows (Table 7.3). Thus, stresses in mechanical/civil

or electrical systems may come from the efforts (or loads) like force, torque, pressure, or voltage [23].

Table 7.3 Energy flow in the multi-port physical system

Modules Effort, e(t) Flow, f(t)

Mechanical translation Force, F(t) Velocity, V(t)

Mechanical rotation Torque, (t) Angular velocity, (t)

Compressor, Pump Pressure difference, P(t) Volume flow rate, Q(t)

Electric Voltage, V(t) Current, i(t)

For a mechanical/civil system, when replacing stress with effort, the time-to-failure can be modified as

Ea Ea
TF=A ( S )
−n
exp ( )
kT
−λ
=A ( e ) exp ( )
kT

(7-17)

where  = power index or damage coefficient

Because the material strength degrades slowly, it may require long times to test a module until failure

occurs. The main hurdles to finding wear induced failures and overstressed failures are the testing

time and cost. To solve these issues, the reliability engineer often prefers testing under severe

conditions. Due to overstress failures of the module can be easily found with parametric ALT.

The more the accelerated conditions, the shorter the testing time will be. This concept is critical to

performing accelerated life tests, but the range of the accelerated life tests will be determined by

whether the conditions in the accelerated tests are the same to that in normally found in the field.
 Fig.7.14. Strain-Stress Curve in mild steel

The stress-strain curve is a way to visualize behavior of material when it is subjected to load (See

Figure 7.14). A result of stresses in the vertical axis has the corresponding strains along the horizontal

axis. Mild steel subjected to loads passes specification limits (proportional limit), operating limits

(elastic limit), yield point, ultimate stress point into fracture (destruct limit). In accelerated testing, the

appropriate accelerated stress levels (S1 or e1) will typically fall outside the specification limits but

inside the operating limits.

In accelerated life tests, when a module has been tested for a number of hours under the accelerated

stressed condition, one wants to know the equivalent operation time at the normal stress condition.

The equivalent operation time is obtained from the multiplication of the inverse of acceleration factor

and normal (or actual) operation time.

From the time-to-failure in Eq. (7-17), the acceleration factor can be defined as the ratio between the

proper accelerated stress levels and normal stress levels. The acceleration factor (AF) also can be

modified to include the effort concepts:

 n λ
S Ea 1 1 e Ea 1 1
AF= 1
S0 ( )[ ( −
k T 0 T1
= 1
e0 )] ( ) [ ( −
k T0 T 1 )] (7-18)

where n is stress dependence,  is cumulative damage exponent.

It is very important to note that the acceleration factor is very special, in that the acceleration factor is

the independent coefficient A. This means that even though the time-to-failure TF must be expressed

as a distribution of time-to-failure, the acceleration factor is unique. AF depends on the kinetic value

(λ, Ea) and not on part-to-part variation.

The first term is the outside effort (or load) and the second is the internal energy in Eq. (7-18). Under

severe conditions, the outside higher load drops the energy barrier and the accelerated (or high)

temperature activates the material elements. In the end, the material degrades and fails. The equation

has two parameters which are temperature and effort. Using a three-level test under accelerated

conditions, these parameters can be obtained. And the quantified value, activation energy (eV), is

called the reaction rate due to temperature rises.

Under severe conditions, the duty effect with repetitive stress (or load) involves the on/off cycles,

which shortens module lifetime [24]. The equation needed to determine the sample size for the

parametric ALT is discussed in the next section.

7.6.2 Derivation of General Sample Size Equation

Due to the cost and time limit, it is difficult to test large samples for reliability testing of product. If the

fewer components are tested, the greater the confidence interval is, the results of a statistical analysis

will become more uncertain. For a more precise result it is necessary that enough samples are tested.

This, however, can increase the time and cost (or effort) involved in testing. Thus, it is important to

develop the sample size equation with acceleration factor in Eq. (7-20), which allow the core testing

methods for securing reliability information as inexpensively as possible in the reliability-embedded

product developing process.

For the lifetime assessment, the confidence levels are necessary because it is not possible to gather

the lifetimes of several sample size. In statistics, the failure behavior of the limited sample may

strongly deviate from the actual failure behavior of the population itself. The core concept in statistics

offers a further help through the confidence levels, which can specify the confidence of the test results

and estimate the failure behavior of the population.

In statistical test planning the first step involves determining how the sample size should be extracted

of the inspection lots or population. The test samples are chosen randomly for a representative test

sample. The sample size is connected with the confidence levels and the statistical range of the

measured failure values. Another important point establishes a suitable test strategy – complete tests,

incomplete (censored) test, and accelerated testing for shortening times.

The best statistical case is a complete test that all test samples of the population are subjected to a

lifetime test. This means that the test is run until the last element has failed. Thus, failure times for all

elements are available for further assessment. However, people should remember why the lifetime

testing in company is completed. That is, for new product, the missing design parameters is found

before market launching,

In order to reduce the time and effort involved in a lifetime testing, it is reasonable to carry out

censored tests or the accelerated testing. The tests are carried out until a certain predetermined

lifetime or until a certain number of failed components have been reached with accelerated condition.

If fewer or limited parts are censored, the statistical assessment becomes more uncertain. If more

accurate result is required, a sufficient quantity of parts is tested. In this case the cost and time will be

demanded. Thus, to save the testing time, parametric accelerated life testing in mechanical/civil

system has to be developed.

From various developed methods to determine sample size, the Weibayes analysis is well known and

widely accepted method. However, its mathematical complexity makes it difficult to apply it directly to

determined sample size. Failures (r ≥ 1) need to be distinguished from no failure (r=0) cases. Hence,

it is necessary to develop a simplified sample size equation from the Weibayes analysis.

The Cumulative Distribution Function (CDF) in Weilbull can be expressed as:

 β

F ( t )=1−e
− ( ηt )
(7-19)

The Weibull reliability function, R(t), is expressed as:

R ( t ) =e
− ( tη )
(7-20)

In statistics, maximum-likelihood estimation (MLE) is a method of estimating the parameters of a

statistical model - some unknown mean and variance that are given to a data set. Maximum likelihood

selects the set of values of the model parameters that maximizes the likelihood function. The

characteristic life MLE from the Maximum Likelihood Estimation can be derived as:

n
t iβ
η βMLE =∑
i=1 r (7-21)

If the confidence level is 100(1 - ) and the number of failure is r  1, the characteristic life, , would

be estimated from Eq. (7-20),

n
2r 2
ηαβ = 2
⋅ηβMLE = 2 ⋅∑ t βi
χ α ( 2 r +2 ) χ α ( 2r +2 ) i =1 for r  1 (7-22)

Presuming there is no failures, p-value is  and In(1/) is mathematically equivalent to Chi-Squared

 ❑❑2 ( 2 )
value, . The characteristic life , would be represented as:
2

n n
2 1
ηαβ == 2
⋅∑ t iβ = ⋅∑ t iβ
χ α ( 2 ) i=1 1
ln i=1
α , for r = 0 (7-23)

Thus, Eq. (7-22) is established for all cases r  0 and can be redefined as follows:

n
2
ηαβ = 2
⋅∑ t βi
χ α ( 2 r +2 ) i =1 for r  0

(7-24)

To evaluate the Weibull reliability function in Eq. (7-24), the characteristic life can be converted into LB

life as follows:

L BX β

R ( t ) =e
−( ) =1−x
η
(7-25)

After logarithmic transformation, Eq. (7-25) can be expressed as:

1
(
LβBX = ln
1−x
⋅η β ) (7-26)

If the estimated characteristic life of p-value ,  , in Eq. (7-24), is substituted into Eq. (7-26), we
obtain the BX life equation:

n
2 1
LβBX = 2
χ α ( 2r + 2 )
⋅ ln( ⋅∑ t βi
1−x i =1 ) (7-27)

If the sample size is large enough, the planned testing time will proceed as:

n
∑ t iβ≃n⋅h β
i =1 (7-28)

The estimated lifetime (LBX) in test should be longer than the targeted lifetime (L*BX)

2 1
LβBX ≃
χ α2 ( 2r + 2 ) (
⋅ ln
1−x )
⋅n⋅ h β ≥L¿BXβ
(7-29)

Then, sample size equation is expressed as follows:

χ α2 ( 2 r +2 ) β
¿
1 LBX
n≥
2
⋅
1
⋅
h ( )
( ln
1−x ) (7-30)

However, most lifetime testing has insufficient samples. The allowed number of failures would not

have as much as that of the sample size.

 n r
∑ t iβ =
∑ t βi +( n−r ) h β≥( n−r ) hβ
i =1 i=1 (7-31)

If Eq. (7-31) is substituted into Eq. (7-27), BX life equation can be modified as follows:

2 1
LβBX ≥
χ α2 ( 2r + 2 ) (
⋅ ln
1−x )
⋅( n−r ) h β ≥L¿BXβ
(7-32)

Then, sample size equation with the number of failure can also be modified as:

χ α2 ( 2 r +2 ) 1 L¿BX β
n≥
2
⋅
1
⋅
h ( )+r
( ln
1−x )
(7-33)

From the generalized sample size Eq. (7-33), we can proceed reliability testing (or parametric ALT

testing) under any failure conditions (r  0). Consequently it also confirms whether the failure

mechanism and the test method are proper.

7.6.3 Derivation of Approximate Sample Size Equation

χ 2α ( 2 r +2 )
As seen in Table 7.4, for a 60% confidence level, the first term in Eq. (7-33) can be
2
approximated to (r + 1) [25]. And if the cumulative failure rate, x, is below about 20 percent, the
 1
denominator of the second term ln approximates to x by Taylor expansion.
1−X

χ 20.4 ( 2 r+ 2 )
Table 7.4. Characteristics of at =60% confidence level
2

χ 20.4 ( 2 r+ 2 ) χ 2α ( 2 r +2 )
r 1−¿ ≈ r +1 1−¿
2 2
0 0.4 0.92 1 0.63

1 0.4 2.02 2 0.59

2 0.4 3.11 3 0.58

3 0.4 4.18 4 0.57

Then the general sample size Eq. (7-33) can be approximated as follows:

¿ β
1 LBX
n≥( r+1 )⋅ ⋅
x h ( ) +r
(7-34)

If the acceleration factors in Eq. (7-19) are added into the planned testing time h, Eq. (7-35) will be

modified as:

¿β
1 LBX
n≥( r +1 )⋅ ⋅
(
x AF⋅h a
+r
) (7-35)

The normal operating cycles of the product in its lifetime are calculated under the expected customer
usage conditions. If failed number, targeted lifetime, accelerated factor, and cumulative failure rate are

determined, the required actual testing cycles under the accelerated conditions can be obtained from

Eq. (7-35). ALT equipment in mechanical/civil system will be designed based on the load analysis and

the operating mechanism of the product. Using parametric ALT with approximated sample size of an

acceleration factor, the failed samples in the design phase can be found. From the required cycles (or

Reliability Quantitative (RQ) test specifications), ha, it determines whether the reliability target is

achieved. For example, without considering the acceleration factor, the calculation results of two

sample size equations are presented in Table 7.5.

Table 7.5. The calculated sample size with h = 1,080 hour testing time
Sample Size
β Failure number Eq. (7-33)
Eq. (7-34)
by Minitab

2 0 3 3

2 1 7 7

3 0 1 1

3 1 3 3

If the estimated failure rate from the reliability testing is not bigger than the targeted failure rate ( *),

the number of sample size (n) might also be obtained. The estimated failure rate with a common

sense level of confidence () can be described as:

¿ r +1
λ ≥λ≃
n⋅( AF⋅ha )
(7-36)

By solving Eq. (7-37), we can also obtain the sample size

 1 1
n≥( r +1 )⋅ ¿⋅
λ AF⋅ha
(7-37)

Multiplying the targeted lifetime (L*Bx) into the numerator and denominator of Eq. (7-38), we can yield

another sample size equation.

¿ ¿ 1
1 L Bx 1 L Bx
n≥( r +1 )⋅ ¿ ⋅
λ ⋅L Bx AF⋅ha
¿
=( r +1 )⋅ ⋅
x AF⋅ha ( ) (7-38)

Here, we know that * L*Bx is transformed into the cumulative failure rate x.

We can see two equations for sample size that have a similar form – Eq. (7-35) and (7-38). It is

interesting that the exponent of the third term for two equations is 1 or , which is greater than 1 for

wear-out failure. Because the sample size equation for the failure rate is included and the allowed

failed numbers r is 0, the sample size equation Eq. (7-35) for the lifetime might be a generalized

equation to achieve the reliability target.

If the testing time of an item (h) is more than the targeted lifetime (L*Bx), the reduction factor R is close

to 1. The generalized equation for sample size in Eq. (7-35) might be rewritten as follows:

1
n≥( r +1 )⋅
x (7-39)

And if the targeted reliability for module is allocated to B1 10 years, the targeted lifetime (L*BX) is easily

obtained from the calculation by hand. For refrigerator, the number of operating cycles for one day

was 5; the worst case was 9. So the targeted lifetime for ten years might be 32,850 cycles.
And the other type of sample size equation that is derived by Wasserman [26] can be expressed as:

2 2 2 2 β
χ α ( 2 r+ 2 ) χ α ( 2 r +2 ) χ α ( 2r +2 ) χ α ( 2 r +2 ) 1 LBX
n=−
2 m β ln R L
=
2 m β ln R L −1
=
2 m β ln ( 1−F−1
L )
=
2
⋅
ln (1−F−1 )
L
⋅ ( )
h
(7-

40)

where m ≅ h/ LBX , n ≫ r

When r = 0, sample size equation can be obtained as:

2 β
ln ( 1−C ) −ln ( 1−C ) ln ( 1−C )−1 ln α −1 χ α (2) 1 L BX
n= = = =
m β ln R L − m β ln R L m β ln R −1 m β ln R
L L
−1
=
2
⋅
ln ( 1− F−1 )
L
( )
⋅
h
(7-

41)

So Wasserman’s sample size equation Eq. (7-41) is similar to Eq. (7-35).

Especially, the ratio between product lifetime versus the testing time in Eq. (7-34) can be defined as

reduction factor. It can be used to determine if accelerated life testing is proper. That is,

¿ β ¿ β
LBX LBX
R= ( ) (
h
=
AF⋅h a ) (7-42)

To effectively proceed the parametric accelerated life testing, we have to find the severe conditions

that will increase the accelerated factor (AF) and the shape factor β. At that time the location and
shape of the failed product in both market and parameter ALT results are similar. If the actual testing

time ha is longer than the testing time that is specified in the reliability target, the reduction faction will

be less than one. So we can obtain the accelerated conditions that can decrease the testing time and

sample size number.

7.7 The reliability design of mechanical system and its verification

7.7.1 Introduction

Completing the design of a new product requires two kinds of activities — managerial and technical

skills. Managerial skill includes adopting a process improvement approach, such as Capability

Maturity Model Integration (CMMI), and controlling quality, which Japan has pursued for over sixty

years (starting with training by Dr. W. Edwards Deming at the Japanese Union of Scientists and

Engineers)[27]. Technical skill involves using a product-specific validation and verification approach.

CMMI has been developed by the Software Engineering Institute at Carnegie Mellon University [28]. It

outlines five organizational levels; from lowest to highest, these are initial, managed, defined,

quantitatively managed, and optimizing. The purpose of assessing the level of organization is to raise

it to the highest level, at which developing engineer and manager including CEO expect to produce

perfect products. This is not an exact approach, but a technique to back up and assess the principal

manufacturing process.

Quality control is mainly related to manufacturing. Its focus is how to assure that item variations are

within the tolerances of already determined specifications. Therefore, quality control methods are

dimensionally different than the verification of new product designs, since the product developer

should establish the necessary specifications for new products as well as their tolerances. Quality

control is not generally an activity in the design area, but a necessary activity in the manufacturing

field in Figure 7.15.

 Fig.7.15 Confusing quality control: 1) R&D, 2) QA (Performance/Reliability), 3) Manufacture QC

Let’s consider product-specific verification as a technical skill. Generally, engineers check numerous

design items when developing new products. In the NASA Systems Engineering Handbook [29], there

are sixteen activities under the heading “Verification Procedures,” almost all of which involve testing.

The keywords include identification of test configuration, test objective, test criteria, test equipment,

and location of test activity. Similarly, verification of software includes test strategy, test plan, test

procedure, test scenario sorting deficiencies, and so on. But these are general comments or

recommendations that may vary according to the activity and the test article, and therefore are not

mandatory. Of course the test is required if applicable specifications exist, but that is not sufficient.

From a verification viewpoint, NASA Handbook addresses tasks used to test products, but does not

establish the detailed specification standards as the frame of reference in which these tasks might be

carried out. It is not acceptable for verifiers to use their discretion when verifying product performance.

Carefully established specifications prevent verification activities to deviate from the determined

process. Thus, when failure occurs, it is possible to determine whether the specifications are

inappropriate or whether verifiers are incorrectly conforming to the specifications. Sometimes we can

also identify omissions in the verification specifications.

Verification specifications should be established over the full range of functions fitted to each product.

A thorough use of available technology and related measures to address issues might be applied at

an early stage of product development. Why don’t developing engineer and manager require

verification specifications for each product? The reason lies in the engineers’ answers. Product-

assurance specialists may insist that all related verification activities are included in a “thick

document.” And they may add that the activities performed are completely reviewed and revised by

related specialists. Furthermore, if developing engineer and manager figure out the technical details,

he would understand all he wanted to know the design details like differences between quality defects

and failures. However, there is gap between design engineer and manager including CEO. Especially,

manager can’t understand the complex situations when engineers are in design. This is a kind of trap.

If the technical details become specification, the situation will change (Table 4.1).

7.7.2 Reliability Quantitative (RQ) specifications

Technology concepts related product could be explained with common sense, although some

concepts of new technology take time to be understood. Everyone understands new product design

concepts if the related staff explains them in everyday communication. Fortunately, the technology

concepts related to design verification is not difficult to grasp because we, as consumers, use product.

For example, the technology concepts to reduce the noise level of a car engine would be difficult to

understand but assessing the performance improvement would be easy since we can hear it. The

concepts of design verification related specification are less complex than that of design itself, and

should be easily grasped by developing engineer and manager. If developing engineer and manager

have difficulty understanding the verification document, there are illogical sections not to be explained

with common sense when the engineers write it. The necessary logic of “thick document” is clarified

for any layman to be understood and controversial as seen in Table 7.6.

Three kinds of simple logic might be addressed in the verification document. Firstly, the information is

divided into two activities: verification-specifications establishment and related execution. Secondly,

procedures for how to extract anticipated issues in a new product need to be addressed in the

verification specifications, avoiding omissions of necessary specifications and adding some details
pertaining to the product for a complete set of specs. Thirdly, verification specifications should be

classified into categories according to technological fields in order for related specialists to review

their accuracy. Verification specifications might be clearly presented, providing brief summaries to

clarify the entire specification concepts.

Table 7.6. Double Ambiguity of Product Quality

Meaning of
basic
quality

Conformance to
Concept Product Life Failure Rate
Specifications

Percent/Year Percent
Unit Year
Percent/Hour ppm

Probability Weibull distribution Exponential distribution Normal distribution

Activities Design change or establishing specification Inspection, Screening

Additionally, there are two other issues involved in establishing verification specifications. First of all,

we think that new products can check the combinations of specifications used for similar products.

This is a misunderstanding. Potential problems inherent in new products cannot be identified using old

specifications. The new product incorporates innovative structures, new materials, and different

software for upgrading performance and decreasing cost. These cannot be adequately tested using

existing specifications.

By using the previous specifications, new failure mechanisms is not easy to be identified for products

that have the design modifications. In addition to updating the specifications, we should also consider

what new testing might be effective. For example, is it possible to apply the test specifications for the

Boeing 777 fuselage made of aluminum alloys to the Boeing 787 Dreamliner fuselage, which
incorporates new materials, like CFRP? Obviously, we know that the previous test specs would be

improper.

The other issue is that reliability quantitative specifications that can use the parametric ALT as one of

methodology mentioned in previous sections include estimating item lifetime. Reliability disasters

caused by the design missing during customer use could tarnish the company’s reputation. But most

people consider this task beyond the scope of possibility. Generally obtaining quantitative results in

reliability analysis is very difficult. Reliability specialist Patrick O’Connor wrote in Practical Reliability

Engineering that there are basically three kinds of situations - small, moderately large, and very large

numbers of factors and interactions [30]. A small number of factors can be predicted with the physical

model. A large number can be predicted with statistical models. Predictive power diminishes, however,

in the case of a moderately large number of factors pertaining to reliability.

Reliability prediction is a necessary task to be undertaken. Let’s look at a product in the standpoint of

reliability problems. We know that there are a few sites in product that are weaker than other sites.

Reliability specialists can presume the location of the weakest site and/or its failure mechanisms,

though they don’t know whether the failure will actually happen in the targeted lifetime, or how high

the failure rate would be. So if we extract one or two failure sites in the product, mostly in a given

module or unit, and classify their failure mechanisms into two categories of reliability – lifetime LB and

failure rate  within lifetime - the factors related to reliability estimation are decreased, and the cases

pertaining to moderately large factors become small-factor-number cases. Thus we can make

quantitative estimations about reliability issues - mainly lifetime under normal conditions. This is the

simple explanation to understand developing engineer and manager.

 Fig.7.16. Reliability Index; BX Life (LBx)

Let’s describe in commonsense terms the basic concepts of the required statistics and methods

pertaining to establish the quantitative lifetime specification, which developing engineer and manager

can easily understand the BX Life as reliability quantitative specifications in Figure 7.16. For instance,

take automobiles. Assume that we test one hundred cars in Germany for ten years and find no trouble

(10 years, 160,000km). We can conclude that the car’s failure rate is below 1% per ten years, which is

called “B1 life 10 years,” When we conclude that the car’s failure rate is below 1% per ten years, its

confidence level reaches around 60%, called the commonsense level of confidence. Of course, we

cannot test products for ten years before market release. So we make the accelerated vehicle testing

by imposing heavy loads and high temperature until we reach an acceleration factor of ten. This will

reduce the test period by one-tenth, or one year. Thus we test one hundred items for one year (16,000

km), or one week without stoppage (7 days  24 hours  100km/h = 16,800km). The next step is to

reduce the sample size.

Then, if you increase the testing time, the items would achieve a sufficiently degraded state and many

would fail after the test; therefore, we can greatly reduce the sample size, because one or two failed

samples would yield enough data to identify the problem area and make corrective action plans.

Increasing the test time by four times, or to one month, reduces the necessary sample size by the

square of the inverse of the test-time multiplier, to one-sixteenth (square of one quarter), or six
engines. The final test specification, then, is that six engines should be tested for one month under

elevated load and temperature conditions with the criterion that no failure is found. This concept,

called Parametric Accelerated Life Testing, is the key to reliability quantitative specifications.

We also cannot guarantee the behavior of a product over ten years under the extreme environments.

These test conditions would be appropriate to mechanical/civil system (or components), like power

engines, but the test conditions are not fit to assess the degradation of paint on the automobile body.

In addition to testing the new engine, we should devise quantitative test methods for other

components - new electrical components (including batteries), electronic control units, lighting

systems, or coating materials. According to the identified failure mechanisms, testing must be

conducted by subassembly to heighten acceleration.

Without such quantitative lifetime testing, we can’t identify all the failures influencing the product’s

lifetime because there would be unanticipated failures. For example, at prototype testing, the lifetime

of a tub in a washing machine was lengthened at first parametric ALT by the missing structural design

changes – a corner radius increase, rib insertion, and so on. The final parametric ALT, however,

showed a weakening of strength in the plastic due to a chemical reaction; the problem was solved by

changing the release agent of the injection molding process—something no one could have been

predicted as the solution. Note that this method reveals exact failure modes, including totally

unexpected ones, that other methods, like FMEA (failure mode and effects analysis), cannot identify.

For the CFRP of the 787 Dreamliner, the failure mechanism is a kind of delamination, which can be

found in pressure/humidity/temperature cycling and ultraviolet irradiation testing. If the acceleration

factor for testing this is calculated using an adequate life-stress model (time-to-failure model) and the

sample size is determined according to the B X life target, then quantitative results can be derived.

Note that we should check the possibility of failure due to various overstresses, such as bird strikes,

with sufficiently degraded samples. For the electrical systems in the 787 engineers should incorporate

the same components used in other commercial airplanes (and different combinations of them),

assessing the possibilities of overstress failures under reliability marginal stresses, since they can

assure lifetime reliability. But for new components like the lithium-ion battery, the failure mechanics as

well as the stresses produced in the aircraft environment have been changed. Thus, we cannot

presume what kind of failure mechanisms would occur due to chemical reaction until the projected
lifetime reaches. Generally, chemical failure mechanisms are delicate and thus difficult to identify and

reproduce, which means that the acceleration conditions and related factors can be hard to

determine. Thus they should test until lifetime under almost normal conditions, and the behavior of

sufficiently degraded components should be checked under the rated stresses and overstresses. The

media have reported various accidents or disasters due to unanticipated failure mechanisms in

chemical items, such as the fires occurring in the Sony lithium-ion battery in notebook computers in

2005, Firestone tires causing Ford Explorer rollovers in the 1970s, wire bundles incorporating silver-

plated copper wire leading to fire in the Apollo 1 cabin in 1967, and so on.

7.7.3 Conceptual framework of specifications for quality assurance

Returning to the subject of establishing verification specifications, there are plenty of specifications

that have few explanations technically. It is difficult to find articles that explain to establish

specifications; there have been a few research studies about it. So let’s consider how to anticipate

issues in a new product and to configure a series of verification specifications responding to them,

and how to develop specifications that will identify these issues accurately.

Here is one such methodology. First, select an existing product to be compared with the new one. All

its relevant specifications are listed except the unnecessary specifications. Secondly, because

inevitably the similar but older product has the intractable problems to be listed, we must devise new

specifications to address these issues. Ongoing problems indicate that any counter-measures have

not resolved the real cause because of the inadequate analysis. Nonetheless, the original design idea

may be faulty. The existing product would be solved by using precise problem analysis, and the new

product would be handled by identifying and fixing the problem before releasing the next model. To

correct the existing problems in similar products, it is important to add the verification specifications.

Sometimes the potential problems of the subassemblies manufactured by a new supplier also might

be considered.

Thirdly, the newly designed portions—those that differ from the current comparable product--should

be listed and the potential issues related to them should be predicted. Verification specifications need

to be devised to address these problems. Especially note that all items incorporating new chemical
materials should be tested to item lifetime under new quantitative specifications because a new kind

of wear-out failure could occur near the item lifetime. Moreover, it is very difficult to computer-simulate

and clarify diverse chemical reactions over an item’s life cycle.

Finally, the new product will also have performance fundamentals unique to it, which sometimes

provide a competitive edge over competitors. Such comparative advantages in performance

fundamentals might be checked with the newly established specifications.

A complete set of verification specifications in Figure 7.17 might be summarized as four types of data:

(1) all the verification specifications for the comparable product(s); (2) specifications to fix existing

problems in the comparable product; (3) specifications that deal with the potential issues in the newly

designed portions; and (4) specifications checking newly incorporated performance features. The

specifications responding to the latter three categories are all established anew. The purpose of

sectioning potential issues in a new product is to check whether necessary issues have been omitted.

Fig.7.17. Complete testing sets of verification specifications for Quality Assurance

All specifications enumerated according to this model would be classified, initially, into two groups:

performance specifications and reliability specifications. If an issue to be identified is related to

material rupture or degradation over time, it is a reliability issue; if not, it is a performance issue. The

specifications are further divided into four categories: Usual Performance specifications (UP), Special

Performance specifications (SP), Reliability Marginal test specifications (RM), and Reliability

Quantitative test specifications (RQ).

UPs check the expected performance by the usual operator or consumer. SPs check performance

under extraordinary environments, such as tropical heat or elevated electromagnetic fields. RMs are

used for identifying physical changes under severe or peculiar conditions, including unusual usage

environments like electrostatic overstress or lightning surges. Finally, RQs are for reviewing the

product state under normal conditions and for estimating the product lifetime, the B X life (lifetime of the

cumulative failure rate X %), and the annual failure rate within lifetime. Note that the lifetime index

MTTF refers to the time-point at which about 60% of the production lot fails, which is an unacceptable

rate.

Parametric accelerated life testing mentioned in the previous sections uses the sample size equation

with acceleration factor. It also is a process that helps designers find the optimal design parameters,

which can help them better estimate expected lifetime LB, failure rate of module , and determine the

overall product reliability. Reliability quantitative (RQ) test specifications are used to estimate the

required lifetime (or cycle) if reliability target of product - the cumulative failure rate X % and lifetime is

given. Parametric accelerated life testing (ALT) might be related to the RQ test specifications. And the

examples of parametric ALT will be discussed with the Chapter 8.

7.8 Testing equipment for quality and reliability

7.8.1 Introduction

In today’s competitive market, more companies are looking to application specific automatic testing

equipment versus functional testing methods. That’s because the traditional testing process did not
apply for complex systems such as aero and automotive engines. High product performance and

reliability are a basic requirement and sometimes the only difference between products of various

manufacturers. Test equipment verifies the performance and reliability of mechanical, electrical,

hydraulic and pneumatic products. These include tool testers, hi-pot testers, power cord and power

supply cord testers, automatic test equipments for a variety of purpose, and leakage current testers.

Product quality is a critical aspect for companies who are struggling to retain customers in these days

of fast eroding brand loyalty. Testing equipment companies designs and builds production test

equipment. They specialize in R&D test equipment, authentication test equipment and quality control

test equipment for mass production. Test equipment is categorized as overall performance test,

durability (life) test, accelerated test, safety test and environmental test, etc (See Figure 7.18).

Fig. 7.18 Reliability assessment concept for developing testing equipment

Test equipment has multi-disciplinary systems that are incorporating machine design, material

science, industrial engineering, statistics, electrical & electronics, and computing system. As product

development requires substantially high level of performance and reliability in the limited developing
time of product, equipment for testing the performance and reliability of product is growing at a

significant scale.

In the stage of the detail design, testing equipment companies are to develop the test equipment that

is applicable to be gratified at the specifications of end users. They provide the latest state-of-the-art

test equipment to rental centers, electrical service facilities, manufacturers and OEMs. In over half

century they have learned what end users want. Their experiences are reflected in a number of

important concepts in hardware and software designs of test equipment. When a request for a test

equipment development is received, testing equipment companies promises to provide customized

support by analyzing the necessary requirements for installation and trial runs of the developed

equipment.

Fig. 7.19. Considerations for developing testing equipment

The Quality, Safety, and Life of product could be increased by reliability testing. Product reliability

testing is a specialized field that requires deep understanding of the product and a state of the art

infrastructure to deliver the goods. Product reliability testing equipment is to help companies to test

reliability of their products. Thus, reliability testing equipment should have a user requirement –
purpose, required power, testing items, control precision, data processing speed, automation level,

software processing ability, maintainability, equipment maintenance cost, spare part, and necessary

budget (See Figure 7.19).

Testing equipment insures the reliability, safety and performance of products they manufacture, use,

service or rent. As product technology advances, testing equipment are required to 1) make products

reliable, 2) meet the international standards, and 3) offer the Product/Parts Reliability, Failure

Analysis, Test Structure (Design, Verification and Test), Technology Qualification Support for product,

and environmental Measurement Services (humidity, temperature, etc.).

Consequently, testing equipment type for product R&D Development can be classified as: 1) Testing

Equipment for General Performance, 2) Testing Equipment for Durability (Life), 3) Testing Equipment

for Accelerated Testing, 4) Testing Equipment for (Combined) Environment, 5) Testing Equipment for

Quality Control of Mass Production, 5) Testing Equipment for Maintenance and Repair. As seen in

Figure 7.20, there are a variety type of testing equipment and their company in global that cannot be

quantified.

Today testing equipment companies also look to custom designed and manufactured automatic

testing equipment that can functionally test new units that employ advanced technologies. By going

beyond simple parametric testing that limits the use of commercial off the shelf testers, specialty-built

functional automatic testing equipment helps guarantee high intrinsic availability and long-lived

performance “to spec” in the field, thereby facilitating the acceptance and success of new

technologies in the marketplace.

They often specialize in selling test equipment and offering a variety of services to survive the

marketplace. They buy, sell, lease sell all kind of new, refurbished, and used equipments. They also

buy networking equipment, used test equipment and used measurement equipment from leading

manufacturers. Whether end users are any reseller, they offer a cost effective solution that will save

time and money.

Fig. 7.20. Type of quality testing equipment
Fig. 7.21. Procedure of Testing Equipment Development
7.8.2 Procedure of Testing Equipment Development (example: solenoid valve tester)

Development procedure of testing equipment can be briefly summarized in Figure 7.21. For example,

the test equipment of solenoid valve tester in nuclear power plant will be suggested. The testing

equipment would test the intended functionality of product and its reliability.

Step 1) Characteristic Study of Product to be tested

Fig. 7.22. Usage of solenoid valve in nuclear power plant

A solenoid valve is operated by an electric current through a solenoid. For more than 440 nuclear

power plants in the world, a solenoid valve has equipped nearly every plant. As seen in Fig.7.22,

nuclear-qualified and critical solenoid valves have the following applications:

• Emergency core cooling systems

• Emergency generator systems

• Steam generator feed-water systems

• Containment sampling systems

• Auxiliary feed-water systems

• Liquid radiation waste systems

• Turbine bypass systems

Nuclear-qualified solenoid valves are indispensable parts of any nuclear plant safety application. Each

has passed the most rigorous testing for nuclear equipment and environmental qualification (EQ).

These solenoid valves are produced with a high degree of designed-in quality and proven

performance.

Additionally, solenoid valves offers desirable product advantages such as diodes that provide simple

surge protection for control, quick-disconnect connectors for increased safety and reduced

maintenance, and radiation-resistant elastomers for long life.

In the 1950s solenoid valves were onboard the first nuclear powered submarine, the USS Nautilus.

Later, solenoid valves protected the earliest commercial nuclear power plants. In 1978 solenoid

valves were among the first and only to be nuclear-qualified to IEEE and RCC-E specifications.

Solenoid valves from specialized nuclear line are specifically designed for environments with high

radiation, temperature, and seismic requirements.

Consequentially, test equipment for a solenoid valve should satisfy the following specifications:

 ISO 6358: 1989 (E) Pneumatic fluid power – Components using compressible fluids –

Determination of flow-rate characteristics

 IEEE-323: 2003 - Standard for Qualifying Class 1E Equipment for Nuclear Power Generating

Stations

Step 2) Collect Related Technical Information/Data

As seen in Figs. 7.23 and 7.24, nuclear solenoid valves meet the rigorous demands and high

expectations of the nuclear industry. They have applications for nuclear qualified 2, 3, and 4-way

solenoid valves. Especially, nuclear 2-way valves are qualified for mild environmental applications as

defined in IEEE-323-2003. The qualification program consisted of a series of four sequential aging

simulation phases (thermal, wear, radiation, and vibration).

 Fig. 7.23. Type of solenoid valve for nuclear power plant

Fig. 7.24. Structure of solenoid valve for nuclear power plant

• IEEE-323: 2003 - Standard for Qualifying Class 1E Equipment for Nuclear Power Generating

Stations.

Qualification consists of subjecting solenoid valve to the following tests as required by the previously

noted IEEE-323 specifications.

A. Thermal aging

B. Wear aging

C. Pressurization aging

D. Radiation aging

E. Vibration aging

F. Seismic event simulation

G. Radiation event simulation

H. LOCA/MSLB/HELB environmental simulation

All solenoid valves are designed with the following special features:

1. Type N Construction (NS Series), Class H (NP Series) coil insulation.

2. Elastomers (gaskets, O-rings, discs): all materials designed to meet high radiation and high

temperature degradation effects.

3. Specially designed solenoid enclosures to withstand Loss-of-Coolant-Accident (LOCA)

environment.

4. Designed to meet seismic loading.

2-way Nuclear Power (NP) solenoid valves are widely used for pilot control of diaphragm and cylinder

actuated valves (and other applications) used in nuclear power plants. Selection of the proper valve

for a specific application is of paramount importance. This engineering information section describes

principles of operation, types of solenoid valves, and types of solenoid enclosures, and materials to

assist you in the proper selection of a valve.

A nuclear solenoid valve is a combination of two basic functional units: (1) a solenoid, consisting of a

coil and a magnetic plunger (or core); and (2) a valve body containing an orifice in which a disc is

positioned to stop or allow flow. The valve is opened or closed by movement of the magnetic plunger

(or core), which is drawn into the solenoid when the coil is energized. Solenoid valves feature a

packless construction. The solenoid is mounted directly on the valve and the core assembly is

enclosed in a sealed tube inside the solenoid coil. This construction provides a compact, leak-tight

assembly, without the need of a stuffing box or sliding stem seal.

Direct-acting solenoid valves operate from zero kPa (no minimum pressure is required for the valve to

operate), to the individual valve’s maximum rated pressure. Because of the wide range of sizes,

construction materials, and pressures, direct-acting qualified valves in brass or stainless steel are

found to the many applications found in nuclear power plants. Two 2-way direct acting types are

available as follows: normally Closed: closed when de-energized and open when energized. Normally

Open: open when de-energized and closed when energized.

Step 3) Design of Test Mechanism

Method to determinate flow rate characteristic of the solenoid valve is based on increasing upstream

pressure while the pressurized air goes through a mass flow sensor. The method of standard ISO

6358 is explained to two equations which describe the flow rate through the orifices.

p2
T0 for ≤b choked flow (7-43)
q m=Cp1 ρ0
√ T1
2
p1

p2

q m=Cp1 ρ0
√ ( )
T0
T1 √ 1−
p1
1−b
−b for
p2
p1
≻b subsonic flow (7-44)
Custom equipment was manufactured for measuring the flow rate characteristics based on

„increasing the upstream pressure”. Part of the valve holding the valve nozzle was replaced with the

special equipment. With this setup it was possible to set the nozzle to a required fixed position.

Without modifications this could not be done.

a) Schematic of measuring the flow characteristics of solenoid valve (1 – pressurized air source, 2 –

pressure regulator, 3/4 – pressure sensor, 5 – flow meter)

b) Dependence flow rate on the upstream pressure.

Fig.7.25. Measuring flow rate characteristic by increasing upstream pressure

Figure 7.25 (a) shows the measuring station which was used for measurement and Figure 7.25 (b)

shows the curvature of dependence flow rate on the upstream pressure. This method has been

applied only on solenoid valve because preparing special equipment for holding the nozzle in a

constant position is very expensive and work intensive.

Step 4~13) Making the solenoid testing equipment

These steps consist of optimal setting of driving actuator and loading actuator, selection of various

sensors, design power (electric/hydraulic) circuit, design sequence circuit, design operating mode,

design automatic stop mode, electronic control system design, computer ↔ DAQ & control unit, DAQ

& control unit ↔ sensor, electric power switching circuit design, fabrication of system hardware,

fabrication of operating unit, fabrication of controller, computer programming by LabVIEW and

MATLAB, and combine system hardware and software (See Figure 7.26 through 7.28).
Fig.7.26. Schematic control diagram of measuring flow rate characteristics of solenoid valve
 Fig.7.27. Solenoid Valve Tester

Fig.7.28. Appearance of Electronic Controller

For creating a simulation of pneumatic fluid power — determination of flow-rate characteristics of

solenoid valve using compressible fluids, MATLAB® Simulink may be used. It makes possible to

compare a lot of measured data with mathematical models, which was a great contribution to the

work. Measurements made with the two valves were compared to theoretical values.
 Fig.7.29. Electrical power connection (example)

Fig.7.30. Simulation schematic of a tank charge in MATLAB® Simulink

Figure 7.30 shows two Simulink models created to determine sonic conductance C and the critical

pressure ratio b from the measured data. The flow rate characteristic was measured by increasing

upstream pressure as depicted in Figure 7.30.

Step 14 ~ 15) System simulation, test run and calibration

 Fig.7.31. Dependence of flow rate to the upstream pressure – data from the ISO 6358 method

measurement

The result is shown in Figure 7.31 Measured data are represented as circles. Line curvature

represents theoretical approximation obtained as a numerical solution searching sonic conductance C

and the critical pressure ratio b. The results are C=6.329 10-9 m3/(s*Pa), b=-0.1527, which can be

obtained from MATLAB® and indicates the solenoid valve characteristics. And this method might be

applied only on solenoid valve because preparation of the special equipment for holding the nozzle in

constant position is very expensive and work intensive.

Values of critical pressure ratio b for each final pressure in the tank were determined based on

knowledge that the linear part of the curvature of the charge is describe by equation (1) and the

behaviors of flow rate in the second part of the curvature is described by equation (2). Of note is that

the critical pressure ratio b expresses the divide of the downstream and the upstream pressure which

flow becomes choked.

For determination of the critical pressure ratio a derivative of smoothed measured data was made,

and the point where the derivative exchanges determines the mentioned critical pressure ratio. The

values of sonic conductance C were determined from the equation which describes the flow rate

through the orifices for subsonic flow.

References

[1] 1990 IEEE Standard Glossary of Software Engineering Terminology. IEEE STD 610.12-1990.

Standards Coordinating Committee of the Computer Society of IEEE: New York.

[2] Kreyszig E (2006) Advanced Engineering Mathematics. 9th ed. John Wiley and Son: NJ: 683.

[3] Taguchi G (1978) Off-line and On-line Quality Control Systems. Proceedings of the International

Conference on Quality Control Tokyo, Japan.

[4] Taguchi G Shih-Chung T (1992).Introduction to Quality Engineering: Bringing Quality Engineering

Upstream. ASME: New York.

[5] Ashley S (1992) Applying Taguchi’s Quality Engineering to Technology Development. Mechanical

Engineering

[6] Wilkins J (2000) Putting Taguchi Methods to Work to Solve Design Flaws. Quality Progress 33(5):

55-59.

[7] Phadke MS (1989) Quality Engineering Using Robust Design. Prentice Hall: NJ.

[8] Byrne D Taguchi S (1987) Taguchi Approach to Parameter Design. Quality Progress. 20(12): 19-

26.

[9] Woo S Pecht M (2008) Failure Analysis and Redesign of a Helix Upper Dispenser. Engineering

Failure Analysis 15 (4): 642–653.

[10] Woo S O’Neal D Pecht M. (2009) Improving the Reliability of a Water Dispenser Lever in a

Refrigerator Subjected to Repetitive Stresses. Engineering Failure Analysis 16 (5): 1597–1606.

[11] Woo S O’Neal D Pecht (2009) M Design of a Hinge Kit System in a Kimchi Refrigerator Receiving
Repetitive Stresses. Engineering Failure Analysis 16 (5): 1655–1665.

[12] Woo S Ryu D and Pecht M (2009) Design Evaluation of a French Refrigerator Drawer System

Subjected to Repeated Food Storage Loads. Engineering Failure Analysis 16 (7): 2224–2234.

[13] Woo S O’Neal D and Pecht M (2010) Failure Analysis and Redesign of the Evaporator Tubing in

a Kimchi Refrigerator. Engineering Failure Analysis 17(2): 369-379.

[14] Woo S O’Neal D and Pecht M (2010) Reliability design of a reciprocating compressor suction

reed valve in a common refrigerator subjected to repetitive pressure loads. Engineering Failure

Analysis 7(4): 979-991.

[15] Woo S Pecht M and O’Neal D (2009) Reliability Design and Case Study of a Refrigerator

Compressor Subjected to Repetitive Loads. International Journal of Refrigeration 32 (3): 478-486

[16] Woo S O’Neal D (2011) Pecht M. Reliability design of residential sized refrigerators subjected to

repetitive random vibration loads during rail transport. Engineering Failure Analysis 18(5): 1322–1332.

[17] Woo S Park J and Pecht M (2011) Reliability design and case study of refrigerator parts

subjected to repetitive loads under consumer usage conditions. Engineering Failure Analysis

18(7):1818-1830.

[18] Woo S Park J Yoon J Jeon H (2012) The Reliability Design and Its Direct Effect on the Energy

Efficiency. Energy Efficiency - The Innovative Ways for Smart Energy, the Future Towards Modern

Utilities. InTech: Chapter 11.

[19] Woo S (2015) The reliability design of mechanical system and its Parametric ALT. Handbook of

Materials Failure Analysis with Case Studies from the Chemicals. Concrete and Power Industries.

Elsevier: Chapter 11 pp. 259-276.

[20] Woo S. and O’Neal D (2016) Improving the Reliability of a Domestic Refrigerator Compressor

Subjected to Repetitive Loading. Engineering 8: 99-115.

[21] Woo S and O’Neal D (2016) Reliability Design of the Hinge Kit System Subjected to Repetitive

Loading in a Commercial Refrigerator. Challenge Journal of Structural Mechanics 2(2): 75-84.

[22] McPherson J (1989) Accelerated Testing. Packaging, Electronic Materials Handbook ASM
International 1: 887-894.

[23] Karnopp DC Margolis DL Rosenberg RC. (2012) System Dynamics: Modeling, Simulation, and

Control of Mechatronic Systems. 5th ed. John Wiley & Sons: New York.

[24] Ajiki T Sugimoto M Higuchi H (1979) A new cyclic biased THB power dissipating ICs.

Proceedings  of the 17th  International Reliability Physics Symposium San Diego CA.

[25] Ryu D Chang S (2005) Novel concept for reliability technology. Microelectronics Reliability 45(3):

611–22.

[26] Wasserman G (2003) Reliability Verification, Testing, and Analysis in Engineering Design. Marcel

Dekker p. 228.

[27] Deming WE (1950) Elementary principles of the statistical control of quality. Japan: JUSE.

[28] CMMI Product Team (2002) Capability Maturity Model Integration (CMMI) Version 1.1,

Continuous Representation. Report CMU/SEI-2002-TR-011, Software Engineering Institute Pittsburgh

PA: Software Engineering;.

[29] NASA (2007) System engineering handbook. Washington: NASA Headquarters p. 92 (NASA/SP-

2007-6105 Rev 1).

[30] O’Connor P (2002) Practical reliability engineering. New York: Wiley p. 159.
Chapter 8
Parametric ALT and its Case Studies

Seong-woo Woo

Email: [email protected]

Reliability Association of Korea, Seoul, Korea (Republic of)

Abstracts: In this chapter, parametric ALT and its casestudies that can be applicable to a variety of

mechanical product will be discussed. Because parametric ALTdescribedin Chapter 7 is suggestedas

the methodology of the reliability-embedded product developing process, it is important for engineer to

figure out how to apply this method to the mechanical system. Here examples of mechanical systems

areapplicable to airplane, automobiles, construction equipment, washing machines, and vacuum

cleaners. To meet thetargeted reliability of mechanical product (or module), parametric ALTcan

identifythe missing controllable design parameters. After a tailored series of accelerated life tests, new

product will satisfy the reliability target because there is no missing design parameter.

Keywords: Parametric ALT,Casestudies, Mechanical engineering system.

8-1 Failure analysis and redesign ofice-maker

The basic function of a refrigerator is to store fresh and/or frozen foods. Today refrigerators

alsoprovide other functions - dispensing ice and water. As the number of refrigerator parts and their

functionsincrease, market pressure for product cost reductionleads to the use of cheaper parts.At that

time refrigerator functions are consistently reliable during customer usage. The refrigerator canbe

designed for reliability by determining proper parameters and their levels.

However, minor design parameters may be neglected in the design review, resulting in product failure

in use.Products with minor design flaws may result in recalls and loss of brand name value.
Furthermore, productliability law requires manufacturers to design products more safely in the

European Union and the United States. Preventing such outcomes is a major objective of the product

development process – design, production,shipping and field testing.

Conventional methods, such as product inspection, rarely identify the reliabilityproblems occurring in

market use. Instead, optimally designing for reliability requiresthe extensive testing at each

development step. As a result, the cost of quality assurance andappraisal can increase significantly.As

a solution, most global companies focus on accelerating life testing (ALT). ALT can help shortenthe

product development cycles and identify diverse design flaws. ALT should be performed with sufficient

samples and testingtime, with equipment designed to match expected product loads.

a) SBS refrigerator (b)Mechanical parts of the ice bucket assembly

Fig.8.1.SBS refrigerator and ice bucket assembly

Fig. 8.1 shows the SBS refrigerator with ice dispenser and the mechanical parts of the ice bucket

assembly.The assembly consists of the bucket case, helix support, helix dispenser clamp, blade
dispenser, helix upperdispenser, and blade, as shown in Fig. 8.1b.The helix upper dispenser in the ice

bucket of refrigerators with ice dispenser systems hasbeen fracturing in field, causing loss of the

dispensing function(seeFigure8.2 and 8.3). Thus reproducing the failure mode to assess how

toprevent the fracture of the helix upper dispenser was critical. The data on failed products in

themarketplace are important to understand the use environment of customer of the product and

helping to pinpointroot causes.

Fig. 8.2.Robust design schematic of ice maker.

Fig.8.3. A damaged product after use

To investigate the reliability of a helix upper dispenser,using robust design schematic, Bond-graph,

and state equations analyzed ‘‘uncontrollable’’ mechanicalload conditions of an ice bucket assembly

New ALT methodologies was proposed for robust designs.

Field data indicated that the damaged products may havetwo structural design flaws: (1) a 2 mm gap

between the blade dispenser and the helix upper dispenser, and (2)a weld line around the impact area

of the helix upper dispenser. Due to the gap, the rotating blade dispenserimpacts the fixed helix upper

dispenser. Because of the weld line, a crack may occur. The temperature of theproduct was below

-20C.

Fig. 8.4.Bond graph of ice bucket assembly.

As seen in Fig. 8.4, the bond graph can be conventional in state space representation to group terms

by state variables.The modeling of ice bucket assembly can be expressed as

di a / dt −R a / La 0 i a 1/ La
e a+ 1 T D
[ dω / dt][ =
mk a −B/ J ][ ] [ ] [ ]
ω
+
0 −1/ J
(8-1)
The mechanical stress (or life) of the ice bucket assembly depends on the disturbance load TPulse in

Eq. (8-1). Theaccelerated life testing applies the stress between low and high to the breakdown

stress. The life–stress model(LS model) can be modified as

−n
TF=A ( S ) =A ( T D )−λ
(8-2)

The acceleration factor (AF) can be derived as

n λ
S T
S0 ( ) ( )
AF= 1 = 1
T0

(8-3)

The ice dispenser of customer is used an averageof approximately 3–18 times per day. Under

maximum use for 10 years, the dispenser incurs about 65,700 usage cycles. Data from the motor

company specifies that normal torque is 0.69 kN cm and maximumtorque is 1.47 kNcm. Assuming the

quotient n = 2, the acceleration factor is approximately 5 in Eq. (8-3).

The test cycles and test sample numbers calculated in Eq. (7-35) were 42,000 cycles and 10 pieces,

respectively. Theparametric ALT was designed to ensure a B1 of 10 years life with about a 60% level

of confidence that it would fail lessthan once during 42,000 cycles.

 (a)Equipment used in accelerated life testing

(b) Duty cycles

Fig.8.5. Equipment used in accelerated life testing and Duty cycles of disturbance load TPulse
 (a) Failed product in field (b) Failed sample in accelerated life testing

Fig. 8.6.Failed product in field and ALT

Figure 8.5 shows the ALT equipment for the reproductionof the failed structural parts in the field and

the duty cycles for the disturbance load TD.Figure 8.6 shows the failed product in the field and a

sample after accelerated life testing. In the photo, the shapeand location of the broken pieces in the

failed market product are identical to those in the ALT results. Figure 8.7represents the graphical

analysis of the ALT results and field data on a Weibull plot. For the shape parameter,the estimated

value in the first ALT is 2.0. However, the final value obtained on the Weibull plot was 4.8.As the ratio

of characteristics life, 1/2, gives the acceleration factor, AF is approximately 2.2 on the Weibullplot.
 Fig.8.7.Field data and results of ALT on Weibull chart.

These methodologies are valid to reproduce the fielded failures because (1) the locationand shape of

the fractures in both market and ALT results are extremely similar; and (2) on the Weibull,the shape

parameters of the ALT results, 1 and market data, 2, are very similar.The reduction factor R also is

0.001 from the experiment data – product lifetime, acceleration factor, actual mission cycles, and

shape parameter. Consequently, we know that this parameter ALT is effective to save the testing time

and sample size

The fracturing and cracking of both the fielded products and the ALT results occur in the contact area

ofthe blade dispenser. These structural flaws generate the concentrated mechanical stress when the

bladedispenser, made of stainless steel, meets the polycarbonate helix upper dispenser at a right

angle. Dueto the 2 mm gap between the blade dispenser and helix upper dispenser and the impact

(1.47 kN cm) ofthe blade dispenser, the concentrated stress of the blade dispenser is approximately
36.9 kPa, based on finite element analysis. Under -20C, it is particularly fragile due to the weld line

near the impact area of the helixupper dispenser(Figure 8.8).

Fig.8.8. Structure of helix upper dispenser

Table 8.1 show the improved design of the helix upper dispenser based on the ALT results. Failure

analysis identified the root cause of the failed product as the 2 mm gap between the blade

dispenserand the helix upper dispenser, and the weld line. To improve the reliability of the newly

designed helix upper dispenser, a second ALT was implementedwith a key controllable design

improvement – no gap in the samples. Based on the first ALT, the AFand  values in the second ALT

were 2.2 and 4.8. The test cycles and test sample numbercalculated in Eq. (7-35) were 54,000 cycles

and 6 pieces respectively. For the second ALT, all samples were failed within 54,000 cycles. In the

second ALT results the failed test samples were still foundinmission test cycles.

For the failed samples, the key controllable design improvement in the third ALT was to add ribs on

theside and front of the impact area. These redesigned samples were implemented for thethird ALT.

The test cycles and test sample calculated in Eq. (7-35) were 54,000 cycles and 6 pieces,

respectively.TheALT was designed to ensure a B1 of 10 years life with about a 60% level of

confidence that it would fail lessthan once during 54,000 cycles.In the third ALT results, the samples
did not crack and fracture until 75,000 cycles of testing.Consequently, the improved helix upper

dispenser will meet the reliability target – B1 10 years.

Table 8.1 shows the results obtained from the third ALT. The B1 life of the redesigned samples was 14

years. When the design of the current product was compared with that of the newly designed one,

theB1 life expanded about fourteen times, from 1.4 years to 14 years. The design improvements of

eliminating thegap and reinforcing the ribs were very effective in enhancing the reliability of the

samples (See Figure 8.9).

Table 8.1 Results of ALT

1st ALT 2nd ALT 3rdALT

Initial Design Second Design Final Design

In 54,000 170 cycles: 1/10 (10%)

17,000 cycles: 1/6 (17%)
cycles, no 5200 cycles: 1/10 (20%)
25,000 cycles: 3/6 (67%) 54,000 cycles: OK
crack and 7880 cycles: 2/10 (40%)
28,200 cycles: 1/6 (83%) Max 75,000 cycles: OK
fracture of 8800 cycles: 2/10 (60%)
38,000 cycles: 1/6 (100%)
helix 11,600 cycles: 4/10(100%)

Helix
upper
dispenser
structure

Material PC + SUS (t = 1.2)

and PC + SUS (t = 1.2) Roundness corner of torsional Gap: 0 mm
specificati Gap C1: 2mm→ 0mm shaft C2: R0.5mm → R2.0mm Added rib on side and front
on of helix
Fig.8.9.Result of ALT plotted in Weibull chart.
8-2 Residential Sized Refrigerators during Transportation
Figure 8.10 shows a typical residential sized French door refrigerator and the mechanical

compartment at the bottomrear of the refrigerator. As refrigerators were transported to the final

destinations by rail,they were subjected to random vibrations from the train. These vibrations were

continually transmitted to therefrigerator (or machine compartment) while train was moving.The

connecting tubes in the mechanical compartments of refrigerators were fracturing and thecompressor

rubber mounts were tearing. Because the tubes were fracturing, refrigerant was leaking out of the

tubes,which resulted in the refrigerator losing its ability to either cool or freeze products. Field data

indicated that the damaged products might havehad design flaws. The design flaws combined with

the repetitive random loads could cause failure.

Based on the field data, the rail transportation was expectedto move a refrigerator 7200 kmfrom Los

Angeles to Bostonin 7 days (L*B). For its machine compartment (or module), B1 life should be kept for

the transported distance (See Figure 8.11).

(a) French Door Refrigerator (b) Machine compartment: (1) compressor, (2) rubber,
(3) connecting tubes, and (4) fan and condenser

Fig. 8.10.French Door Refrigerator and machine compartment (or module)

 (a) Failed locations in the field

(b) Failed connecting tubes in the mechanical compartment

Fig.8.11.Fracture of the refrigerator connecting tubes in the field.

A random vibration in refrigeration system is motion which is non-deterministic. Refrigerator is

subjected to ride on a rough road or rail, wave height on the water. A measurement of the acceleration

spectral density is the usual way to specify random vibration. As seen in Figure 8.12, a refrigerator

subjected to base is random vibrations and their power spectral density.

(a) A refrigerator subjected to base random vibrations

(b) Typical intermodal random vibration in the United States

Fig.8.12. Refrigerators subjected to base random vibrations and their power spectral density
 Fig.8.13.A simplified model of the refrigerator subjected to repetitive random vibrations

A refrigerator subjected to random vibration during transportation can be modeled using the one-

degree-of-freedom vehicle model (see Figure 8.13). The equivalent model of refrigerator is simplified

as:

m ẍ +c ẋ+kx=ky +c ẏ (8.4)

The force transmitted to the refrigerator can be expressed as force transmissibility Q. That is,

FT 1+ ( 2 ζr )2
Q=
kY
=r 2
[ ( 1−r 2 )2 + ( 2 ζr )2 ] (8.5)
The acceleration factor (AF) can be expressed as the product of the amplitude ratio of acceleration R

and force transmissibilityQ. That is,

S 1 n F 1 λ a1 F T λ
AF=
S0 ( ) ( )( )
=
F0
=
a0 kY
=( R×Q )λ
(8-6)

For natural frequency (r = 1.0) and small damping ratio ( = 0.1), the force transmissibility Qhad a

value of approximately 5.1 and the amplitude ratio of accelerationR was 4.17. Using a stress

dependence of 2.0, the acceleration factor in Eq. (8-6) was found to be approximately 452.0(Table

8.2).

Table 8.2 ALT conditions in refrigerator

System conditions Worst Case ALT AF

Transmissibility, 5.1
- 5.1
Q (r =1.0,  =0.1) (From Eq. 8-5)

Amplitude ratio of acceleration,

0.24g 1g 4.17
R (a1/a0)

Total AF (=()2) 452

Suppose that the shape parameterwas 6.41 based on field data and the allowed failed numbersr was

0, the test time and the number of samples from Eq.(7-35) would be 40min and 3 pieces for the first

ALT. To meet the reliability target B1, there needs to be no fracturedsample at the connecting tube of

the refrigerator in 40min that might be the Reliability quantitative (RQ) test specifications (Figure

8.14).
 Fig.8.14. Field data and results of accelerated life test on Weibull chart

For the first ALT, the connecting tubes in the mechanical compartments of three samples at 20 min

were fracturing and thecompressor rubber mounts were tearing during x-axis vibration tests. The

estimated lifetimeLB1 was approximately 3 days and estimatedfailure rate of the design samples  was

2.9%/day.The shape and location of the failure in the ALT weresimilar to those seen in the field. The

reduction factor R also is 0.013 from the acceleration factor = 452 and shape parameter = 6.13.

Consequently, we know that this parameter ALT is effective to save the testing time and sample size

(Figure 8.15).
 (a) Field (b) 1st ALT Results

Fig.8.15. Failure of refrigerator tubes in the field and 1st ALT result

The modified design parameters for the compressor compartment (or module) was modified as

follows: (1) the shape of compressor rubber mount(C1: gap reduction, 1.2 → 0.5mm), (2)the shape of

the connecting tube design (C2)(Figure 8.16).

 (a) Shape of compressor rubber (polymer) mount

(b) Shape of the connecting tube design

Fig.8.16. Modified design parameters of machine department (or module)

With these modified parameters, a second ALT was carried out and there were no problem at 40 min

and 60 min. The estimated lifetimeLB1 was more than 7 days and the estimatedfailure rate of the

design samples was less than 0.14%/day.Over the course of the two ALTs, refrigerators with the
targeted B1 lifewere expected to survive without failure during cross country rail transportin the

US.Table 8.3 shows a summary of the results of the ALTs, respectively.

Table 8.3 Results of ALT.

1st ALT 2nd ALT

Initial Design Second Design

In 45 min fracture
of the connecting
20 min: 1/3 Fracture 45 min: 3/3 OK
tube in
40 min: 2/3 Fracture 60 min: 3/3 OK
refrigerator is less
than 1

Machine room in
refrigerator

C1 Shape of the compressor rubber

Material & Spec
C2 Connecting tube design

8-3 Water dispenser lever in a refrigerator

Figure8.17 shows the Bottom Mounted Freezer (BMF) refrigerator with the newly designed water

dispenserthat consists of the dispenser cover (1), spring (2), and dispenser lever (3). As the consumer

presses the lever, the dispenser system will supply water. To properly work this function, the dispenser

system needs to be designed to handle the operating conditions subjected to it by the consumers who

purchase and use the BMF refrigerator (Figure 8.17 &Figure 8.18).
(a) BMF refrigerator (b) Mechanical parts of the dispenser lever assembly: Dispenser cover (1),

Spring (2) and Dispenser lever (3)

Fig.8.17.BMF refrigerator and dispenser assembly

Fig.8.18.Robust design schematic of water dispensing

In the field, the dispenser lever in the refrigerators had been fracturing, causing loss of the dispensing

function. The field data on the failed products were important for understanding the use environment
of consumers and helping to pinpoint design changes that needed to be made in the product.The

dispenser system of a bottom-mounted refrigerator in field were cracking and fracturing

underunknown consumer usage conditions. The damaged products might have had structural

designflaws, including sharp corner angles resulting in stress risers in high stress areas. The design

flaws combined with the repetitiveloads on the dispenser lever could cause a crack to occur(Figure

8.19).

Fig.8.19A damaged product after use

The mechanical lever assembly of the water dispensing system consisted of many mechanical

structural parts - the dispenser cover, spring, and dispenser lever. Dependingon the consumer usage

conditions, the lever assembly experienced repetitive mechanical loads in the water

dispensingprocess.Figure 8.20 shows the functional design concept ofthe mechanical dispensing

system. As a cup presses on the lever to dispense water, water will dispense.The number of water

dispensing cycles will be influenced by consumer usage conditions. In the United States, the
typicalconsumer requires a BMF refrigerator to dispense water from four up to 20 times a day.

Fig.8.20. Design concept of mechanical dispensing system

Because the stress of the lever hinge depends on the applied force of the consumer, the life-stress

model (LS model) can be modified as

−n −λ
TF=A ( S ) =A ( F )
(8-7)

The acceleration factor (AF) can be derived as

S 1 n F1 λ
AF=
S0( ) ( )
=
F0

(8-8)
The dispenser is used on average 4–20times per day. With a life cycle design point of 10 years, the

dispenser incurs about 73,000 usage cycles.The applied force is 19.6 N which is the maximum force

applied by the typical consumer. Doubling the applied force forthe ALT to 39.2 N and using a stress

dependence of 2.0, the acceleration factor is found to be approximately four in Eq. (8-8).

The test cycles and test sample calculated in Eq. (7-35)were 56,000 cyclesand 8 pieces, respectively.

The ALT was designed to ensure a B1 of 10 years life with about a 60% level of confidence that

itwould fail less than once during 56,000 cycles. Figure 8.21(a) shows the experimental setup of the

ALT with labeled equipmentfor the robust design of the dispenser. Figure 8.21(b) shows the duty

cycles for the pushing force F.

(a) Test equipment of water dispenserused in accelerated life testing

 (b) Duty cycles of repetitive load F

Fig.8.21.Equipment used in accelerated life testing and Duty cycles of repetitive load F
An air cylinder controlled the pushing force, F of the cup. When the start button in the controllerpanel

gave the start signal, the air cylinder with the mug-shape cup pressed the dispenser lever. At this

point, the cupimpacted the dispenser lever at the maximum mechanical force of 39.2 N.

(a) Failed product in Field (b) Failed sample in ALT.

Fig.8.22.Failed products in field and ALT.

Figure8.22 shows the failed product from the field and from the accelerated life testing. In the photos,

the shape and location of the failure in the ALT were similar to those seen in the field because of the

stress raiser such as lever corner with no rounding. The reduction factor R also is 0.009 from the

acceleration factor = 5 and shape parameter = 3.5. Consequently, we know that this parameter ALT is

effective to save the testing time and sample size.These stress raisers in lever like no rounding should

be improved to meet the reliability target of lever in the design phase.

Figure 8.23also shows the photograph of the ALT results and field data and Weibull plot. The shape

parameter in the first ALT was estimated at 2.0. For the finaldesign, the shape parameter was

obtained from the Weibull plot and was determined to be 3.5.These methodologies were valid in
pinpointing the weak designs responsible for failures in the field and supported by twofindings in the

data. The location and shape also, from the Weibull plot, the shape parameters of the ALT, 1, and

market data,2, were found to be similar.

Fig.8.23.Photograph of the ALT results and field data and Weibull plot

The fracture of the dispenser lever in both the field products and the ALT test specimens occurred in

both the front cornerof the lever and the hinge (Figure 8.24). The repetitive applied force in

combination with the structural flaws may have caused cracking and fracture of the dispenser lever.

The design flaw of sharp corners/angles resulting in stress risers in high stress areas can becorrected

by implementing fillets on the hinge rib and front corner as well as increasing the hinge rib thickness.
Through a finite element analysis, it was determined that the concentrated stresses resulting in

fracture at the shaft hinge and the front corner were 8.37 MPa and 5.66 MPa, respectively.

Fig.8.24.Structure of failing dispenser lever in field

The confirmed values of AF and  in Figure 8.23 were 4.0 and 3.5, respectively. The recalculated test

cycles and sample sizecalculated in Eq. (7-35)were 56,000 and 8 EA, respectively. To meet the

reliability target, three ALTs were performed to obtain the designparameters and their proper levels. In

the three ALTs the dispenser lever cracked and or fractured at the front corner ofthe lever and at the

hinge in the first test, at the front corner of the lever in the second test, and at the front of the leverin

the third test.

Tables 8.4 shows the results of the design parameters confirmed from a tailored set of ALTs and a

dispenser lever with high fatigue strengthwas redesigned by parametric ALTs (See Tables 8.5). With

these modified parameters,the BMF refrigerator can repetitively dispense water for a longer period

without failure.Based on the modified design parameters, corrective measures taken to increase the

life cycle of the dispenser systemincluded: (1) increase the hinge rib rounding, C1, from 0.0 mm to 2.0

mm; (2) increase the front corner rounding, C2, from0.0 mmto 1.5 mm; (3) increase the front side
rounding, C3, from 0.0 mmto 11.0 mm; (4) increase the hinge rib thickness, C4,from 1.0 mm to 1.8

mm; and (5) increase the front lever thickness, C5, from 3.0 mm to 4.0 mm.
 Table 8.4Redesigned dispenser lever

Rib1C1: T1mm→T1.8mm (first ALT)

Rib2 C2: T3mm→T4 mm (third ALT) Fillet2 C4: R0mm→R1.5mm (first ALT)
Fillet1 C3: R0mm→R1.5mm (first Fillet3 C5: R0mm →R8mm (first ALT)
ALT)→R2.0mm (second ALT)
 Table 8.5 Results of ALT.

1st ALT 2nd ALT 3rd ALT

Initial Design Second Design Final Design

In 56,000
cycles,
56,000 cycles: 8/8 OK 56,000 cycles: 8/8 OK
fracture of 52,000 cycles: 2/8 fracture
67,500 cycles: 1/8 fracture 68,000 cycles: 1/8 fracture
dispenser 74,000 cycles: 6/8 OK
92,000 cycles: 7/8 OK 92,000 cycles: 7/8 OK
is lessthan
1.

Dispenser
lever
structure

Rib1 T1.0mm →T1.8mm

Material
Fillet 1 R0.0mm →R1.5
and Fillet 1 R1.5mm→R2.0mm
mm Rib2 T3.0mm →T4.0mm
specificati Fillet 2 R8.0 mm→R11.0 mm
Fillet 2 R0.0 mm
on
→R1.5mm

Figure8.25 shows the graphical results of ALT plotted in a Weibull chart. Applying the new design

parameters to the finiteelement analysis the stress concentrations in the shaft hinge decreased from

8.37 MPa to 6.82 MPa and decreased in thefront corner from 5.66 MPa to 3.31 MPa. Over the course

of the three ALTs the B1 life of the samples increased from 8.3 yearsto over 10.0 years.
Fig.8.25.Results of ALT plotted in Weibull chart
8-4 Refrigerator compressor subjected to repetitive loads

(a) A vapor-compression refrigeration cycle

(b) Parameter diagram of refrigeration cycle

Fig.8.26.A vapor-compression refrigeration cycle and its parameter diagram

A refrigerator consists of a compressor, a condenser, a capillarytube and an evaporator. The

refrigerant enters thecompressor at a low pressure. It then leaves the compressorand enters the

condenser at some elevated pressure; therefrigerant is condensed as heat is transferred to

thesurroundings. The refrigerant then leaves the condenser asa high-pressure liquid. The pressure of

the liquid is decreasedas it flows through the expansion valves, and as a result, someof the liquid

flashes into cold vapor. The remaining liquid ata low pressure and temperature is vaporized in the

evaporatoras heat is transferred from the fresh/freezer compartment.This vapor then reenters the

compressor. The main function of the refrigerator is toprovide cold air from the evaporator to the

freezer andrefrigerator compartments(Figure 8.26).

A capillary tube controls the flow in the refrigeration system and drops the high pressure of the

refrigerant in the condenser to the low pressure in the evaporator. In a refrigeration cycle design, it is

necessary to determine both the condensing pressure, P c and the evaporating pressure, P e. These

pressures depend on ambient conditions, customer usage conditions, and heat exchanger capacity in

the initial design stage.

To derive the life-stress model and acceleration factor, the time to failure (TF) can be estimated from

the McPherson’s derivation:

Ea
−n
TF=A ( S ) exp ( )
kT
(8-9)

To use Eq. (8-9) for accelerated testing, it needs to bemodified and put into a more applicable form. A

refrigeration system operates on the basicvapor-compression refrigeration cycle. The compressor

receives refrigerant from the low-side (evaporator) and then compresses and transfers the refrigerant

to thehigh-side (condenser) of the system. The capillary tube controls the flow in a refrigeration

system and drops the high pressure of the refrigerant in the condenserto the low pressure in the

evaporator. In a refrigeration cycle design, it is necessary to determine both the condensingpressure,

Pc and evaporating pressure, Pe. (See Figure 8.26(a)).

The mass flow rate of refrigerant in a compressor can be modeled as

 ηv
ṁ=PD×
v suc

(8-10)

The mass flow rate of refrigerant in a capillary tube can be modeled as

P3 0 .5

[ ]
−∫P ρ dP
2
ṁcap= A
2 ρ
D
f m ΔL+ln 2
( )
ρ3

(8-11)

By conservation of mass, the mass flow rate can be determined as:

ṁ=ṁcap (8-12)

The energy balance in the condenser can be described as

Qc =ṁ ( h1−h2 ) =( T c −T o ) / R c

(8-13)

The energy balance in the evaporator can be described as

 Qc =ṁ ( h4 −h 3 ) =( T i−T e ) / R e

(8-14)

When nonlinear Eq. (8-12) through (8-14) are solved, the mass flow rate, ṁ, evaporator
temperature,Te, and condenser temperature, T c can be obtained. Because the saturation pressure,

Psat, is a function of temperature, the evaporator pressure, P e (or condenserpressure Pc), can be

obtained as:

Pe=f ( T e ) or Pc =f ( T c )
(8-15)

One source of stress in a refrigerationsystem may come from the pressure difference between suction

pressure,Psuc, and discharge pressure, Pdis.

For the theoretical single-stage cycle, the stress of the compressor depends on the

pressuredifference suction pressure, Psuc, and discharge pressure, Pdis. That is,

ΔP=Pdis−P suc ≃Pc −Pe (8-16)

By repeating the on and off cycles, the life of compressorshortens. The oil lubrication then relieves the

stressful wearand extends the compressor life.Because the stress of the compressor depends on

thepressure difference of the refrigerator cycle, the life-stressmodel can be modified as

Ea
TF=A ( ΔP )
−n
exp ( )
kT (8-17)
 Fig.8.27. Redesigned compressor and crankshaft

Figure8.27shows a redesigned crankshaft developed to reducenoise and improve energy efficiency of

compressors inside-by-side (SBS) refrigerators. For these applications, thecompressor needs to be

designed robustly to operate undera wide range of customer usage conditions.

The acceleration factor (AF) from Eq. (8-17) can be derived as

S 1 n ΔP 1 λ
Ea 1 1
AF=
S0( ) ( )[ (
=
ΔP 0
−
k T0 T1 )] (8-18)

The normal number ofoperating cycles for one day was approximately ten; the worstcase was twenty-

four. Under the worst case, the objectivecompressor cycles for ten years would be 87,600 cycles.

The normalpressure was 1.07 MPa at 42 C and the compressor dometemperature was 90 C. It was

measured after T type thermocouple pierced into the top compressor. For the acceleratedtesting, the

acceleration factor (AF) for pressure at 1.96 MPawas 3.37 and for the compressor with a 120 C

dome temperaturewas 3.92 with a quotient, m, of 2. The total AF wasapproximately 13.2 (Table 8.6).
 Table 8.6ALT conditions in a vapor compression cycles

System conditions Worst case ALT AF

High side 1.07 1.96

Pressure, MPa Low side 0.0 0.0

ΔP 1.07 1.96 3.36

(=(/)2)

Temp., ℃ Dome Temp. 90 120 3.92 

Total AF (=  X ) - 13.2

The parameter design criterion of the newly designedcompressor can be more than the target life of

B1 ten years.Assuming the shape parameter  was 1.9, thetest cycles and test sample numbers

calculated in Eq. (7-35)were18,000 cycles and 30 pieces, respectively. The ALT was designedto

ensure a B1 of ten years life with about a sixty-percent level ofconfidence that itwould fail less

thanonceduring 18,000 cycles.

Figure8.28 shows the ALT equipment used for the life testing inthe laboratory. Figure 8.29 shows the

duty cycles for the repetitivepressure difference, P.For the ALT experiments, a simplified vapor-

compressionrefrigeration system was fabricated. It consisted of an evaporator,compressor,

condenser, and capillary tube. The inlet tothe condenser section was at the top and the condenser

outletwas at the bottom.

 (a) Adrawing of the test system

(b) Photograph

Fig.8.28. Equipment used in accelerated life testing. (a) Adrawing of the test system. (b) Photograph.
 Fig.8.29. Duty cycles of repetitive pressure difference on thecompressor.

The condenser inlet was constructed with quick coupling and had a high-side pressure gauge. A

tengram refrigerator dryer was installed vertically at the condenser inlet. A thermal switch was

attached to the condenser tubing at the top of the condenser coil to control the condenser fan. The

evaporator inlet was at the bottom. At a location near the evaporator outlet, pressure gauges were

installed to enable access to the low side for evacuation and refrigerant charging.

The condenser outlet was connected to the evaporator outlet with a capillary tube. The compressor

was mounted on rubber pads and was connected to the condenser inlet and evaporator outlet. A fan

and two 60 Watt lamps maintained the room temperature within an insulated (fiberglass) box. A

thermal switch attached on the compressor top controlled a 51m 3/h axial fan compressor, condenser,

and capillary tube. The inlet to the condenser section was at the top and the condenser outlet

In SBS units sold it was found that the crankshafts of somecompressors were locking. Locking refers

to the inability ofthe electric stator to rotate the crankshaft, due to a failure ofone more components

within the compressor under a range of unknown customer usage conditions. Fielddata indicated that

the damaged products may have hada design flaw – oil lubrication problems. Due to this designflaw,

the repetitive loads could create undue wear on thecrankshaft and cause the compressor to lock.
 (a) Failed product in field (b) Failed sample in 1st ALT

Fig.8.30.Failed product in field and 1stALT.

Figure 8.30 shows the crankshaft of a locked-up compressor fromthe field and a sample from the

accelerated life testing. In thephoto, the shape and location of the parts in the failed productfrom the

field were similar to those in the ALT results. Figure8.31represents the graphical analysis of the ALT

results and fielddata on a Weibull plot. For the shape parameter, the estimatedvalue in the previous

ALT was 1.9.It was concluded that the methodologies used were valid inpinpointing the weaknesses

in the original design of the unitssold in the market because (1) the location and shape of thelocking

crankshaft from both the field and ALT were similar;and (2) on the Weilbull, the shape parameters of

the ALTresults, 1 and market data, 2, are very similar. The reduction factor R also is 0.15 from the

acceleration factor = 13.2 and shape parameter = 1.9. Consequently, we know that this parameter

ALT is effective to save the testing time and sample size.

 Fig.8.31.Field data and results of ALT on Weibull chart.

When both the locked compressors from the field and theALT compressor were cut apart, severe

wear was found inregions of the crankshaft where there was no lubrication – thefriction area between

shaft and connecting rod, and alsothe friction area between crankshaft and block. The locking ofthe

compressor resulted from several design problems. Therewas (1) no oil lubrication in some regions of

the crankshaft (Figure8.32); (2) a low starting RPM (1650 RPM) (Figure8.32); and, (3)a crankshaft

made from material with a wide range of hardness(FCD450) (Figure 8.33).

Fig.8.32.No lubrication region in crankshaft and low starting RPM (1650 RPM).
 Fig.8.33.A large variation of hardness (FCD450) in crankshaft.

The vital parameters in the design phase of the ALT werethe lack of an oil lubrication region, low

starting RPM, andweak crankshaft material. These compressor design flawsmay cause the

compressor to lock up suddenly when subjectedto repetitive loads.The parameter design criterion of

the newly designedsamples was more than the target life, B1, of ten years. Theconfirmed values  on

Weibull chart was 1.9. When the secondALT and third ALT proceeded, the recalculated test cycles

andsample sizecalculated in Eq. (7-35)were 18,000 and 30 pieces, respectively.Based on the B1 life

of ten years, the first, second, and third ALTswere performed to obtain the design parameters and

properlevels. The compressor failure in the first ALT was due thecompressor locking. In the second

ALT, it was due to interferencebetween the crankshaft and a thrust washer. During thethird ALT, no

problems were found with the compressor.

Fig.8.34.Redesigned crankshaft in first ALT

 Fig.8.35. Redesigned crankshaft in second ALT.

To improve the lubrication problems in the crankshaft, itwas redesigned as the relocated lubrication

holes, new grooveand new shaft material FCD500 (Figure 8.34). To avoid the wearbetween

crankshaft and washer, the minimum clearance wasincreased from 0.141mm to 0.480mm (Figure

8.35). With thesemodified design parameters, the SBS refrigerators can operatein the process of on

and off repetitively with a B1 life of tenyears life.

The modified design parameters, with the corrective actionplans, included (1) the modification of the

oil lubricationregion, C1; (2) increasing the starting RPM, C2, from1650 to 2050; (3) changing the

crankshaft material, C3, fromFCD450 to FCD500; and (4) modifying the thrust washer dimension, C4,

(See Table 8.7).

Table 8.7 provides a summary of the ALT results. Figure8.36 showthe results of ALT plotted in a

Weibull chart. With the improveddesign parameters, the B1 life of the samples in the first, secondand

third ALTs lengthen from 3.8 years to over 10.0 years.

Fig.8.36.Results of ALT plotted in Weibull chart

 Table 8.7 Results of ALT.

1st ALT 2nd ALT 3rd ALT

Initial Design Second Design Final Design

In 18,000
10,504 cycles: 2/30
cycles,lock 18,000 cycles: 2/30 wear 18,000 cycles: 30/30 OK
Locking
ing is less 18,000 cycles: 28/30 OK 20,000 cycles: 30/30 OK
18,000 cycles: 28/30 OK
than 1.

Crank
shaft
structure

Material FCD450/FCD450
and One New Groove Modification of washer
specificati Location modification of oil dimension
on supply holes
8-5 Hinge kit system (HKS) in a Kimchi refrigerator

(a) Kimchi Refrigerator (b) Mechanical parts of the hinge kit system: (1) kit

cover,

(2) oil damper, (4) spring, (6) shaft, and (7) oil damper

Fig.8.37Kimchi refrigerator and hinge kit assembly

Figure 8.37 shows the Kimchi refrigerator with the newly designed hinge kit system. When a

consumer closes the door, theywant to close it conveniently and comfortable. For this function, the

hinge kit system needs to be designed to handle theoperating conditions subjected to it by the

consumers who purchase and use the Kimchi refrigerator. The hinge kit assemblyconsists of the kit

cover, shaft, spring, oil damper, and kit housing, as shown in Figure 8.37(b).In the field, the hinge kit

assembly in the refrigerators had been fracturing, causing the door not to close easily. Thus, thedata

on the failed products in the field were important for understanding the usage environment of

consumers and helpingto pinpoint design changes that needed to be made in the product.
 Fig.8.38.Damaged products after use.

In the field, parts of the hinge kit system of a Kimchi refrigerator were failing due to cracking and

fracturing (Figure 8.38) underunknown consumer usage conditions. Field data indicated that the

damaged products might have had structural designflaws, including sharp corner angles and not

enough enforced ribs resulting in stress risers in high stress areas. These designflaws combined with

the repetitive loads on the hinge kit system could cause a crack to occur, and thus cause failure.

The mechanical hinge kit assembly of the door closing function consisted of many mechanical

structural parts. Dependingon the consumer usage conditions, the hinge kit assembly receives

repetitive mechanical loads when the door is closed. Doorclosing involves two mechanical processes:

(1) the consumer opens the door to take out the stored food and (2) they thenclose the door by force.
 Fig. 8.39 Design concept of mechanical hinge kit system in the accelerated testing

Fig.8.40. Robust design schematic of hinge kit system

Figure 8.39 shows the functional design concept ofthe mechanical hinge kit system in the accelerated

testing. Figure 8.40shows the robust design schematic overview of the hinge kit system. As the

consumer presses the refrigerator door, the hinge kit system helps to close the doorsmoothly. The

stress due to the weight momentum of the door is concentrated on the hinge kit system.The number

of door closing cycles will be influenced by consumer usage conditions. In the Korean domestic

market, thetypical consumer requires a Kimchi refrigerator the door system to open and close

between three and ten times a day.The moment balance around the door system with an accelerated

weight and the hinge kit system can be represented as

The moment balance around the HKS without accelerated weight in Fig. 8.39 can be represented as

M 0 =W door ×b=T 0 =F0 ×R (8-19)

The moment balance around the HKS with an accelerated weight can be represented as

M 1=M 0 +M A =W door ×b+W A ×a=T 1=F 1 ×R

 (8-20)

Because F0 is impact force in normal conditions and F 1 is impact force in accelerated weight, the

stress on the HKS depends on the applied impact. Under the same temperature and efforts concept,

the life-stress model (LS model) and can be modified as

−n −λ
TF=A ( S ) =AT −n = A ( F×R ) (8-21)

The acceleration factor (AF) can be derived as

n λ λ λ
S T F ×R F
S0 ( ) ( ) (
AF = 1 = 1 = 1
T0 F 0 ×R
= 1
F0 ) ( ) (8-22)

The closing of the door occurs an estimated average 3–10 times per day. With a life cycle design

point of 10 years, the hinge kit incurs about 36,500 usage cycles. For the worst case, the applied

force around the hinge kit is 1.10 kN which is the maximum force applied by the typical consumer. The

applied force for the ALT with accelerated weight is 2.76 kN. Using a stress dependence of 2.0, the

acceleration factor is found to be approximately 6.3 in Eq. (8-22).

 (a) Equipment used in accelerated life testing

(b) Duty cycles of repetitive load F

Fig.8.41.Equipment used in accelerated life testing and duty cycles of repetitive load F

For the reliability target B1 of 10 years, the test cycles and test sample numbers calculated in Eq. (7-

35)were 34,000 cyclesand six pieces, respectively. The ALT was designedto ensure a B1 of ten years

life with about a sixty-percent level ofconfidence that itwould fail less thanonceduring 34,000 cycles.

Figure 41(a) shows the experimental setup of the ALT with labeled equipment for therobust design of

the hinge kit system. Figure 41(b) shows the duty cycles for the impact force F.

The control panel on the top started or stopped the equipment, and indicated the completed test

cycles and the test periods,such as sample on/off time. The door closing force F was controlled by the
accelerated load applied to the door. Whenthe start button in the controller panel gave the start signal,

the simple hand-shaped arms held and lifted the Kimchi refrigeratordoor. At this point it impacted the

hinge kit with the maximum mechanical impact force due to the accelerated weight(2.76 kN).

(a)Failed product from the field (b)Accelerated life testing

Fig.8.42Failed products in field and ALT

Figure 8.42 shows the failed product from the field and from the accelerated life testing, respectively.

In the photos inFigure 8.42, the shape and location of the failure in the ALT were similar to those seen

in the field. Figure 8.43 represents the graphicalanalysis of the ALT results and field data on a Weibull

plot. The shape parameter in the first ALT was estimated at 2.0. For thefinal design, the shape

parameter was obtained from the Weibull plot and was determined to be 2.1.

These methodologies were valid in pinpointing the weak designs responsible for failures in the field

and supported by twofindings in the data. The location and shape from the Weibull plot, the shape

parameters of the ALT (1) and market data (2)were found to be similar. The reduction factor R also

is 0.016 from the experiment data – product lifetime, acceleration factor, actual mission cycles, and

shape parameter. Consequently, we know that this parameter ALT is effective to decrease the testing

time and sample size.

The fracture of the hinge kit in both the field products and the ALT test specimens occurred in the

housing of the kit (Figure 8.44(a)). The oil damper leaked oil in the hinge kit assembly (Figure

8.44(b)). The repetitive applied force in combination withthe structural flaws may have caused the

fracturing of the hinge kit housing and the leak of the oil damper. The concentratedstresses of the

housing hinge kit were approximately 21.2 MPa, based on finite element analysis. The stress risers in
highstress areas resulted from the design flaws of sharp corners/angles, housing notches, and poorly

enforced ribs.

Fig.8.43Field data and results of ALT on Weibull chart.

(a) The fracture of the hinge kit (b)The leaked oil damper

Fig.8.44Structure of failing hinge kit system in accelerated testing

 (a) Housing hinge kit structure

(b) Redesigned housing hinge kit structure

Fig.8.45.Structure of newly designed hinge kit system.

 (a) Old Design (b) New Design

Fig.8.46.Sealing structure of redesigned oil damper

The correctiveaction plans was to implement fillets, add the enforced ribs, and remove the notching

on the housing of the hinge kit(Figure 8.45). Applying the new design parameters to the finite element

analysis, the stress concentrations in the housing of hingekit decreased from 21.2 MPa to 18.9 MPa.

The sealing structure of the oil damper had a 0.5 mm gap in the O-ring/Teflon/O-ring assembly. Due to

the wear andimpact, this sealing with the gap leaked easily. The sealing structure of the redesigned oil

damper has no gap with Teflon/O-ring/Teflon (Figure 8.46). The parameter design criterion of the

newly designed samples was more than the target life of B1 of 10 years.The confirmed values of AF

and  in Figure 8.43 were 6.3 and 2.1, respectively. The test cycles and sample size recalculated in

Eq. (7-35)were 41,000 and six pieces, respectively. Based on the targeted BX and sample size, three

ALTs were performed to obtain the designparameters and their proper levels. In the second ALTs the

fracture of hinge kit cover occurs due to the repetitive impactstresses and its weak material. The cover

housing of hinge kit assembly was modified by the material change from the plasticsto the Al die-

casting (Figure8.47).
 (a) Cover housing structure

(b) Redesigned cover housing structure

Fig.8.47.Redesigned cover housing structure

The levels of the modified design parameters with corrective action plans were (1) the modification of

the housing hingekit (Figure 8.45); (2) the modification of the oil sealing structure (see Figure8.46); (3)

the material change of the cover housing (see Figure 8.47).

Table 8.8 shows the summary of the results of theALTs, respectively. With these modified parameters,

the Kimchi refrigerator can smoothly close the doors for a longer periodwithout failure. Figure 8.48

shows the graphical results of the ALT plotted in a Weibull chart. Over the course of the three ALTs the

B1 life of thesamples increased from 8.3 years to over 10.0 years.

 Table 8.8 Results of ALT.

1st ALT 2nd ALT 3rd ALT

Initial Design Second Design Final Design

In 41,000
7800 cycles: 1/6 crack
cycles,
3340 cycles: 2/6 crack 9200 cycles: 3/6 crack 41,000 cycles: 6/6 OK 74,000
fracture is
15,000 cycles: 4/6 crack 14,000 cycles: 1/6 crack cycles: 6/6 OK
less than
26,200 cycles: 1/6 crack
one.

Hinge kit
structure

C1: Redesigned housing hinge

Material and
kit C3: Plastic →Al die casting
specification
C2: Oil damper
Fig.8.48. Results of ALT plotted in Weibull chart
8-6 Refrigerator drawer system

(a)French refrigerator (b)Mechanical parts of the drawer

Fig.8.49.Refrigerator and drawer assembly. (a) French refrigerator (b) Mechanical parts of the

drawer:handle ①, drawer ②, slide rail ③, and pocket box ④

Fig.8.50. A damaged product after use

Figure 8.49 shows a refrigerator with the newly designed drawer and handle system and itsparts. In

the field, the refrigerator drawer and handle system had been failing, causingconsumers to replace

their refrigerators (Figure 8.50). The specific causes of failures of therefrigerator drawers during

operation were repetitive stress and/or the consumer improperusage. Field data indicated that the

damaged products had structural design flaws,including sharp corner angles and weak ribs that

resulted in stress risers in high stress areas.

A consumer stores food in a refrigerator to have convenient access to fresh food. Puttingfood in the

refrigerator drawer involves opening the drawer to store or takeout food, closingthe drawer by force.

Depending on the consumer usage conditions, the drawer and handleparts receive repetitive

mechanical loads when the consumer opens and closes the drawer.

Figure 8.51 shows the functional design concept of the drawer and handle system. The stressdue to

the weight load of the food is concentrated on the handle and support slide rail of thedrawer. Thus, the

drawer must be designed to endure these repetitive stresses.The force balance around the drawer

and handle system cans be expressed as:

Because the stress of the drawer and handle system depends on the food weight, the life stress

model (LS model) can be modified as follows:

TF=A ( S )−n =A ( F draw )−λ= A ( μW load )−λ (8-23)

The acceleration factor (AF) can be derived as

n λ λ λ
S F μW 1 W1
S0 ( ) ( ) ( ) ( )
AF = 1 = 1 =
F0 μW 0
=
W0 (8-24)
 (a) Design concept of mechanical drawer and handle system

(b) Parameter diagram of drawer and handle system

Fig.8.51. Functional design concept of the drawer and handle system

The normalnumber of operating cycles for one day was approximately 5; the worst case was 9.

Underthe worst case, the objective drawer open/close cycles for ten years would be 32,850 cycles.

For the worst case, the food weight force on the handle of the drawer was 0.34 kN. Theapplied food

weight force for the ALT was 0.68 kN. With a quotient, n, of 2, the total AF was approximately 4.0

using Eq. (8-24).The parameter design criterion of the newly designed drawer can be more than the

targetlife of B1 of 10 years. Assuming the shape parameter β was 2.0, the test cyclesand test sample
numberscalculated in Eq. (7-35)were 67,000 cycles and 3 pieces,respectively. The ALT was designed

to ensure a B1 life of 10 years with about a 60% level ofconfidence that it would fail less than once

during 67,000 cycles.

(a) ALT equipment and controller

(b) Duty cycles of repetitive food weight force on the drawer

Fig.8.52. ALT equipment and duty cycles.

Figure 8.52 shows ALT equipment and duty cycles for the repetitive food weight force, Fdraw.For the

ALT experiments, the control panel on top of the testing equipment started andstopped the drawer
during the mission cycles. The food load, F, was controlled by theaccelerated weight load in the

drawer storage. When a button on the control panel waspushed, mechanical arms and hands pushed

and pulled the drawer.

(a) Failed product in field (b) Failed sample in first accelerated life testing

Fig.8.53.Failed products in field and first ALT

 Fig.8.54. Failed slide rails in second ALT

Fig.8.55. Field data and results of 1st ALT on Weibull chart.

The fracture of the drawer in the first and second ALTs occurred in the handle and slide rails (Figure

8.53(b) and Figure 8.54). These design flawsin the handle and slide rails can result in a fracture when

the repetitive food load is applied. To prevent the fracture problem and release the repetitive stresses,

the handle and slide rails were redesigned. The corrective action plan for the design parameters

included: (1) increasing the width of the reinforced handle, C1, from 90mm to 122mm; (2) increasing

the handle hooker size, C2, from 8mm to 19mm; (3) increasing the rail fastening screw number, C3,

from 1 to 2; (4) adding an inner chamber and plastic material, C4, from HIPS to ABS; (5) thickening

the boss, C5, from 2.0mm to 3.0mm; (6) adding a new support rib, C6 (Table 8.9).

The parameter design criterion of the newly designed samples was more than the reliability target life,

B1, of ten years. The confirmed value, β, on the Weibull chart in Figure 8.55 was 3.1. The reduction

factor R also is 0.0014 from the acceleration factor = 4 and shape parameter = 3.13. Consequently,

we know that this parameter ALT is effective to save the testing time and sample size.

For the second ALT, the test cycles and sample sizerecalculatedin Eq. (7-35)were 32,000 and 3

pieces, respectively. In the third ALT, no problems were found with the drawer after 32,000 cycles and

65,000 cycles. We therefore concluded that the modified design parameters were effective.Figure

8.56 shows the results of the 1 stALT and 3rdALT plotted in a Weibull chart. Table 8.10 provides a

summary of the ALT results. With the improved design parameters, B1 life of the samples in the third

ALT was lengthened to more than 10.0 years.

 Table 8.9 Redesigned handle and right/left slide rail

Handle Right/left slide rail

C3: Rail fastening screw number 1→2 (2ndALT)

C4: Chamfer: Corner chamfer
C1: Width L90mm → L122mm (1stALT)
Plastic material HIPS → ABS (2ndALT)
C2: Width L8mm → L19mm (1stALT)
C5: Boss thickness 2.0 → 3.0 mm (2ndALT)
C6: New support rib (2ndALT)
Fig.8.56.Results of 1st ALT and 3rd ALT plotted in Weibull chart
 Table 8.10 Results of ALTs

1st ALT 2nd ALT 3rd ALT

Initial Design Second Design Final Design

In 32,000
7,500 cycles: 2/3 crack 32,000 cycles: 3/3 OK
cycles, 16,000 cycles: 2/3 crack
12,000 cycles: 12,000No problem 65,000 cycles: 3/3 OK
fracture is less
than one.

Hinge kit
-
structure

Rib1: new support rib

Material and Width1: L90 →L122 boss: 2.0 → 3.0 mm
specification Width2: L8→L19.0 Chamfer1: Corner
Material: HIPS →ABS
8-7 Compressor suction reed valve

Fig. 8.57. Fracture of the compressor suction reed valve in the field.

In the field, the suction reed valve in the compressor of the commercial refrigerator had been

fracturing, causingloss of the cooling function(Figure 8.57). The data on the failed products in the field

were important for understanding how consumersused the refrigerators and pinpointing design

changes that needed to be made to the product.The suction reed valves open and close to allow

refrigerant to flow into the compressor during the intake cycle of the piston.Due to design flaws and

repetitive stresses, the suction reed valves of domestic refrigerator compressors used in the field were

crackingand fracturing, leading to failure of the compressor.

The fracture started in the void of the suction reed valve and propagated to the end (Figure 8.58).

Specific customer usage conditions and load patterns leading to the failures were unknown. Because

thecompressor would lock up when the valve failed, the function of refrigerator was lost and

customers would ask to have therefrigerator replaced. To solve this problem, it was very important to

reproduce the field failure mode of the suction reedvalve in the laboratory.
Fig. 8.58 Fractography of the compressor suction reed valve on SEM.
(a) Compressor system in a vapor-compression refrigeration cycle

(b) Parameter diagram of compressor system

Fig. 8.59.Schematic diagram for a compressor system.

(a) Mechanical parts of the reciprocating compressor: crankshaft, piston, valve plate (1),suction reed

valve (2)

(b) Valve plate (1) and suction reed valve (2)

Fig. 8.60.Reciprocating compressor and parts.

A refrigerator compressor assembly is a simple mechanical system that operates according to the

basic principles of thermodynamics.The compressor receives refrigerant from the low-side

(evaporator) and then compresses and transfers it to thehigh-side (condenser) of the system. Most
compressor manufacturers are making every effort to develop more efficient,high-volumetric

compressors. For these applications, the compressor needs to be designed robustly to operate under

awide range of customer usage conditions. The compressor assembly in the refrigerator in question

consists of many mechanicalparts, including the crankshaft, piston, valve plate (1), and suction reed

valve (2) (see Figure 8.59 and 8.60).

Fig.8.61. Structure of suction reed and valve plate

Analysis of the failed compressors from the field led to the postulate that there were two structural

design flaws: (1) thesuction reed valve had an overlap with the valve plate; and (2) the valve plate had

a sharp edge (Figure8.61). When the suction reed valveimpacted the valve plate over a long enough

period of time, it would fracture.

The stress of the compressor depends on the pressuredifference suction pressure, Psuc, and

discharge pressure, Pdis. That is,

ΔP=Pdis −P suc ≃P c−P e (8-25)

For a refrigeration system, the time-to-failure, TF, can be modified as

Ea Ea
−n
TF=A ( S ) exp ( )
kT
−λ
=A ( ΔP ) exp ( )
kT
(8-26)
The acceleration factor (AF) can be modified to include the load from Eq. (8-26):

n λ
S Ea 1 1 ΔP 1 Ea 1 1
AF = 1
S0( )[ ( −
k T 0 T1
=
)] ( ) [ (
ΔP0
−
k T0 T 1 )] (8-27)

The system was subjected to 22 on–off cycles per day under normal operating conditions. A worst

case scenariowas also simulated with 98 on–off cycles per day. Under the worst case conditions, the

compressor operation for 10 yearswould be 357,700 cycles.

From the test data of the worst case, normal pressure was 1.27 MPa and the compressor dome

temperature was 90 C. Foraccelerated life testing, the acceleration factor (AF) for pressure was 2.94

MPa and the compressor dome temperature was120 C. With a quotient, n, of 2, the total AF was

calculated using Eq. (8-26) to be 20.9 (See Table 8.11).

Table 8.11 ALT conditions in a vapor-compression cycles

Worst case, ALT,

System conditions AF
gauge gauge

High side 13.0 30.0

5.3 
Pressure, kg/cm2 Low side 0.0 0.0
(=(/)2)
ΔP 13 30

Temp., ℃ Dome Temp. 90 120 3.92 

Total AF (=  x ) - 20.9

With a shape parameter,, of 1.9, the test cycles and test sample numberscalculatedin Eq. (7-35)

were 40,000cycles and 20 pieces, respectively. The ALT was designed to assure a B1 of 10 years

with about a 60% level of confidence that nounit would fail during 40,000 cycles.
 Fig. 8.62. Equipment for the compressor accelerated life tests.

For the ALT experiments, a simplified vapor compression refrigeration cycle was fabricated. It

consisted of an evaporator, compressor, condenser, and capillary tube. A fan and two 60-W lamps

maintained the temperature within the insulated (fiberglass) box. A thermal switch attached on the

compressor top controlled a 51 m3/h axial fan. The test conditions and test limits were set up on the

control board. As the test began, the high-side and low-side pressures could be observed on the

pressure gauge (See Figure 8.62).

(a) Field (b) 1st ALT Results

Fig. 8.63.Failure of suction reed valve in marketplace and first ALT result.
One sample in the first ALT (n = 20) failed after 8687 cycles. The confirmed value, , based on field

data was 1.9. The shapes and locations of the failures in samples from the first ALT and the field were

similar (Figure8.63). The reduction factor R also is 0.2 from the acceleration factor = 20.9 and shape

parameter = 1.89. Consequently, we know that this parameter ALT is effective to save the testing time

and sample size.

The fracture of the suction reed valve came from its weak structure. It had the following

characteristics: (1) an overlap with the valve plate; (2) weak material (0.178t); and (3) a sharp edge on

the valve plate, previously mentioned in Figure8.61.

When the suction reed valve impacted the valve plate continually, it will suddenly fracture. The

dominant failure mode of the compressor was leakage and locking due to the cracking and fracturing

of the suction reed valve.

d = 0.73 mm 1.25mm SANDVIK 20C 0.178t (Carbon Steel)

Adding Ball Peening && Brush Process SANDVIK 7C 0.178t (Stainless Steel)

Adding tumbling process

(a) Valve plate (b) Suction reed valve

Fig.8.64. Redesigned suction reed and valve plate

Figure 8.64 shows the redesigned suction reed valve and the valve plate. The valve controls the

refrigerant gas during the process of suction and compression in the compressor. The suction reed

valve required high bending/impact fatigue properties. The modified design parameters were: (1)

increasing the trespan size of the valve plate from 0.73 mm to 1.25 mm, C1; (2) changing the material

property from carbon steel (20C) to stainless steel (7C), C2; (3) adding a ball peening and tumbling

process during the treating of suction reed valve, C3.

It would appear that the ALT methodology was valid for reproducing the failure found in the field. First,

the location and shape of the fractured suction reed valves from the field and those in the ALT results

were similar. Figure 8.65 and Table 8.12 show the graphical results of an ALT plotted in a Weibull

chart andthe summary of the results of the ALTs, respectively.

Fig. 8.65.Result of ALTs plotted in Weibull chart

 Table 8.12 Results of the ALTs.

1st ALT 2nd ALT

Initial Design Second Design

In 23,000 cycles, crack

8,687 cycles: 1/20 23,000 cycles: 60/60 OK
of suction reed valve is
8,687 cycles: 19/20 stop 29,000 cycles: 60/60 OK
less than 1

Suction reed valve and

plate structure

<Plate plate>

<Suction reed valve>

C1: Trespan size d: 0.73 mm→1.25 mm

Material and C2: Adding ball peening and rush process
specification C3: SANDVIK 20C: 0.178t →0.203t
C4: Extending tumbling: 4 h→14 h
8-8 Failure analysis and redesign of the evaporator tubing

(a) Kimchi Refrigerator b) Mechanical parts of the hinge kit system: Inner Case

(1), Evaporator tubing (2), Lokring (3), and Cotton adhesive tape (4)

Fig. 8.66.Kimchi refrigerator (a) and the cooling evaporator assembly (b).

Figure 8.66 shows the Kimchi refrigerator with the cooling aluminum evaporator tubing suggested for

cost saving. When a consumer stores the food in the refrigerator, the refrigerant flows through the

evaporator tubing in the cooling enclosure to maintain a constant temperature and preserve the

freshness of the food. To perform this function, the tube in the evaporator need to be designed to

reliably work under the operating conditions it is subjected to by the consumers who purchase and

use the Kimchi refrigerator. The evaporator tube assembly in the cooling enclosure consists of an

inner case (1), evaporator tubing (2), Lokring (3), and adhesive tape (4), as shown in Figure 8.66 (b).

In the field, the evaporator tubing in the refrigerators had been pitting, causing loss of the refrigerant in

the system and resulting in the loss of cooling in the refrigerator. The data on the failed products in the

field were important for understanding the usage environment of consumers and pinpointing design

changes that needed to be made to the product (Figure 8.67).

 (a) Pitted evaporator tube in field

(b) X-Ray Photography showing a pitting corrosion on the evaporator tube

Fig. 8.67. A damaged product after use

Field data indicated that the damaged products might have had design flaws. The design flaws

combined with the repetitive loads could cause failure. The pitted surfaces of a failed specimen from

the field were characterized by a scanning electron microscopy (SEM) and EDX spectrum (Figure

8.68). We found a concentration of the chlorine in the pitted surface (Table 8.13). When Ion Liquid

Chromatography (ILC) was used to measure the chlorine concentration, the result for the tubing

having had the cotton adhesive tape was 14 PPM. In contrast, the chlorine concentration for tubing

having had the generic transparent tape was 1.33 PPM. It was theorized that the high chlorine

concentration found on the surface must have come from the cotton adhesive tape.

As mentioned in Figure 8.66, the evaporator tubing assembly in the cooling enclosure of the Kimchi

refrigerator consists of many mechanical parts. Depending on the consumer usage conditions, the

evaporator tubing experienced repetitive thermal duty loads due to the normal on/off cycling of the

compressor to satisfy the thermal load in the refrigerator. Because the refrigerant temperatures are

often below the dew point temperature of the air, condensation can form on the external surface of the

tubing.

Table 8.13 Chemical composition of the no pitting and pitting surfaces.

No Pitting Pitting

Weight Atomic Weight Atomic

O 11.95 18.65 25.82 37.38

Al 87.29 80.74 68.28 58.61
Cl 0.33 0.23 3.69 2.41
Si 0.42 0.38 0.66 0.55
Ca 0.70 0.40
K 0.50 0.30
Na 0.34 0.34
Totals 100.00 100.00
 (a) No Pitting

(b) Pitting

Fig. 8.68.SEM fractography showing a pitting corrosion on the evaporator tube.

 Fig. 8.69.Robust design schematic of a cooling enclosure system.

Fig. 8.70.An accelerating corrosion in the crevice due to low PH, high Cl - concentration, de-

passivation and IR drop.

Figure 8.69 shows a robust design schematic overview of the cooling evaporator system. Figure 8.70

shows the failure mechanism of the crevice (or pitting) corrosion that occurs because of the reaction

between the cotton adhesive tape and the aluminum evaporator tubing. As a Kimchi refrigerator

operates, water acts as an electrolyte and will condense between the cotton adhesive tape and the

aluminum tubing. The crevice (or pitting) corrosion will begin.

The crevice (or pitting) corrosion mechanism on the aluminum evaporator tubing can be summarized

as: (1) passive film breakdown by Cl - attack; (2) rapid metal dissolution: Al→Al+3 + 3e-; (3) electro-

migration of Cl into pit; (4) acidification by hydrolysis reaction: Al +3 + 3H2O→Al(OH)3↓+ 3H+; (5) large

cathode: external surface, small anode area: pit; (6) the large voltage drop (i.e., ‘‘IR” drop, according

to Ohm’s Law V = I  R, where R is the equivalent path resistance and I is the average current)

between the pit and the external surface is the driving force for propagation of pitting.

The number of Kimchi refrigerator operation cycles is influenced by specific consumer usage

conditions. In the Korean domestic market, the compressor can be expected to cycle on and off 22–

98 times a day to maintain the proper temperature inside the refrigerator.

Because the corrosion stress of the evaporator tubing depends on the corrosive load (F) that can be

expressed as the concentration of the chlorine, the life-stress model (LS model) can be modified as

−n −λ −λ (8-28)
TF=A ( S ) =A ( F ) = A ( Cl % )

The acceleration factor (AF) can be derived as

n λ λ
S F Cl 1 %
S0 ( ) ( ) ( )
AF = 1 = 1 =
F0 Cl 0 % (8-29)

The compressor in a Kimchi refrigerator is expected to cycle on average 22–98 times per day. With a

life cycle design point of 10 years, the Kimchi refrigerator incurs 358,000 cycles. The chlorine

concentration of the cotton adhesive tape was 14 PPM. To accelerate the pitting of the evaporator

tubing, the chlorine concentration of the cotton tape was adjusted to approximately 140 PPM by

adding some salt. Using a stress dependence of 2.0, the acceleration factor was found to be

approximately 100 in Eq. (8-27).

 (a) Kimchi refrigerators in testing with 0.2 M NaCl water solution on evaporator

(b) Duty cycles of repetitive corrosive load F

Fig. 8.71.Kimchi refrigerators in accelerated life testing and duty cycles of repetitive corrosive load F

For B1 life of 10 years, the test cycles and test sample numbers with the shape parameter =6.41

calculated in Eq. (7-35) were 4700 cycles and 18 pieces, respectively. The ALT was designed to

ensure a B1 of 10 years life with about a sixty percent level of confidence that it would fail less than

once during 4700 cycles. Figure 8.71(a) shows the Kimchi refrigerators in accelerated life testing and

an evaporator tubing in the enclosure contained a 0.2 M NaCl water solution. Figure 8.71(b) shows

the duty cycles for the corrosive force (F) due to the chlorine concentration.
(a) Failed product from the field (b) Accelerated life testing

Fig. 8.72.Failed products in field and ALT.

Fig. 8.73. Field data and results of ALT on Weibull chart.

Figure 8.72 shows the failed product from the field and from the accelerated life testing respectively.

In the photos, the shape and location of the failure in the ALT were similar to those seen in the field.

Figure 8.73 shows a graphical analysis of the ALT results and field data on a Weibull plot. These

methodologies were valid in pinpointing the weak designs responsible for failures in the field and were

supported by two findings in the data. The location and shape also, from the Weibull plot, the shape

parameters of the ALT, (1), and market data, (2), were found to be similar.

The pitting of the evaporator tubing in both the field products and the ALT test specimens occurred in

the inlet/outlet of the evaporator tubing (Figure 8.74). Based on the modified design parameters,

corrective measures taken to increase the life cycle of the evaporator tubing system included: (1)

extending the length of the contraction tube (C1) from 50.0 mm to 200.0 mm; (2) replacing the cotton

adhesive tape (C2) with the generic transparent tape.

Figure 8.75 shows a redesigned evaporator tubing with high corrosive fatigue strength. The confirmed

values of AF and  in Figure 8.73 were 100.0 and 6.41, respectively. The test cycles and sample size

recalculated in Eq. (7-35) were 5300 and 8 EA, respectively. Based on the target BX life, two ALTs

were performed to obtain the design parameters and their proper levels. In the two ALTs the outlet of

the evaporator tubing was pitted in the first test and was not pitted in the second test.

The repetitive corrosive force in combination with the high chlorine concentration of the cotton tape

and the crevice between the cotton adhesive tape and the evaporator tubing contained the condensed

water as an electrolyte may have been pitting.

With these modified parameters, the Kimchi refrigerator can reserve the food for a longer period

without failure. Figure 8.76 and Table 8.14 show the graphical results of ALT plotted in a Weibull chart

and the summary of the results of the ALTs, respectively. Over the course of the two ALTs the B1 life of

the samples increased by over 10.0 years.

Fig. 8.74. Structure of pitting the corrosion tubing in field and the ALT test specimens
 Fig. 8.75. A redesigned evaporator tubing

Fig. 8.76. Results of ALT plotted in Weibull chart

 Table 8.14 Results of ALT.

1st ALT 2nd ALT

Initial Design Second Design

In 5300 cycles, 1130 cycles: 1/18 pitting

Corrosion 1160 cycles: 2/18 pitting
5300 cycles: 8/8 OK
of evaporator pipe is 1680 cycles: 4/18 pitting
less than 1 1680 cycles: 11/18 OK

Evaporator pipe
structure

Length of the contraction tube,

Material and C1: 50.0 mm → 200.0 mm
specification Adhesive tape type,
C2: cotton type → generic transparent tape
8-9 Compressor with redesigned rotor and stator

Fig. 8.77. Reciprocating compressors with redesigned rotor and stator

Fig. 8.78. Parameter diagram of refrigeration cycle

A refrigerator system, which operates using the basic principles of thermodynamics, consists of a

compressor, a condenser, a capillary tube, and an evaporator. The vapor compression refrigeration

cycle receives work from the compressor and transfers heat from the evaporator to the condenser.

The main function of the refrigerator is to provide cold air from the evaporator to the freezer and

refrigerator compartments. Consequently, it keeps the stored food fresh.

To improve its energy efficiency, designer would choose the good performance of compressor. Figure

8.77 shows a reciprocating compressor with redesigned rotor and stator. The redesign was developed

to improve the energy efficiency and reduce the noise from the compressors in a side-by-side (SBS)

refrigerator. For these applications, the compressor needed to be designed robustly to operate under

a wide range of consumer usage conditions (Figure 8.78).

(a) Compressor stopping noise recorded with torso

(b) Reciprocating compressor and the design flaws of suspension spring

Fig. 8.79.Stopping noise of the reciprocating compressor.

As seen in Figure 8.79, the reciprocating compressor in the refrigerators had been making noise in

the field, causing the consumer to request replacement of their refrigerator. One of the specific causes

of compressor failure during operation was the compressor suspension spring. When the sound level

during compressor shutdown of problematic refrigerators in the field was recorded, the result was

approximately 46 dB (6.2 sones). The design flaws of the suspension spring in the problematic

compressor were the number of turns and the mounting spring diameter. When the compressor would

stop suddenly, the spring sometimes would not grab the stator frame tightly and would cause the

noise.

After identifying the missing control parameters related to the newly designed compressor system, it

was important to modify the defective compressor either through redesign of components or change

the material used in the components. Failure analysis of marketplace data and accelerated life testing

(ALT) can help to confirm the missing key control parameters and their levels in a newly designed

compressor system.

In a refrigeration cycle design, it is necessary to determine both the condensing pressure, Pc, and the

evaporating pressure, Pe. One indicator of the internal stresses on components in a compressor

depends on the pressure difference between suction pressure, Psuc, and discharge pressure, Pdis,

previously mentioned in Eq. (8-16).

The acceleration factor (AF) can be derived as,

n λ
S Ea 1 1 ΔP 1 Ea 1 1
AF = 1
S0 ( )[ ( −
k T 0 T1
=
)] ( ) [ (
ΔP0
−
k T0 T 1 )] (8-30)

The normal number of operating cycles for 1 day was approximately 24; the worst case was 74.

Under the worst case, the objective compressor cycles for 10 years would be 270,100 cycles. From

the ASHRAE Handbook test data for R600a, the normal pressure was 0.40 MPa at 42 C and the

compressor dome temperature was 64 C. For the accelerated testing, the acceleration factor (AF) for

pressure at 1.96 MPa was 12.6 and for the compressor with a 110 C dome temperature was 2.31

with a quotient, n, of 2. The total AF was approximately 29.2 (Table 8.15).

 Table 8.15 ALT conditions in a vapor compression cycles for R600a.

System conditions Worst case ALT AF

High side 0.40 1.39

Pressure, MPa Low side 0.02 0.4

ΔP 0.38 1.35
12.6 
Temp., ℃ Dome Temp. 64 110 2.31 

Total AF (=  X ) - 29.3

(a) Duty cycles of repetitive pressure difference on the compressor.

(b) Equipment used in Accelerated life testing.

Fig. 8.80. Duty cycles and equipment used in Accelerated life testing
The parameter design criterion of the newly designed compressor can be more than the target life of

B1 10 years. Assuming the shape parameter  was 1.9, the test cycles and test sample numbers

calculated in Eq. (7-35) were 9300 cycles and 100 pieces, respectively. The ALT was designed to

ensure a B1 of 10 years life with about a 60% level of confidence that it would fail less than once

during 9300 cycles. Fig. 8.80(a) shows the duty cycles for the repetitive pressure difference P. For

the ALT experiments, a simplified vapor compression refrigeration system was fabricated (See Fig.

8.80(b)).

(a) Noise in ALT equipment

 (b) Noise in refrigerator

Fig. 8.81. Failed products in first ALT

Figure 8.81 shows the stopping noise and vibration of a compressor from the accelerated life testing.

In the chart, the peak noise level and vibration of a normal sample in the compressor were 52 dB and

0.08 G when it stopped. On the other hand, for the failed sample #1, the peak noise levels and

vibration were 65 dB and 0.52 G. For the failed sample #2, the peak noise levels and vibration were

70 dB and 0.60 G. Considering that the vibration specifications called for less than 0.2 G, the failed

sample vibrations violated the specification. When the problematic samples in ALT equipment were

mounted on the test refrigerator, the vibration also was reproduced with 0.25 G and violated the

specification. In the field consumer would request the failed samples to be replaced. Figure 8.82

represents the graphical analysis of the ALT results and field data on a Weibull plot. For the shape

parameter, the estimated value on the chart was 1.9.

Fig. 8.82 Field data and results of ALT on Weibull chart

(a) Modified inspection jig

 (b) Gap between the stator frame and the upper shell

Fig. 8.83.Modified inspection jig and gaps.

When the failed samples were cut apart, a scratch was found inside the upper shell of compressor

where the stator frame had hit the shell. The gap between the frame and the shell was measured to

be 2.9 mm. The design gap specification should have been more than 6 mm to avoid the compressor

hitting the shell for the worst case. It was concluded that the stopping noise came from the hitting (or

interference) between the stator frame and the upper shell. Thus, the tests pinpointed the design

flaws in compressor (See Figure 8.83 a). For the shape parameter, the estimated value on the chart

was 1.9 from the graphical analysis of the ALT results and field data on a Weibull plot. The vital

missing parameter in the design phase of the ALT was a gap between the stator frame and the upper

shell. These design flaws may make noise when the compressor stops suddenly. To reduce the noise

problems in the frame, the shape of the stator frame were redesigned. As the test setup of the

compressor assembly was modified to have more than a 6 mm gap, the gap size increased from 2.9

mm to 7.5 mm (Fig. 8.82b and 8.84).

 Fig. 8.84.Redesigned stator frame in second ALT.

The parameter design criterion of the newly designed samples was more than the target life, B1, of 10

years. The confirmed value, , on the Weibull chart was 1.9. When the second ALT proceeded, the

test cycles and sample size recalculated in Eq. (7-35) were 9300 and 100 pieces, respectively. In the

second ALT, no problems were found with the compressor in 9300 cycles and 20,000 cycles. We

expect that the modified design parameters are effective.

Table 8.16 provides a summary of the ALT results. Figure 8.85 shows the results of ALT plotted in a

Weibull chart. With the improved design parameters, the B1 life of the samples in the second ALT

lengthens more than 10.0 years.

Fig. 8.85. Results of ALT plotted in Weibull chart

 Table 8.16 Results of ALT.

1st ALT 2nd ALT

Initial Design Second Design

In 9300 cycles, locking is less 100 cycles: 2/100 noise 9300 cycles: 100/100 OK

than 1 100 cycles: 98/100 OK 20,000 cycles: 100/100 OK

Compressor structure

Material and specification C1: Modification of the frame shape

8-10 French refrigerator drawer system

Fig. 8.86 French refrigerator and drawer assembly: Vegetable Box (1), guide rail (2), center support

(3)

Figure 8.86 shows the French refrigerator with the newly designed drawer system. When a consumer

put food inside the refrigerator, they want to have convenient access to it and have the food stay

fresh. For this to occur, the draw system needs to be designed to withstand the operating conditions it

is subjected to by users. The drawer assembly consists of a box, left/right of the guide rail, and a

support center, as shown in Figure 8.86 (b).

Fig. 8.87.A damaged products after use.

In the field, parts of the drawer system of a French refrigerator were failing due to cracking and

fracturing under unknown consumer usage conditions. Thus, the data on the failed products in the

field were important for understanding the usage environment of consumers and helping to pinpoint

design changes that needed to be made in the product (Figure 8.87).

Field data indicated that the damaged products might have had structural design flaws, including

sharp corner angles and weak ribs that resulted in stress risers in high stress areas. These design

flaws that were combined with the repetitive loads on the drawer system could cause a crack to occur,

and thus cause failure.

The drawer assembly consists of many mechanical structural parts. Depending on the consumer

usage conditions, the drawer assembly receives repetitive mechanical loads when the drawer is open

and closed. Putting and storing food in the drawer involves two mechanical processes: (1) the

consumer opens the drawer to store or take out the stored food and (2) they then close the drawer by

force.

Figure 8.88 shows the robust design schematic overview and the functional design concept of the

drawer system. As the consumer stores the food, the drawer system helps to keep the food fresh. The

stress due to the weight load of the food is concentrated on the drawer box and its support rails. And

thus it is important to overcome these repetitive stresses when designing the drawer.

The number of drawer open and close cycles will be influenced by consumer usage conditions. In the

United States, the typical consumer requires the drawer system of a French refrigerator to open and

close between five and nine times a day.

The force balance around the drawer system can be represented

F box =μ W load (8-31)

as Because the stress of the drawer system depends on the applied force from the foods weight, the

life-stress model (LS model) can be expressed as

TF=A ( S ) =A ( F box )− λ= A ( μW ; oad )− λ

−n
(8-32)

(a) Design concept of mechanical drawer system

(b) Robust design schematic of drawer system.

Fig. 8.88. Design concept and robust design schematic of mechanical drawer system
The acceleration factor (AF) can be derived as

n λ λ λ
S F μW 1 W1
S0 ( ) ( ) ( ) ( )
AF = 1 = 1 =
F0 μW 0
=
W0 (8-33)

The open and closing of the drawer system occurs an estimated average five to nine times per day.

With a life cycle design point of 10 years, the drawer would occur about 32,900 usage cycles. For the

worst case, the weight force on the drawer is 0.59 kN which is the maximum force applied by the

typical consumer. The applied weight force for the ALT was 1.17 kN. Using a stress dependence of

2.0, the acceleration factor was found to be approximately 4.0 using Eq. (8-31).

For B1 life, the test cycles and test sample numbers calculated in Eq. (7-35) were 22,000 cycles and

six pieces, respectively. The ALT was designed to ensure a B1 life of 10 years life with about a 60%

level of confidence that it would fail less than once during 22,000 cycles.

(a) Equipment used in accelerated life testing.

 (b) Duty cycles of repetitive load F on the drawer system

Fig. 8.89. Equipment used in accelerated life testing and duty cycles

Figure 8.89 shows the experimental setup of the ALT with the test equipment and the duty cycles for

the opening and closing force F. The control panel on the top of the testing equipment started and

stopped the equipment, and indicated the completed test cycles and the test periods, such as sample

on/off time. The drawer opening and closing force, F, was controlled by the accelerated weight load in

the drawer system. When the start button in the controller panel gave the start signal, the simple

hand-shaped arms held the drawer system. The arms then pushed and pulled the drawer with the

accelerated weight force (1.17 kN).

Fig. 8.90. Failed products in field (left) and 2nd ALT (right)
Figure 8.90 shows the failed product from the field and from the accelerated life testing, respectively.

In the photos, the shape and location of the failure in the ALT were similar to those seen in the field.

Figure 8.91 represents the graphical analysis of the ALT results and field data on a Weibull plot.

Fig. 8.91. Field data and results of ALT on Weibull chart.

The shape parameter in the first ALT was estimated at 2.0. For the final design, the shape parameter

was obtained from the Weibull plot and was determined to be 3.6. These methodologies were valid in

pinpointing the weak designs responsible for failures in the field and were supported by two findings in

the data. In the photo, the shape and location of the broken pieces in the failed market product are
identical to those in the ALT results. And the shape parameters of the ALT (1) and market data (2)

were found to be similar from the Weibull plot. The reduction factor R also is 0.034 from the

acceleration factor = 4.0 and shape parameter = 1.9. Consequently, we know that this parameter ALT

is effective to save the testing time and sample size.

Fig. 8.92. Structural problems of the left, right, and center support rails in Loading.
Initially when the accelerated load of 12 kg was put into drawer, the center support rail was bent and

the rollers on the left and right rail were broken away (Figure 8.92). The design flaws of the bent

center rail and the breakaway roller resulted in drawers not sliding. The rail systems could be

corrected by adding reinforced ribs on the center support rail as well as extruding the roller support to

7 mm (Table 8.17).

Fig. 8.93.Structure of failing drawer system in field.

The fracture of the drawer in both the field products and the ALT test specimens occurred in the

intersection areas of the box and its cover (Figure 8.93). The repetitive food loading forces in

combination with the structural design flaws may have caused the fracturing of the drawer. The design

flaws of no corner rounding and poorly enforced ribs resulted in the high stress areas. These flaws

can be corrected by implementing the fillets and thickening the enforced ribs (Table 8.17).

Fig. 8.94. Structural problems of the left/right rail (left) and center support rails (right) in 1 st ALT.
 Table 8.17 Redesigned box and center support rail.

Box Center support rail

C1: Rib1 T2.0 mm → T3.0 mm C3: Rib2 New Added Rib

C2: Fillet R0.0mm → R1.0mm C4: Extending Rib1 L0.0mm → L2.0mm

Guide Rail (Left/Right)

C5: Rib3 (New Added Rib, loading test)

C6: Extruder roller L0.0mm → L7.0mm
C7: Fillet R3mm → R4mm
C8: Rib4 New added back rib
The confirmed values of AF and  in Fig. 8.91 were 4.0 and 3.6, respectively. The test cycles and

sample size recalculated in Eq. (7-35) were 22,000 and six pieces, respectively. Based on the

targeted Bx, three ALTs were performed to obtain the design parameters and their proper levels. Due

to repetitive i stresses, the left and right rails of the drawer system cracked (Figure 8.94(a)) and the

roller of the support center was sunken (Figure 8.94(b)) in the first ALTs. Thus, a rib extruded 2 mm

from the center support rail. And the left and light rail systems were corrected by design changes such

as corner rounding and inserting ribs (Table 8.17).

Table 8.18 gives a summary of the results of the ALTs, respectively. . Figure 8.95 shows the results of

ALT plotted in a Weibull chart. With these modified parameters, the French refrigerator can smoothly

open and close the drawers for a longer period without failure.

Table 8.18 Results of ALT.

1st ALT 2nd ALT 3rd ALT

Initial Design Second Design Final Design

In 22,000
cycles, 15,000 cycles: 2/6 Fail
3800 cycles: 3/6 Fail 22,000 cycles: 6/6 OK
fracturing 28,000 cycles: 1/6 Fail
3800 cycles: 3/6 OK 45,000 cycles: 6/6 OK
is less 28,000 cycles: 3/6 OK
than 1

Drawer
Structure

Redesigned rail
Material C1: Rib3 new added rib
Redesigned box
and C2: Extrude1: L 0.0mm→L
C5: Rib1 T2.0mm→T3.0mm
specificati 7.0mm
C6: Fillet1 R0.0mm→R1.0mm
on C3: Fillet2: R3mm→R4mm
C4: Rib4: new added back
Fig. 8.95 Results of ALT plotted in Weibull chart
Chapter 9
Parametric ALT: A Powerful Tool for future engineering
development

Seong-woo Woo

Email: [email protected]

Reliability Association of Korea, Seoul, Korea (Republic of)

Abstracts: This chapter will discuss the concept of system engineering. Mechanical product is

developing under the principle of system engineering. Product reliability become one of the product

requirements. So when mechanical system with the sophisticated technology put into plan, product

reliability in the established design process should be implemented with reliability methodology like

parameter ALT. If not, new product will be faced with quality problems. To settle down them, company

will pay the quality costs.

Keywords: System engineering (SE)

 Fig. 9.1 The Systems Engineering Process [1]

Today new product such as automobiles, construction equipment, machine tools, airplane, domestic

appliance, and bridges were designed under the principle of System Engineering (SE). SE is an

interdisciplinary field of engineering on how complex engineering projects should be designed and

managed over product life cycles. Issues such as reliability, logistics, coordination of different teams

(requirement management), evaluation measurements and different disciplines become more difficult

when dealing with large but complex projects. In systems engineering all aspects of a system are

considered, and integrated into a whole (Figure 9.1).

Company also would like to survive the limitless competition through the new technology

development. Because there are a lot of things in the design phase for short developing duration,

products often have inherent design problems. Due to their reliability disasters, engineers have

become a critical factor to consider reliability in designing the product. The reliability should consider

factor early in the design phase. The basic question is how to consider the reliability concept in the

established design process. The company should have new quantitative developing process that

considers the reliability factor in parallel with the established design process. If not, company will
confront numerous recalls in the market.

Reliability disasters might come from the faulty components that have the missing design parameters

not concerned in the process of R&D. When subjected to the wearout stress or overstress under the

end user operating or environmental conditions, the problematic components mounted in product

cause failure. New product should be developed in the quantitative developing process that is

included in 1) reliability target, 2) reliability testing and Weibull analysis, 3) design feedback, and 4)

the analysis of the field failure data.

A new methodology for reliability design therefore is required to prevent the reliability disasters in the

mechanical/civil system. The traditional qualitative methods – Capability Maturity Model Integration

(CMMI), FMEA and FTA is to look for the design problems on the documents. They only carry out to

gather the design ideas or past experience by the representatives – planning, design, and production.

Consequently, they often miss the chance to find a critical data in the design phase. The parametric

ALT would be an alternative quantitative method to search out the missing design data because it

uses the ALT plan, load analysis, and accelerated testing.

All mechanical products are fabricated from a multiple of structure to carry out the customer-required

functions, which will tend to degrade or break down abruptly by random loads in the field. When

mechanical/civil products are subjected to random loads, they would start the void in material (or

design failures), propagate, and rupture it. If failure such as fatigue or fracture occurs, the product

may no longer meet the required product functionality. To avoid failure, mechanical system should be

designed to robustly withstand a variety of loads in a lifetime.

To accomplish the reliability design of modules in mechanical/civil product, the basic concepts of

parametric ALT were discussed: – 1) Setting overall parametric ALT plan of product, 2) Failure

mechanics, design and reliability testing, 3) Parametric accelerated life testing with an acceleration

factor, and 4) Derivation of the sample size equation in Chapter 7. The failure modes and

mechanisms of the mechanical system in the field and parametric ALT may come from the missing

design parameters or design flaws not considered in the design process. In the design phase the

mechanical products should reveal the design flaws and establish action plans. To do it, the detail

case studies on the design flaws were suggested in Chapter 8.

 Fig. 9.2 Reliability–embedded design process

With the study of missing parameters in the design phase of the mechanical system, the parametric

ALTs can be successful in proving a more reliable product or module with significantly longer life. The

product or module with the modified design parameters will meet the reliability target. This reliability

design methodologies will provide the reliability quantitative (RQ) test specifications of a mechanical

structure that includes several assembly subjected to repetitive stresses under customer usage

conditions. As a result, reliability-embedded design process will save the design modification cost

because the problem number decreases (Figure 9.2).

There are a variety of other structural systems – appliance, automobiles, airplane, machine tool,

construction equipment, washing machines, and vacuum cleaners. For improving the reliability design

of these systems, the missing controllable design parameters need to be identified to meet the

targeted product (or module) reliability. And these principles of parametric ALT also are applicable to

the area of civil engineering to design the construction structure. It is recommended that the missing

controllable design parameters on these systems be further studied for reliability design of product in
lifetime.

References

[1] Defense of Department (2001) Systems Engineering Fundamentals. Defense Acquisition

University Press: Fort Belvoir, Virginia p.31