Notes on Process Capability

Professor Graeme Britton, 2000

Process Capability
These notes are a brief introduction to process capability. For further information refer to Bothe in the
bibliography.
In order to carry out process planning a process planner must be able to predict how manufacturing
processes will perform. This is achieved by measuring the process capability of the processes. Before the
capability of a process can be measured the process must be stable. In other words, prediction of future
performance is not possible without a stable process. The performance of a process can be measured using
the dimensions of parts produced by the process. Parts made by the process will vary slightly in size, i.e., in
the values of the dimensions. The frequency distribution of the values will form a statistical distribution, a
normal distribution, Figure 1. A process is stable if the parameters of the statistical distribution dont
change, i.e., the mean and variance are constant, Figure 2.
99.73% of parts lie within the 6 limits

(6)
Number of
parts of a
specified size

large part

small part

Size of parts, e.g., shaft diameter

Figure 1.: Variables Measurement of Process Performance

constant mean and variance

Past

Future
Present

Figure 2.: Prediction Requires Constant Process Parameters

Notes on Process Capability

Professor Graeme Britton, 2000
Control charts are used to ensure that a process is statistically stable. Two main types of control charts are
used for variables measurement: X bar chart and R (or s) chart. The X bar chart controls the process mean
and the R chart controls the variance. Control charts are based on sampling distributions. They plot and
track the statistics of samples of parts, not individual parts. The X bar tracks the sample means and the R
chart the sample ranges, Figure 3.

X
Chart

averages

upper control limit

lower control limit

Range
Chart

ranges

upper control limit

lower control limit

Figure 3.: Variables Control Charts
Control limits are placed on the charts. Samples are taken at regular intervals and plotted on the charts.
Provided the sample values fall within the control limits no action is taken. If a sample value falls outside
the limits, action is taken to find out why and to rectify the situation, if necessary. A process is said to be in
a state of control if there at least 25 sub-groups (points) between out-of-control conditions. This condition is
necessary but not sufficient to define process stability. In addition, the points between control limits must be
random, due solely to common causes of variation. Non-random patterns are:

runs: 7 or more consecutive points on one side of the chart centreline,

trends: 7 or more points move consecutively up or down,

cycles: repeated patterns,

hugging the centreline: 12 consecutive points lying in the middle third of the control chart,

hugging the control limits: a large number of points near the control limits and very few near the
centreline.

Control charts are used to ensure process stability rather than inspection based on individual parts, because
they are more sensitive to changes in process parameters, Figure 4. In this example, the process mean
moves an amount equal to 6. If inspection is used there is only a 50% change of detecting this very large
shift in process mean. On the other hand, if a control chart is used there is a 99.865% change of detecting
the change.

Notes on Process Capability

Professor Graeme Britton, 2000

X
LSL

=0
=2 USL LSL
+6

-6
X=0
X=1

X
LCL

=6
=2 USL
+6

-6

99.865%

6 shift in mean
UCL

-3 X=0 +3

50%

X=6

LCL
-3

UCL

X=1

+3 X=6

Figure 4.: Control Charts Better at Detecting Change Than 100% Inspection
(Reproduced from Bothe 1997)
The power of control charts is shown in Figure 5. Note that they are very good at detecting large shifts in
the process mean, but not good at detecting small shifts in the mean. Hence long-term process variability
will, in general, be larger than the short-term variability, Figure 6.

Probability that
sub-group
average falls
outside a control
limit

Shift in process average (in units)

Figure 5.: Power of Control Charts

(Reproduced from Bothe 1997)

Notes on Process Capability

Professor Graeme Britton, 2000

Figure 6.: Long-term Variability is Greater than Short-term

Variability
(Reproduced from Bothe 1997)
Process capability is defined as the ability of a process to satisfy customer requirements. The most
common quantitative definition of process capability is the process spread, or 6, when the process is
statistically stable. This definition just measures a process parameter and does not relate this to the customer
requirements. The minimum process capability occurs when the limits of the process spread exactly match
the limits of the design specification, Figure 7. LSL and USL mean design lower and upper specification
limit respectively.

minimum capability to meet customer specifications

occurs when process limits match design limits
USL

LSL
0.135%

99.73% within limits

bilateral specification

LSL

0.135% 0.135%

99.865% above limit

unilateral specification

Figure 7.: Minimum Process Capability

(Reproduced from Bothe 1997)
Process capability can be defined using a short-term measure of variation or a long-term measure of
variation. In addition, it can be measured in terms of process spread only (potential capability) or process
spread and mean (performance capability). A process has potential capability if it could meet the minimum
requirements of process capability by shifting the mean. It has performance capability if it is actually
meeting the minimum requirements (that is, the mean is correctly positioned relative to the design
specification limits).

Notes on Process Capability

Professor Graeme Britton, 2000
It is not economically feasible to monitor all process parameters for all product characteristics, so instead,
only the critical characteristics are chosen for control. The critical characteristics may be specified by the
customer, determined by discussing with the customer or determined from customer complaints. Once the
characteristic has been chosen it is necessary to specify the capability goal, the target capability to be
achieved by the process. Often this is expressed as a percentage of conforming parts, but for many
operations this is no longer acceptable; it is necessary to work in defective parts per million. Figure 8 shows
why 99% conforming parts may not be acceptable for some processes. The goal should be set taking into
account the costs of process scrap, rework and litigation, the cost of improving the process, and the stability
of the process.
In the USA, this capability goal would result in
these events occurring DAILY
no electricity
for 14 minutes

5,280,000 cheques deducted

from the wrong bank accounts

2 plane crashes at
each major airport

18,921,600 phone calls

misplaced by
telecommunication services

Figure 8.: Unacceptable Process Capability Goal

The steps for carrying out a process capability study are:

verify the process stability

estimate the process parameters (see appendix)

measure the process capability (remember to check for normality)

compare the actual capability to the desired capability

make a decision about process changes

report results and make recommendations

There are over 60 different measures of capability (Bothe, 1997), which can be classified into short- or longterm measures and potential or performance measures. In practice, companies use more than one measure in
order to properly assess a process. For process planning we use a potential capability measure because the
factory is not yet making parts for the customer. It is also preferable to use a measure of long-term variation
as this covers all machines and operators and hence does not require the process planner to allocate a process
to a specific machine and operator (which is referred to as quality scheduling).

Notes on Process Capability

Professor Graeme Britton, 2000
Before we look at the relationship between process tolerance and process capability lets consider the main
sources of variation in manufacturing. These are:

The Machine

clearance between moving parts of slideways and bearings

geometric errors in slideways, bearings and lead-screws

dynamic stiffness

resolution of the measuring system

resolution of the positioning system

thermal stability

The Cutting Tool

tool wear

variation of tool size and cutting geometry

rigidity of the tool and support

thermal stability

Fixture

variation between duplicate fixtures

variation in location

wear and contamination of locating surfaces

deflection of locators and fixture

thermal stability

Workpiece

variation in physical and chemical properties

variation of workpiece size

rigidity of workpiece

thermal stability

stress relaxation

Coolant

variation of flow

variation of temperature

contamination

Notes on Process Capability

Professor Graeme Britton, 2000

degradation

Operator
particularly if the finished size is under the direct control of the operator

Environmental Conditions
changes in temperature affect the machine, fixture and tool geometry and hysteresis in moving parts,
e.g., slideways and bearings

Process variables
changes in process variables, such as feed and depth of cut, have a direct effect on workpiece size and
geometric variation

The dimensional variations can be classified into two groups: those that are random, unpredictable and
cannot be controlled, and those that are time dependent or capable of being controlled. Two examples of the
first group are the effects of hysteresis and random variations of the chip forming process. The second
group includes the effects of tool and fixture wear over a period of time, which can be measured and
compensated for, or predicted by a tool wear management system.
Process capability is defined as 6 , where is the standard deviation of a process population distribution
that is in a state of control. It represents the best a manufacturing process can do under existing factory
practices. For a machining process using cutting tools, e.g., turning or milling, the process capability is
defined by the statistical distribution produced by a newly sharpened tool.
In practice, the variability of a manufacturing process will be greater than the process capability because of
cutting tool wear, temperature variations during the day, size and geometric variations between fixtures, and
machine wear. Cutting tool wear has the most significant effect on variability. Figure 9 illustrates the
changes in mean value and standard deviation produced by tool wear. We shall use the term process
tolerance to refer to the actual (planned) overall variability of a process. The minimum value of a process
tolerance is the process capability. Process tolerance is the fundamental measurement unit for tolerance
planning and control. It is conveniently expressed as an equal bilateral value, e.g., 0.01mm, which defines
the 6 limits. This definition of process tolerance guarantees that at least 99.73% of parts will lie within the
process tolerance limits. Note that this definition excludes the process mean. This is acceptable because
virtually all manufacturing processes, certainly all machining processes, can be adjusted. The aim during
process planning is to ensure that the process spread and tolerance stackups lie within the specification
limits. The process means can be adjusted on the shop floor later on when the parts are actually being
produced.

Notes on Process Capability

Professor Graeme Britton, 2000

Upper limit of process tolerance

Shift in variance

Process
tolerance
New tool & fixture

Process
capability
t1

an
me
n
i
ft
Sh i

Worn tool
& fixture

Lower limit of process tolerance

t2

Time
Figure 9.: Relationship Between Process Tolerance and
Process Capability
The maximum value of a process tolerance depends upon the material being processed, the fixture
repeatability, the cutting conditions, and the factory practice for controlling tool wear. The latter is
controlled by controlling the time interval between tool offset adjustment or replacement. Three methods
are commonly used:

Allow the tool to wear without adjustment, and replace the tool once the tool life or wear-land reaches a
predetermined value. The preset tool life value may be determined using tool wear or surface finish as
the criterion.

Allow the tool to wear with adjustment and replacement at predetermined time intervals, using a PREcontrol chart to adjust the process during these time intervals. This will reduce the overall variability.

Use a statistical control chart or a computerized tool management system to control the process and to
determine when the tool needs to be adjusted or replaced. If a computerized tool management system is
used it is possible to correct for known sources of variation and hence to reduce the overall variability.

Some processes are surface roughness controlled rather than tolerance controlled, e.g., all finishing
processes. In these cases the process tolerance is a result of the process conditions that are set to produce the
required surface roughness. For processes that can produce different qualities of surface roughness there
will be a set of tolerance values, with each value corresponding to a specified value of roughness. For
example, consider finish turning: the surface roughness for turning can be controlled by varying the cutting
speed and feedrate and/or by using a finishing tool. The final surface roughness depends on the process
conditions, the finishing tool, and the control technique. The tolerance will also be affected by these factors.

Notes on Process Capability

Professor Graeme Britton, 2000
What about manufacturing processes whose variables produce a non-normal distribution?
characteristics with known non-normal distributions are listed below:

geometric dimensions (taper, flatness, concentricity, perpendicularity, angularity, roundnesss,

straightness, squareness, parallelism, hole location)

warpage

shrinkage

surface finish

Some

Geometric tolerances are refinements of dimensional tolerances and so are taken into account anyway,
unless it is necessary to consider tolerance stackups for geometric tolerances (not covered in this course).
The other characteristics may affect the process tolerance. For a non-normal distribution it is possible to
define an equivalent process capability. This is the 6 value for which at least 99.73% of the parts of the
non-normal distribution will lie between the 6 limits. This value can then be used to specify the process
tolerance.
Thus far, we have discussed process tolerances as if each process had a single tolerance value. In fact this is
not true. Modern machines are multi-functional, e.g., CNC machining centres. Different parts of a machine
will wear at different rates and effect each function differently. The actual tolerance variation of a
workpiece depends on the machine functions that are used to machine the workpiece. For example, wear in
bearings supporting rotation of a machine spindle will effect tolerances during contour milling. However,
there will be little or no effect during planar milling because the spindle is in a fixed position during the
machining cycle. The tolerance variation for each function needs to be monitored and controlled. During
process planning the appropriate tolerances on working dimensions need to be selected. This is almost
always dependent on the detailed knowledge of the machine operator.

Notes on Process Capability

Professor Graeme Britton, 2000

Appendix
Estimation of Standard Deviation Using Range
Formula applies to X bar and R charts.

est=Rmean/d2
where d2 depends on sample size and is given below:

Sample size n

d2 factor

1,128

1,693

2,059

2,326

2,534

2,704

2,847

2,970

10

3,078

10