STATISTICAL METHODS FOR INDUSTRIAL PROCESS CONTROL

David Drain

INTRODUCTION TO STATISTICAL PROCESS CONTROL

Statistical process control (SPC) is a set of statistical and management tools
that cause processes to function in a more predictable and profitable manner. It
enhances process stability and improves product quality indirectly; product
quality assurance is not the goal of a process control system.
Process monitors or process parameters are measurements made of
quantities that are thought to be indicators of process quality. A parameter
target is the ideal value for the parameter; material manufactured on target
exhibit an optimum of value for the customer and profit for the producer.
Targets are defined on the basis of process characterization, experimentation,
simulations, and experience. Targets may be defined for both attributes of the
final product, and for process parameters measured during the creation of a
product.
Specification limits are the outermost bounds of acceptable performance
with respect to a product characteristic. Upper and lower specifications limits
are based on quality or economic considerations they have no relation to the
actual performance of the process. Specification limits are closely related to
disposition limits and scrap limits.
The success of SPC rests upon a shrewd observation by Shewhart (1931)
regarding process variations. There are two basic kinds of variation:
Common cause variation (intrinsic process variation) is part of the
process itself. It produces differences from the target that are relatively
small in magnitude, and exhibit no systematic patterns over time. It
produces parameter values that are normally distributed.
Special cause variation (result of some assignable cause) an unusual
occurrence that results in a momentary large difference from target or a
systematic patter over time.

A process influenced by special causes is said to be unstable. It must be

investigated and repaired.

SPC detects special causes with the aid of control limits (boundaries of normal
process operation computed on the basis of process history and statistical
theory). A stable process will nearly always produce measurements within its
control limits.

Ideal process: stable, almost always produces material within

specification limits
Promising process: stable, but sometimes produces material out of
specification limits
o If process if off-target, a simple recipe change may be sufficient to
move it back to target and produce consistently good material
o If process is centered at target, intrinsic process variation must be
reduced (through research and experimentation to find equipment
or procedure corrections necessary)

Essential components of SPC

Successful SPC has five working parts: measurements, calculations, control
charts, trend rules, and response plans.
Measurements: cornerstone of process control. Is the most expensive
and time-consuming part of a process control system
Calculations: necessary to process and summarize measurements
Control charts: plot of summarized measurements in time order, with
control limits (boundaries for summary measurements)
Trend rules: aid to control chart interpretation. Eight on one side is
generated by a special cause
Response plan (corrective action plan): systematic way of responding to
indications of process instability. It leads the operator through a series of
predetermined actions that should usually help them find the special
cause, correct it, restart the process, and document the actions taken.

Benefits and costs of SPC

Benefits
Improves product quality

Costs
People must be trained to
implement and use SPC

Product uniform and reliable

Control systems must be

established

Enhances productivity

Metrology equipment has to be

pruchased and maintained

Costs reduces (rework, scrap, testing,

inspection)
Provide basis for continuous process
improvement
Give workers more control over their own
work

STATISTICAL PROCESS CONTROL IMPLEMENTATION

Statistical process control (SPC) is essential to manufacturing success without
it, customers receive inferior products, manufacturers waste their money, and
workers waste their time.

Select key process parameters

Key process parameters are those vital few indicators of process health that
must be carefully monitored and controlled to maintain process stability.
Critically important measurements can be logically classified into key product
characteristics and key process parameters.
Key product characteristics
Key product characteristic is some measurable property of the product that
is absolutely vital to the customer. Measurements of this type usually have
carefully researched specification limits which are documented in product
descriptions, and parts with all key product characteristics within specifications
should function in nearly any customer application.
Key process parameters are measured on equipment or material being
processed, and they exert significant influence on some key product
characteristic. They should have well-characterized targets, and they usually
also have specification limits so that material with a high probability of
eventual failure can be scrapped early in the production process.
A fundamental difference between key product characteristics and key process
parameters is the level and timeliness of control that can be exerted over each
type of measurement.
Key process parameters can be measured inside factory during
productions, they can be changed simultaneously with production of
material (input parameters)
Key product characteristics can be measured only on finished products,
measurements occurs long after the actions that created the device
(output parameters)
SPC is used to control key process parameters and reduce their variation from
target.

As a process is more thoroughly characterized, these influences are discovered,

and process control emphasis can move up the chain of cause and effect until
the only parameters measured are those closest to the process step they
influence first.
The exercise of choosing key product characteristics has benefit in itself
because it requires that customer desires are clearly understood. The further
discovery of key process parameters corresponding to each key product
characteristic enhances process understanding and provides a clear path for
improvement. The distinction between key product characteristics and key
process parameters is important because it discourages the futile practice of
applying statistical process control techniques to key product characteristics,
especially those that are influenced by process steps occurring long before the
final product test.
The discovery of key product characteristics begins by compiling a
comprehensive list of important product characteristics. Key characteristics can
be chosen by assigning a high, medium or low judgment to each of the
three criteria. The method takes three factors into account when choosing KPC:
Probability distribution of the characteristic
Measurement capability
Effectiveness of screen for bad material
Risk or loss faced by customer when inferior material is allowed to pass
A product characteristic is key if it exhibits a high probability of badness, has
a high risk of escape when bad, and causes significant damage to the customer
when bad material is released.

Key process parameters

KPC are kept on target only by controlling the KPP that influence them. Some
candidate process parameters will be obvious from the physics and chemistry
of the process. Other candidates may be found by examining a process flow
diagram actions at each step of a complex process.
A thorough review of a process will almost surely turn up more candidate
parameters than are reasonable to control. Some of the criteria that help in
assessing and comparing the suitability of candidate parameters for SPC:
1. Influence on key product characteristics. Can be found through
linear regression analysis of historical data.
2. Variance in routine manufacturing. KPP must have a practically
important impact on some key product characteristics.
A systematic way to compare the relative influences of candidates parameters
is to compute an influence index based on estimated coefficients from linear
regression analyses (1) and candidate parameter standard deviation ():

I =| 1 |
Candidates with the largest influence indices are most likely to exert an
important influence during routine manufacturing.

3. Propensity for excursions. Parameters that stray far from target

rarely but catastrophically may also be key
4. Extent of control over the candidate process parameter. A
parameter that can be easily controlled is a better choice than one that
is difficult to control
5. Time between detection of a problem and correction. Parameters
that can immediately predict and lead to the correction are better
choices
6. Influence on other important process parameters. A parameter
that, thorugh its own variation, transmits variance to other process
parameters may be a key process parameter.
7. Cost and quality of measurement. Attributes data are generally
inferior to variable data.

Design a data collection system and collect data

Select a sampling plan

The primary goal of SPC is to reduce the variation of KPP from their targets.
Samples are taken from the process so that its behavior can be gauged and
appropriate action taken depending on process stability. There is no universally
ideal sampling plan. Economic aspects of sampling are also unique to each
individual process, but some general guidelines can be applied to reduce costs.
The number of measurements taken in each sample also has a significant
effect on sampling cost and quality: larger samples cost more, but usually
obtain better information.
Design the Data Collection System
A data collection system includes the computer program, the people making
the measurements, and the tools they use to record the data. The ordered
checklist below should help in designing a data collection system:
Item
Observations
Do operators know what they are
Measurement and calculation
supposed to do?
requirements reside in operations
specification
Specification should detail: when, where,
how, and precisely what should be done
afterward
How do they know they have done the
task correctly?
Do they have the means to perform the
task?

Operators require sufficient feedback to

know if they are doing it correctly
Measurement equipment is accessible,
operators have enough time to take
required measurements

Are there incentives to reward

performance? And consequences if the
task is not performed?
Do they know how to do the task?

Taking data must not seem like

punishment; measurement equipment
should be capable and reliable
Take time to train operators to use SPC

Collect data
Before a process control system can be constructed, some information about
process behavior must be obtained. Only those machine operators who will
eventually take measurements in production should make measurements
during this initial data collection period.
Select summary measures and control charts
A summary measure is the result of simple calculations based on some raw
data. The most common summary measure is the sample mean (average) for a
group of measurements. Summary measures give important information about
the process without consuming the time and causing the distraction which
would result from plotting each individual measurement.
Summary measures are plotted on control charts, and decisions regarding the
stability of the process are based upon them, rather than raw data. The
7

particular summary measures chosen for a process control system depend on

properties of that system: type of measurements being taken, sampling
scheme, and process variance components. The purpose of a process control
system is to reduce process parameter variation from target, so summary
measures are chosen to give concise but sufficient information about this
aspect of process performance.

X-bar chart
The X-bar chart is used to control a process by tracking the mean of a set of
measurements taken at the same time (subgroup). The centerline of the X-bar
chart is the average of all measurements taken. If the process can be adjusted
to run on target, and typically runs near target, a common practice is to set the
centerline to this target. Control limits are fixed at three standard deviations of
the subgroup means from the centerline.

Trend rules
1. The one point out of 3-sigma rule states that the process should be
considered unstable.
2. The two of three points outside of 2-sigma rule decrees the process
unstable whenever two of three points fall outside of the same 2-sigma
line.
3. The four of five points outside of 1-sigma rule calls the process unstable
when four of the last five points fall outside of the same 1-sigma line
4. The eight points on one side of the centerline rule is violated if eight
successive points fall on the same side of the centerline.