Process Capability & Statistical Quality Control

Outline:
 Basic Statistics
 Process Variation
 Process Capability
 Process Control Procedures
 Variable data
 X-bar chart and R-chart
 Attribute data
 p-chart

 Acceptance Sampling
 Operating Characteristic Curve

Dr. Saydam, Lecture 1

Focus
 This technical note on statistical quality control (SQC) covers the
quantitative aspects of quality management
 SQC is a number of different techniques designed to evaluate
quality from a conformance view
 How are we doing in meeting specifications?
 SQC can be applied to both manufacturing and service
processes
 SQC techniques usually involve periodic sampling of the process
and analysis of data
 Sample size
 Number of samples
 SQC techniques are looking for variance
 Most processes produce variance in output
 we need to monitor the variance (and the mean also) and correct
processes when they get out of range

Dr. Saydam, Lecture 1

Basic Statistics

Normal Distributions have a

mean μ( ) and a standard μ
 +
deviation (σ)
99.7%
For a sample of N observations:
N
Mean X  x
i 1
i
N where:
xi = Observed value
N = Total number of observed values

 x
2
Standard Deviation i
X 
 i 1

Dr. Saydam, Lecture 1

Statistics and Probability

 SQC relies oncentral limit theorem andnormal dist .

 We establish the Upper Control Limits (UCL) and the Lower
Control Limits (LCL) with plus or minus 3 standard deviations.
Based on this we can expect 99.7% of our sample observations
to fall within these limits.
 Acceptance sampling relies on Binomial and Hyper geometric
probability concepts

  + +

/2 /2
99.7%

 = Prob. of LCL UCL

Type I error
Dr. Saydam, Lecture 1
 Basic Stats

 Using SQC,
 samples of a process output are taken, and
 sample statistics are calculated

 The purpose ofsampling is to find when the

process has changed in somenonrandom way
 The reason for the change can then be quickly
determined and corrected

 This allows us to detect changes in the actual

distribution process

Dr. Saydam, Lecture 1

Variation
Random (common ) variation is inherent in the
production process.

Assignable variation is caused by factors that

can be clearly identified and possibly managed

 Using a saw to cut 2.1 meter long boards as a

sampleprocess
 Discuss random vs. assignable variation

 Generally, when variation is reduced, quality

improves.
 It is impossible to have zero variability. T or F ?

Dr. Saydam, Lecture 1

 Taguchi’s View of Variation
Traditional view is that quality within the LS and US is good and that
the cost of quality outside this range is constant, where Taguchi
views costs as increasing as variability increases, so seek to achieve
zero defects and that will truly minimize quality costs.

High High

Incremental Incremental
Cost of Cost of
Variability Variability

Zero Zero

Lower Target Upper Lower Target Upper

Spec Spec Spec Spec Spec Spec
Exhibits
TN8.1 &
Traditional View TN8.2 Taguchi’s View
Dr. Saydam, Lecture 1
Process Capability

 Tolerance (specification, design) Limits

 Bearing diameter 1.250 +- 0.005 inches
 LTL = 1.245 inches UTL = 1.255 inches

 Process Limits
 The actual distribution from the process
 Run the process to make 100 bearings, compute the mean
and std. dev. (and plot/graph the complete results)
 Suppose, mean = 1.250, std. dev = 0.002

 How do they relate to one another?

Dr. Saydam, Lecture 1

Tolerance Limits vs. Process Capability

Specification Width

Actual Process Width

Specification Width

Actual Process Width

Dr. Saydam, Lecture 1

 Process Capability Example
 Design Specs: Bearing diameter 1.250 +- 0.005 inches
 LTL = 1.245 inches UTL = 1.255 inches
 The actual distribution from the process  mean = 1.250, s = 0.002
 +- 3s limits  1.250 - 3(0.002)  [1.244, 1.256]

 Anew process,  std. dev. = 0.00083

Dr. Saydam, Lecture 1

Process Capability Index, Cpk
 Capability Index shows how well parts being produced fit
into design limit specifications
 

 Compute the Cpk for the bearing example.

 Old process, mean = 1.250, s = 0.002

 What is the probability of producing defective bearings?

 New process, mean = 1.250, s= 0.00083, re-compute the Cpk

 When the computed (sample) mean = design (target) mean,

what does that imply?

Dr. Saydam, Lecture 1

The Cereal Box Example

 Recall the cereal example. Consumer Reports has just

published an article that shows that we frequently have less
than 15 ounces of cereal in a box.
 Let’s assume that the government says that we must be within ±
5 percent of the weight advertised on the box.
 Upper Tolerance Limit = 16 + 0.05(16) = 16.8 ounces
 Lower Tolerance Limit = 16 – 0.05(16) = 15.2 ounces

 We go out and buy 1,000 boxes of cereal and find that they
weight an average of 15.875 ounces with a standard deviation of
0.529 ounces.

Dr. Saydam, Lecture 1

Cereal Box Process Capability
 Specification or Tolerance Limits
 Upper Spec = 16.8 oz,  X  LTL UTL X 
C  Min  ; 
 Lower Spec = 15.2 oz
 3 3
pk

 Observed Weight
 Mean = 15.875 oz, Std Dev = 0.529 oz

 What does a Cpk of 0.4253 mean?

 Many companies look for a Cpk of 1.3 or better… 6-Sigma

company wants 2.0!
Dr. Saydam, Lecture 1
Types of Statistical Sampling

1. Sampling to accept or reject the immediate lot of

product at hand (Acceptance Sampling).
 Attribute (Binary; Yes/No; Go/No-go information)
 Defectives refers to the acceptability of product across a
range of characteristics.
 Defects refers to the number of defects per unit which may
be higher than the number of defectives.
 p -chart application
2. Sampling to determine if the process is within
acceptable limits (Statistical Process Control)
 Variable (Continuous)
 Usually measured by the mean and the standard deviation.
 X-bar and R chart applications

Dr. Saydam, Lecture 1

Evidence for Investigation…

Dr. Saydam, Lecture 1

Control Limits

 If we establish control limits at +/-

3 standard deviations, then we
would expect 99.7% of our
observations to fall within these
limits
x
UCL
LCL UCL

LC L

UCL

LC L

Dr. Saydam, Lecture 1

Attribute Measurements ( p-Chart)

 Item is “good” or “bad”

 Collect data, compute average fraction bad (defective)
and std. dev. using: Total Number of Defectives
p =
Total Number of Observatio ns

p (1 - p )
sp =
n

 The, UCL, LCL using:

UCL = p Z sp

LCL = p - Z sp
 Excel time!

Dr. Saydam, Lecture 1

Variable Measurements (x-Bar and R Charts)
 A variable of the item is measured (e.g., weight, length,
salt content in a bag of chips)
 Note that the item (sample) is not declared good or bad
 Since the actual the standard deviation of the process
is not known (and it may indeed fluctuate also) we
usethe sample data to compute the UCL & LCL

x Chart Control Limits R Chart Control Limits

UCL = x + A 2
R UCL = D 4R

LCL = x - A 2
R LCL = D 3R

 For 3-sigma limits, factors A2 , D3 , and D4 and are given in Exhibit

9A.6, p. 341
 Excel time!

Dr. Saydam, Lecture 1

Acceptance Sampling vs. SPC

 Sampling to accept or reject the immediate lot of

product at hand (Acceptance Sampling).
 Determine quality level
 Ensure quality is within predetermined (agreed) level

 Sampling to determine if the process is within

acceptable limits - Statistical Process Control (SPC)

3
Dr. Saydam, Lecture 1
Acceptance Sampling

 Advantages  Disadvantages
 Economy  Risks of accepting “bad” lots
 Less handling damage and rejecting “good” lots
 Fewer inspectors  Added planning and
 Upgrading of the inspection documentation
job  Sample provides less
 Applicability to destructive information than 100-percent
testing inspection
 Entire lot rejection
(motivation for improvement)

Dr. Saydam, Lecture 1

A Single Sampling Plan
 A Single Sampling Plan simply requires two
parameters to be determined:

1. n the sample size (how many units to sample

from a lot)

2. c the maximum number of defective items that

can be found in the sample before the lot is
rejected.

Dr. Saydam, Lecture 1

RISK

RISKS for the producer and consumer in sampling

plans:
 Acceptable Quality Level (AQL)
 Max. acceptable percentage of defectives defined by
producer.
  (Producer’s risk)
 The probability of rejecting a good lot.
 Lot Tolerance Percent Defective (LTPD)
 Percentage of defectives that defines consumer’s rejection
point.
 (Consumer’s risk)
 The probability of accepting a bad lot.

Dr. Saydam, Lecture 1

 Operating Characteristic Curve
The OCC brings the concepts of producer’s risk, consumer’s risk,
sample size, and maximum defects allowed together
Probability of acceptance

1
0.9  = .05 (producer’s risk)
0.8
0.7 n = 99
0.6 c=4 The shape or
0.5 slope of the curve
0.4 is dependent on a
0.3  =.10 particular
combination of
0.2 (consumer’s risk) the four
0.1 parameters
0
1 2 3 4 5 6 7 8 9 10 11 12
AQL LTPD
Percent defective
9
Dr. Saydam, Lecture 1
 Example: Acceptance Sampling
Zypercom, a manufacturer of video interfaces,
purchases printed wiring boards from an outside
vender, Procard. Procard has set an acceptable
quality level of 1% and accepts a 5% risk of rejecting
lots at or below this level. Zypercom considers lots
with 3% defectives to be unacceptable and will
assume a 10% risk of accepting a defective lot.

Develop a sampling plan for Zypercom and

determine a rule to be followed by the receiving
inspection personnel.

10
Dr. Saydam, Lecture 1
 Developing A Single Sampling Plan
 Determine:
 AQL? ?
 LTPD? ?
 Divide LTPD by AQL  0.03/0.01  3
 Then find the value for “c” by selecting the value in the TN8.10 “n(AQL)”
column that is equal to or just greater than the ratio above (3).
 Thus, c  6
 From the row with c6, get  nAQL  3.286 and divide it by AQL
 3.286/0.01  328.6, round up to 329,  n  329

Sampling Plan:
c LTPD/AQL n AQL c LTPD/AQL n AQL
Take a random 0 44.890 0.052 5 3.549 2.613
sample of 329 units 1 10.946 0.355 6 3.206 3.286
from a lot.
Reject the lot if more 2 6.509 0.818 7 2.957 3.981
than 6 units are 3 4.890 1.366 8 2.768 4.695
defective. 4 4.057 1.970 9 2.618 5.426
Dr. Saydam, Lecture 1