VW - 10131 - en

Klass.-Nr.
11 00 5 February 2005
Norm vor Anwendung auf Aktualität prüfen / Check standard for current issue prior to usage.
Process Capability Investigation for VW
Measurable Characteristics 101 31
Konzernnorm
Descriptors: process capability investigation, process data analysis, capability index, quality
capability, managed process, capable process, process time model
Contents
Page
1 Purpose and scope............................................................................................................ 2
2 Principle of the process capability investigation ................................................................. 3
3 Theoretical principles......................................................................................................... 4
3.1 Process time models ......................................................................................................... 4
3.2 Distribution models............................................................................................................ 9
3.2.1 Normal distribution............................................................................................................. 9
3.2.2 Absolute value distribution, type I .................................................................................... 10
3.2.3 Type II absolute value distribution (Rayleigh distribution) ................................................ 12
3.2.4 Extended normal distribution ........................................................................................... 13
3.2.5 Mixed distribution ............................................................................................................ 14
3.3 Determination of capability .............................................................................................. 15
3.3.1 Determination of capability for defined distribution models .............................................. 16
3.3.2 Determination of capability for undefined distribution models .......................................... 25
3.4 Limit values for process capability ................................................................................... 28
3.5 Statistical tests ................................................................................................................ 30
4 Carrying out a process capability investigation ................................................................ 35
4.1 Test equipment use ......................................................................................................... 40
4.2 Sampling ......................................................................................................................... 40
4.3 Special regulation for restricted PFI................................................................................. 41
4.4 Data analysis................................................................................................................... 41
4.4.1 Selection of the expected distribution model.................................................................... 41
4.4.2 Test for outliers................................................................................................................ 42
4.4.3 Take outliers out of the calculation of statistics ................................................................ 42
4.4.4 Test for change of process control................................................................................... 42
4.4.5 Test for change of process location ................................................................................. 42
4.4.6 Evaluation according to process time models A............................................................... 43
4.4.7 Test for deviation from the instantaneous normal distribution .......................................... 44
4.4.8 Evaluation according to process time models C .............................................................. 44
4.4.9 Evaluation according to process time model B ................................................................ 45
4.4.10 Evaluation according to process time model D ................................................................ 45
4.5 Documentation ................................................................................................................ 46
4.6 Results evaluation ........................................................................................................... 47
4.7 Process improvement...................................................................................................... 49
4.8 Handling of unmanaged or incapable processes ............................................................. 49
5 Examples ........................................................................................................................ 50
6 Referenced standards ..................................................................................................... 60
7 Referenced literature ....................................................................................................... 60
8 Keyword index................................................................................................................. 61
Form FE 41 - 01.03
Page 1 of 64
Fachverantwortung/Responsibility Normung/Standards (EZTD, 1733)
K-QS-1/2 Neupert Tel.: -79298 Bestenreiner Tel.: +49-5361-9-26513 Sobanski
Confidential. All rights reserved. No part of this document may be transmitted or reproduced without prior permission of a Standards Department of the Volkswagen Group.
Parties to a contract can only obtain this standard via the B2B supplier platform “www.vwgroupsuppy.com”.
 VOLKSWAGEN AG
Page 2
VW 101 31: 2005-02
Introduction
An evaluation of the process capability regarding observable measurable production characteristics
is an important prerequisite for fulfillment of the specified quality requirements. However, for many
practical cases of the capability investigation, to date no standards or uniform Group regulations
exist, so in identical cases completely different capability evaluations could result. This standard
was developed in order to make it possible to carry out the capability investigation under uniform
rules for all practical cases and thus ensure the comparability of the results in the Volkswagen
Group.
This standard contains all of the theoretical principles, as a self-contained whole, that are needed
for use and understanding. Only the statistical tests that are already described in detail in
standards or standard works in the literature of statistics will be indicated with references.
To carry out a process capability investigation according to this standard, a computer program is
required in which the algorithms described are implemented. If such a computer program is
available, the user can basically concentrate on the rules described in Section 4 and look up
theoretical principles if necessary. The most important sub-sections in it are:
─ 4.2 Sampling
─ 4.5 Documentation
─ 4.6 Results evaluation
Section 5 also lists examples that can be used to help with the results evaluation.
1 Purpose and scope

The goal of a process capability investigation (PCI) is mainly the evaluation and the documentation
of the quality capability of a production process under real standard production conditions.
The PFI is also used for
─ Recognition of systematic influences, the prerequisite for targeted process improvement
─ Evaluation of the effect of process improvement measures
─ Specification of the control limits from quality control cards
─ Test whether the prerequisites for effective process control are given within the tolerance zone
How and under which conditions process capability investigation are to be carried out is the object
of this standard. It is applicable to any continuous1 (measurable) production characteristics or
process parameters2.
1 In the following, the designation “measurable characteristics” is used for this characteristic type instead of the designation
recommended by DIN “continuous characteristics”, since this has become established at VW.
2 In the following, production characteristics and process parameters are referred to as characteristic.
Page 3
VW 101 31: 2005-02
2 Principle of the process capability investigation

Because of the random influences different values result for a considered characteristic during the
production process for like parts. These characteristic values3 vary across a certain range due to
systematic influences, depending on production quality. Therefore there is a test of how well the
distribution of the characteristic values fits into the tolerance zone defined by the designer (Figure
1). The evaluation of this is expressed by the capability indexes cp and cpk (c as in capability),
whereby only the process dispersion is taken into consideration in the cp value and the process
location is additionally taken into consideration in the cpk value. These statistics must be at least as
great as the defined limit values in order to fulfill the requirement for a secure process.
To determine the capability indexes regarding the characteristic considered, an adequately large
number of random samples are taken from ongoing production, each typically with a size of 3 or 5,
over a suitable period of time for all influences from the categories machine, operator, material,
environment and method to take effect, if possible. From this total ramdom sample, the situation µ
and dispersion rang elimits x0.135% and x99.865% are estimated fort he population of characteristic
values (theoretically infinite quantity) and compared to the tolerance interval [Gu, Go] (Figure 1).
In this case, the dispersion range limits are specified in such a way that the percentage of
characteristic values outside the dispersion range is pe = 0,135% on both sides.
In addition, the time curve of a process and the distribution of the characteristic values are
analyzed using statistical tests to obtain information on systematic process influences.
Tolerance
Toleranz
definierte
Defined Prozeßstreubreite
process dispersion range
Frequency
Häufigkeit
pe = 0,135% pe = 0,135%
Gu x 0,135% µ x 99,865% Go
0 2 4 6 8 10 12
Characteristic
Merkmalswert value
Figure 1 - Example of a distribution of characteristic values within a defined tolerance zone
3 The term characteristic value should not be confused with the term measured value, since in contrast to the first, the latter contains an
uncertainty.
Page 4
VW 101 31: 2005-02
3 Theoretical principles
3.1 Process time models

An important criterion for evaluating a production process is its behavior over time. From this, a
decision can be made about whether the process is managed or not, whereby obtaining a
managed process is the prerequisite for goal-oriented process improvement.
A process can be considered managed if it has a consistent behavior, its behavior over time
follows a known regularity or if changes lie within known limits.
The process behavior over time can essentially be described using the statistics for location and
dispersion of the instantaneous distribution of characteristic values, the instantaneous mean value
µ m and the instantaneous standard deviation σm4.
According to DIN 55319 the process behavior over time is first divided into 4 basic cases, which
are identified in Table 15.
Table 1 - Identification of the process time models

Location µ m(t) Location µ m(t)
constant not constant
Dispersion σm(t)
A C
constant
Dispersion σm(t)
B D
not constant
In addition, depending on the type of instantaneous distribution, process time model A is divided
into
A1: Normal distribution
A2: Defined, but not normal distribution model
A2*: Undefined distribution model6
In addition, depending on the type of time-dependent location and the resulting distribution,
process time model C with instantaneous normal distribution is divided into7
C1: Non-systematic location changes8 and resulting normal distribution
C3: Systematic location change (trend and cycles) and extended normal distribution
C4: Systematic and non-systematic location changes and no resulting normal distribution
Figures 2 to 9 show the process time models schematically.
4 Further statistics include the skew and kurtosis of a distribution. These are not required for characterizing the process time models and
therefore are considered no further.
5 The process time models described previously in VDA Vol. 4, Part 1, Securing Quality before Introduction into Standard Production are
identified according to another system.
6 The process time model A2* was defined as an expansion to DIN 55319, in order, for example, to be able to differentiate an incorrect
sampling from different tools of a machine from the correct case.
7 Process time model C2 defined in DIN 55319 is not used in this standard since there is no statistical test that can differentiate it from
model C4 and the most important practical cases can already be described using models C1, C3 and C4.
8 Non-systematic situation changes shall not be confused with the random deviations from the values of a characteristic of the current
process situation. This is a case of systematic deviations from the process situation, that do not exhibit any regularity.
Page 5
VW 101 31: 2005-02
Process time model A1:

Location: constant
Variation: constant
Instantaneous distribution: normal distribution
Resulting distribution: normal distribution
Example: length dimensions of shafts
resultierende Verteilung
resulting distribution
Time
Zeit
it
Ze
Frequency
Häufigkeit
Häufigkeit
Merkmalswert
Merkmalswert
Characteristic value
Figure 2 - Process time model A1
Process time model A2:

Location: constant
Variation: constant
Instantaneous distribution: defined distribution model except for normal distribution
Resulting distribution: defined distribution model except for normal distribution
Example: deviations of radial runout
resultierende
Verteilung
Zeit
Time
it
Ze
Frequency
Häufigkeit
Häufigkeit
Characteristic
Merkmalswert
Merkmalswert value
Figure 3 - Process time model A2

Page 6
VW 101 31: 2005-02
Process time model A2*:

Location: constant
Variation: constant
Instantaneous distribution: undefined distribution model
Resulting distribution: undefined distribution model
Example: mixed distribution due to sampling from 2 tools
resultierende
Verteilung
Zeit
Time
it
Ze
Frequency
Häufigkeit
Häufigkeit
Merkmalswertvalue
Characteristic
Merkmalswert
Figure 4 - Process time model A2*
Process time model B:

Location: constant
Variation: not constant
Example: increase in dispersion due to wear of a machine bearing
Time
Zeit
it
Ze
Häufigkeit
Häufigkeit
Frequency
Merkmalswertvalue
Characteristic
Merkmalswert
Figure 5 - Process time model B

Page 7
VW 101 31: 2005-02
Process time model C1:

Location: non-systematic changes
Variation: constant
Resulting distribution: normal distribution
Example: process with jumps due to frequent tooling change
Time
Zeit
it
Ze
Häufigkeit
Frequency
Häufigkeit
Characteristic
Merkmalswertvalue
Merkmalswert
Figure 6 - Process time model C1

Location: systematic change (trend and cycles)
Variation: constant
Resulting distribution: extended normal distribution
Example: process with trend due to tool wear
resultingVerteilung
resultierende distribution
Time
Zeit
it
Ze
Frequency
Häufigkeit
Häufigkeit
Characteristic
Merkmalswert
Merkmalswert value

Page 8
VW 101 31: 2005-02

Location: systematic and non-systematic changes
Variation: constant
Resulting distribution: mixed distribution
Example: process with jumps due to batch change
Time
Zeit
it
Ze
Frequency
Häufigkeit
Häufigkeit
Merkmalswert
Merkmalswert
Process time model D:

Location: not constant
Variation: not constant
Instantaneous distribution: arbitrary distribution
Example: unmanaged process
Time
Zeit
it
Ze
Frequency
Häufigkeit
Häufigkeit
Characteristic
Merkmalswert
Merkmalswertvalue
Figure 9 - Process time model D

Page 9
VW 101 31: 2005-02
3.2 Distribution models

The distribution of characteristic values can be described by a distribution model for most types of
production characteristics. This means that production characteristics with two-sided tolerance
zones, e.g. length dimensions, diameters and torques can be based on a normal distribution.
In contrast, the variation behavior of production characteristics with upper limited tolerance zone
only can be described by absolute value distributions of type I or type II. For example, the type I
absolute value distribution can be used as the basis for the characteristic types parallelism and
asymmetry and the type II absolute value distribution can be used for the characteristic types
position and coaxiality.
In addition to the distribution models for processes with constant location and variation (process
time model A) there are distribution models that describe the variation behavior of processes with
constant variation but not constant location (process time models C). This means that the extended
normal distribution can be used in the case of a process with trend (model C3) and mixed
distribution can be used in the case of a process with batch jumps (model C4).
3.2.1 Normal distribution

The function of the probability density (density function for short) of a normal distribution that
is shown graphically in Figure 10 is:
1  x −µ 
2
−  
fN (x ) =
1 2 σ 
⋅e (2.1)
σ ⋅ 2π
with the parameters mean value µ and standard deviation σ, that identify the location and width
of a distribution, whereby the square of the standard deviation σ2 is identified as variance.
Wahrscheinlichkeitsdichte
Probability density
Wendepunkte
Turning points
pe = 0,135% pe = 0,135%
-4
µ-4σ -3
µ-3σ -2
µ-2σ -1
µ-σ 0µ 1
µ+σ 2
µ+2σ 3
µ+3σ 4
µ+4σ
Merkmalswert
Figure 10 - Function of the probability density of a normal distribution

Page 10
VW 101 31: 2005-02
The portion of the area below the graph in Figure 10 within a considered time interval can be
interpreted as probability. The probability of finding a characteristic value x in a population that is as
high as a considered limit value xg is thus specified by the integral function, the distribution
function. For normal distribution, this is
xg
FN (xg ) = ∫ fN (x ) ⋅ dx (2.2)
−∞
where
∞
∫ fN (x ) ⋅ dx = 1 (2.3)
−∞
and f N ( x ) ≥ 0 for all values x.

With the transformation
x−µ
u= (2.4)
σ
(2.1) produces the probability density of the standardized normal distribution
1
− u2
ϕ (u ) =
1
⋅e 2 (2.5)
2π
and the distribution function
ug
Φ (u g ) = ∫ ϕ (u ) ⋅ du (2.6)
−∞
with the standard deviation σ =1
3.2.2 Absolute value distribution, type I

The type I absolute value distribution results by convolution of the density function of a normal
distribution at the zero point, where the function values on the left of the convolution are added to
those on the right.
The density function and the distribution function of the type I absolute value distribution are thus
 − 1  x − µ N  
2 2
1  x + µN 
1  2  σ N  −  
 
fB1 (x ) =
2 σ
⋅ e +e  N   for x ≥ 0 (2.7)
σ N ⋅ 2π  
 
 xg − µN   x + µN 
FB1 (x g ) = Φ   +Φ g
  σ
 −1
 (2.8)
 σN   N 
where
µ N: mean value of the original normal distribution that identifies a systematic zero point shift
σN: standard deviation of the original normal distribution
Φ: distribution function of the standardized normal distribution
Page 11
VW 101 31: 2005-02
Figure 11 shows density functions that result from convolution of the density of the normal
distribution at different zero point shifts
µN = 0 µN = 1σN µN = 2σN µN = 3σN

Probability density
0 1σN 2σN 3σN 4σN 5σN 6σN

Merkmalswertvalue
Characteristic
Figure 11 - Density functions of the type I absolute value distribution with different zero
point shifts
Mean value and variance of the type I absolute value distribution are:
2
1 µ 
 µ   − µN  2 ⋅ σ N −  N
2σN


µ = µ N ⋅ Φ  N  − Φ    +
 ⋅e 
(2.9)
 σN   σN  2π
σ 2 = σ N2 + µ N2 − µ 2 (2.10)
For the case of a zero point shift µ N = 0, the following result from (2.9) and (2.10)
2 ⋅σ N
µ= (2.11)
2π
 2 2
σ 2 = 1 −  ⋅σN (2.12)
 π
As Figure 11 shows, with increasing zero point shift, the type I absolute value distribution
approaches a normal distribution. Thus for the case
µ
≥3 (2.13)
σ
the type I absolute value distribution can be replaced with a normal distribution with good
approximation.
Page 12
VW 101 31: 2005-02
3.2.3 Type II absolute value distribution (Rayleigh distribution)

The type II absolute value distribution results from the vectorial values of the orthogonal
components x and y of a two-dimensional normal distribution, where equal standard deviations are
assumed for the components. This case is present for many production characteristics in the form
of radial deviations r from a considered point or a considered axis.
The density function and the distribution function of the type II absolute value distribution are
generally
1  z 2 +r 2  z ⋅r ⋅cos α
−   2π
r 
f B 2 (r ) =
2  σ N2 σ N2
⋅e 
⋅ ∫e dα for r ≥ 0 (2.14)
2π ⋅ σ N2 0
rg
FB 2 (rg ) = ∫ fB 2 (r ) ⋅ dr (2.15)
0
where
σN: Standard deviation of the orthogonal components x and y, from which the radial deviation r
of a reference point or reference axis results
z: Eccentricity: distance between coordinate origin and frequency midpoint
Figure 12 displays density functions of the type II absolute value distribution that result in units of
σN for different eccentricities.
z=0 z = 1σN z = 2σN z = 3σN

Probability density
0 1σN 2σN 3σN 4σN 5σN 6σN

Merkmalswert
Figure 12 – Density functions of the type II absolute value distribution with different
eccentricities
Mean value and variance of the type II absolute value distribution are:
Page 13
VW 101 31: 2005-02
∞
µ = ∫ fB 2 (r ) ⋅ r ⋅ dr (2.16)
0
σ 2 = 2σ N2 + z 2 − µ 2
(2.17)
For an eccentricity z = 0, (2.14) and (2.15) density function and distribution function of the Weibull
distribution, with the shape parameter value 2 yield
2
1 r 
−  

r
fB 2 (r ) = ⋅e 2σN 
(2.18)
σ N2
2
1  rg 
−  
FB 2 (rg ) = 1 − e
2 σN 

(2.19)
and from them, in turn, mean value and variance
π
µ = σN ⋅ (2.20)
2
 π
σ 2 = 2 −  ⋅σN
2
(2.21)
 2
As Figure 12 shows, with increasing eccentricity, the type II absolute value distribution approaches
a normal distribution. Thus for the case
µ
≥6 (2.22)
σ
the type II absolute value distribution can be replaced with a normal distribution with good
approximation.
3.2.4 Extended normal distribution

The extended normal distribution approximately results as the resulting distribution from the
superimposition of instantaneous normal distributions that have equal standard deviations but with
mean values that shift increasingly in a linear manner over time with respect to the initial location
(see process time model C3 in Section 3.1).
The density function of the extended normal distribution is:
 − 1  x − µ N1 
2
 2  σ N 
e x < µ N1
1
f EN ( x ) = ⋅ 1 for µ N1 ≤ x ≤ µ N 2 (2.23)
2π ⋅ σ N + µ N 2 − µ N1  −  N2 
 − µ 
2
x > µN 2
1 x
e 2  σ N 

where
µ N1 , µ N 2 : mean values of the normal distributions that limit the extended normal
distribution
σN : standard deviation of the normal distributions that limit the extended normal
distribution
Page 14
VW 101 31: 2005-02
An example of an extended normal distribution is shown in Figure 13
Probability density
σN σN
µN1 µN2
Merkmalswert
Figure 13 - Density function of an extended normal distribution
3.2.5 Mixed distribution

Mixed distribution (see also process time model C4 in Section 3.1) results from superimposing
several weighted normal distributions that have the same standard deviations but different mean
values, as follows:
2
1  x −µ l 
L −  
f M* (x ) = ∑ g l ⋅ e
2 σm 

(2.24)
l =1
with the weighting coefficients

tl
gl = (2.25)
tg
where
L : number of superimposed normal distribution
µ l : mean value of the Ith normal distribution
σ m : standard deviation of the superimposed instantaneous normal distributions
t l : Ith time range with constant process location
t g : total examination period
With the normalizing factor

∞
c= ∫ fM (x ) ⋅ dx
*
(2.26)
−∞
Page 15
VW 101 31: 2005-02
(2.24) yields the probability density of the mixed distribution

1 *
fM ( x ) = ⋅ fM (x ) (2.27)
c
An example of a mixed distribution is shown in Figure 14

Probability density
σk
σk
µ1 µ2
Merkmalswert
Figure 14 - Density function of a mixed distribution from two normal distributions with
different mean values
3.3 Determination of capability

The capability indexes cp and cpk indicate how well the process results comply with the tolerance
zone of a considered characteristic. In this case, only the process dispersion is taken into
consideration in the cp value. The process location is taken into consideration by the cpk value.
Because of this, on the one hand, there can be an expression of what value is possible in an ideal
process location and, on the other, a comparison of the two values makes it possible to express
how much the process location deviates from the specified value. The higher the capability indexes
determined, the better the process.
There are different evaluation formulas, which have to be selected appropriately in the individual
case, for determining the capability indexes. Since the determination of the capability indexes can
only be carried out using random sample values, the results only represent estimates of the values
to be determined for the population and are thus identified by a roof symbol9.
9 The identification of the estimated capability characteristic values by a roof symbol is only important for understanding the theory, so
they can be dispensed with during the evaluations in practice.
Page 16
VW 101 31: 2005-02
3.3.1 Determination of capability for defined distribution models
3.3.1.1 Capability indexes

For a production characteristic to be examined that has random sample values that are not in
contradiction to a theoretically expected distribution model, the capability indexes are estimated
according to the following formulas:
Capability indexes for characteristics with two-sided tolerance zones (according to DIN 55319,
measuring method M4), e.g. for length dimension:
Go − Gu
ĉ p = (3.1)
x̂ 99 ,865 % − x̂ 0 ,135 %
 Go − µˆ µˆ − Gu 
ĉ pk = min  ;  (3.2)
 x̂ 99 ,865% − µˆ µˆ − x̂ 0 ,135% 
Capability indexes for characteristics with upper limited tolerance zone and natural lower limit value
zero, e.g. for radial runout deviation:
Go
ĉ p = (3.3)
x̂ 99 ,865 % − x̂ 0 ,135 %
Go − µˆ
ĉ pk = (3.4)
x̂ 99 ,865 % − µˆ
Capability index for characteristics with lower limited tolerance zone only, e.g. for tensile strength:
µˆ − Gu
ĉ pk = (3.5)
µˆ − x̂ 0 ,135%
where
Go, Gu : Upper and lower limiting value, respectively
µ̂ : estimated mean value of the process in the examination period
x̂ 0 ,135 % , x̂ 99 ,865 % : Estimated values for dispersion range limits (quantiles, below which the
specified percentage p of measured values lie)
Another useful statistic for evaluating a process for characteristics with two-sided tolerance zones
is the instantaneous capability index
Go − Gu
ĉ *p = (3.6)
6 ⋅ σˆ m
where
σ̂ m : estimated instantaneous (or also inner) standard deviation of the process
This makes it possible to indicate what capability index could be achieved under the ideal
conditions of a constant and centered process. This value is thus especially important for setting up
the control limits of quality control cards, e.g. for processes with trend or batch jumps.
Page 17
VW 101 31: 2005-02
3.3.1.2 Estimation of the statistics
The statistics mean value µ and standard deviation σ of a population can be estimated,
independently of the distribution model, from the measured values of m random samples according
to expectation, by
1 m
µ̂ = x = ⋅∑nj ⋅ xj (3.7)
n g j =1
( )
m
1
⋅ ∑nj ⋅ x j − x
2
σ̂ = s = (3.8)
m − 1 j =1
with the mean values of the single random samples

n
1 j
xj = ⋅ ∑ x ji (3.9)
n j i =1
and the effective total random sample size (i.e. without outliers)
m
ng = ∑ n j (3.10)
j =1
where
m: number of random samples
nj : effective size of the jth random sample (i.e. without outliers)
xji : ith measured value of the jth random sample
The instantaneous standard deviation is determined by
1 m 2
σ̂ m = s m = ⋅ ∑s j (3.11)
m j =1
with the variances of the single random samples
n
⋅ ∑ (x ji − x j )2
1 j
s 2j = (3.12)
n j − 1 i =1
In the case of data to be evaluated in the form of frequency distributions of classified

measurements and ranges, e.g. from manually recorded control cards with a class subdivision of
the value range and measured value recordings (tally), the statistics µ and σ can be estimated by
1 K
µ̂ = x = ⋅ ∑ ak ⋅ x k (3.13)
n g k =1
( )
K
1
⋅ ∑ ak ⋅ x k − x
2
σ̂ = s = (3.14)
n g − 1 k =1
where
Page 18
VW 101 31: 2005-02
x k : mean value of the kth class

a k : absolute frequency of the measured values in the kth class
K : maximum number of measured value classes
The estimation of the instantaneous standard deviation is carried out with the assumption of an
instantaneous normal distribution from the classified ranges Rk of the single random samples by
R
σˆ m = (3.15)
dn
where
1 Κ
R= ⋅ ∑ ak ⋅ Rk (3.16)
m k =0
R k = k ⋅ ∆x (3.17)
d n = d n − (d n − d n − 1 ) ⋅ (n − n ) (3.18)
ng
n= (3.19)
m
n −1 < n ≤ n (3.20)
ak : absolute frequency of the ranges in the kth class
Κ : maximum class range
k : class quantity from the difference between the classes with maximum and minimum
measured value of a random sample (class range)
∆x : Class width
d n : Expected value of the w distribution (see also [1])
The value dn is given in Table 2 for a few random sample sizes
Table 2 - Expected value of the w distribution in relationship to n

ng dn
2 1,128
3 1,693
4 2,059
5 2,326
6 2,534
7 2,704
8 2,847
9 2,970
10 3,078
Page 19
VW 101 31: 2005-02
3.3.1.3 Estimation of the dispersion range limits

The dispersion range limits depend on the distribution model and are estimated as follows:
Dispersion range limits for normal distribution:

If the normal distribution model is the fitting distribution model, the values µ̂ and σ̂ determined
according to (3.7) and (3.8) or (3.13) and (3.14) respectively result as estimated values for the
dispersion range limits
x̂ 99 ,865 % 
 = µˆ ± 3σˆ (3.21)
x̂ 0 ,135 % 
which when used in formulas (3.1) and (3.2), result in the classic formulas for calculating the
capability indexes.
Dispersion range limits of the type I absolute value distribution:

To determine the dispersion range limits for a type I absolute value distribution, first the statistics µ
and σ are estimated according to formulas (3.7) and (3.8) or (3.13) and (3.14) respectively.
For the case µˆ σˆ < 3 , the estimated statistics µ and σ are used to estimate the parameter values
µ N and σN to be determined of the type I absolute value distribution to be fitted, in the following
manner:
Equation (2.9) yields the function
2
1  µˆ 
µˆ  µˆ  µˆ N   µˆ   − µˆ N  2 −  N
2  σˆ N


= ψ B1  N  = ⋅ Φ  N  − Φ    +
 ⋅e 
(3.22)
σˆ N  σˆ N  σˆ N   σˆ N   σˆ N  2π
With equation (2.10), this yields the following function
 µˆ N 
ψ B1  
µˆ  µˆ  σˆ
 N
= ϑ B1  N  = (3.23)
σˆ  σˆ N  2 2
 µˆ    µˆ 
1 +  N  − ψ B1  N  
 
 σˆ N    σˆ N 
Equations (2.11) and (2.12) result in the condition
µˆ 2
≥ = 1,3236 (3.24)
σˆ π −2
Under the condition (3.24), the parameter values to be determined of the type I absolute value
distribution can be estimated using
Page 20
VW 101 31: 2005-02
 µˆ 
2
1+ 
 σˆ 
σˆ N = σˆ ⋅ 2
(3.25)
  µˆ  
1 +  ξ B1   
  σˆ  
 µˆ 
µˆ N = ξ B1   ⋅ σˆ N (3.26)
 σˆ 
where
 µˆ 
ξ B1   : inverse function of (3.23)
 σˆ 
In the case where the ratio µˆ σˆ is lower than the limit value 1.3236 from condition (3.24) because
of random deviations of the random sample statistics, the ratio µˆ σˆ is set to this limit value, at
which the following parameter values result:
µˆ N = 0 and according to formula (2.12)
π
σˆ N = ⋅ σˆ = 1,659 ⋅ σˆ (3.27)
π −2
The connection between the parameter values µN and σN of the type I absolute value distribution
and the statistics µ and σ is shown graphically in Figure 15, related to σ
3,5
parameter
3,0
Verteilungsparameter
µN / σ
2,5
relativerdistribution
2,0
1,5
σN / σ
1,0
Relative
0,5
0,0
1,2 1,4 1,6 1,8 2,0 2,2 2,4 2,6 2,8 3,0
Relative Lage µ / σ
location
relative
Figure 15 - Relative parameter values of the type I absolute value distribution in relationship
to the relative location
Page 21
VW 101 31: 2005-02
For the fitted type I absolute value distribution, the dispersion range limits can then be determined
numerically and their relationships with the relative location are shown in Figure 16
7
range limit
6 x99,865% / σ
Streubereichsgrenze
5
relative dispersion
3
Relative
1 x0,135% / σ
0
1,2 1,4 1,6 1,8 2,0 2,2 2,4 2,6 2,8 3,0
relative
location
Figure 16 - Relative dispersion range limits of the type I absolute value distribution in
relationship to the relative location
For direct determination of the statistics µ N and σN from the statistics µ and σ, for
1,3236 ≤ µˆ σˆ < 3 the following approximation can also be used as an inverse function of (3.23)
with adequate precision (error related to σ less than 0.01)
 µˆ   µˆ  µˆ
0 ,286 1,05
 
ξ B1   ≅ 1,64 ⋅  − 1,3236  + 0 ,634 ⋅  − 1,3236  (3.28)
 σˆ   σˆ   σˆ 
In addition, the determination of the dispersion range limits for 1,3236 ≤ µˆ σˆ < 3 is carried out
directly with the use of the following approximation (error related to σ less than 0,02)
  µˆ 
0 ,67
 µˆ 
x̂ 99 ,865 %  − 2 ,47 ⋅  − 1,3236  + 2 ,505 ⋅  − 1,3236  + 5 ,3171
   σˆ   σˆ 
 ≅ σˆ ⋅ 4
(3.29)
 µˆ 
x̂ 0 ,135 %   0 ,018 ⋅  − 1,3236  + 0 ,0028
  σˆ 

For µˆ σˆ ≥ 3 the calculation of the dispersion range limits is carried out according to formula
(3.21).
Dispersion range limits of the type II absolute value distribution:

To determine the dispersion range limits for a type II absolute value distribution, first the statistics µ
and σ are estimated according to formulas (3.7) and (3.8) or (3.13) and (3.14) respectively.
For the case µˆ σˆ < 6 , the estimated statistics µ and σ are used to estimate the parameter values
z and σN to be determined of the type II absolute value distribution to be fitted, in the following
manner:
Page 22
VW 101 31: 2005-02
Equations (2.14) and (2.16) yield the function

2
1  ẑ 
∞
µˆ  ẑ  −   − v
1 2 2π ẑ
⋅v ⋅cos α 
⋅ ∫  v 2 ⋅ e 2 ⋅ ∫ e σ N dα  ⋅ dv
1 2  σˆ N 
= ψ B 2   = ⋅e 
(3.30)
σˆ N  σˆ N  2π 0
 
0 
where
r
v= (3.31)
σˆ N
With equation (2.17), this yields the following function
 ẑ 
ψ B 2  
µˆ  ẑ   σˆ N 
= ϑ B 2   = (3.32)
σˆ  σˆ N  2

2
 ẑ   ẑ 
2 +   − ψ B 2   

 σˆ N    σˆ N 
Equations (2.20) and (2.21) result in the condition
µˆ π
≥ = 1,9131 (3.33)
σˆ 4 −π
Under the condition (3.33), the parameter values to be determined of the type II absolute value
distribution can be estimated using
 µˆ 
2
1+ 
σˆ N = σˆ ⋅  σˆ 
2
(3.34)
  µˆ  
2 +  ξ B 2   
  σˆ  
 µˆ 
ẑ = ξ B 2   ⋅ σˆ N (3.35)
 σˆ 
where
 µˆ 
ξ B 2   : inverse function of (3.32)
 σˆ 
In the case where the ratio µˆ σˆ is lower than the limit value 1,9131 from condition (3.33) because
of random deviations of the random sample statistics, the ratio µˆ σˆ is set to this limit value, at
which the following parameter values result:
ẑ = 0 and according to formula (2.21)
2
σˆ N = ⋅ σˆ = 1,526 ⋅ σˆ (3.36)
4 −π
Page 23
VW 101 31: 2005-02
The connection between the parameter values z and σN of the type II absolute value distribution
and the statistics µ and σ is shown graphically in Figure 17, related to σ
7
parameter
Verteilungsparameter
5
z/σ
relativerdistribution
2
Relative
σN / σ
0
1,5 2,0 2,5 3,0 3,5 4,0 4,5 5,0 5,5 6,0
location
relative
Figure 17 - Relative parameter values of the type II absolute value distribution in

For the fitted type II absolute value distribution, the dispersion range limits can then be determined
numerically and their relationships with the relative location are shown in Figure 18
9
parameter
8
x99,865% / σ
Streubereichsgrenze
6
relativedistribution
3
Relative
2
x0,135% / σ
1
0
1,5 2,0 2,5 3,0 3,5 4,0 4,5 5,0 5,5 6,0
relative
location
Figure 18 - Relative dispersion range limits of the type II absolute value distribution in
Page 24
VW 101 31: 2005-02
For direct determination of the statistics z and σN from the statistics µ and σ, for 1,9131 ≤ µˆ σˆ < 6
the following approximation can also be used as an inverse function of (3.32) with adequate
precision (error related to σ less than 0,02)
 µˆ   µˆ  µˆ
0 ,313 1,22
 
ξ B1   ≅ 2 ,1 ⋅  − 1,9131  + 0 ,466 ⋅  − 1,9131  (3.37)
 σˆ   σˆ   σˆ 
In addition, the determination of the dispersion range limits for 1,9131 ≤ µˆ σˆ < 6 is carried out
directly with the use of the following approximation (error related to σ less than 0,03)
  µˆ 
0 ,62
 µˆ 
0 ,9
x̂ 99 ,865%   − 1,64 ⋅  − 1,9131  + 2 ,075 ⋅  − 1,9131  + 5 ,5485
   σˆ   σˆ 
 ≅ σˆ ⋅  (3.38)
  µˆ  µˆ
0 ,9
    
x̂ 0 ,135%  2 ,6 ⋅ exp −  − 1,7031  ⋅ 0 ,8  + 1,425 ⋅  − 1,9131  − 2 ,1206
   σˆ    σˆ 

For µˆ σˆ ≥ 6 the calculation of the dispersion range limits is carried out according to formula
(3.21).
Dispersion range limits of the extended normal distribution:

In the case of an extended normal distribution, the dispersion range limits can be estimated directly
according to the following formulas:
x̂ 99 ,865% = µˆ N 2 + u ⋅ σˆ N
1 −α
*
2
x̂ 0 ,135% = µˆ N1 − u ⋅ σˆ N (3.39)
1 −α
*
2
where
u : quantile of the normal distribution
1−α
*
2
where
α *  µˆ − µˆ N1 
= 1 + N2 ⋅ 0 ,00135 (3.40)
2  2π ⋅ σˆ N 
and
5
µˆ N1 = µˆ N 2 = µˆ for σˆ ≤ ⋅ σˆ m
4
µˆ N 2   3  5
 = µˆ ± 3 ⋅  σˆ − σˆ m − ⋅ σˆ m 
2 2
for σˆ > ⋅ σˆ m (3.41)
µˆ N1   4  4
 5σˆ 2 + σˆ m2 5


σˆ ≤ ⋅ σˆ m
σˆ N = 6 for 4 (3.42)
5
 47 ⋅ σˆ σˆ > ⋅ σˆ m
 32 m 4
The estimation of the standard deviation σ of the extended normal distribution and σm of the
instantaneous normal distribution is carried out according to formula (3.8) or (3.14) or (3.11) or
(3.15) respectively.
Page 25
VW 101 31: 2005-02
Dispersion range limits of the mixed distribution:

In the case of a mixed distribution, first the points of significant changes in the process location
(jump points) are determined using a statistical test (see Section 3.5). Then the mean values within
these points are estimated according to
j ol n j
1
µ̂ l = x l =
ml
∑ ∑ x ji (3.43)
j = j ul i =1
and the instantaneous standard deviation σ m is estimated according to formula (3.15), with
m l = j ol − j ul + 1 (3.44)
where
m l : number of random samples within the Ith range of constant process location
j ol , j ul : random sample number of the upper/lower limit of the Ith range of a constant process
location
By superimposition of the L normal distributions with the mean values µ̂ l and the standard
deviation σ̂ m , the mixed deviation results according to formulas (2.24) to (2.27), from which the
dispersion range limits can then be determined numerically.
For the case where no jump points can be determined, instead of using formula (2.24), a
superimposition is carried out of L = m normal distributions with the mean value µˆ j = x j by
 x − µˆ j
2
n 
m −  
f M* (x ) = ∑ e
2 ⋅(n −1 )  σˆ m 

(3.45)
j =1
3.3.2 Determination of capability for undefined distribution models

If no fitting distribution model can be assigned to a production characteristic or the measured
values from the random samples contradict the assumed distribution model, a distribution-free
estimate of the capability index is carried out according to the range method, in the following
modified form under consideration of the random sample size10:
Capability indexes for characteristics with two-sided tolerance zones:

Go − Gu
ĉ p = (3.46)
x̂ o − x̂ u
G − x̂ 50 % x̂ 50% − Gu 
ĉ pk = min  o ;  (3.47)
 x̂ o − x̂ 50 % x̂ 50% − x̂ u 
10 In this case, it is simply a matter of a somewhat different representation of the same method of calculation as in the previous
Volkswagen operating equipment specification BV 1.40 for characteristics without normal distribution.
Page 26
VW 101 31: 2005-02
Capability indexes for characteristics with upper limited tolerance zone and natural lower limit zero:
Go
ĉ p = (3.48)
x̂ o − x̂ u
Go − x̂ 50%
ĉ pk = (3.49)
x̂ o − x̂ 50%
Capability index for characteristics with lower limited tolerance zone only:
x̂ 50% − Gu
ĉ pk = (3.50)
x̂ 50% − x̂ u
where
x̂o , x̂u : estimated values of the upper and lower dispersion range limits
x̂50% : estimated value of the 50% quantile
In the case of single values

x̂50% = ~
x (3.51)
where
~
x : median value, the value that lies in the middle of an ordered sequence of measurements
The dispersion range limits are estimated with

x̂ o  R
 = xc ± k ⋅ (3.52)
x̂ u  2
where
x max + x min
xc = (3.53)
2
R = x max − x min : range (3.54)
x max , x min : maximum and minimum measured value, respectively of the effective total random
sample
By the correction factor

6
k= (3.55)
dn
the effective total random sample size ng will be taken into consideration, where
d n : expected value of the w distribution11
11 Strictly speaking, a normally distributed population of the individual values is a prerequisite for the expected value of the w
distribution. However, this prerequisite is not taken into consideration due to the lack of a suitable method for the distribution-free
calculation of the capability characteristic values.
Page 27
VW 101 31: 2005-02
The value dn is given in Table 3 for a few total random sample sizes ng.
Table 3 - Expected value of the w distribution in relationship to ng

ng dn
30 4,09
35 4,21
40 4,32
45 4,42
50 4,50
60 4,65
70 4,77
80 4,87
100 5,04
125 5,21
150 5,34
200 5,55
250 5,71
For random sample sizes that are greater than 20, the expected values of the w distribution can be
determined according to the following approximation formula:
d n ≅ 1,748 ⋅ (ln(n g ))
0 ,693
(3.56)
In the case of data to be evaluated in the form of frequency distributions of classified measured
values and ranges
x̂o , x̂u are the upper limit, lower limit, respectively, for the top/bottom occupied class
and
n − Auk
n
x̂ 50% = x uk + 2 ⋅ ∆x for Auk < ≤ Auk + a k (3.57)
ak 2
where
x uk : lower limit of the kth class
∆x : class width
a k : absolute frequency of the measured values in the kth class
Auk : absolute cumulative frequency of the measured values to the lower limit of the kth class
Page 28
VW 101 31: 2005-02
3.4 Limit values for process capability

To obtain the process capability for a considered characteristic, the capability indexes determined
must fulfill the following requirement with respect to the specified limit values c p ;limit and c pk ;limit :
for characteristics with two-sided tolerance zones:

cˆ p ≥ c p;limit and cˆ pk ≥ c pk ; limit (4.1)
for characteristics with one-sided tolerance zone only12:
cˆ pk ≥ c pk ; limit (4.2)
with an effective total sample size of n g ≥ 125
Unless otherwise agreed, the following capability limit values apply:

c p;limit = 1,33
c pk ;limit = 1,33
In cases where, with reasonable effort, an examination can only be carried out with an effective
total random sample size lower than 125, the resulting uncertainty of the capability indexes
determined must be taken into consideration by appylying correspondingly higher limit values as
follows.
The determination of limit values for effective total random sample sizes lower than 125 is related
here to limit values that comply with the requirement from (4.1) or (4.2) for the population to be
investigated with 95% probability (lower confidence range limits). With the assumption of a
normally distributed population, these result from the upper confidence range limit of the standard
deviation
124
σ o = σˆ ⋅ (4.3)
χ 52%;124
and the statistical percentage range for the process dispersion

x 99 ,865%   1  124
 = µˆ ± u 99 ,865% ⋅ 1 + ⋅ ⋅ σˆ (4.4)
x 0 ,135%   2 ⋅ 125  χ 52%;124
where
u 99 ,865 % = 3 ,0 : quantile of the standardized normal distribution
χ 52%;124 = 99 ,3 : quantile of the chi-square distribution with a degree of freedom of f = 124 (see
also [1])
12 Although no limit value for Cp is specified for characteristics with upper limited tolerance zone only, a Cp value is to be determined
and specified, since from the comparison of this value to the Cpk value, it is possible to recognize whether an undesirable situation
shift exists. For example, if the Cpk value is higher than the Cp value, the distribution lies closer to the natural limit value zero than to
the maximum value.
Page 29
VW 101 31: 2005-02
By transformation and substitution in the evaluation formulas (3.1) and (3.2), this results in the
capability limit values for the population
χ 52%;124
c p ≥ c p;limit ⋅ = 0,895 ⋅ c p;limit (4.5)
124
1 χ 52%;124
c pk ≥ c pk ;limit ⋅ ⋅ = 0,891 ⋅ c pk ;limit (4.6)
1+ 1 124
250
Thus, for the effective total random sample sizes n g < 125 , the following fitted capability limit
values13 result
f
cˆ p ≥ c p;limit ⋅ 0,895 ⋅ (4.7)
χ 2
5%; f
 1  f
cˆ pk ≥ c pk ;limit ⋅ 0,891 ⋅ 1 + ⋅ (4.8)
 2 ⋅ ng  χ 5%; f
2
 
with the degree of freedom
f = ng − 1 (4.9)
Example:
For specified capability limit values of c p ;limit = 1,33 , c pk ;limit = 1,33 and an effective total random
sample size of n g = 50 , according to the formulas (4.7) to (4.9) the following fitted limit values
result for characteristics with two-sided tolerance zones:
50 − 1
ĉ p ≥ 1,33 ⋅ 0 ,895 ⋅ = 1,43
33 ,9
 1  50 − 1
ĉ pk ≥ 1,33 ⋅ 0 ,891 ⋅ 1 + ⋅ = 1,44
 2 ⋅ 50  33 ,9
Since for a characteristic with two-sided tolerance zones, the Cp value cannot be lower than the
Cpk value, in this case, the limit value for cp is equated to the limit value for cpk.
13 The determination of the adjusted capability limit values using formulas (4.7) to (4.9) is also used for populations without normal
distribution, since there are no other methods for these at this time and thus at least a usable consideration of a total random sample
size is carried out that is lower than 125.
Page 30
VW 101 31: 2005-02
3.5 Statistical tests

Ideally, the measured values recorded in a process capability investigation should not have any
─ unexpectedly great deviation of single measured values (outliers) in comparison to the
dispersion of other measured values,
─ significant change of the process location and dispersion,
─ significant deviation from the expected distribution model.
Otherwise additional systematic influences on production have to be taken into consideration. The
causes should then be known for this behavior and their effect should be acceptable in order to
fulfill the requirement of a secure manufacturing process.
Therefore, to test the above mentioned criteria the appropriate statistical tests are to be used
during a process capability investigation. Since these tests are described in detail in standards and
standard statistics references, they will only be indicated by the following references:
─ Test for outliers using the distribution-free test according to Hampel in modified form (see
VW 10133)
─ Test for change of the process location using the distribution-free Kruskal-Wallis test (see
[2])
─ Test for change of process dispersion using the Cochran test14 (see [1])
─ Test for deviation from the normal distribution according to Epps-Pulley (see ISO 5479)
─ Test for deviation from instantaneous normal distribution using the extended Shapiro-Wilk
test15 (see ISO 5479, Section 9)
─ Test for deviation from any specified distribution model using the chi-square test (see [1])
If the only data to be evaluated available are classified measured values and ranges, then
─ for testing the process location, a modified form of the simple variance analysis16 (for simple
variance analysis, see [3])
─ for testing the processing dispersion, a modified form of the Cochran test
can be used in which the variances of the single random samples for the test variables are
determined from the ranges analogously to formula (3.15). However, a test for deviation from the
instantaneous normal distribution can then not be used.
The statistical tests all proceed according to the following system:
─ Setting up the null hypothesis H0 and the alternative hypothesis H1, e.g.
H0: The population of the measured values of the characteristic considered has normal
distribution
H1: The population of the measured values of the characteristic considered does not have
normal distribution
─ Determination of the confidence level γ = 1- α or probability of error α (usually γ = 95%,
99% or 99,9%)
─ Setting up the formula for the test variable
14 Strictly speaking, an instantaneous normal distribution is the prerequisite for the Cochran test. However, the corresponding
distribution-free Levene test has proven to be unsuitable for simulations. With increasing random sample quantity, it too often
indicated dispersion changes which did not exist. In addition, only random samples with the same sizes are used for this test.
15 Although in ISO 5479 the extended Shapiro-Wilk test has the prerequisite n > 8, the test is also applicable for smaller random sample
sizes n > 3, as simulations of it showed.
16 For this test, in fact, an instantaneous normal distribution is required, but since for this type of data no other test is currently available,
this requirement cannot be taken into consideration. Because of this only the accuracy changes somewhat in the case of an
instantaneous distribution that is not normal.
Page 31
VW 101 31: 2005-02
─ Calculating the test value from the random sample values according to the test variable
formula
─ Determination of the threshold value (quantile) of the test distribution
─ Comparison of the test value to the threshold value for decision of whether a contradiction to
the null hypothesis exists and thus the alternative hypothesis is valid
It should be noted that in a statistical test with the specified confidence level γ possibly only a
contradiction to the null hypothesis can be proven, e.g. that a significant deviation of the measured
values from a normally distributed population exists. If no contradiction to the null hypothesis
results from the test result, this is not a confirmation of the validity of the null hypothesis. In this
case, the specified confidence level does not allow proving that a normally distributed population
exists, for example. Then, analogously to the legal principle "if in doubt, find for the defendant," a
decision is simply made for assumption of the null hypothesis.
The probability of error α indicates the risk of rejecting the null hypothesis on the basis of the test
result, although it applies (α risk). However, not just any small value can be specified for the
probability of error, since that would increase e.g. the risk of not discovering an actual deviation
from a normal distribution (β risk).
In the case where there is a process with constant dispersion, but not constant location, additional
statistical tests are necessary to be able to assign the time curve of the location to the
corresponding process time model C1, C3 or C4 (see Section 3.1), as well as to assign the
corresponding distribution model to the resulting distribution. This assignment is possible with the
use of the tests described in the following:
Test for trend under consideration of location jumps

In the case of constant dispersion, but not constant location, first there is a test of whether a trend
exists. Generally in the test period, processes with trend do not exhibit a monotonically increasing
or falling time curve. It is much more the case that its curve is interrupted by location jumps that
result from corrective measures after the control limits of quality control cards have been exceeded
(so-called sawtooth curve, see Figure 19). A test for trend over these jump points would lead to an
unusable result. Therefore, in the first step, the areas must be determined that lie between the
jump points.
value
Messwert
Measured
0 5 10 15 20 25 30 35 40 45 50
Stichprobe
Sampling
Figure 19 - Example of a process curve with trend and location correction

Page 32
VW 101 31: 2005-02
Test for location jump:
H0: location does not change or only changes steadily

H1: location changes with jumps
Specified probability of error: α = 1%
Test variable for all random sample numbers j from 1 to m:
(∆x j )2 = (x j +2 + x j +1 + x j − x j −1 − x j −2 − x j −3 )2 (5.1)
where
x j : mean value of the jth random sample size n
Threshold value:
6 27
E = χ 12−α ;1 ⋅ ⋅ σˆ m2 + ⋅ σˆ R2 (5.2)
n 4
where
σˆ R2 = σˆ 2 − σˆ m2 (5.3)
σ̂ 2 : total variance estimated according to (3.8)

σ̂ m2 : instantaneous variance estimated according to formula (3.11)
χ 2
1−α ;1 : quantile of the chi-square distribution with degree of freedom 1
H1 applies, if jth test value
(∆x j )2 >E (5.4)
From j, the lth jump point is determined using
j +2 {
(∆x j +2 ) = max (∆x j +2 ) ,(∆x j +1 ) , (∆x j )
2 2 2 2
}
2
{
J l =  j + 1 for (∆x j +1 ) = max (∆x j + 2 ) , (∆x j +1 ) , (∆x j )
 2 2 2
} (5.5)
 j
 {
(∆x j )2 = max (∆x j +2 )2 , (∆x j +1 )2 , (∆x j )2}
Thus the lth area between two jump points
J ul ≤ j ≤ J ol (5.6)
is specified by
 1 l =1
J ul =  for (5.7)
J l −1 + 1 l >1
J l − 2 l <L
J ol =  for (5.8)
 m l =L
Page 33
VW 101 31: 2005-02
where
L: number of areas free of jump points
In the second step, the test for trend occurs. To do this, with a random sample quantity m l ≥ 7 ,
first the constant a0 and the slope a1 of the regression lines for the areas free of jump points
determined are estimated as follows:
â0 = x l − â1 ⋅ j l (5.9)
s xJ
â1 = (5.10)
s J2
where
1  J ol 
s xJ = ⋅  ∑ j ⋅ x j − ml ⋅ j l ⋅ x l  : covariance (5.11)
m l − 1  j = Jul 

1  J ol 
s J2 = ⋅  ∑ j 2 − m l ⋅ j l2  : variance of the random sample numbers (5.12)
m l − 1  j = Jul 

j l : mean value of the random sample numbers
x l : mean value of the random sample mean values
Then a test for deviation of the regression line slope from zero is carried out:
H 0 : a1 = 0
H1 : a1 ≠ 0
Test variable:
â1
t= (5.13)
s a1
where
1 s r2
s a21 = ⋅ : variance of the line slope (5.14)
(ml − 1) s J2
∑ (x j )
J ol
1 2
s r2 = − x̂ j : variance of the residues (5.15)
fl j = J ul
x̂ j = â0 + â1 ⋅ j : estimated values for linear trend (5.16)
f l = m l − 2 : degree of freedom (5.17)

Page 34
VW 101 31: 2005-02
H1 is valid, if test value

t > t1−α ; fl
(5.18)
2
where
t1−α ; fl
: quantile of the t distribution (see also [1])
2
To prevent erroneous determination of a trend with a location jump, the following test for linearity
deviation is also carried out:
H 0 : trend linear
H1 : trend non-linear
Test variable:
s r2
F =n⋅ (5.19)
s K2
H1 is valid, if
F > F1 −α ; f1 ; f2 (5.20)
where
F1−α ;f1 ;f2 : quantile of the f distribution (see also [1]) where
f1 = m l − 2 (5.21)
f 2 = m l ⋅ (n − 1) (5.22)
If linear trends are found for at least two-thirds of all areas free of jump points, then process time
model C3 is assumed.
If, after the test described for trend, process time model C3 is not assumed, the measured values
recorded are tested for deviation from a normally distributed population. If no contradiction to the
hypothesis of a normal distribution results, process time model C1 is assumed. Otherwise,
process time model C4 is assumed, whereby the test for location jumps is carried out for
determining the mixed distribution up to the threshold value
6
E = χ 12−α ;1 ⋅ ⋅ σˆ m2 (5.23)
n
in the same way as for the test for trend.
Page 35
VW 101 31: 2005-02
4 Carrying out a process capability investigation

A process capability investigation (PFI) is carried out according to the sequence shown in Figures
20 to 24.
Start
Start
4.1Prüfmittelanwendung
4.1 Test equipment use
4.2 Sampling
4.2 Stichprobenentnahme
Conditions for
Bedingungen zur no
MFI
PFUfulfilled
erfüllt
yes 4.3
4.3Special regulationfür
Sonderregelung for
restricted PFIPFU
eingeschränkte
4.44.4 Data analysis
Datenauswertung
4.5Dokumentation
4.5 Documentation
4.6 Results evaluation

Ergebnisbeurteilung
ja Auswertungswie
Evaluation repeat
4.7
4.7 Process
Prozessverbesserung
improvement
derholung
yes
no
Prozess
Process managed no machbare
Feasible
beherrscht und
and capable Prozessverbess
process
fähig erung
improvement
yes no
4.8 Handling
4.8 Behandlung
of unmanaged
nicht or
beherrschter
incapable
oder
processes
nicht fähiger
Prozesse
Ende
End
Figure 20 - Sequence of a process capability investigation

Page 36
VW 101 31: 2005-02
4.4 Data analysis
4.4.1 Selection of the expected

distribution model
4.4.2 Test for outliers
yes 4.4.3 Take outliers out of the

Outliers present calculation of statistics
no
4.4.4 Test for change of process
control
4.4.5 Test for change of process

location
Change of yes Change of no

dispersion location
no yes
Change of yes
location
4.4.7 Test
4.4.7 Test for
aufdeviation from
Abweichung
instantaneous normal distribution
von momentaner
no
Normalverteilung
Instantaneously no
normally distributed
4.4.6 Evaluation according to yes 4.4.10 Evaluation according

process time models A to process time model D
4.4.8 Evaluation according to 4.4.9 Evaluation according

process time models C to process time model B
Continued at 4
Figure 21 - Sequence of data analysis

Page 37
VW 101 31: 2005-02
4.4.6 Evaluation according

to process time model A
4.4.6.1 Test for deviation from the

specified distribution model
yes
Deviation from the process
model
no
no 4.4.6.4 Evaluation according

normal distribution to process model A2*
yes
4.4.6.3 Evaluation according
to process model A2

to process model A1
Continued at 4.4
Figure 22 - Sequence of the evaluation according to process time models A

Page 38
VW 101 31: 2005-02
4.4.8 Evaluation according

to process time models C
4.4.8.1 Test for trend
yes
Trend present
no
4.4.8.2 Test for deviation from

resulting normal distribution
Deviation from normal yes 4.4.8.5 Evaluation according

distribution to process model C3
no
to process model C4

to process model C1
Continued at 4.4
Figure 23 - Sequence of the evaluation according to process time models C

Page 39
VW 101 31: 2005-02
Outliers present yes
no Outliers due to no
yes
incorrect
measurements
Process time no
model A
yes
Deviation from yes

distribution model
no Different no
distribution
model possible
yes
Evaluation repeat
Continued at 4
Cause known and
yes effect acceptable
Capability indexes yes

less than limit values
Process not managed
Continued at 4
no
Process not capable
Continued at 4
Process managed + capable

Continued at 4
Figure 24 - Sequence of the results evaluation

Page 40
VW 101 31: 2005-02
4.1 Test equipment use

Test equipment shall only be used for the PFI that has been released by the responsible
department for the planned test process.
4.2 Sampling
A PFI relates only to one production characteristic or one process parameter. Generally the single
measurements of the random samples are recorded for evaluation. In the case of quality control
cards that are recorded manually (tally), with recorded measured values in a class subdivision of
the value range and the class ranges determined from the single random samples, the frequency
distributions of the classified measured values and ranges can be recorded instead17.
In order to be able to record all possible influences acting under standard production conditions,
the following conditions shall be complied with during sampling:
─ The random samples shall be taken from ongoing production under standard production
conditions (i.e. with machine broken in and cycle time and machine adjusting parameter
values as in standard production) at regular intervals in the same size each time18. In this case,
in contrast to the MFI, all effects that result from changes in machine parameter values, tooling
changes and machine malfunctions shall be taken into consideration19.
─ During the random sample, the parts with the considered characteristic or the values of the
considered process parameters shall be taken as much as possible in direct sequence20. The
random sample size n must be at least 3 (recommended size n = 5)
─ The random sample quantity m must be at least 6
─ The total random sample size should generally be m ⋅ n ≥ 125 . If this random sample size is
difficult to obtain for economic or technical reasons, a smaller one is also permissible. Then the
corresponding higher limit values according to Table 5 or the formulas (4.7) to (4.9) must be
complied with. However, the effective total random sample size (without outliers) must be at
least 30.
─ The random sample frequency must be high enough so that at least once, a sampling of the
specified size n is carried out within a time interval in which no systematic influences on the
process are expected, i.e. at least one random sample per
– shift and employee
– change of the machine parameter values
– tool life and
– material batch
─ The testing period, over which the sampling of the m random samples extends, must contain
at least the following influences (recommended period 1 to 2 weeks):
– 3 changes of shift
– 3 changes of employees
– a change of the tooling or change of the machine parameter values and
– a change of the material batch
17 Since this form of manual data recording still occurs frequently in practice, it is taken into consideration in this standard, although the
possibilities of process analysis are somewhat restricted by it. However, the requirement is that the classified values of the ranges of
the individual random samples were also recorded.
18 When taking random samples, the measurement of the parts should be carried out as promptly as possible. However, if an
accumulation of several random samples before the measurement recording cannot be prevented, an assignment of the parts to the
random samples in their time sequence is to be made possible. In contrast, the sampling sequence of the individual parts of a
random sample does not have to be considered.
19 However, the measurements that are necessary for machine setup or for a corrective measure are not taken into consideration in the
PFI.
20 The form of sampling sometimes carried out in practice, in which each nth part, e.g. each 10th part, is taken from the ongoing
standard production, is not suitable for the process capability investigation.
Page 41
VW 101 31: 2005-02
─ The measuring method must be specified before the investigation and agreed upon between
supplier and customer. It shall not change during the test period.
4.3 Special regulation for restricted PFI

If the conditions named in 4.2 for sampling cannot be met completely, especially during start of
production for determining the control limits of quality control cards, in justified cases a restricted
PFI can be carried out, whereby the justifications are to be documented under the note "Restricted
PFI".
4.4 Data analysis
4.4.1 Selection of the expected distribution model

For data evaluation, first the distribution model expected for the characteristic type has to be
defined. For the most important types of production characteristics (see also VW 01056), the
assigned distribution models shall be taken from Table 4.
Table 4 - Assignment of characteristic types and distribution models

Characteristic type Distribution model
Length dimensions N
Diameter, radius N
Straightness B1
Flatness B1
Roundness B1
Cylindricity B1
Profile of any line B1
Profile of any surface B1
Parallelism B1
Squareness B1
Slope (angularity) B1
Position B2
Coaxiality / concentricity B2
Symmetry B1
Radial runout B2
Axial runout B2
Roughness B1
Unbalance B2
Torque N
Legend: N: Normal distribution

B1: Type I absolute value distribution
B2: Type II absolute value distribution (Rayleigh distribution)
Page 42
VW 101 31: 2005-02
For characteristic types not listed, in most cases an assignment of a distribution can be carried out
according to the following rules:
─ a normal distribution for characteristics with two-sided tolerance zones or one-sided tolerance
zone only
─ and a type I or II absolute value distribution for characteristics with an upper limited tolerance
zone only
4.4.2 Test for outliers

First a determination of whether the measured values recorded contain outliers is made using the
distribution-free outlier test according to VW 101 33. Outliers are measured values that lie so far
from the other measured values that it is highly probable that they do not come from the same
population as the remaining values, e.g. erroneous measurements. The outlier test shall be carried
out with a confidence level of 99%.
4.4.3 Take outliers out of the calculation of statistics

If outliers are found, these are not considered in the calculation of the statistics21. However, these
outliers must not be deleted, rather they are to be indicated accordingly in the graphic
representation of the single value curve and their number is to be indicated in the documentation.
4.4.4 Test for change of process control
Using the Cochran test (see [1]), it is examined whether the process exhibits a significant change
of dispersion in the course of the test period. A significant change of dispersion can occur, e.g. due
to wear on a machine bearing. This test shall be carried out with a confidence level of 95%.
If only the frequency distributions of classified measured values and ranges were recorded, the
modified form of the Cochran test (see Section 3.5) can be used with a confidence level of 95%.
4.4.5 Test for change of process location

With the use of the distribution-free test according to Kruskal and Wallis (see [2]), it is determined
whether the process location has changed significantly in the course of the test period. A significant
change of the process location can occur, e.g. by a change of material batches or by tooling wear.
This test shall be used with a confidence level of 95%.
If only the frequency distributions of classified measured values and ranges were recorded, the
modified form of the simple variance analysis (see Section 3.5) can be used with a confidence level
of 95%.
21 However, from the random samples with outliers, the remaining measured values will be taken into consideration for determining the
statistics. The different random sample sizes resulting from this will be taken into consideration in formulas (3.7) to (3.20) for
estimation of the statistics.
Page 43
VW 101 31: 2005-02
4.4.6 Evaluation according to process time models A

If the measured values recorded do not exhibit any significant change of the process location and
dispersion in the test period, then the evaluation is to be carried out according to process time
models A.
4.4.6.1 Test for deviation from the specified distribution model

First the recorded measured values are to be tested to see whether they exhibit a significant
deviation from the distribution model that was defined for the characteristic involved. To do this, in
the case of a specified normal distribution, the Epps-Pulley test (see ISO 5479) shall be used and
in the case of a different specified model, e.g. with a type I or II absolute value distribution, the chi-
square test (see [1]) shall be used with a confidence level of 95%.
A deviation from the specified distribution model can occur, e.g. due to sampling from different
tools (mixed distribution, see also Section 5. Examples, Figure 28).
4.4.6.2 Evaluation according to process time model A1

In the case of a specified normal distribution, in which the recorded measured values exhibit no
significant deviation from the distribution model, the behavior over time is identified by process time
model A1 and the calculation of the capability indexes shall be carried out according to formulas
(3.1) to (3.5), depending on the tolerance type, whereby the dispersion range limits are determined
according to (3.21) (see also Section 5. Examples, Figure 25).
4.4.6.3 Evaluation according to process time model A2

In the case of a specified type I or II absolute value distribution (or possibly another non-normal
distribution) in which the recorded measured values do not exhibit any significant deviation from the
distribution model, the behavior over time is to be identified by process time model A2 and the
calculation of the capability indexes shall be carried out according to formulas (3.1) to (3.5),
depending on the tolerance type, whereby the statistics of the distribution to be fitted is determined
according to formulas (3.25) and (3.26) or (3.34) respectively and (3.35) using the approximated
function (3.28) or (3.37) and the dispersion range limits can be calculated according to the
approximated functions (3.29) or (3.37) respectively (see also Section 5. Examples, Figures 26 and
27).
4.4.6.4 Evaluation according to process time model A2*
If the statistical test results in a contradiction between the recorded measured values and the
specified distribution model, or if no fitting distribution model can be found for the considered
production characteristic, the behavior over time is to be identified by the process time model A2*
and a distribution-free calculation of the capability indexes is carried out according to (3.46) to
(3.57), depending on the tolerance type (see also Section 5. Examples, Figure 28).
Page 44
VW 101 31: 2005-02
4.4.7 Test for deviation from the instantaneous normal distribution

If there is a significant change of the location, but no significant change of the dispersion of the
process, the extended Shapiro-Wilk test (see ISO 5479, Section 9) is used to test whether the
recorded measured values exhibit a significant difference from an instantaneous normally
distributed population of characteristic values. This test shall be carried out with a confidence level
of 95%.
The cause of a significant deviation from an instantaneous normally distributed population can be,
e.g. sampling from different tools.
To use this test, the measurement resolution (accuracy of reading) must be sufficiently precise,
since otherwise there is a high risk that the test will result in a contradiction to the instantaneous
normally distributed population of the characteristic values22 although no such contradiction exists.
Therefore, the ratio of instantaneous standard deviation and measured value must be at least 5.
Otherwise this test must not to used, and according to the principle "if in doubt, find for the
defendant," a normal distribution is assumed if a normal distribution was selected as a model or
was approximated according to the criteria (2.13) or (2.22) respectively.
For a specified type I or II absolute value distribution (or possibly a different non-normal
distribution), in which no approximation has been carried out using a normal distribution according
to the criteria (2.13) or (2.22) respectively, a significant deviation from an instantaneous normal
distribution has to be taken into consideration, even if no significant deviation results from the test.
If only the frequency distributions of classified measured values and ranges were recorded, this
test cannot be used, so the procedure is carried out as described previously.
4.4.8 Evaluation according to process time models C

If the statistical tests prove a significant change of the location, but not in the dispersion, of a
process in the test period and if there is no contradiction to the assumption of an instantaneous
normal distribution, the evaluation is carried out according to the process time models C.
4.4.8.1 Test for trend

First the test described in Section 3.5 is used to determine whether the measurements recorded
exhibit a trend (possibly with location jumps due to corrections). A trend can occur, e.g. due to tool
wear that increases over time.
If only the frequency distributions of classified measurements and ranges were recorded, this test
cannot be used. However, the option of deciding whether a trend exists based on a visual
assessment of the curve of the associated measured values recorded manually in control cards is
given.
4.4.8.2 Test for deviation from resulting normal distribution

A determination shall be made using the test according to Epps-Pulley (see ISO 5479) of whether
the measured values recorded exhibit a significant deviation from a resulting normal distribution.
This test shall be carried out with a confidence level of 95%.
22 The term characteristic value should not be confused with the term measured value, since in contrast to the first, the latter contains
an uncertainty.
Page 45
VW 101 31: 2005-02
4.4.8.3 Evaluation according to process time model C1

If no trend can be documented and no contradiction to the assumption of a resulting normal
distribution results from the measured values recorded, the behavior over time is to be identified by
the process time model C1 and the calculation of the capability indexes according to formulas (3.1)
to (3.5) shall be done depending on the tolerance type, whereby the dispersion range limits are to
be determined according to (3.21) (see also Section 5. Examples, Figure 29).

If no trend can be documented and the measured values recorded exhibit a contradiction to the
assumption of a resulting normal distribution, the behavior over time is to be identified by process
time model C4 and the calculation of the capability indexes is carried out according to formulas
(3.1) to (3.5), depending on the tolerance type, whereby to determine the dispersion range limits, a
mixed distribution that results from the superimposition of several normal distributions according to
formulas (2.24) to (2.27) with the mean values according to formulas (3.43) and (3.44) shall be
fitted (see also Section 5. Examples, Figure 31).
If only the frequency distributions of classified measurements and ranges were recorded, the fitting
of a mixed distribution is not possible, so instead of that, a distribution-free evaluation is to be
carried out according to formulas (3.46) to (3.50), depending on the tolerance type, whereby the
upper limit of the highest and the lower limit of the lowest class are to be specified as dispersion
range limits and the median value is calculated according to formula (3.57).

If a trend can be documented, the behavior over time is to be identified by process time model C3
and the calculation of the capability indexes is carried out according to formulas (3.1) to (3.5),
depending on the tolerance type, with an extended normal distribution according to (2.23), whereby
the dispersion range limits are to be determined from the formulas (3.39) to (3.42) (see also
Section 5. Examples, Figure 30).
4.4.9 Evaluation according to process time model B

In the case of a significant change of the dispersion, but no significant change of the location, of a
process, the behavior over time is identified by process time model B and a distribution-free
calculation of the capability indexes is carried out according to the formulas (3.46) to (3.57),
depending on the tolerance type (see also Section 5. Examples, Figure 32).
4.4.10 Evaluation according to process time model D

In the case of a significant change of the process location and dispersion or in the case of a
significant change of the process location and a contradiction to an instantaneous normal
distribution23, the behavior over time is identified by process time model D and a distribution-free
calculation of the capability indexes is carried out according to formulas (3.46) to (3.57), depending
on the tolerance type (see also Section 5. Examples, Figure 33).
23 For instantaneous distributions that are not normal, generally significant situation changes also involve dispersion changes in the
cases in which no significant change of the dispersion was determined by the Cochran test, so the process curve is to be described
using process time model D (see example in Figure 34).
Page 46
VW 101 31: 2005-02
4.5 Documentation
The documentation of a process capability investigation regarding a characteristic must contain at
least the following information and representations:
Header data:
─ Department, coordinator and preparation date
─ Information regarding the part
─ Designation, nominal dimension and tolerance zone of the characteristic
─ Machine data
─ Test equipment data
─ Examination period
Results:
─ Graphical representation of the single value curve with the random sample mean values with
limit lines of the tolerance zone (as long as single values were recorded)
─ Histogram with the fitted distribution model, limit lines of the tolerance zone and dispersion
range, as well as mean value or median value line (see examples in Section 5, center
diagrams)
─ Representation in the probability grid with the fitted distribution model, limit lines of the
tolerance zone and dispersion range, as well as mean value or median value line (see
examples in Section 5, lower diagrams)
─ Single random sample size
─ Random sample quantity and frequency
─ Effective total random sample size
─ Number of outliers determined
─ Estimated value of the process location
─ Estimated values of the dispersion range limits or the estimated value of the dispersion range
─ Distribution model used
─ Process time model
─ Instantaneous capability index Cp* for the case of assigned process time models C (to 2 digits
after the decimal)
─ Capability indexes Cp and Cpk (to 2 digits after the decimal)
─ Required limit values for Cp and Cpk
References and notes:

─ if necessary, reference to restricted PFI and description of the restrictions
─ if necessary, special events during the test period
Page 47
VW 101 31: 2005-02

Whether a process is evaluated as managed and capable with respect to a considered
characteristic depends on the results evaluation, as follows:
If outliers result during the evaluation, the measured values recorded are to be examined more
closely and if possible their cause is to be clarified. Outliers shall only be caused by incorrect
measurements or be wrongly identified as such using the outlier test based on the specified
probability of error of 1%. Otherwise, the process is to be evaluated as unmanaged. If more than
5% of the recorded measured values or more than 3 values are identified as outliers, there shall be
a test of whether the test process is faulty. The PFI shall then be repeated if necessary.
If the measurements recorded exhibit significant changes in the process location or dispersion,
generally its cause must be known, the behavior over time must follow a known regularity or the
changes must lie within known limits. In addition, the effect should be acceptable in order to be
able to evaluate the process as managed.
In most cases, for processes that are identified by process time model C1, C3 or C4, the location
changes can be explained by tooling wear, tooling change, material batch change or other known
systematic influences, so that with acceptable effect, these processes can be evaluated as
managed (see also Section 5. Examples, Figures 29 to 31).
In contrast, processes that are identified by process time model B or D are generally to be
evaluated as unmanaged, since a change of dispersion usually indicates deterioration in
production quality that can not be corrected using machine controls, unless it is a case of decrease
of dispersion due to an improvement measure (see also Section 5. Examples, Figures 32 to 34).
In addition, processes that are identified by process time model A2* are evaluated as unmanaged if
the cause of the significant deviation from the specified distribution model is unknown and no other
distribution model can be assigned24 to the characteristic considered. If a different distribution
model can be considered for the characteristic, then the evaluation shall be repeated with this
distribution model.
Unless otherwise agreed, the capability indexes determined with an effective total random sample
size of n g ≥ 125 (i.e. without outliers) must meet the requirement
ĉ p ≥ 1,33 and ĉ pk ≥ 1,33 for a characteristic with two-sided tolerance zones
ĉ pk ≥ 1,33 for a characteristic with one-sided tolerance zone only

in order to be able to evaluate the process as capable. In this case, for comparison to the limit
values, the capability indexes determined are to be rounded to two digits after the decimal so that,
e.g. a determined value of cˆ pk = 1,32645 , still meets the requirement after the resulting rounding to
1.33.
With an effective total random sample size of 30 ≤ ng < 125, correspondingly higher capability limit
values shall be complied with. The fitted limit values are given in Table 5 for a few random
sample sizes. If there is agreement on other limit values on the basis of n g ≥ 125 , the
corresponding fitted limit values shall be determined according to formulas (4.7) to (4.9).
24 In the case of a very small tolerance (T < 5 µm) there is frequently a significant deviation from the specified distribution model
because of a restricted measured value resolution
Page 48
VW 101 31: 2005-02
Table 5 - Fitted limit values for process capability for 30 ≤ ng < 125
ng ĉ p and ĉ pk ≥
30 1,54
35 1,51
40 1,48
45 1,46
50 1,44
60 1,41
70 1,39
80 1,37
100 1,35
125 1,33
If a capability index results that is lower than the corresponding limit value, the process shall be
evaluated as incapable.
With a determined capability index of c pk ≥ 2.0 (corresponds to 12 σ) and for characteristics with
two-sided tolerance zones of c p ≥ 2.0 , the process with respect to the characteristic considered
can also be evaluated as capable independently of a significant change of the location or
dispersion or deviation from the specified distribution model.
For characteristics with two-sided tolerance zones, the goal is a Cpk value that deviates as little as
possible from the Cp value since this characterizes a good centering of the process location. In
contrast, for characteristics with upper limited tolerance zone only, the goal is a Cpk value that is
higher than the Cp value since this characterizes a process that lies closer to the natural limit zero
than to the upper limiting value.
Although no other limit value is specified for the instantaneous capability index ĉ *p , it should be
at least 1.67 so that the capability requirement can be met even during necessary process
changes, e.g. for processes with trend.
The results of a PFI are important not only for evaluating a process, but also for determining the
control limits of quality control cards for ongoing monitoring and control of a process.
Page 49
VW 101 31: 2005-02
4.7 Process improvement

In the case of an unmanaged or incapable process with respect to a characteristic considered,
measures are to be taken for process improvement so far as these are feasible under economic
considerations. To do this, the significant interfering influences on the process have to be identified
(e.g. using DOE statistical test methodology).
While incapable processes that are identified by process time model A1 or A2 often reach the limit
of economic feasibility for process improvement, if the tolerance of the characteristic values is less
than 5 µm, such processes that are identified by the other process time models can usually be
improved with reasonable effort.
4.8 Handling of unmanaged or incapable processes
If a managed and capable process for the characteristic considered cannot be achieved with
economically reasonable improvement measures in a reasonable time, the reasons are to be
documented. Then, with the use of statistical tolerance calculation, a test according to
VW 010 57 of whether, under the condition of a managed process, a tolerance extension to the
required zone is possible for the characteristics considered shall be performed. However, if the
tolerance cannot be extended, suitable measures for process monitoring shall be agreed upon and
documented, e.g. a 100% monitoring.
Page 50
VW 101 31: 2005-02
5 Examples
The following Figures 25 to 34 show evaluations of different simulated processes.
Measured value
Random sample no.

Relative frequency
Measured value
Relative cumulative frequency in %
Measured value
Desired value µo 20 Estimated mean value 20.039
Upper limiting value. 20,2 Upper dispersion range limit 20,153
Lower limiting value 19,8 Lower dispersion range limit 19,926
Single random sample size 5 Distribution model N
Random sample quantity m 51 Process time model A1
Total random sample size 255 Estimated Cp* instantaneous value 1,78
Number of outliers 0 Estimated Cp value 1,76
Limit value for Cp and Cpk 1,33 Estimated Cpk value 1,41
Figure 25 - Managed and capable process with constant location and dispersion and
instantaneous normal distribution, e.g. with length dimensions of shafts
Page 51
VW 101 31: 2005-02
Measured value
Random sample no.

Relative frequency
Measured value
Measured value
Desired value µo 0 Estimated mean value 0,077
Upper limiting value 0,4 Upper dispersion range limit 0,280
Lower limiting value 0 Lower dispersion range limit 0,000
Single random sample size 5 Distribution model B1
Total random sample size 255 Estimated Cp* instantaneous value
instantaneous absolute value distribution B1, e.g. with parallelism deviations
Page 52
VW 101 31: 2005-02
Measured value
Random sample no.

Relative frequency
Measured value
Measured value
Lower limiting value 0 Lower dispersion range limit 0,004
Single random sample size 5 Distribution model B2
instantaneous absolute value distribution B2, e.g. with radial runout deviations of shafts
Page 53
VW 101 31: 2005-02
Measured value
Random sample no.

Relative frequency
Measured value
Measured value
Single random sample size 5 Distribution model
Random sample quantity m 51 Process time model A2*
Figure 28 - Incapable process with constant location and dispersion but instantaneous
mixed distribution, e.g. by sampling from 2 tools
Page 54
VW 101 31: 2005-02
Measured value
Random sample no.

Relative frequency
Measured value
Measured value
Single random sample size 5 Distribution model N
Random sample quantity m 51 Process time model C1
Figure 29 - Managed and capable process with frequent location jumps but constant
dispersion and instantaneous normal distribution, e.g. by frequent tooling changes
Page 55
VW 101 31: 2005-02
Measured value
Random sample no.

Relative frequency
Measured value
Measured value
Single random sample size 5 Distribution model EN
Figure 30 - Managed and capable process with trend, constant dispersion and
instantaneous normal distribution, e.g. in shaft diameter due to tooling wear
Page 56
VW 101 31: 2005-02
Measured value
Random sample no.

Relative frequency
Measured value
Measured value
Single random sample size 5 Distribution model MV
Figure 31 - Managed and capable process with location jumps, constant dispersion and
instantaneous normal distribution, e.g. due to material batch change
Page 57
VW 101 31: 2005-02
Measured value
Random sample no.

Relative frequency
Measured value
Measured value
Random sample quantity m 51 Process time model B
Limit value for Cp and Cpk 1.33 Estimated Cpk value 1.05
Figure 32 - Unmanaged and incapable process with constant location but dispersion
change, e.g. due to machine bearing wear
Page 58
VW 101 31: 2005-02
Measured value
Random sample no.

Relative frequency
Measured value
Measured value
Random sample quantity m 51 Process time model D
Figure 33 - Unmanaged and incapable process with location and dispersion changes
Page 59
VW 101 31: 2005-02
Measured value
Random sample no.

Relative frequency
Measured value
Measured value
Random sample quantity m 51 Process time model D
Figure 34 - Unmanaged and incapable process with correlating location and dispersion
changes with an instantaneous absolute value distribution B2, e.g. trend with radial runout
deviations
Page 60
VW 101 31: 2005-02
6 Referenced standards25
VW 010 56 Drawings; Form and Position Tolerances
VW 010 57 Statistic Tolerance Calculation of Dimension Chains
VW 101 33 Test auf Ausreißer (Test for Outliers – currently only available in German)
DIN 55319 Quality Capability Statistics (QCS)
ISO 5479 Statistical Interpretation of Data – Tests for Departure from the Normal Distribution
7 Referenced literature
[1] Graf, Henning, Stange, Willrich, Formeln und Tabellen der angewandten mathematischen
Statistik, Springer publishing group, third edition 1987
[2] Kühlmeyer M., Statistische Auswertungsmethoden für Ingenieure, Springer publishing group,
2001
[3] DGQ Document No. 16-32, SPC 2 – Qualitätsregelkartentechnik, fourth edition, 1992
25 In this Section, terminological inconsistencies may occur as the original titles are used.
Page 61
VW 101 31: 2005-02
8 Keyword index
Key word Page
A
Absolute frequency a 18, 27
Absolute cumulative frequency A 27
α-Risk 31
Alternative hypothesis 30
Fitted capability limit values 29, 48
Outlier 30, 42
Confidence level 30, 42
B
Conditions for sampling 40
Managed process 4, 47
Beta risk 31
Absolute value distribution, type I 10, 19
Absolute value distribution, type II 12, 21
C
Capability 3
Chi-square distribution 28
Cochran test 30, 42
D
Data analysis 41
Density function f(x) 9
Documentation 46
E
Effective total random sample size ng 17, 40
Simple variance analysis 30
Restricted process capability 41
investigation
Epps-Pulley test 30, 43
Results evaluation 47
Expected value for the w-distribution dn 18
Extended normal distribution 7, 13
Eccentricity z 12
F
Determination of capability 15
Capability indexes cp and cpk 3, 15
Capability limit values 28, 47
Degree of freedom f 28, 34
F-distribution 34
G
Total random sample size 17
Total variance 32
Limit values for process capability 28, 47
H
Hampel test 30
Page 62
VW 101 31: 2005-02
Key word Page

Frequency distribution 30, 40
Histogram 46
Upper limiting valueGo 16
I
Probability of error α 30
K
Class width ∆x 18
Classified measured values / ranges 17, 27
Correction factor k 26
Covariance 33
Kruskal-Wallis test 30
L
Location 3
Location jump 31
Linear trend 34
Linearity deviation 34
M
Median value 26
Characteristic type 9, 41
Characteristic value 3
Measuring method 40
Lower limiting value Gu 16
Mixed distribution 14, 25
Mean value µ 17
Instantaneous distribution 4
Instantaneous standard deviation σm 17
Instantaneous capability index cp* 16, 48
Instantaneous variance 32
N
normal distribution 4, 9
Null hypothesis 30
Zero offset 11
P
Parameter of a distribution 9
Process behavior 4
Process improvement 49
Process time modelss 4
Test variable 30
Test equipment use 40
Q
Quantile 16, 28
- of the standardized normal distribution 28
- of the chi-square distribution 28
- of the t-distribution 34
- of the F-distribution 34
R
Rayleigh distribution 12
Radial deviation r 12
Regression line 33
Page 63
VW 101 31: 2005-02
Key word Page

Residues 33
Resulting distribution 5
Rounding of capability values 47
S
Sawtooth curve 31
Estimation / estimated value 15, 16
Threshold value 31
Standard production conditions 40
Shapiro-Wilk test 30
Significant change / deviations 31
Range R 18, 26
Jump point 31
Standard deviation σ 9, 17
Standardized normal distribution 10
- U Transformation 10
- Distribution function Φ(ug) 10
- Probability density function φ(u) 10
statistics 17
Statistical tests 30
Statistical percentage range 28
Statistical tolerance calculation 49
Random sample quantity m 40
Sampling 40
Random sample frequency 40
Random sample number 33
Random sample size n 40
Dispersion range limits x0.135%, x99.865% 19
- of normal distribution 19
- of absolute value distribution, type I 19
- of absolute value distribution, type II 21
- of extended normal distribution 24
- of mixed distribution 25
T
Test
- for outliers 30, 42
- for deviation from normal distribution 30, 43
- for instantaneous normal distribution 30, 44
- for resulting normal distribution 44
- for specified distribution model 43
- for change of process location 30, 42
- for change of process variation 30, 42
- for location jump 32
- for linearity deviation 34
- for slope of the regression lines 33
- for trend 33, 44
Tolerance expansion 49
Tolerance zone 3
Toleranced characteristic 16
Page 64
VW 101 31: 2005-02
Key word Page

- on one side, upward 16, 26
- on one side, downward 16, 26
- on two sides 16, 25
Trend 31
t-distribution 34
U
Examination period 40
V
Variance σ2 9
Distribution 3
Distribution-free estimate 25
Distribution function F(x) 10
Distribution model 9
Confidence range limit 28
W
Probability p 10
Probability density function f(x) 9
Probability grid 46
w-Distribution 18, 27
Weibull distribution 13
Z
Process behavior over time 4
Random influences 3

VW - 10131 - en

Uploaded by

Copyright:

Available Formats

VW - 10131 - en

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

VW - 10131 - en

Uploaded by

Copyright:

Available Formats

Klass.-Nr.

1 Purpose and scope

2 Principle of the process capability investigation

Figure 1 - Example of a distribution of characteristic values within a defined tolerance zone

3.1 Process time models

Table 1 - Identification of the process time models

Figures 2 to 9 show the process time models schematically.

Process time model A1:

Figure 2 - Process time model A1

Process time model A2:

Figure 3 - Process time model A2

Process time model A2*:

Figure 4 - Process time model A2*

Process time model B:

Figure 5 - Process time model B

Process time model C1:

Figure 6 - Process time model C1

Process time model C3:

Figure 7 - Process time model C3

Process time model C4:

Figure 8 - Process time model C4

Process time model D:

Figure 9 - Process time model D

3.2 Distribution models

3.2.1 Normal distribution

Figure 10 - Function of the probability density of a normal distribution

and f N ( x ) ≥ 0 for all values x.

with the standard deviation σ =1

3.2.2 Absolute value distribution, type I

µN = 0 µN = 1σN µN = 2σN µN = 3σN

0 1σN 2σN 3σN 4σN 5σN 6σN

3.2.3 Type II absolute value distribution (Rayleigh distribution)

z=0 z = 1σN z = 2σN z = 3σN

0 1σN 2σN 3σN 4σN 5σN 6σN

and from them, in turn, mean value and variance

3.2.4 Extended normal distribution

An example of an extended normal distribution is shown in Figure 13

Figure 13 - Density function of an extended normal distribution

3.2.5 Mixed distribution

with the weighting coefficients

With the normalizing factor

(2.24) yields the probability density of the mixed distribution

An example of a mixed distribution is shown in Figure 14

3.3 Determination of capability

3.3.1 Determination of capability for defined distribution models

3.3.1.1 Capability indexes

3.3.1.2 Estimation of the statistics

with the mean values of the single random samples

The instantaneous standard deviation is determined by

In the case of data to be evaluated in the form of frequency distributions of classified

x k : mean value of the kth class

The value dn is given in Table 2 for a few random sample sizes

Table 2 - Expected value of the w distribution in relationship to n

3.3.1.3 Estimation of the dispersion range limits

Dispersion range limits for normal distribution:

Dispersion range limits of the type I absolute value distribution:

With equation (2.10), this yields the following function

Equations (2.11) and (2.12) result in the condition

Dispersion range limits of the type II absolute value distribution: