Process Capability

Process Capability calculations indicate the ability of a process to meet specification limits.

When to use it: Situations when it is useful.

How to understand it : Details of how it works.
Examples : Some examples of usage.
How to do it : Step-by-step instructions on using it.
Practical variations : Variants and variation.

Notes
The investigation of a process and its capability is often called a Process Capability Study,
or PCS.

When to use it
Use it when setting up a process, to ensure it can meet its specification limits.
Use it when setting specification limits, to ensure they are neither too wide nor too
narrow.
Use it when investigating a process that is not meeting its specification limits.
Use it only when the process is stable and has a Normal distribution.

Fig. 1. Using Process Capability in problem solving

How to understand it
The outputs of any process will vary, as discussed in the chapter on Variation, and it is
common for specification limits to be defined such that if the measured output of the
process exceeds the specified limits, the process is deemed to have failed. The term
'specification limits' is most commonly used for the dimensions of a manufactured item, but
can be used in any process. Thus, for example, the specification limits for the time a
telesales operator may take to answer a customer call may be between zero and five
seconds.
The results of most processes will vary around a central v alue, as described in Chapter 5,
and the 'capability' of the process is defined as the spread of results around this value, with
high capability occurring when process results group closely around it. Thus a process that
can produce parts to within 0.001mm of a target value is more capable than one which can
only produce them to within 0.015mm.
The most common measure of this spread is standard deviation, and 'Process Capability'
may be defined as the range between three standard deviations either side of t he average.

Fig. 1. Sta nda rd deviation

Specifications are often defined separately from the process that is being measured and
without a great deal of consideration of how easily the process can meet them. This can
result in either many failures and rejects or effectively redundant specificatio ns, as the
variation in the process fits badly or well within the specified limits.
Specification limits and process capability thus need to be considered together. The limits
still cannot be too tight as if calibration is done under ideal conditions, proc ess distribution
may subsequently drift or spread, for example as a result of wear in a machine tool.

Fig. 2. Natural distribution fit within defined specification limits

A common Process Capability measure, Cp (often called a Process Capability Index),
indicates how well the process distribution fits within its specification limits, and is simply
the ratio of the specification width to the variation width. Thus, in the figure above ,
processes (a) and (b) have Cp greater than one, (c) is equal to one and (d) is less than
one.
The problem w ith Cp is that it does not take account of how well the process distribution is
centered within its limits, which can result in a process with both a low Cp and many
rejects. The solution to this is a second measure, Cpk, which measures a similar ratio, but
considers only the variation half that is closest to the specification limits, as in the figure
below. Thus Cp and Cpk, taken together, give a measure of both the potential and
centering of the process distribution w ithin the specification limits.
Process Capability measures are only as good as the data used, and there is plenty of
opportunity for misinterpretation. In particular, Process Capability measurement is based
on three important assumptions which are thus preconditions for valid calculations:

1. The process is in a state of statistical control, and there are no special causes of
variation. The implication of this is that before Cp and Cpk can be measured, special
causes must be found and eliminated. This may be done using the Control Chart over
a period of time long enough to give confidence that this has been successfully
completed.
2. The process distribution is bell-shaped or 'Normal', which allows the width of the
distribution to be calculated as six times the standard deviation. In practice, there
are many situations where the distribution is not normal, and in Process Capability
measurement the Central Limit Theorem does not act to normalize this, as it does
when using a Control Chart.

Fig. 3. Capability measurement

3. The measured data is representative of the process. This means that it should be a
randomly selected and large sample, taken over a long period. Samples taken over a
 short period can suffer from a limited range of changes in either external seasonal
effects or internal process variables, such as humidity or tool wear.
When interpreting values of Cpk, there are three signif icant regions which may be
considered, and a general rule is given in the table below . The value of 3 as a 'total
confidence' limit may be lowered if measurements are taken as the average of sample
batches. This commonly happens when Cpk is measured using the same data that is used
to plot the Control Chart (e.g. the confidence limit reduces to 2 for the common sample size
of 4).
In the broader sense, studying Process Capability is more than just
measuring Cp and Cpk; it involves understanding the statistical performance and
operational working of the process. Most importantly, it means understanding what causes
variation within the process, under what conditions, and how these variables interact. The
purpose of doing this is to enable confident process improvement that steadily reduces
variation.

Table 1. C pk values a nd capability

Value of Cpk Capability Action

Less than 1 Incapable Improve by reducing

common causes of
variation in proc ess
variables. Use 100%
inspection.

Between 1 and 3 Capable Do nothing or some

process improvement.
Dependent on sample
size.

Greater than 3 Very capable Do nothing or reduc e

specification limits. No
inspection necessary.

Example
A company producing kitchen worktops, specified the length of one range at 120 c m ± 0.25
cm. As a normal part of production monitoring, Control Charts were kept for such specified
measures and all output was inspected against specification limits. A reques t from the sales
team for tighter limits prompted the question of whether the limits could be reduced to ±
0.1c m, as this could result in significant orders from a new customer who preferred not to
shave worktops to fit.
The production supervisor used the Control Chart data to draw a Histogram and check for
Normal distribution, then calculate Cpk, across samples from a week's work, as in the figure
below . This showed that the process was currently so capable that inspection could be
dropped and the new limits still met. This turned out to be as a result of old specification
limits coupled with several recent process improvements.
As a result, the new orders were achieved and savings were also made on inspection costs.

Fig. 1. Process Capa bility example

Other examples
A sales completion team, aiming to reduce the shipment time of urgent orders,
studies the process, including plotting control charts and using them identifying
variables and eliminating common causes of variation. They then measure the
process capability and use the results to set delivery targets for each of a set of
geographical areas. Know ing their capability, they then guarantee delivery times to
customers.
A lathe operator doing an ongoing process improvement of his work learns the
variables involved and how to measure them. He then uses a Control Chart to help
stabilize the process. Work pieces are still sometimes failing specification limits, so
 he uses a Process Capability index measure to identify how capable the overall
process is of meeting limits. He finds that the process is centered, but only just
capable. Further process improvements help to improve this situation.
An auctioneer studies the pricing variation on different categories of cars, and uses a
Process Capability study to help determine pricing limits. She then uses this data to
start bidding prices and advise potential sellers.

How to do it
1. Identify the objectives of measuring the Process Capability. Typical reasons include:
o Prioritization of process improvement projects.
o Investigation of causes of specification limit failure.
o Setting realistic specification limits for process.
o Determining results of improvement s made.

2. Def ine the process that is to be investigated. Typically this will encompass all actions
and variables that will affect the item to be measured and which may be changed.
For example, if measuring the elasticity of a rubber cord, both the molding and
curing processes may be included in the study, when both are believed to contain
variables that affect the elasticity.

3. Identify the actual measure to make. This is usually easy to find, as it has
specification limits already defined.
Also define the process of measurement in order that measurements may be
consistently made. This may include such activities as verification of measurement
equipment accuracy, training of people doing the measurement, etc.

4. Take sample measures of the process at random times over a period which is long
enough for all variables within the process to travel through their full range of
possible values. Also take enough measurements to enable this range to be
identif iable, as detailed in step 5.
The measurements should be made under normal working conditions, because if
machines and other items in the process are set up specially for measurements to
be made, the result will only show the capability of the process under 'best case'
conditions.

5. Verify that the process is in a 'state of statistical control', with no special causes of
variation. This can be performed by plotting Control Charts and checking for
significant trends or points outside the control limits. If special causes are found,
they must be eliminated before Process Capability indices can be measured.
Make this as accurate as possible by using Average and Range (X-bar, R) Control
Charts, plotting at least 25 points, for example with 100 measurements and a
sample size of 4. If so many measurements are not possible, use Individuals and
Moving Range (X, MR) Control Charts with at least 31 measurements.
6. Verify that the process has a Normal (bell-shaped) distribution, for e xample by
plotting a Histogram and inspecting its shape.

7. Calculate the Capability Indices, using the same data that was used for drawing the
Control Charts, as below.
 Fig. 1. Capability Index ca lculation
A common short-cut method of estimating the standard deviation is used in this
example, where a table value is combined with the average range. This is useful
where subgroup, average and range measurements are already made for a Control
Chart. If an Individuals Control Chart is used, then the standard deviation can be
calculated as shown in Chapter 5.

8. Interpret and act on the results. Thus, in the figure above, Cp shows that the
process is reasonably capable of meeting specification, but Cpk shows that it also is
performing off-center. The process may then be examined to determine whether it
can be permanently centered and whether the variation can be reduced some more.

Practical variations
Cpm is an alternative Process Capability index which gives a measure in a single
calculation. The figure below. shows how to calculate Cpm, using the same data as
before.

Fig. 1. Cpm ca lculation

Set up an optimum environment before measuring the Process Capability, rather

than measuring under normal working conditions. For example, use new tools, most
skilled people, graded materials, slower execution, etc. This will give a measure of
the potential of the process. The difference between this and the measure taken
from the normal working Process Capability will give some indication of the possible
improvement that may be made.