Subhash Dixit
Control Charts
While every process displays Variation, some processes display controlled variation, while other processes display uncontrolled variation (Walter Shewhart). Controlled Variation is characterised by a stable and consistent pattern of variation over time. Associated with Common Causes. Process A shows controlled variation.
X-Bar Chart for Process A X-Bar Chart for Process A
UCL=77.20 UCL=77.20 75
X -B a rC h a rt fo rP ro ce ssB
8 0 U C L = 7 7 .2 7
Sample Mean
7 0
X= 7 0 .9 8 L C L = 6 4 .7 0
6 0
5 0
Special Causes
1 0
1 5
2 0
2 5
S a m p leN u m b e r
75
70
X=70.91 X=70.91 70
65
65 0 0 5 5 1 0 1 0 1 5 1 5 20 20 25
LCL=64.62 LCL=64.62
25
Uncontrolled Variation is characterised by variation that changes over time. Associated with Special Causes. Process B shows uncontrolled variation.
In-Statistical-Control
A process is said to be operating in a state of statistical control (or short in-statistical-control) when only sources of variation are from common causes.
Interpreting Data
Specifications may be used to define when one is in trouble with regard to the voice of the customer, specifications do nothing to describe or define the voice of the process.
It was Walter Shewhart who first gave a simple and effective way to define the voice of the process --- he called it a Control Chart. UCL LCL Time A Control Chart begins with a time series graph. A central line (X-bar) is added as a visual reference for detecting shifts or trends. (Process location). Control limits (UCL & LCL computed from the data) are placed equidistant on either side of the central line. (Process dispersion).
Metric
Average
Control Limits
Time Types of Charts: Individuals & Moving Range Charts X-Bar & R Charts p Charts (& np Charts) c Charts (& u Charts)
Individuals Charts
Appropriate questions for Individuals and Moving Range (I and MR) Charts take the form of How does the short term variation compare to long-term variation?
Time
We are interested in: - the average of the process - the short-term variation - the long-term variation
Individual Value
38 Mean=37.38 37
UCLX ) UCLX = = X X+ + (3 (3 mR mR /d /d2 2) = = 37.38 37.38 + + (( 3 3 0.5195 0.5195 // 1.128)= 1.128)= 38.76 38.76
36 0 10 20 30 40
LCL=35.99
LCLX ) LCLX = = X X -- (3 (3 mR mR // d d2 2) = = 37.38 37.38 -- (3 (3 0.5195 0.5195 // 1.128) 1.128) = = 35.99 35.99
Subgroup
Moving Range
UCL=1.697
UCLR UCLR = = 3.267 3.267 mR mR = = 3.267 3.267 0.5195 0.5195 = = 1.697 1.697
1 mR=0.5195 0 LCL=0
=1.128 n n= =2 2 (difference (difference of of two two measures) measures) then then d d2 2 =1.128
I and MR Chart
39 UCL=38.76
38 Mean=37.38 37
How are the control limits of the individual values derived? What comparison is being made on the Individual chart?
36 0 10 20 30 40
LCL=35.99
Subgroup
M o ving Ra nge
1 mR=0.5195 0 LCL=0
Step 1: Calculate moving range. To calculate each moving range, subtract each measurement from the previous one. There will be no moving range for the first observation on the chart Moving range = Positive value of (Yn - Yn-1) Step 2: Calculate average of individuals data and average of moving range Step 3: Calculate the control limits UCLR = Upper Control Limit for Range = 3.267 (mR) UCLX = Upper Control Limit for Individuals = X + (3 mR / d2) LCLX = Lower Control Limit for Individuals = X - (3 mR / d2) Where d2 = 1.128 (for 2 samples) Step 4: Plot the data, the averages and the control limits in 2 charts Step 5: Analyse the data Averages
X X
mR mR
Formulas:
UCLR = 3.267 mR
UCLX = X + 3 LCLX = X 3
mR d2
mR d2
d2 = 1.128 when the moving range is based on the range between successive numbers.
Step 1: Calculate moving range. To calculate each moving range, subtract each measurement from the previous one. There will be no moving range for the first observation on the chart Moving range = Positive value of (Yn - Yn-1) Step 2: Calculate average of individuals data and average of moving range Step 3: Calculate the control limits UCLR = Upper Control Limit for Range = 3.267 (mR) UCLX = Upper Control Limit for Individuals = X + (3 mR / d2) LCLX = Lower Control Limit for Individuals = X - (3 mR / d2) Where d2 = 1.128 (for 2 samples) Step 4: Plot the data, the averages and the control limits in 2 charts Step 5: Analyse the data
X X
mR mR
Averages
Sub-grouping
The concept of subgrouping is one of the most important components of the control chart method. Shewharts principle is to organise (classify, stratify, group, etc.) data from the process in a way that ensures the greatest similarity among the data in each subgroup and the greatest difference among the data in different subgroups. The aim of rational subgrouping is to include only common causes of variation within subgroups, with all special causes of variation occurring between subgroups. The purpose of understanding the within and between variation is to determine where to work, i.e. we reduce the number of potential variables considerably when we determine the variation is within or between.
x1 + x 2 + K + x k 1 k x = xi = k i =1 k R1 + R 2 + K + Rk 1 k R = Ri = k i =1 k UCL x = X + A 2 R LCL x = X A 2 R
UCLR = D 4 R
LCLR = D3 R
AVERAGES
Sample Mean
1 45
1 1
BETWEEN BETWEEN
40 1 0 10 1 1 1 1
LCLX=38.91
UCL = X + A2 R UCLX X = X + A 2 R = = 42.74 42.74 + + 0.729 0.729 5.25 5.25 = = 46.56 46.56
35 Subgroup
20
Range: the Highest value minus the lowest value WITHIN each subgroup
UCLR=11.98
LCL = X - A2 R LCLX X = X - A 2 R = = 42.74 42.74 -- 0.729 0.729 5.25 5.25 = = 38.91 38.91 UCL = D4 R UCLR R = D 4 R = = 2.282 2.282 5.25 5.25 = = 11.98 11.98
Sample Range
10
WITHIN WITHIN
R=5.250
UCL = X + A2 R UCLX X = X + A 2 R
UCL=242.8
= = 212.4 212.4 + + 0.483 0.483 62.6 62.6 = = 242.8 242.8 LCL = X - A2 R LCLX X = X - A 2 R = = 212.5 212.5 -- 0.483 0.483 62.6 62.6 = = 182.3 182.3
Sample M ean
BETWEEN BETWEEN
Mean=212.5
LCL=182.3 1 2 3 4 5
Subgroup
150
UCL = D4 R UCLR R = D 4 R
UCL=125.4
Sample Rang e
100
WITHIN WITHIN
R=62.6 LCL=0
50
Between Group
X
Within Group
UCLR _ R LCLR
-X UCL = X -X LCL
Exercise
Based on the previous I and MR analysis, the team observes potential sources of variation in operator and round. Evaluating product performance has proved difficult with the large variation present. The team decides they must first understand within and between variation for these sources before continuing with the evaluation. How might your team change the way you are looking at the data to gain insight to differences in the sources of variation? Plan your experiment then run it and look at your data. Make it give up information related to these sources of variation.
Sampling Plans
Sampling Plans: A Control Chart is only as useful as the sampling used to collect data. Consider a hypothetical process where all measurements on a key characteristic are available. The output from this hypothetical process is shown below.
TIME
X Chart Range Chart Examines Variation Within Subgroups
Upper Control Limit
Sampling Strategies
The purpose of sampling is to reduce the cost and time of checking every unit. The goal is to capture the voice of the process efficiently. The charts on the following page are the output from a typical process. The goal is to make the output consistent, but you dont know whether to work on xs within time period or between time period. You dont know about the points you dont measure, yet you need sampling to understand the state of the process. Sketch the control charts for each sampling plan and subgrouping strategy. (Space is provided for you to suggest and draw charts for two sampling schemes.) Which of the charts best displays the voice of the process?
TIME
TIME
Create charts by subgrouping the first and last point of each time period.
Range Chart
Upper Control Limit
Create charts by subgrouping the first two and last two points of each time period.
Range Chart
Upper Control Limit
Centre Line(R)
Centre Line(R)
X Chart
Upper Control Limit
X Chart
Upper Control Limit
Centre Line(X)
Lower Control Limit
Centre Line(X)
Lower Control Limit
Shift 1
TIME
Create charts by subgrouping the last two points and first two points of each shift (4 points total per group)
Range Chart
Upper Control Limit
X Chart
Upper Control Limit
X Chart
Upper Control Limit
TIME
TIME
X Chart
Upper Control Limit
X Chart
Upper Control Limit
TIME
TIME
X Chart
Upper Control Limit
X Chart
Upper Control Limit
Expansion of Sub-grouping
Examine this output from a hypothetical process. The '' indicate output from one machine. The '*' indicate output from another machine.
Time Assignment: Describe some sampling & graphing plans that might detect differences in machine.
* * **
*** *
* * ** * ** ** ** *
* * ** * * * * * * *** *
* * ** * * ** * * ** * ** * *
**
* *** ** * * * * * * ** ** **
* * * * * *
* * ** * ** *
* *
To aid in understanding a process, the intent of control charts on end-product parameters is to learn about sources of variation. The size and frequency of subgroups, the number of subgroups, and the sampling locations are a function of the state of process knowledge. Subgroups are determined by the process variables identified on the process map and by cause and effect diagrams.
Subgroup size, sampling frequency and time period are largely determined by considering the possible frequency of change of those process variables and causes. Sources of variation potentially causing systematic differences in product parameters are also large drivers of sampling and subgrouping plans.
p(# of defective items; Sample size is not constant) c (# of defects; Constant sample size) u(# of defects; Sample size is not constant)
Attribute
No
Yes
Is the sample size constant?
Yes No
Variable
p Chart
No
U Chart
Yes
Yes
np or p Chart
Does it make sense to subgroup for Within group variation analysis? No Yes Can subgroup averages No be conveniently calculated? Yes Subgroup size 9 or more? Yes Can sample standard deviation be calculated? Yes No No
C or U Chart
Me-R Chart
X-mR Chart
Examples: Batches, customer demand by request date, too little data, etc.
Xbar-R Chart
Xbar-R Chart
Assumptions: Rational Subgrouping performed if using subgroups. The measurement system has been assessed and is appropriate. Adapted from AIAG SPC Manual
Xbar-s Chart
np-Chart
Charts number of items in a sample having an attribute Each subgroup is the same size Purposes
Find the average number having the attribute Find subgroups where the number is out of control relative to other subgroups
30
Sample Count
20
1 UCL=19.96
10
NP=12
LCL=4.043 0 0 10 20
Sample Number
An example is inspection
Attribute is nonconformance to some requirement Periodic samples are checked Sample size is always the same Assume nonconforming
Assumptions
Items in a subgroup are distinct Each item either has or does not have the attribute
p-Chart
Charts proportion of items having an attribute - the number having the attribute divided by the number of items Each subgroup can be a different size Purposes
Find the average probability of having the attribute Find subgroups where the proportion is out of control relative to other subgroups
P Chart for Rejects
UCL=0.3324 0.3
Proportion
0.0 0 10 20
LCL=0.004728
Sample Number
Caution: All proportions are ratios, but not all ratios are proportions
Incorrect uses:
Same assumptions as
Todays rework divided by todays production Number of defects on all invoices divided by the number of invoices
Example p-Chart
Each month, the company receives a different number of complaints Each complaint requires a report be written within 15 days For some number of the complaints, the reports are not finished by 15 days P-Chart shows the proportion of each months complaints where the report was not completed on time Conclusion Overall, 16% of the reports were not finished on time Process improvements seem to have reduced the proportion, continue to monitor Unusually high action is necessary to uncover the cause
P Chart for Unfinished Reports
0.4
0.3
UCL=0.2950
Proportion
0.0 0 5 10
Sample Number
C-Chart
Charts the number of events within a constant area of opportunity C Chart for Blemish Purposes Find the average number of events 8 3.0SL=7.677 7 Find subgroups where the number is 6 out of control relative to other subgroups 5 4 Assumptions 3 C=2.725 Count the number of discrete events 2 The discrete events occur within some 1 well-defined region of space or time or 0 LB=0 product The events occur independently of 0 10 20 30 40 Sample Number each other, and in proportion to the size of the area of opportunity (it Example: Number of blemishes doesnt matter which portion of the continuum is chosen to be the area of A roll of material is 1 meter wide opportunity) A 2 meter section is taken from the The size of the area of opportunity is end of each roll the same for all samples For each section, the number of ALL the events are counted; as opposed to inspection, where things blemishes is counted and charted might be rejected upon encountering the first event, terminating the count
Sample Count
U-Chart
U Chart for NumLeaks
0.4
0.3 UCL=0.2955 Charts the number of events per unit sampled in each sub0.2 group (uneven subgroup sizes) U=0.1379 0.1 Same assumptions as C0.0 LCL=0 Chart, except: 0 10 20 The area of opportunity can Sample Number varying in size Example: Leaks per radiator The sizes dont vary Number of radiators varies each day randomly, but vary All leaks are counted on all radiators Some radiators might have zero deliberately or inherently leaks Chart the average number of Area of opportunity represents all events per area the surface area of radiators that
Sample Count
were produced that day Chart shows the average number of leaks per radiator by day
Dont RECORD it if you are not going to GRAPH it Dont GRAPH it if you are not going to ANALYSE it Dont ANALYSE it if you are not going to take ACTION
Forms of action from Control Charts: 1. Proactive tool Real time tool for process control. Anticipate trends before they become a problem. 2. Predictive tool Process or product designers will know how to specify new products or processes. 3. Reactive tool When rejects or failures occur try to understand why and focus problem solving on the right sources of variation.
Case Study
Reference :
How To Teach Others To Apply Statistical Thinking, By Roger Hoerl, Galen Britz, Don Emerling, Lynne Hare, Janice Shade
Objective: Analyse the Case Study as if you were the VP of Sales Whats Going On Here?
Time
How To Teach Others To Apply Statistical Thinking by Roger Hoerl (GE), Galen Britz, Don Emerling, Lynne Hare, Janice Shade in Quality Progress - June 1997
Individual Value
X=1000
Individual Value
Q4 1992
10
X=1196
1996
-3.0SL=632.4
-3.0SL=857.3
Subgroup 600
GLORY
1
20
Subgroup
GLORY
20
Moving Range
3.0SL=451.6
Moving Range
500
3.0SL=416.0
R=138.2 -3.0SL=0.00E+
R=127.3 -3.0SL=0.00E+
Individual Value
Individual Value
X=519.5
-3.0SL=369.4 0 10
Subgroup
GLORY
20
Subgroup 200
BLASTED
20
Moving Range
Moving Range
300
3.0SL=334.3
3.0SL=184.5
Individual Value
Individual Value
1 3.0SL=750.2 X=656.7
Subgroup
BLASTED
20
Subgroup 150
BLASTED
20
3.0SL=114.9
3.0SL=140.0
Moving Range
Moving Range
100
100 50 0
50 R=35.16 0 -3.0SL=0.00E+
R=42.84 -3.0SL=0.00E+
The VP gave GLORY for selling into the next quarters demand, no improvement, and loss of market share. The VP BLASTED two regions for excellent year over year growth, and one region for holding even in a tough competitive situation. This will drive terrible behaviours in the sales force. The best performers may look elsewhere for work. The marginal performers have been reinforced to sell into next month, continue mediocre performance, and hope for luck.
Final Thoughts
Failure Failureto touse useControl ControlCharts Chartsto toanalyse analysedata datais is one oneof ofthe thebest bestways waysknown knownto tomankind mankindto to increase increasecosts, costs,waste wasteeffort, effort,and andlower lowermorale. morale. --Dr. Dr.Donald DonaldJ. J.Wheeler Wheeler
People respond to the way they are measured. Using data correctly can drive GREAT results. Using data incorrectly drives CHAOS. Dont be part of the PROBLEM - Be part of the SOLUTION
Summary
We evaluate metrics to look for major sources of variation Concept of Within & Between variation is crucial. Control charting helps with this investigation. R Chart is used to display Within Variation X-bar Chart displays the Between Variation Subgrouping is a key concept. I mR chart can be used anytime. There are specific charts for attribute data.
1999-2003
1996-98
1994-96
1988
1985-87
Mr Jack Welch, GE: "Six Sigma is the most important concept we have ever adopted, and it will bring us 70 to 110 MUSD savings in the coming 5 years."
Motorola 1987
Air Products American Express Ford Motor Johnson Control Johnson & Johnson J.P Morgan Chase LG Group Ericsson Maytag Compaq Navistar Dow Chemical DuPont NCR Deere Nokia Lockheed Martin Philips Raytheon NEC Praxair PACCAR Seagate Tech Samsung Electronics Siemens Solectron Sony Sumitomo Toshiba United Technologies Whirlpool US Postal Service 1997 1999
Amazon.com Black & Decker Bosch Canon Caterpiller Compac Daimler Chrysler Dell Computor Delphi Aut. Syst. Eaton Corp. Flextronix Int Heller Financial ITT Internat. Johnson Controls Pilkington Polaroid Rexam Beverage Roche Diagno Samsung Sear Sun Microsystem Textron + +
2000
Benchmarking Sigma 6 5 4 3 2 1 PPMO 3.4 233 6210 66,807 COPQ < 10% 10-15% TYPE World Class
You get . . One hour of unsafe drinking water every month. Two unsafe plane landings per day at OHare International Airport in Chicago. 16,000 pieces of mail lost by the U.S. Postal Service every hour. 500 incorrect surgical operations each week. 50 newborn babies dropped at birth by doctors each day. 22,000 checks deducted from the wrong bank accounts each hour.
Unconscious incompetence
Conscious incompetence
Conscious competence
Unconscious competence
00Customer Centricity
Sigma Rating
2 3 4 5 6 Process Capability
PPM