Certification Course on

Quality Assurance and Statistical Quality Techniques

Course Level A Process Capability Code 0.07 b: 7 Process Capability
Issue No.: 01
Effective Date: 15-04-2014

PROCESS CAPABILITY

Process capability refers to the normal behavior of a process when operating in a state of statistical
control. It refers to the inherent ability of a process to produce similar parts for a sustained period of
time under a given set of conditions

Why is process capability important? Because it allows one to quantify how well a process can produce
acceptable product. As a result, a manager or engineer can prioritize needed process improvements and
identify those processes that do not need immediate process improvements. Process capability studies
indicate if a process is capable of producing virtually all conforming products. If the process is capable,
then statistical process controls can be used to monitor the process and conventional acceptance efforts
can be reduced or eliminated entirely. This not only yields great cost savings in eliminating non-value
added inspections but also eliminates scrap, rework and increases customer satisfaction. The benefits of
performing process capability studies are certainly worth the effort in the long run.

Sometimes processes may start out as capable, but change over time to show more variability. In
addition, the process mean may also shift placing the process too close to one of the specification limits.
Both increases in process variability and shifting of the mean may result in once capable processes
becoming incapable.
.
Measuring Process Capability

It is a statistical measure of inherent process variability and is defined as the length of natural process
interval (plus minus 3 sigma process spread). It represents the variation that is obtained when the
process operates under chance causes only. Any further reduction in the variability may need
modification in the process or a technological upgradation in the process.

The lower and upper natural process limits are obtained as α and (100 – α ) percentiles of the
distribution and these include (100 - 2α) percent of the items when the process is in a state of statistical
control.

For a normally distributed process, the natural process limits are taken as + 3σ and they correspond to a
α = 0.135. Even for non-normal distributions, the percentile values may be taken corresponding to α =
0.135 therefore Pα = P 0.1,35. In such a case, the natural process limits will not necessarily be symmetrical
around the process mean. The natural process interval (also called the process spread), namely, the
interval between the upper and the lower natural process limits, covers 99.73 percent of items
produced.

Fig 1

Estimating Process Capability :

Process capability can be expressed as percent nonconforming or in terms of the natural spread related
to the specification spread. The most commonly used indices are Cp, Cpk, Pp, Ppk

Factors influencing process variation

Following are the major factors that influence the process variation:
a) Condition of the machine or equipment,
b) Skill of the operator,
c) Type of operation and operational conditions,
d) Raw materials used,
e) Environment, and
t) Measurement process.
Planning process capability study

A process capability study is a systematic procedure for determining the capability of a process. The
process capability study may include studies to improve the process and in turn the capability of the
process. Process capability studies are usually performed as part of a process approval or a process
optimization or process improvement effort.

The following points should be taken care of before taking up the process capability study :
a) The study should be conducted by well trained and experienced personnel,
b) In order to estimate process capability, one must know the location, spread, and shape of the
process distribution. These parameters regularly change in an out-of-control process. Therefore,
only use Process Capability indices if the process is in control for an extended period.
c) The order of production should be maintained. The samples drawn for measurement should be
identified with the process line, order of production, etc. This is important for all studies, that the
collected data are traceable so that unexpected values can be investigated. A logbook would be
suitable for recording all process settings and to monitor all events during the study such as
adjustments or temperature fluctuations.
d) Details of input materials, operational conditions, process lines and other facilities should be
recorded,
e) The process should be operating under its normal operating conditions unless the study is being
conducted for investigating specific abnormalities.
f) The data collection system needs to ensure that appropriate data is collected. It is preferable to
collect at least 60 data values for each critical parameter.
g) Significant digits for measuring values should be the number of significant digits required as per the
specification limits plus one extra significant digit to assure that process stability can be evaluated.
h) The applicable specification limits, which may be either one-sided (lower or upper) or two-sided
should be recorded.

Steps involved

a) Identify the product characteristic to be studied e.g dimension, strength, concentration

b) Define the process and its operating conditions.( If there is a change in those conditions it will
necessitate a fresh process study). This includes identification of specific machines, input raw
material quality, machine settings, process control parameters including environmental conditions,
persons involved where individual skills influence process results

c) The data collected by sampling are plotted as a histogram. This will indicate the type of distribution,
any obvious skew, long tails, measurement errors etc

d) Test the process for normality. A normality test is carried out to see that the data displays normal
distribution. If data is non normal , alternative methods are chosen for conducting the capability
analysis

a) Test the process for stability and for being under statistical control. . A number of samples must be
plotted in a control chart used in connection with the examination before the result can be
calculated. Outliers are observations in a set of data that appears to be inconsistent with the
remainder of that dataset. If it is suspected that such values do not belong to the same basic
population as the other collected values, their validity must be investigated. If present, such values
can lead to false conclusions and do not reflect the actual performance of the process. For Process
capability analysis, the subgroup containing the outliers as observed on the X bar or Range chart,
should be eliminated and the data re-plotted.
b) Estimate the inherent process variability + 3 σ limits
c) Identify the specification limits for the chosen characteristic
d) Select an appropriate measure of process capability Cp, Cpk, Pp, Ppk

Process capability and process performance

For a given characteristic, process capability and process performance describe the distribution of the
observations of a process over time. These enable prediction of the ability of the process to produce
results in accordance with the given specifications or tolerances.

Process capability and performance, are similar analyses, about making examinations of the variation
patterns a process produces during a given period of time. However they differ as shown in Figures
below:

Fig 2

Process Capability is a measure of process variation which is within small sub-groups and assumes that
the variation between the sub- group is zero or tending to zero. It is generally a short term study of
process behavior. The standard deviation of a process is estimated within sub-groups by using control
charts.
̅
σc =
 Fig 3

Process performance is measured when the process is indicating out of control sub groups or when we
do not have any idea that the process is within statistical control, and have no means to verify it.
Process performance is normally calculated for one of batches or for quick estimation of the ongoing
process. It can also be used to decide whether or not a process capabilty study is required. It is based on
the total variation and is given by the standard formula used for calculating the standard deviation of
any population as the variation around the overall population mean.

σp = √ ̅) n-1)

The calculated Process Performance and Process Capability indices will likely be quite similar when the
process is in statistical control

Indices

Process Capability index Cp - Quantity that describes the capability in relation to given specifications.
The process capability index is expressed as the value of the specification interval divided by a measure
of the length of the reference interval for a process in a state of statistical control. The term potential
process capability index is often used for this index. It is often used during the product design stage and
the pilot production phase.
Cp = =

Fig 4

Minimum Process Capability Index Cpk - This is a process capability index that describes the process
actual performance by accounting for a shift in the mean of the process towards either the upper or
lower specification limit. It is often used during the pilot production phase and during routine
production phase.

Cpk = or = whichever is smaller

Therefore Cpk can never be greater than Cp

Single sided limits

Sometimes a specification is given which has only one limit, e.g. a maximum (or minimum) value. In
these circumstances it will only be possible to compute a Cpk index based on the upper (or lower)
capability index.

In such cases Cpk = (Cpu, Cpl),

as applicable
Where:

Cpu =

Cpl =

Fig 5
The table below informs how relative measures of cp and cpk tell us what is happening to the process:

Cpk
Low High

Low Variability High Will not happen

Expected NC will be high
Cp
High Process setting has shifted. Favorable(>2)

Process performance Index Pp and Ppk

Pp Compares total engineering tolerance with process performance + 3 sigma range

Pp = =

Where = ̅)
√

Ppk Compares minimum difference of Specification limit from process mean with 3 sigma range

Ppk = = Minimum (Ppu, Ppl),

Where: Ppu =

Ppl =

One sided limits also apply to Ppk similar to Cpk

Relationship of indices with non-conformance

The table below provides information of how the different values of the Cp index impact the level of
non-conforming products generated
 Defective parts per
Capability Ratio Cp Status of + 3 Sigma limits
million

0.50 133,614
Exceeds spec range
0.75 24,449

1.00 Equals spec range 2,700

1.30 96
Uses less than spec range
2.00 0.0018

The Cpk ratio will yield similar level of non-conformance on one side of the probability distribution
(divided by half), I,e a Cpk value of 1.00 will yield 1350 ppm non conformance

The Process performance requirements are normally defined by the customer. When no requirements
are defined, the following are the recommended achievement levels

For statistically controlled processes a Cpk value > 1.33 should be achieved

For unstable processes, a Ppk value >1.67 should be achieved

Two MS Excel Templates have been provided with this course: a) Template titled Process Capability
short term where short term data can be fed and entered for quick estimation. B) Template titled
Process Capability with X bar, where systematic subgroup wise data is entered to calculate both process
capability as well as Process performance. Data for practice has been provided in the Practical exercises
Worksheet.

Normality Test for Data

Many of the procedures used in statistics, including process capability tests require that the data be
normally distributed. Control charts have an underlying normality assumption in the calculation of the
control limits. If the data are not by nature, normally distributed and a control chart for averages or
individual values is used, points outside the control limits may be due to the skew of the data and not
due to special-cause variation.

It is therefore necessary to establish before taking up process capability studies that the data is normally
distributed. There are several ways in which this can be done:

a) By constructing a histogram from data taken from the process. A histogram like we see below is
a clear indication that the data is normally distributed
 Fig 6

The histograms below are examples of data that are clearly non normal and should not be taken up for
normal process capability studies

Fig 7
b) Normal probability plot

A probability plot (also called the cumulative or normal probability plot), is a graph of the relative
cumulative frequencies of the data, using a specific plotting convention.
In the probability plot the relative cumulative frequencies are plotted on special graph paper (called
normal probability paper). This paper has a special y-axis (vertical) scale that has been chosen so that
data that are normally distributed will tend to yield a straight line. The x-axis (horizontal) scale is just a
regular linear scale. Each ordered pair of the data point and its cumulative probability is plotted.
Step 1: Arrange all data in ascending order as in column 2 of the table below
Step 2: Assign a rank to each value starting from 1
Step 3: Calculate probability using the formula: Y = 100* {( i – 3/8) / (n + 1/4)}
Where i is the rank and n is the total number
Step 4 : Plot the x and y points on the probability plot paper
Draw a trend line dividing the plot points equally on both sides to get the best fit
If all points are closely clustered around this point, the data set is normally distributed.

Rank Data in Probability

ascending position (y)
(i)
order (x)

1 230 3.62
2 240 9.42
3 250 15.22
4 250 21.01
5 260 26.81
6 260 32.61
7 270 38.41
8 270 44.20
9 280 50.00
10 280 55.80
11 280 61.59
12 280 67.39
13 290 73.19
14 300 78.99
15 300 84.78
16 310 90.58
17 320 96.38
Fig 8
If the data is normal or near normal, then the plotted points will tend to fall on a straight line as we see
on the graph above. . If it is not, the points will deviate from a straight line, generally in some systematic
manner. The decision whether or not to reject the hypothesized model is subjective but with some
experience a reasonably good assessment can be made.

The plot below is clearly indicative of non- normal data, the actual distribution is shown alongside.

Fig 9 Fig 10

The Chi Square goodness of Fit Test

Another method where we want to determine the presence or absence of normality at given confidence
level say 0.5, we use the The Chi Square goodness of Fit Test. ‘It works on the principle of comparing an
actual data distribution with a standard normal distribution around the same mean and standard
deviation and calculating a test statistic for the differences. The test is very logical as it generates an
expected frequency for each class interval or band in which we divide our data and compares it with the
actually observed frequency
Step 1: Organize the data into histogram,. Select the class intervals and the starting and end bins in a
manner that most of the bins have frequency of 5 or more
Step 2: Calculate the cumulative probability distribution for upper and lower values of each class interval
to get the probability band for the interval. Use Excel formula NORM.DIST(x, mean, standard deviation,
cumulative)
Step 3 : Calculate expected frequency for the band as a product of probability band and total number of
samples in data set
Step 4 : Calculate Chi Square value for each band by the formula:
(Expected – Observed)2/Expected
Step 5: Add all Chi Square values to get the Chi Square test statistic
Compare with Chi Square statistic for confidence level of 0.05 and the known degrees of freedom
(number of frequency bands-3) or calculate p value (Excel formula CHI.DIST(Chi square value, df)
Example : Data set : 168 values; Rrange:1.10 to 1.70; Mean 1.40; Standard deviation 0.:112
Frequency band Obs Frequency Cum Norm Probability Exp Chi Sq Stat
Probability Band Frequency (Exp-Obs)2/Exp
<=1.14 2 0.00985 0.00985 1.65 0.072
(1.14 - 1.17] 1 0.01953 0.00968 1.63 0.242
(1.17 - 1.2] 5 0.03633 0.01680 2.82 1.680
(1.2 - 1.23] 3 0.06346 0.02713 4.56 0.533
(1.23 - 1.26] 7 0.10425 0.04078 6.85 0.003
(1.26 - 1.29] 12 0.16130 0.05706 9.59 0.608
(1.29 - 1.32] 12 0.23560 0.07430 12.48 0.019
(1.32 - 1.35] 15 0.32566 0.09006 15.13 0.001
(1.35 - 1.38] 18 0.42727 0.10161 17.07 0.051
(1.38 - 1.41] 20 0.53398 0.10671 17.93 0.240
(1.41 - 1.44] 16 0.63828 0.10430 17.52 0.132
(1.44 - 1.47] 14 0.73318 0.09490 15.94 0.237
(1.47 - 1.5] 14 0.81355 0.08037 13.50 0.018
(1.5 - 1.53] 9 0.87691 0.06335 10.64 0.254
(1.53 - 1.56] 6 0.92339 0.04648 7.81 0.419
(1.56 - 1.59] 6 0.95514 0.03175 5.33 0.083
(1.59 - 1.62] 4 0.97532 0.02018 3.39 0.110
(1.62 - 1.65] 2 0.98726 0.01194 2.01 0.000
>1.65 2 100.00000 0.01274 2.14 0.009
168 168.00 4.711

Step 1: The data was organized in Bands with class interval of 0.03,. The starting and end class intervals
and the band size were selected so that most frequencies are more than 5, at least in the middle
ranges.. This improves the accuracy of the test.
Step 2: Third Column gives the cumulative probability for normal distribution using the higher end of the
class interval as x, Mean as 1.40 and SD as 0.112. The fourth column gives the probability band for the
class interval by subtracting the previous value of the cumulative probability
Step 3 : Fifth column gives the calculated Expected frequency of a perfect normal distribution by
multiplying the value in column 4 with 168 the total number of samples.
Step 4: the Chi Square test statistic is calculated by the formula shown using values in second and fifth
columns
Step 5 : the overall test value for Ki square is calculated by summing all the values in the last column .
We now have to test whether the value of test statistic for is significant or not
The following table gives the significance calculation. MS Excel formulas are used for getting the values
of Chir square and p.
Parameter Formula Value

)
Test Statistic, x ∑ 4.711

Degrees of Freedom,
n -3 19 - 3 16
Df

Confidence level, α 95 % 0.05

MS Excel formula:
P value CHISQ.DIST(4.711,16,TRUE) 0.00299
CHISQ.DIST(x, df, cumulative)

Critical Chi Square MS Excel formula

CHISQ.INV(0.05,16) 7.962
value CHISQ.INV(probability, df)

Inference : a) As p value is < 0.05, data follows normal distribution

Alternate inference : As observed Test statistic < critical Chi square value, data follows normal
distribution