Chapter One

Introduction

1.1 Introductory Comments

Nowadays engineers are facing new and emerging challenges. Examples of these
challenges include the intensive use of computational simulations and virtual prototyping,
the application of new technologies into complex systems, the requirements of high
quality and reliability, and the reliable decision-making under uncertain design
environments. Conventional design methods are inadequate to deal with these challenges.
Therefore, nontraditional design methods such as probabilistic engineering design have
been increasingly developed and used in industrial applications.

How to deal with uncertainty in engineering is one of the challenges. Uncertainty is

ubiquitous in any engineering systems and at any stages of product development.
Examples of uncertainty include manufacturing imprecision, usage variation, and
imperfect knowledge. To manage various uncertainties, engineers have increasingly
applied probabilistic and statistical methods as an integral part in analysis and design
activities. However, because graduating engineers typically are not readily equipped with
practical knowledge of probabilistic and statistical methods, companies have to provide
them with intensive internal training. It is essential for modern engineers to be armed
with basic probabilistic and statistical tools for solving complex engineering problems in
the face of uncertainty.

This course aims at exploring computational methodologies for engineering design under
uncertainty. It is intended for undergraduate seniors and graduate students who are
interested in statistical/probabilistic methods and design optimization in engineering
analysis and design. It covers reliability analysis, analytical robustness assessment, robust
design, reliability-based design, and their engineering applications. Associated outcomes
include 1) an ability to model uncertainties for engineering analysis and design, 2) an
ability to apply knowledge of statistics and probability to engineering design, 3) an ability
to integrate robust design and reliability-based design with CAD/CAE simulations,
design optimization, and Design of Experiments (DOE), and 4) an ability to use
probabilistic and statistical methods for Design for Six Sigma.

The following specific topics will be covered:

1. Uncertainty modeling fundamentals of probability and statistics

2. Uncertainty analysis analysis methodologies that quantify the effect of
uncertainty on design performances
1 Chapter 1 Introduction

Reliability analysis
Monte Carlo simulation
Sensitivity analysis
Robustness assessment
3. Probabilistic design design methodologies that manage and mitigate the effect
of uncertainty
Introduction to design optimization
Reliability-based design
Robust design
Integrated probabilistic design
4. Case study in Industry (automotive, structural, and mechanical applications)

In this chapter, the basic concepts of engineering design and design process will be
discussed first, followed by the introduction to the fundamentals of probabilistic
engineering design. In the subsequent chapters, we will primarily discuss three major
topics: 1) uncertainty modeling with the application of probability theory, 2) uncertainty
analysis that quantifies the effects of uncertainty, and 3) probabilistic engineering design
that manages and mitigates the effects of uncertainty.

1.2 Engineering Design and Engineering Design Process

There are many definitions about engineering design. Two of them are listed below.

Engineering design is a process that establishes and defines solutions to new engineering
problems, which have been solved before, or new solutions to engineering problems,
which have previously been solved, in a different way. The key word is new problems or
new solutions.

Engineering design is the process of devising a system, component, or process to meet

desired needs [1]. It is a decision-making process (often iterative), in which the basic
science, mathematics, and engineering sciences are applied to convert resources
optimally to meet a stated objective. The key of this definition is decision-making.

A general engineering design process involves the following major phases [1-4].

Phase 1: Problem definition to collect customer needs, clarify design objectives,

establish user requirements, and identify constraints.

Phase 2: Conceptual design to establish functions and design specifications, generate

design concepts (alternatives), evaluate design concepts, and select the best design
concepts.

1
2 Probabilistic Engineering Design

Phase 3: Embodiment design to engineer a solution principle for the selected design
concepts by determining the general arrangements and preliminary shapes and materials
of all components. Embodiment design is also called preliminary design.

Phase 4: Detail design to specify all the details of the final design and produce
manufacturing drawings and documentation.

The engineering design process is demonstrated in Fig. 1.1. The input of the design
process is the customer needs, and the output of the design process is the final design,
including manufacturing specifications and all the documentations. The process is
dynamic and iterative. Rework is needed among the design phases before a satisfactory
final design is reached.

Phase 1 Phase 2
Problem clarification Conceptual design

Phase 3 Phase 4
Embodiment design Detailed design

Figure 1.1 General Design Process

1.3 Design and Analysis

Engineering analysis is the study of a design, especially for a product, to understand or
model its performance under the conditions of its normal use. It is typically performed on
potential designs before they are built or when a product does not meet an expectation.
The analysis creates an understanding that allows for improvements in the design and
corrections of performance problems. In a design process, analysis is performed to check
if an existing design satisfies all the design requirements. Table 1.1 lists the differences
between design and analysis.

For example, if a task is to determine a mechanism system that satisfies the functional
requirement y f ( x ) , in which x and y are input and output rotational displacements,
respectively. It is a design task because there are multiple solutions to this new problem,

2
3 Chapter 1 Introduction

and many decisions need to be made, such as the materials, the design options, the
geometry, dimensions, etc. Figure 1.2 shows several possible design options to the
problem, including a four-bar linkage, a cam-follower mechanism, and a pair of gears.

Table 1.1 Differences between Design and Analysis

Design Analysis
A decision making process A problem solving process

Solutions to new problems or new solutions Solutions to existing problems

to existing problems
More than one solution Only one unique solution

x y

Figure 1.2 Multiple Potential Solutions

Let us look at the reverse problem. A four-bar linkage mechanism has been identified as
shown in Figure 1.3. The task is to find the output angle y given the input angle x. This
task is an analysis problem because there is only one unique solution to this existing
system, and it is a problem solving process where algebraic equations derived from
kinematics are used to find the solution.

c
a x y

3
4 Probabilistic Engineering Design

Figure 1.3 Four-Bar Mechanism

We have seen the differences between design and analysis. In a real engineering design
process, design and analysis are also tied to each other. A design involves a number of
analyses as shown in Fig. 1.4. After having generated a number of design concepts,
engineers perform analyses on the design concepts. Then they use the analysis results to
make decisions on selecting the best design concepts in terms of design performance.
After the concept selection, engineers make more decisions in order to detail and refine
the selected design. If the design is not considered satisfactory, they will use the analysis
results to improve and update the design by making necessary changes on material
selections, configurations, component interfaces, parameters, and so on. The process
iterates until a satisfactory design is identified. During this process, numerous decisions
are made.

Initial design

Analysis

Analysis results
(Design performances)
Updated design
Yes
Satisfactory? Design results

No

Modify the design

Figure 1.4 Analysis in A Design Process

On the other hand, a single analysis may also contain other design tasks. For example,
solving a mathematical equation is an analysis problem. Designing and selecting
algorithms to solve the equation is a design problem.

1.4 Analysis Model

With the advancements of computational tools and the demand of shortening product
design to market, engineers increasingly rely on mathematical and simulation models.
These models can provide a flexible and cheap means to explore and examine design
alternatives before physical deployment. With this fundamental paradigm shift, product

4
5 Chapter 1 Introduction

development is moving toward an engineering process where decisions are heavily based
on computational simulations with decreasing physical experiments.

x Analysis Model y
y = g(x)

Figure 1.5 A Design/Analysis Model

An analysis model is shown in Fig. 1.5 and is given by

y g (x) (1.1)

In Eq. 1.1, x is a vector of input variables. x may contain design variables (e.g. the
diameter of a shaft) that can be controlled and changed during the design process, or
design parameters (e.g. the temperature of the environment) that can not be controlled. In
general, x is the mixture of design variables and design parameters. y is an output or
response variable which is dependent on x. y is usually a design performance, such as the
cost and maximum stress.

g () is the functional relationship between input x and output y. In complex engineering

design, g () may not have an analytic formula. The output is obtained through numerical
calculations or simulations. This kind of model is usually called a black-box model.
Examples of black-box models include those of finite element analysis, dynamics
simulation, and computational fluid dynamics. In product development such as a vehicle
design, sophisticated engineering computer models are eminent. Different from a
scientific model that is to fit extant data, an engineering model is primarily used to
predict future performances (behaviors) before a physical product is made.

Analysis models are important for many reasons. (1) Significant upfront design decision-
making occurs prior to the availability of physical prototypes. Such design-making relies
heavily on the predictions of design performances from the models. (2) Physical testing
can be expensive, time consuming, harmful, or even, in some situations, prohibitive. (3)
Engineers use models to gain some insights into certain phenomena, which may be
lacking from physical experiment due to measurement system limitations or its
practicality.

In this class, we will focus on model-based engineering design where engineers use
analysis models to predict product performances and make design decisions.

5
6 Probabilistic Engineering Design

1.5 Where Does Uncertainty Come From?

Uncertainty can be viewed as the difference between the present state of knowledge and
the complete knowledge (see Fig. 1.6). In the context of model-based design, uncertainty
is the difference between the model prediction and reality. Uncertainty is usually
classified into aleatory and epistemic types.
Uncertainty
Epistemic uncertainty Aleatory uncertainty

Complete Present state of Complete

ignorance knowledge knowledge
Knowledge

Model structure uncertainty Parameter uncertainty

Figure 1.6 Uncertainty types

Aleatory uncertainty, also termed as objective or stochastic uncertainty, describes the

inherent variation associated with a physical system or environment. This kind of
uncertainty arises from complex physical phenomena, including variations in
temperature, material properties, usage conditions, dimensions of a product caused by
manufacturing imprecision, etc. Since uncertainty is a result from natural variability,
aleatory uncertain is irreducible. Aleatory uncertainty is usually modeled as randomly
distributed quantities that can take values in an established or known range, but the exact
values will vary by chance from unit to unit or from time to time.

Epistemic uncertainty is derived from some level of ignorance or incomplete information

about a physical system or environment. Epistemic uncertainty is subjective in nature and
arises primarily from limited knowledge. The key feature is that the fundamental source
of epistemic uncertainty is incomplete information or incomplete knowledge of some
characteristics of the system or the environment. In short, epistemic uncertainty is due to
the lack of knowledge. Epistemic uncertainty is reducible. The degree of uncertainty can
be reduced if more knowledge is acquired or more data are collected.

Uncertainty is also conveniently classified into parameter uncertainty and model

structure uncertainty. Parameter uncertainty is due to limited information or the inherent
variation in the physical system or environment in estimating the characteristics of
parameters. As shown in Fig. 1.6, uncertainty associated with a parameter can be aleatory
(due to the inherent variation) or epistemic (due to limited information). For example, if
the diameter of a shaft varies around its nominal value within the specified tolerance with
a normal distribution, the parameter uncertainty associated with the diameter is aleatory.
In another example, engineers do not have enough information about the coefficient of
friction between two materials. What they can estimate is that the coefficient of friction is
within a range between 0.1 and 0.35. But they do not know how the values of the

6
7 Chapter 1 Introduction

coefficient of friction are distributed within the range. Due to the lack of knowledge, the
parameter uncertainty associated with the coefficient of friction is epistemic. If engineers
perform more analyses or experiments, the estimate of coefficient of friction will be more
precise and the range will be narrower or reduced to a single value. The epistemic
parameter uncertainty will then be reduced or eliminated.
Model structure uncertainty is the uncertainty in the model structure itself, including
uncertainty in the validity of the assumptions underlying the model. The uncertainty
associated with a model structure is a special type of epistemic uncertainty, which comes
from assumptions or a lack of knowledge in the model building process.

The classification of uncertainty is summarized in Fig. 1.7.

Uncertainty

Aleatory Uncertainty Epistemic Uncertainty

Aleatory Parameter Epistemic Parameter Model Structure

Uncertainty Uncertainty Uncertainty

Figure 1.7 Classification of Uncertainty

To better understand the concept of uncertainty, let us look at a simple beam design
example (see Fig. 1.8). The design variables that are to be determined are the cross-
sectional dimensions, including widths b1 and b2, heights h1 and h2, and lengths l1 and l2.
A vertical external force P applies at the tip of the beam. The yield strength of the beam
material is S.

7
8 Probabilistic Engineering Design

P
A A-A
C B

h1 h2

b1
l2 l1 A
b2
Figure 1.8 A Cantilever Beam

To make the design feasible, it is necessary to calculate the maximum stress max and
make sure that it is less than the yield strength S. The analytical model y g (x) for the
design margin is defined by the difference between the strength and stress, namely,

6P(l1 l2 )
y g (x) S max S (1.2)
b2h22

where x b1, b2 , h1, h2 , l1, l2 , S and the maximum stress max is derived from the basic
beam theory as

6 P(l1 l2 )
max (1.3)
b2h22

Parameter uncertainty: Due to the manufacturing imprecision, the dimension variables

b2, h2, l1, and l2 in Eq. 1.2 are random variables, varying within the tolerance around their
nominal values. All the dimension variables are normally distributed, and the parameter
uncertainty associated with them is aleatory. Due to the varying operational environment,
the external force P is also an uncertain variable. If there are adequate data (samples), P
can be described mathematically with a random distribution. In this case, P has aleatory
parameter uncertainty. However, if the data are scarce, P may not be precisely modeled
by a random distribution. Then epistemic uncertainty also exits. Similarly, the material
property, the strength S, is also subject to uncertainty. For the same reason, the parameter
uncertainty of S may be either aleatory or epistemic. Because of the input parameter
uncertainties in x, the model prediction y is also subject to uncertainty.

Model structure uncertainty: Eq. 1.2 is derived from the basic beam theory based on
several idealized assumptions, such as (1) the material is isotropic and homogenous; the
material is also linearly elastic; (2) plane sections remain plane under a load; (3) the
moduli of elasticity in tension and compression are identical; and (4) the support of the
beam at C is perfectly rigid. The assumptions may not be completely valid, and therefore
the prediction of the design margin in Eq. 1.2 will be different from the true value. This
indicates the existence of model structure uncertainty. Model structure uncertainty is a

8
9 Chapter 1 Introduction

special type of epistemic uncertainty. If a more sophisticated model is used, the

prediction will be closer to the reality, and the model uncertainty will then be reduced.

It is worthwhile to note that numerical errors also exist in model-based predictions. If

numerical methods such as finite element analysis are used to solve the maximum stress
max , the solution may not be identical to the solution from the theoretic model in Eq.
1.3. The numerical error is the difference between the numerical solution and the
analytical solution (accurate solution).

The above concepts are further demonstrated by the following vehicle crashworthiness
design example (Fig. 1.9). Finite element models play an integral part in the vehicle
crashworthiness design, which involves uncertain parameters such as those of geometry
(shape, thickness, and tolerances), material properties (elasticity, yield strength, and
damping), and loading. Some parameters (e.g. tolerances) are random variables with
aleatory uncertainty and their probability distributions are precisely known. Other
parameters are epistemically uncertain because the knowledge about them is imprecise.
For example, the most significant epistemic parameter uncertainty is that of the contact
resistance, which is assumed to lie in an interval due to a lack of knowledge [5]. Another
significant uncertainty is the model structure uncertainty, a special type of epistemic
uncertainty, caused by the linear approximation for stress and strain and by other
assumptions [5, 6]. With the model structure uncertainty, proving ground tests (Fig. 1.9)
are required to verify that the model-based design meets the mandated crashworthiness
standards. There are also many sources of uncertainty in the vehicle manufacturing
process and proving ground tests that, in turn, induce experimental uncertainty in the test
results, which also contain both aleatory and epistemic uncertainties.

Figure 1.9 Vehicle Crash Simulation (left) and Proving Ground Test (right)
(Courtesy of Ford Motor Company)

9
10 Probabilistic Engineering Design

Probability theory is commonly used to quantify aleatory uncertainty. Quantifying and

managing epistemic uncertainty (epistemic parameter uncertainty and model structure
uncertainty) needs more advanced mathematical theories [7]. It is still an ongoing
research topic in both academia and industry. In this class, we will primarily focus on
aleatory parameter uncertainty.

1.6 What is Probabilistic Engineering Design?

As discussed above, engineers increasingly rely on analysis (simulation) models. Unlike
scientific models of nature developed to fit extant data, engineering analysis models are
intended to predict future performance of systems. Uncertainties are considerable, and
they cannot be controlled or minimized beyond modest limits. Thus, it is important to
quantify and manage uncertainty inherent in engineering design. Probabilistic
engineering design is a design methodology that meets such a need.

Probabilistic engineering design is a mathematically based design methodology for

producing high quality products. The features of probabilistic engineering design include

Using probability theory to quantify and treat uncertainties

Accommodating uncertainties and mitigating the effects of uncertainties
Ensuring high robustness, reliability, and safety
Integrating optimization with mathematical or CAE (Computer Aided
Engineering) simulation models
Combining Design of Experiments in many engineering applications

Typical methods of probabilistic engineering design include

Reliability-based design (RBD)

Reliability-based design seeks a design that has a probability of failure less than
some acceptable (invariably small) value and therefore ensures that the events that
lead to catastrophe are extremely unlikely. The focus is the higher reliability
(safety) and lower risk.

Robust design
Robust design is a method for improving the quality of a product through
minimizing the effect of uncertainty without eliminating the causes of uncertainty.
The focus is the robustness of the product performance.

Design for Six Sigma (DFSS)

Design for Six Sigma is a comprehensive approach to product development that
links business and consumer needs to critical product attributes, product
functions, detailed designs, tests, and verification. The focus is the product
quality, customer satisfaction, and competitiveness. RBD and robust design are
usually employed in DFSS.

10
11 Chapter 1 Introduction

1.7 The Effects of Uncertainty on Product Performance

Uncertainty associated with parameters, model structures, and numerical errors has a
significant impact on design performances. The ignorance or inappropriate treatment of
uncertainty may lead to

Erroneous decision-making,
Low quality, robustness, reliability, safety,
High risk,
High cost of product-life cycle,
Costly warranty,
Over-designed (conservative) products,
Low customer satisfaction, and
Catastrophic consequences

For example, if a product is not robust, the product performance will be sensitive to the
variation of system inputs. As a result, small variations of system inputs such as the
imprecision of manufacturing may lead to a large variation in product performance. A
large variation in performance means low quality and will consequently results in low
customer satisfaction. In addition, if a product is not reliable, the chance of failure will be
relatively high. Catastrophic events may occur when the product fails.

1.8 Why Does An Engineer Need to Know Probabilistic Engineering

Design?
Traditionally, engineering design has been performed on a deterministic basis as if
everything could be calculated with certainty through formulae or simulations that
modeled nature with absolute precision. Due to this belief, engineers are typically trained
through the courses where the analysis models are formulated ideally in such a way that
there are no variations. Given a deterministic input, there always exists a deterministic
output.

As we have seen previously, uncertainty exists in every engineering system. It impacts

the product performance significantly. A small variation of system input may cause a
huge quality loss. The ignorance of uncertainty may lead to catastrophic failure events.
To accommodate uncertainty, the common practice is the use of a factor of safety. It has
long been customary for engineers to modify their design with arbitrary factors of
safety so that bridges would not collapse and machines would not break down. It is
evident that the use of factors of safety may be either risky (under-designed) or
conservative (over-designed).

11
12 Probabilistic Engineering Design

With the intensive requirement of high product quality, the more complex computer
simulations are increasingly being used, and more complicated decision making is
required during a design process. The need of uncertainty consideration in engineering
design has become imperative. Probabilistic and statistical based design methods, such as
Design of Experiments (DOE), robust design, reliability-based design, and Design for Six
Sigma (DFSS), have been used in industry to meet such a need. A typical example is the
vehicle development process, where vehicle program managers are continually
challenged with tasks with the presence of uncertainties. The typical tasks include
integrating uncertain information across a large number of functional areas, assessing
program risk relative to business goals, and then making program-level decisions. In the
mean time, engineers struggle to develop design alternatives facing with uncertainties in
design and analysis models, manufacturing processes, and environment. They must
provide the program managers with credible, timely, and robust estimates of design
related vehicle performance [8]. For many engineers in other industries, applying
nondeterministic approaches to handle uncertainty has also become a part of their routine
job.

1.9 Uncertainty Management

In this class, we will discuss how to deal with uncertainty in engineering design at three
complementary levels modeling, analysis, and design. The three levels are illustrated in
Fig. 1.10.

Design process - Robust design

- Reliability-based
Level 1 design
Uncertainty Modeling - Design for Six Sigma
Level 3
Design under Uncertainty
- Probability theory
Level 2 - Reliability analysis
- Statistics
Uncertainty Analysis - Robustness assessment
- Other theories
- Sensitivity analysis

Figure 1.10 Managing Uncertainty in a Design Process

1. Level 1 Uncertainty modeling

The task of uncertainty modeling is to quantify uncertainty mathematically. Probability
theory is commonly used for this task. An uncertain quantity is described by a random
variable and is characterized by a probability distribution. Since the distribution is usually
obtained from statistical data, statistics is used to formulate the distribution. The
mathematical structures of uncertain variables at uncertainty modeling level then provide
the input to uncertainty analysis at the next level.

12
13 Chapter 1 Introduction

2. Level 2 Uncertainty analysis

The task of uncertainty analysis is to quantify the uncertainty of design performance
(model output) given the uncertainty of model input. The uncertainty of model input is
modeled at the above modeling level. Uncertainty analysis helps engineers understand
how uncertainty impacts design performance and provide them with tools to evaluate
important design characteristics, such as reliability and robustness. The knowledge from
uncertainty analysis will then be used at the next design level for managing and
mitigating the effects of uncertainty.

3. Level 3 Design under uncertainty

The task of design under uncertainty is to mitigate the effects of uncertainty by making
appropriate decisions. Depending on design needs, the focus may be on the reliability
(safety), robustness, or quality. To make the design cost-effective, the common practice is
to determine optimal design variables at the design stage without eliminating the causes
of uncertainty. The reason is that in many cases, eliminating uncertainty causes is very
expensive. It requires high precision manufacturing and strict quality control. As
indicated in Fig. 1.10, design under uncertainty is an iterative process. During this
process, the design is continually updated until satisfactory design is achieved.
Uncertainty analysis is performed for each updated design. Therefore, the design process
repeatedly calls uncertainty analysis.

1.10 Concluding Remarks

We have reviewed the general engineering design process and introduced important
concepts of uncertainty. In this class, we will primarily use probability theory to model
aleatory uncertainty in probabilistic engineering design. The methods of probabilistic
engineering design introduced in this class can be used in stages of conceptual design,
preliminary deign, and detail design. In the following chapters, we will first present the
basic probability theory from an engineering perspective, for readers who may or may not
be familiar with probability theory. Then we will discuss the basic probabilistic analysis
and design methods that are commonly practiced in industry. Equipped with the basic
tools provided in this class, students will have a basic working knowledge for uncertainty
analysis and design under uncertainty.

Reference

[1] ABET, 2000. Criteria for Accrediting Engineering Programs. New York:
Accreditation Board for Engineering and Technology, Inc.
[2] Dym, C.L., and Little P., 2004, Engineering Design, A Project-Based
Introduction, 2nd Edition, John Wiley & Sons, Inc., U.S.A.
[3] Eggert, R., Engineering Design, 2005, Pearson Prentice Hall, Upper Saddle
River, New Jersey.

13
14 Probabilistic Engineering Design

[4] Ertas, A., and Jones, J., 1996, The Engineering Design Process, John Wiley &
Sons, Inc., U.S.A.
[5] Bayarri, M.J. Berger, J.O., Higdon, D., Kennedy, M.C., Kottas, A., Paulo, R.,
Sacks, J., Cafeo, J.A., Cavendish, J, Lin, C.H., and Tu J., 2000, A Framework for
Validation of Computer Models, Foundations for Verification and Validation in
the 21st Century Workshop, John Hopkins University/Applied Physics
Laboratory, Laurel, Maryland.
[6] Gu, L. and Yang, R. J., 2003, Computer Model Validation in Vehicle Crash
Safety Design, Proceedings of ASME 2003 Design Engineering Technical
Conferences and the Computers and Information in Engineering Conference,
Chicago, Illinois, September 1-4, 2003.
[7] Du, X., 2006, A Unified Uncertainty Analysis Framework by Probability and
Evidence Theories, 2006 ASME DETC Conferences, September 10-13, 2006,
Philadelphia, Pennsylvania.
[8] Cafeo, J.A., Donndelinger, J.A., Lust, R.V., and Mourelatos, Z.P., 2005, The
Need for Nondeterministic Approaches in Automotive Design: A Business
Perspective, in Engineering Design Reliability Handbook, edited by Nilolaidis,
E., Chiocel, D.M., and Singhal, S., CRC Press, Washington D.C.

14