Advanced Statistical Quality Control and Monitoring
Prof. Fugee TsungHong Kong University
of Science and Technology
Quality Lab: http://qlab.ielm.ust.hk
Dr. Fugee Tsung, HKUST 2
Outline
Module A: Overview of SPC; Autocorrelated SPCModule B: Multivariate SPC; Profile MonitoringModule C: Integrating EPC and SPC; Run-to Run ControlModule D: Multistage SPC; Multi-ChartsModule E: Categorical SPC; Nonparametric SPC
Dr. Fugee Tsung, HKUST 3
Ignore
DesensitizeRobust D
esign
Compensate EPC
Reduce/Eliminate
SPC
How to Handle Variation
Dr. Fugee Tsung, HKUST 4
Statistical Process Control (SPC): Role in Six Sigma
SPC is emphasized as the major tool used to control and sustain a Six Sigma process
Define
Improve Analyze
MeasureControl
Dr. Fugee Tsung, HKUST 5
SPC: Basic Idea
3020100
1.5
1.0
0.5
0.0
-0.5
-1.0
-1.5
Sample NumberS
ampl
e M
ean
Mean=0
UCL=1.342
LCL=-1.342
? y
In most SPC applications it is assumed that the quality of a process or product can be adequately represented by the distribution of a univariate quality characteristic or by the general multivariate distribution of a vector consisting of several quality characteristics (Woodall et al. 2004).
),(~
),(~222
211
σμ
σμ
Ny
Ny
i
i
.,...,1 ;,...,2,1 nifor
ifor+=
=τ
τ
SPC for Profiles
Dr. Fugee Tsung, HKUST 7
SPC: A Different Idea
3020100
1.5
1.0
0.5
0.0
-0.5
-1.0
-1.5
Sample NumberS
ampl
e M
ean
Mean=0
UCL=1.342
LCL=-1.342
? yx
In many practical situations, however, the quality of a process or product is better characterized and summarized by a relationship between a response variable and one or more explanatory variables (Woodall et al. 2004).
)( ii xfy =
Dr. Fugee Tsung, HKUST 8
A Profile Example
Dr. Fugee Tsung, HKUST 9
Another Profile Example
SPC for engineering controlled processes
Dr. Fugee Tsung, HKUST 11
A Engineering Controlled Process
3020100
1.5
1.0
0.5
0.0
-0.5
-1.0
-1.5
Sample NumberS
ampl
e M
ean
Mean=0
UCL=1.342
LCL=-1.342
? yxProcess
EPC Controller
iDiI
iPi
iD
i
jiIiPi
yBkyB
kykx
ykykykx
)1(1
1
−+−
+=
∇++= ∑=
Dr. Fugee Tsung, HKUST 12
Example: CMP Run-to-Run Control
Drift in removal rate during processingVariation in incoming thicknessHead-to-head variation in processingUnmeasurable tool characteristics affect product outputIncomplete product measurementsSignificant measurement delay
4
51
2
3
CMPProcess
EPCController
+xt
dt
yt
yt
SPC for Multistage processes
Dr. Fugee Tsung, HKUST 14
A Multistage Process
3020100
1.5
1.0
0.5
0.0
-0.5
-1.0
-1.5
Sample NumberS
ampl
e M
ean
Mean=0
UCL=1.342
LCL=-1.342
? ykyk-1
kkkk
kkkkkk
vyCYwByAy
+=++= −− μ11
Stage 1 Stage NStage kStage k-1
),0(~),,0(~ 22wkvk NwNv σσ
Dr. Fugee Tsung, HKUST 15
Example: Printed Circuit Board Manufacturing
Exposure
Black Oxide
Lay-up
Hot Press
Cutting
Drilling
Inspection
…….
From Ding, Shi, and Ceglarek (2002)
Module A: Overview of Statistical Process Control (SPC)/ Statistical Quality Control (SQC)
Based on Montgomery, Introduction to Statistical Quality Control, 4th edition, John Wiley, 2001.
Dr. Fugee Tsung, HKUST 17
Outline
I. What is SPC
II. Common and special causes of quality
variation
III. Statistical basis of control charts
IV. Introduction of popular control charts
I. What is SPC
Dr. Fugee Tsung, HKUST 19
Statistical process control (SPC)Control charts, plus other problem-solving toolsUseful in monitoring processes, reducing variability through elimination of special causesOn-line technique
Dr. Fugee Tsung, HKUST 20
The Tools of SPCThe primary tool of SPC is the Control ChartThe Control Chart looks similar to a run chart, however the addition of control limits plays an important role.
II. Common and special causes of quality variation
Dr. Fugee Tsung, HKUST 22
Purpose for Using Control Charts
The Control Limits probabilistically separate common cause and special cause variability.
Dr. Fugee Tsung, HKUST 23
Common and Special Causes of Quality Variation
A process that is operating with only common cause/chance causes of variation present is said to be in statistical control.A process that is operating in the presence of special/assignable causes is said to be out of control.The eventual goal of SPC is reduction or elimination of variability in the process by identification of assignable causes.
Dr. Fugee Tsung, HKUST 24
Common and Special Causes of Quality Variation
III. Statistical basis of control charts
Dr. Fugee Tsung, HKUST 26
Statistical Basis of the Control Chart
Basic PrinciplesA typical control chart has control limits set at values such that if the process is in control, nearly all points will lie between the upper control limit (UCL) and the lower control limit (LCL).
Dr. Fugee Tsung, HKUST 27
Statistical Basis of the Control ChartBasic Principles
Dr. Fugee Tsung, HKUST 28
The Out-of-Control SignalThe control limits define a probabilistic level of occurrence of an extreme reading.
Dr. Fugee Tsung, HKUST 29
Statistical Basis of the Control Chart
Out-of-Control SituationsIf at least one point plots beyond the control limits, the process is out of controlIf the points behave in a systematic or nonrandom manner, then the process could be out of control.
Dr. Fugee Tsung, HKUST 30
Hypothesis Testing & Control Charting
Dr. Fugee Tsung, HKUST 31
Statistical Basis of the Control ChartRelationship between hypothesis testing andcontrol charts
We have a process that we assume the true process mean is μ = 74 and the process standard deviation is σ = 0.01. Samples of size 5 are taken giving a standard deviation of the sample average, , as x
0045.0501.0
nx ==σ
=σ
Dr. Fugee Tsung, HKUST 32
Statistical Basis of the Control ChartRelationship between hypothesis testing and control charts
Control limits can be set at 3 standard deviations from the mean.This results in “3-Sigma Control Limits”UCL = 74 + 3(0.0045) = 74.0135
CL= 74LCL = 74 - 3(0.0045) = 73.9865
Dr. Fugee Tsung, HKUST 33
Statistical Basis of the Control ChartRelationship between hypothesis testing and control charts
Choosing the control limits is equivalent to setting up the critical region for testing hypothesisH0: μ = 75H1: μ ≠ 75
Dr. Fugee Tsung, HKUST 34
Statistical Basis of the Control ChartRelationship between the process and the control chart
Dr. Fugee Tsung, HKUST 35
Statistical Basis of the Control ChartImportant uses of the control chart
Most processes do not operate in a state of statistical control.Consequently, the routine and attentive use of control charts will identify assignable causes. If these causes can be eliminated from the process, variability will be reduced and the process will be improved.The control chart only detects assignable causes. Management, operator, and engineering action will be necessary to eliminate the assignable causes.Out-of-control action plans (OCAPs) are an important aspect of successful control chart usage.
Dr. Fugee Tsung, HKUST 36
Statistical Basis of the Control Chart
Types the control chartVariables Control Charts
These charts are applied to data that follow a continuous distribution (measurement data).
Attributes Control ChartsThese charts are applied to data that follow a discrete distribution.
Dr. Fugee Tsung, HKUST 37
Statistical Basis of the Control ChartType of Process Variability
Stationary behavior, uncorrelated dataStationary behavior, autocorrelated dataNonstationary behavior
Dr. Fugee Tsung, HKUST 38
Statistical Basis of the Control Chart
Type of VariabilityMost conventional control charts are effective when the in-control process data is stationary and uncorrelated.
Dr. Fugee Tsung, HKUST 39
Statistical Basis of the Control ChartPopularity of control charts
1) Control charts are a proven technique for improving productivity.
2) Control charts are effective in defect prevention.3) Control charts prevent unnecessary process
adjustment.4) Control charts provide diagnostic information.5) Control charts provide information about process
capability.
Dr. Fugee Tsung, HKUST 40
Choice of Control LimitsGeneral model of a control chart
where L = distance of the control limit from the center line
= mean of the sample statistic, w.= standard deviation of the statistic, w.
WW
W
WW
LLCLLineCenter
LUCL
σ−μ=μ=σ+μ=
Wμ
Wσ
Dr. Fugee Tsung, HKUST 41
Choice of Control Limits“99.7% of the Data”
If approximately 99.7% of the data lies within 3σof the mean (i.e., 99.7% of the data should lie within the control limits), then 1 - 0.997 = 0.003 or 0.3% of the data can fall outside 3σ (or 0.3% of the data lies outside the control limits). (Actually, we should use the more exact value 0.0027)0.0027 is the probability of a Type I error or a false alarm in this situation.
Dr. Fugee Tsung, HKUST 42
Choice of Control Limits
Three-Sigma Limits
The use of 3-sigma limits generally gives good results in practice. If the distribution of the quality characteristic is reasonably well approximated by the normal distribution, then the use of 3-sigma limits is applicable.These limits are often referred to as action limits.
Dr. Fugee Tsung, HKUST 43
Sample Size and Sampling Frequency
In designing a control chart, both the sample size to be selected and the frequency of selection must be specified.Larger samples make it easier to detect small shifts in the process.Current practice tends to favor smaller, more frequent samples.
Dr. Fugee Tsung, HKUST 44
Sample Size and Sampling Frequency
Average Run LengthThe average run length (ARL) is a very important way of determining the appropriate sample size and sampling frequency.Let p = probability that any point exceeds the control limits. Then,
p1ARL =
Dr. Fugee Tsung, HKUST 45
Sample Size and Sampling Frequency
IllustrationConsider a problem with control limits set at 3standard deviations from the mean. The probability that a point plots beyond the control limits is again, 0.0027 (i.e., p = 0.0027). Then the average run length is
3700027.01ARL ==
Dr. Fugee Tsung, HKUST 46
Sample Size and Sampling Frequency
What does the ARL tell us?The average run length gives us the length of time (or number of samples) that should plot in control before a point plots outside the control limits.For our problem, even if the process remains in control, an out-of-control signal will be generated every 370 samples, on average.
Dr. Fugee Tsung, HKUST 47
Sample Size and Sampling Frequency
Average Time to SignalSometimes it is more appropriate to express the performance of the control chart in terms of the average time to signal (ATS). Say that samples are taken at fixed intervals, h hours apart.
)h(ARLATS =
Dr. Fugee Tsung, HKUST 48
Rational Subgroups
Subgroups or samples should be selected so that if assignable causes are present, the chance for differences between subgroups will be maximized, while the chance for differences due to these assignable causes within a subgroup will be minimized.
Dr. Fugee Tsung, HKUST 49
Analysis of Patterns on Control ChartsNonrandom patterns can indicate out-of-control
conditionsPatterns such as cycles, trends, are often of considerable diagnostic value.Look for “runs” - this is a sequence of observations of the same type (all above the center line, or all below the center line)Runs of say 8 observations or more could indicate an out-of-control situation.
Run up: a series of observations are increasingRun down: a series of observations are decreasing
Dr. Fugee Tsung, HKUST 50
Analysis of Patterns on Control ChartsWestern Electric Handbook Rules (Should be used carefully because of the increased risk of false alarms)
A process is considered out of control if any of the following occur:1) One point plots outside the 3-sigma control limits.2) Two out of three consecutive points plot beyond the 2-
sigma warning limits.3) Four out of five consecutive points plot at a distance of 1-
sigma or beyond from the center line.4) Eight consecutive points plot on one side of the center line.
IV. Introduction of popular control charts
1. The Shewhart chart (X-bar & R chart)
Dr. Fugee Tsung, HKUST 53
Notation for variables control chartsn - size of the sample (sometimes called a subgroup) chosen at a point in timem - number of samples selected
= average of the observations in the ith sample (where i = 1, 2, ..., m)
= grand average or “average of the averages (this value is used as the center line of the control chart)
ix
x
Dr. Fugee Tsung, HKUST 54
Notation and valuesRi = range of the values in the ith sample
Ri = xmax - xmin
= average range for all m samplesμ is the true process meanσ is the true process standard deviation
R
Dr. Fugee Tsung, HKUST 55
Statistical Basis of the ChartsAssume the quality characteristic of interest is normally distributed with mean μ, and standard deviation, σ.If x1, x2, …, xn is a sample of size n, then the average of this sample is
is normally distributed with mean, μ, and standard deviation,
nxxxx n21 +++
=L
x
n/x σ=σ
Dr. Fugee Tsung, HKUST 56
Statistical Basis of the ChartsThe probability is 1 - α that any sample mean will fall between
The above can be used as upper and lower control limits on a control chart for sample means, if the process parameters are known.
nZZ
andn
ZZ
2/x2/
2/x2/
σ−μ=σ−μ
σ+μ=σ+μ
αα
αα
Dr. Fugee Tsung, HKUST 57
The Operating Characteristic Function
How well the and R charts can detect process shifts is described by operating characteristic (OC) curves.Consider a process whose mean has shifted from an in-control value by k standard deviations. If the next sample after the shift plots in-control, then you will not detect the shift in the mean. The probability of this occurring is called the α-risk.
x
Dr. Fugee Tsung, HKUST 58
The Operating Characteristic FunctionThe probability of not detecting a shift in the process mean on the first sample is
L= multiple of standard error in the control limitsk = shift in process mean (#of standard deviations).
)nkL()nkL( −−Φ−−Φ=β
Dr. Fugee Tsung, HKUST 59
2. The Cumulative Sum (CUSUM) Charts
Dr. Fugee Tsung, HKUST 61
Basic Principles: The Cusum Control Chart for Monitoring the Process Mean
The cusum chart incorporates all information in the sequence of sample values by plotting the cumulative sums of the deviations of the sample values from a target value.If μ0 is the target for the process mean, is the average of the jth sample, then the cumulative sum control chart is formed by plotting the quantity
∑ μ−==
i
1j0ji )x(C
jx
Dr. Fugee Tsung, HKUST 62
The Tabular or Algorithmic Cusum for Monitoring the Process Mean
Let xi be the ith observation on the processIf the process is in control thenAssume σ is known or can be estimated.Accumulate derivations from the target μ0 above the target with one statistic, C+
Accumulate derivations from the target μ0 below the target with another statistic, C-
C+ and C- are one-sided upper and lower cusums, respectively.
),(N~x 0i σμ
Dr. Fugee Tsung, HKUST 63
The statistics are computed as follows:The Tabular Cusum
starting values areK is the reference value (or allowance or slack value)If either statistic exceed a decision interval H, the process is considered to be out of control. Often taken as a H = 5σ
[ ][ ]−
−−
+−
+
+−−μ=
++μ−=
1ii0i
1i0ii
Cx)k(,0maxC
C)k(x,0maxC
0CC 00 == −+
Dr. Fugee Tsung, HKUST 64
Selecting the reference value, KK is often chosen halfway between the target μ0 and the out-of-control value of the mean μ1 that we are interested in detecting quickly.Shift is expressed in standard deviation units as μ1= μ0+δσ, then K is
22K 01 μ−μ
=σδ
=
3. The Exponentially Weighted Moving Average (EWMA) Charts
Dr. Fugee Tsung, HKUST 66
3. The EWMA chart
Dr. Fugee Tsung, HKUST 67
Dr. Fugee Tsung, HKUST 68
Dr. Fugee Tsung, HKUST 69
Dr. Fugee Tsung, HKUST 70
Other ChartsCuscore Charts (Box and Ramirez, 1992) Reference-Free Cuscore Charts (Han and Tsung, JASA, 2006) General EWMA Charts (Han and Tsung, Annals of Stat, 2004)GLRT Charts
Dr. Fugee Tsung, HKUST 71
Related referencesHan, D., and Tsung, F.(2004), " A Generalized EWMA Control Chart and its
Comparison with the Optimal EWMA, CUSUM and GLR Schemes," The Annals of
Statistics, 32.
Han, D. and Tsung, F. (2005), " Comparison of the Cuscore, GLRT and CUSUM
Control Charts for Detecting a Dynamic Mean Change," The Annals of the Institute of
Statistical Mathematics, 57.
Han, D. and Tsung, F.(2006) " A Reference-Free Cuscore Chart for Dynamic Mean
Change Detection and a Unified Framework for Charting Performance Comparison ,"
Journal of the American Statistical Association, 101, 368-386.
Ou, Y. J., Wu, Z. and Tsung, F. (2012), " Comparison Study of Effectiveness and
Robustness of Control Charts for Monitoring Process Mean," International Journal of
Production Economics, 135, 479-490.