Process Capability for non Normal Data
When normality test failed
• Transform the data.
• Calculate Process Capability using transformed data and spec limit
• Use a percentile-based calculation
1
Process Capability for non Normal Data (exercise)
• Open up Data and Template [Link] file.
• Copy data from Process Capability Non sheet.
• Check Data Normality
• Transformed data
• Analyze
• Fit Model
2
Process Capability for non Normal Data (exercise)
• Check Data Normality
• Transformed data
• Analyze
• Fit Model
• Put Void Data
into Y box
3
1 4
1. Select small red triangle
2. Select
• Factor profiling
• Box Cox Y
Transformation 3
3. Find lamda for smallest SSE
=0.284
4. Select small red triangle
A new row of transformed data
• Save Best
Transformation
4
• Check normality on transformed data.
• Transformed Spec limit (2 mils) according to
formula used.
• Calc CPK
Transformed spec limit
To find out transformed formula
= 2 ^(0.284) – 1
0.402793
1. Goto VoidX Column header
2. Right click
= 0.54014
• Column info…
5
1. Analyze
1. Distribution
2. Click on red triangle
• Process Capability
• Enter transformed
Spec Limit
6
Process Capability for non Normal Data (self exercise)
• Use the same data set as earlier exercise.
• Use the following simple formula Void^(0.284) for
transformation.
• Check normality and re calculate process capability.
• Is data become normal after transformed?
• Is process Capability same as earlier formula?
7
Process Capability for non Normal Data (self exercise - answer)
• Is data become normal after transformed? Yes
• Is process Capability same as earlier formula? Yes
8
Percentile-based Process Capability (Cnp and Cnpk)
Utilize percentiles of the distribution instead of the mean and standard deviation.
• For a normal distribution:
• The interval μ ± 3σ contains over 99% of the measurements
• Generalize and replace this distance with distances based on percentiles
• Replace 6σ with P0.995 – P0.005
• Replace μ with P0.5 = median
USL-LSL USL - P 0.5 P 0.5 - LSL
Cnp = Cnpk = Min ( , )
P0.995 – P 0.005 P 0.995 – P 0.5
P 0.5 – P 0.005
Limitation:
• Need to have a large amount of data to estimate the percentiles.
• At least 100 measurements.
9
Percentile-based Process Capability (Cnp and Cnpk) – exercise.
• Use the same data set as earlier exercise.
• Calculate using percentile method.
• Find P(0.995,0.5,0.005) from quantile table.
USL-LSL USL - P 0.5 P 0.5 - LSL
Cnp = Cnpk = Min ( , )
P0.995 – P 0.005 P0.995 – P 0.5
P 0.5 – P 0.005
P(0.995)= 5.385
Cnp 0.372
P(0.5) = 0.7135 Cnp U 0.275
P(0.005) = 0.007
10
Non-Normal Distributions - Percentile-based Cp and Cpk
Using other percentiles
• Some common percentile pairs
11
Percentile-based Process Capability (Cnp and Cnpk) – exercise.
• Compare Cnpk value with the Cpk calculate using transformed
value in earlier exercise.
• Are they giving the same result ?
Method CnpK (U)
Transformed (best) 0.34
(99.5-0.5%) 0.275
(99.5-0.5)*0.8586 0.236
(99-1%) 0.214
95-5% 0.340
90-10% 0.218
12
Percentile-based Process Capability (Cnp and Cnpk) – Case Study
• Case Study 1
• 4Use set2b data from excel worksheet. (Non Normal Data)
• Perform the same comparison on CpK calculated using Transformed data vs percentile
• Try using Data^3 as transform formula.
• Spec Limit 1.5 to 2.5
• Case Study 2
• Try Using Data ^4 as transform formula.
• Case Study 3
• Try Using Data ^4 as transform formula.
• Try using Spec limit at 1 to 2.5
13
Percentile-based Process Capability (Cnp and Cnpk) – Case Study 1
• Normality check
14
Percentile-based Process Capability (Cnp and Cnpk) – Case Study1
• Transformed Data • Percentile
CnP 1.088139282
Cnpk(U) 2.593984962
Cnpk(L) 0.483920368
Try using Data^3 as transform formula.
Spec Limit 1.5 to 2.5
15
Percentile-based Process Capability (Cnp and Cnpk) – Case Study1
• Transformed Data • Percentile
CnP 1.088139282
Cnpk(U) 2.593984962
Cnpk(L) 0.483920368
Try using Data^4 as transform formula.
Spec Limit 1.5 to 2.5
16
Percentile-based Process Capability (Cnp and Cnpk) – Case Study 3
• SPEC limit 1 to 2.5, Transformed using Data^4
CnP 1.632208923
Cnpk(U) 2.593984962
Cnpk(L) 1.249617152
Discussion
17