By Michelle Paret and Paul Sheehy, Minitab

SOFTWARE/SPC

Being in Control
The fact that variation exists means SPC is critical to monitor process behaviour
Variation surrounds us. The amount of time you spend
commuting to work is slightly different each day. If you
buy two bottles of soda, they may not be filled to precisely
the same height. The light bulbs you purchase dont last the
exact same number of hours. In some cases, variation is
welcomefor instance, almost everyone welcomes a
shorter-than-usual commute.
But variation is not welcome in business and industry.
When you buy a hamburger from a restaurant, or cash a
cheque at the bank, you expect consistency. Slight variation
is inevitable, but we are satisfied as long as it stays within
acceptable limits. The teller who takes your cheque today
may not be the same one you saw last week, but as long
as you receive the correct amount of cash, that variation
is acceptable.
The fact that variation exists makes it imperative to use
statistical process control (SPC) to monitor process behaviour.
We want processes to be stable, consistent, and predictable,
because only then can we ensure the quality of products and
services. Assessing process stability enables us to identify and
eliminate sources of variation that adversely affect a process.
A primary SPC tool for monitoring process stability is the
control chart, first introduced by Dr. Walter A. Shewhart at
Bell Laboratories in the 1920s. A control chart turns
time-ordered data for a particular characteristicsuch as
product weight or call centre hold timeinto a picture that
is easy to understand and react to when necessary. These
specialized time series plots provide a signal when there
are unusual shifts in a process. Statistically speaking, control
charts help you detect nonrandom sources of variation in the
data. In other words, they separate variation due to common
causes from variation due to special causes, where:
Common cause variation is variation that is naturally
inherent in a process, and always present.
Special cause variation represents assignable or unusual
sources of variation that are not typically part of a
process. Special causes can be either detrimental or
beneficial to a process.
Statistical software makes it easy to create and interpret
control charts, but those charts are useless if they arent created
using the right subgroups of your data. Understanding and
choosing rational subgroups before you collect your data and
create control charts is critical, but the concept is frequently
misunderstood. This article demonstrates what can go wrong
when subgroups are improperly assigned.
What are subgroups?
Rational subgroups reflect how your data are collected.
They also represent the inherent variation in your process

at any given time.You typically should collect data so that

each subgroup includes only the variation that is naturally
inherent to the process (common cause variation). Subgroups
should be as free of special-cause variation as possible. The
reason? Rational subgroups help you identify other sources of
variation that may adversely affect the process (special cause
variation). The control limits on a control chart, which help
to signal when a process is unstable, are calculated using the
variability within each subgroup. Therefore, it is important
to select subgroups that represent only the common cause
variation in a process.
For many processes, you can form rational subgroups by
sampling multiple observations that are close together in
time. Examples include parts manufactured at the same time
or created by the same operator. If you sample ten parts at a
time, then the subgroup size is ten. Or, you may sample say,
one part per hour, in which case the subgroup size would be
equal to one.
Case Study
Now we will consider how improperly assigned subgroups
impacted a turbine blade manufacturers control charts and
thus their assessment of their process. The turbine blades
at this facility were manufactured on four- and six-spindle
tracing mills (see Fig. 1). The tracing mill has a master that
is used to generate four or six identical blades equal to
one load that look exactly like the master. Due to the
lengthy four-hour process required to manufacture these
blades, only two loads are completed per shift, for a total of
six loads per day.

Figure 1. A tracing mill manufactures four

turbine blades per load.
(Photo from MachineTools.com)

Engineers discovered issues with the quality of the blades

at final inspection, so they had to find out why the blades
were defective and fix the problem. To seek root causes, they
decided to focus on a key characteristic, the chord length. In
mathematical terms, chord length is the straight line distance
between two ends of an arc. For a curved blade, the chord
length is the distance across the opening of the blade from
corner to corner. Anecdotally, the engineers were informed
that the four-spindle mill, equipment number SCM4, was
exhibiting the worst quality.
The engineers decided to do an in-process evaluation of
chord length using control charts.
Quality Canada

summer 2009

19

SOFTWARE/SPC
The First Analysis
For many processes, you can form rational subgroups by
sampling multiple observations that are close together in time
and produced under similar conditions. The engineers at the
turbine manufacturing facility used this principle to form their
subgroups. Since each load produces four blades made at the
same time by the same operator with the same tooling, the
engineers chose to use each load of four blades as the rational
subgroup, or a subgroup size of four. Data were collected and
entered into Minitab Statistical Software (see Fig. 2).

Spindle 3, or Spindle 4, thereby possibly introducing

an assignable cause of variation.
The Second Analysis
She recommended that the engineers use separate control
charts for each spindle rather than using subgroups across
spindles by load. Since the volume of blades was low, with
one part per spindle every four hours, and the tooling
was changed at the start of each new load, the engineers
recognized that there were no logical subgroups. They
decided to use a subgroup size of one and create separate
I-MR charts for each spindle. The results showed that
Spindles 1, 2, and 3 were all stable, while Spindle 4 was
clearly out of control (see Fig. 4).

Figure 2. The chord length measurements are recorded by spindle per load.

The engineers then created an Xbar-R chart to monitor

the process stability where each subgroup (or row of data)
included the four blades from the same load produced by
the four-spindle tracing mill. The Xbar-R chart (see Fig.
3) showed an extremely stable process where all points fell
within the control limits and no other tests for special causes
failed. The process appeared to be in statistical control.

Figure 3. The Xbar-R chart shows that the process is stable.

A new Black Belt observing the result commented that

perhaps the engineers choice for the rational subgroups was
invalid. She said her Master Black Belt emphasized that, by
design, a rational subgroup must contain only common cause
variation, and should not contain special or assignable cause
variation. An assignable cause is generally indicated if you
can assign a name to a feature that is different for the
samples within a subgroup. In this case, she pointed out,
there were different features for each of the four samples
within a subgroup, namely the spindle on which each blade
was produced. Spindle 1 may be different from Spindle 2,
20

summer 2009

Quality Canada

Figure 4. Creating separate control charts for each spindle reveals a problem
with Spindle 4.

Why did the first Xbar-R chart show that the process
was in control, while the second analysis revealed an issue
with Spindle 4? The variation introduced to each subgroup
by Spindle 4 increased the within-subgroup variation, and
therefore the control limits on both the Xbar and R charts
were falsely inflated. The widened control limits that resulted
from the improper subgroup assignment effectively hid
the problem with Spindle 4, and it would have remained
undetected had a second analysis not been suggested.
This case study clearly demonstrates why rational
subgroups should only contain common cause variation.
If special cause variation is mistakenly introduced into the
subgroups, processes can appear to be stable even when
they are not. In this case, the engineers detected why the
blades were defective only when the rational subgroups were
properly chosen.
The control chart is one of the most important tools in
an SPC toolkit. Used properly, these charts help us separate
natural variation in a process from unusual variation that we
need to pay attention to. However, control charts are only as
good as the data that are used to create them. QC
Michelle Paret, Product Marketing Manager
Paul Sheehy,Technical Training Specialist
Minitab Inc.
www.minitab.com