CHAPTER 18

--

There is no rule for the frequency of taking subgroups, but the frequency should be often enough to detect process changes. The inconveniences of the factory or office lay. out and the cost of taking subgroups must be balanced with the value of the data obtained. In general, it is best to sample quite often at the beginning and reduce the sampling frequency when the data permit. The precontrol rule for the frequency of sampling could also be used. It is based on how often the process is adjusted. If the process is adjusted every hour, then sampling should occur every 10 minutes; if the process is adjusted every 2 hours, then sampling should occur every 20 minutes; if the process is adjusted every 3 hours then sampling should occur every 30 minutes; and so forth. ' Data Collection Assuming that the quality characteristic and the plan for the rational subgroup have been selected, a team member such as a technician can be assigned the task of collecting the data as part of his normal duties. The first-line supervisor and the operator should be informed of the technician's activities; however, no charts or data are posted at the work center at this time. Because of difficulty in the assembq of a gear hub to a shaft using a key and keyway, the project team recommends using X and R charts. The quality characteristic is the shaft keyway depth of 6.35 mm (0.250 in.). Using a rational subgroup of four, a technician obtains five subgroups per day for five days. The samples are measured, the subgroup average and range are calculated, and the results are recorded on the form as shown in Table 18-4. Additional recorded information includes the date, time, and any comments pertaining to the process. For simplicity, individual measurements are coded from 6.00 mm. Thus, the first measurement of 6.35 is recorded as 35. It is necessary to collect a minimum of 25 subgroups of data. A fewer number of subgroups would not provide a sufficient amount of data for the accurate computation of the control limits, and a larger number of subgroups would delay the introduction of the control chart.

Trial Central Lines and Control Limits The central lines for the X and R charts are obtained using the equations

R = r. Rjg
where X
Xi

= average of the subgroup averages (read " X double bar" )

= average of the ith subgroup = number of subgroups

R
Ri

= average of the subgropp ranges

= range of the ith subgroup

1 --Number pala on the fABLE18-4

operator Date 7/23 Comment 1130 X Time 1:45 420 3.45 R 41 40 36 37 32 34 35 46 0.08 0.10 0.06 6.36 6.40 44 34 68 64 59 69 38 6.65 6.39 X4 X3 X2 850 X1 Range New, temporary Average

499 MEASUREMENTS

STATISTICAL

PROCESS CONTROL

Depth of the Keyway (millimeters)

17

40 41 35 28 41 37 38 7/30 6.40 44 43 38 36 39 35 42 43 39 42 39 0.06 0.08 0.07 6.39 6.38 6.41 55 48 56 0.11 6.51 2:35 3:15 2.19 34 29 58 38 7/29 0.12 0.30 6.36 6.42 1100 935 11:35 1020 425 2:00 45 Bad material 925 Dama;;ed oil line

160.25

Trial control limits for the charts are established at shown by the equations

3(J

from the central line, as

DCL x = X + 3crx
LCLI' = X where
3crx

DCl" =
lCl"

3crR

= R - 3crR

DCl:= upper control limit

LCL = lower control limit

ax = population standard deviation of the subgroup averages

a R = population standard deviation of the range

500

CHAPTER 18

In 'p~ctice, the calculations are simplified by using the product of th~ average of the range (R) and a ~~ctor (Az) to replace the three standard deviations (AzR = 30'x) in the equation for the X chart. For the R chart, the range is used to estimate the standard deviation of the range. Therefore, the derived equations are UCLx = X

+ AzR

UCLR LCLR

= D4R = D4R

LCLx

= X - AzR

where Az, D3, and i>4 are factors that vary with the subgroup size and are found in Appendix A. For the X chart, the upper and lower control limits are symmetrical about the central line. Theoretically, the control limits for an R chart should also be symmetrical about the central line. But, for this situation to occur, with subgroup sizes of six or less, the lower control limit would need to have a negative value. Because a negative range is impossible, the lower control limit is located at zero by assigning to D~ the value of zero for subgroups of six or less. When the subgroup size is seven or more, the lower control limit is greater than zero and symmetrical about the central line. However, when the R chart is posted at the work center, it may be more practical to keep the lower control limit at zero. This practice eliminates the difficulty of explaining to the operator that points below the lower controllimit on the R chart are the result of exceptionally good performance rather than poor performance. However, quality personnel should keep their own charts with the lower control limit in its proper location, and any out-of-controllow points should be investigated to determine the reason for the exceptionally good performance. Because subgroup sizes of seven or more are uncommon, the situation occurs infrequently. In order to illustrate the calculations necessary to obtain the trial control limits and the central line, the data concerning the depth of the shaft keyway will be used. From Table 18-4, LX = 160.25, LR = 2.19, and g = 25; thus, the central lines are

x=

L X/g = 160.25/25 = 6.41 mm

R = L Rig = 2.19/25 = 0.0876 mm

From Appendix Table A, the values for the factors for a subgroup size (n) of four are Az = 0.729, D3 = 0, and D4 = 2.282. Trial control limits for the X chart are UCLx

= X + AzR = 6.41 + (0.729)(0.0876)

= 6.47 mm

LCL.x = X - AzR = 6.41 - (0.729)(0.0876) = 6.35 mm

Trial control limits for the R chart are

UCLR

= D4R = (2.282)(0.0876)
= 0.20 mm

LCLR

= D3R = (0)(0.0876)
=Omm

STATISTICAL

PROCESS CONTROL

501

Control charts -

Depth of keyway

\\6.47 I .A 6.41 " e \. I ~ s I

OJ:)

I ..

6.35

"

~ s I

OJ:) <=

0.09 0

LCLR

5
Figure 18-17

10

15
Subgroup number

20

25

X and R Charts for Preliminary Data with Trial Control Limits

Figure 18-17 shows the central lines and the trial control limits for X and R charts for the preliminary data. Revised Central Lines and Control Limits Revised central lines and control limits are established by discarding out-of-control points with assignable causes and recalculating the central lines and control limits. The R chart is analyzed first to determine if it is stable. Because the out-of-control point at subgroup 18 on the R chart has an assignable cause (damaged oil line), it can be discarded from the data. The remaining plotted points indicate a stable process.

CHAPTER 18

The X chart can now be analyzed. Subgroups 4 and 20 had an assignable cause, whereas the out-of-control condition for subgroup 16 did not. It is assumed that subgroup 16's out-of-control state is due to a chance cause and is part of the natural variation of the process. The recalculated values are Xo = 6.40 mm and Ro = 0.079. They are shown in Figure 18-=-18. For illustrative purposes, the trial values are also shown. The limits for both the X and R charts became narrower, as was expected. No change occurred in LCLR because the subgroup size is less than 7. The figure also illustrates a simpler charting technique in that lines are not drawn between the points. Also, Xo and Ro, the standard or reference values, are used to designate the central lines. The preliminary data for the initial 25 subgroups are not plotted with the revised Controllimits. These revised control limits are for reporting the results for future subgroups. To make effective use of the control chart during production, it should be displayed in a conspicuous place, where it can be seen by operators and supervisors.
Control charts Depth of keyway

'" E I E 18:

~ 6.35 ~ 6.41

<l) ...

6.47

UCLx = 6.46

0.20

LCLx

Xo

= 6.40

LCLx =6.34

UCLR

=0.18

E '" 0/) ~

E
I

R
o 0.09

r-

LCLR Ro = 0.08

L _______

L-l
5 Subgroup number 5 10

Figure 18-18 Trial Control Limits and Revised Control Limits for X and R Charts

STATISTICAL

PROCESS CONTROL

50~

Before proceeding to the action step, some final CClmments are appropriate. First, many analysts eliminate this step in the procedure because it appears to be somewhat redundant. However, by discarding out-of-control points with assignable causes, the central line and control limits are more representative of the process. If this step is too complicated for operating personnel, its - elimination would not affect the next step. Second, the central line X(I for the X chart is frequently based on the specifications. In such a case, the procedure is used only to obtain RD. If. in our example problem, the nominal value of the characteristic is 6.38 mm, then Xo is set to that value and the upper and lower control limits are
UCLx

= Xo + Az~ = 6.38 + (0.729)(0.079)

LCL" =

Xo

- AzRo

= 6.44 mm

= 6.38 ..c. (0.729)(0.079) = 6.32 mm

The central line and control limits for the R chart do not change. This modification can be taken only if the process is adjustable. If the process is not adjustable, then the original calculations must be used. Third, it follows that adjustments to the process should be made while taking data. It is not necessary to run nonconforming material while collecting data, because we are primarily intere~ted in obtaining Ro, which is not affected by the process setting. The independence of Xo and Ro provides the rationale for this concept. Fourth, the process determines the central line and control limits. They are not established by design, manufacturing, marketing, or any other department, except for Xo when the process is adjustable.

When control charts are first introduced at a work center, an improvement in the process performance usually occurs. This initial improvement is especially noticeable when the process is dependent on the skill of the operator. Posting a quality control chart appears to be a psychological signal to the operator to improve performance. Most workers want to produce a quality product or service: therefore, when management shows an interest in the quality, the operator responds. Figure 18-19 illustrates the initial improvement that occurred after the introduction of the X and R charts in January. Owing to space limitations, only a representative number of subgroups for each month are shown in the figure. During January the subgroup ranges had less variation and tended to be centered at a slightly lower point. A reduction in the range variation occurred also. Not all the improved performance in January was the result of operator effort. The first-line supervisor initiated a program of tool-wear control, which was a contributing factor. At the end of January new central lines and control limits were calculated using the data from subgroups obtained during the month. It is a good idea, especially when a chart

CHAPTER 18

Control charts 6.46

E E
I

Depth of keyway

1- __ ----------

18 6.40
<1.l

"

DCl

Oil

e
<1.l

-----

lCl

-<

6.34 0.18 ~------------

g
I

E '" 0: ~ E
Oil
<1.l

0.08 0

DCl

January

February

July

Figure 18-19 Continuing Use of Control Charts. Showing Improved Quality

is being initiated, to calculate standard values periodically to see if any changes have occurred. This reevaluation can be done for every 25 or more subgroups, and the results can be compared to the previous values. New control limits for the X and R charts and central line for the R chart were established for the month of February. The central line for the X chart was not changed because it is the nominal value. During the ensuing months, the maintenance department replaced a pair of worn gears, purchasing changed the material supplier, and tooling modified a workholding device. All these improvements were the result of investigations that tracked down the causes for out-of-control conditions or were ideas developed by a project team. The generation of ideas by many different personnel is the most essential ingredient for continuous quality improvement. Ideas from the operator, first-line supervisor, quality assurance, maintenance, manuf~cturing engineering, and industrial engineering should be evaluated. This evaluation or testing of an idea requires 25 or mere subgroups. The control chart will tell if the idea is good, is poor, or has no effect on the process. Quality improvement occurs when the plotted points of the X chart converge on the central line, when the plotted points of the R chart trend downward, or when both actions occur. If a poor idea is tested, then the reverse occurs. Of course, if the idea is neutral, it will have no effect on the plotted point pattern. To speed up the testing of ideas, the taking of subgroups can be compressed in time as long as the data represent the process by accounting for any hourly or day-to-day fluctuations. Only one idea should be tested at a time; otherwise, the results will be confounded.

STATISTICAL

PROCESS CONTROL

505

At the end of June, the periodic evaluation of the past performance showed the need to revise the central lines and the control limits. The performance for the month of July and subsequent months showed a natural pattern of variation and no quality improvement. At that point, no further quality improvement would be possible without a substantial investment in new equipment or equipment modification. Dr. Deming has stated that if he were a banker, he would not lend money to an organization unless statistical methods were used to prove that the money was necessary. This is precisely what the control chart can achieve, provided that all personnel use the chart as a method of quality improvement rather than a monitoring function. When the objective for initiating the charts has been achieved, their use should be discontinued or the frequency of inspection be substantially reduced to a monitoring action by the operator. The median chart is an excellent chart for the monitoring activity. Efforts should then be directed toward the improvement of some other ql!1Jity characteristic. If a project team was involved, it should be recognized and rev.arded for its performance and disbanded. The U.S. Postal Service at Royal Oak, Michigan used a variables control chart to reduce nonconformance in a sorting operation from 32% to less than 6%. Th:, activity resulted in an annual savings of $700,000 and earned the responsible team the 1999 RIT/USA Today Quality Cup for government.

State of Control
When the assignable causes have been eliminated from the process to the extent that the points plotted on the control chart remain within the control limits, the process is in a state of control. No higher degree of uniformity can be attained with the existing process. However, greater uniformity can be attained through a change in the basic process resulting from quality improvement ideas. When a process is in control, there occurs a natural pattern of variation, which is illustrated by the control chart in Figure 18-20. This natural pattern of variation has

LCL

Figure 18-20 Natural Pattern of Variation of a Control Chart

:CHAPTER

18

(1) about 34% of the plotted points on an imaginary band between one standard devia_ tion on both s-ides of the central line, (2) about.l~.5% of the pl~tted points in an imag_ inary band between one and two standard deviatIOns on both sides of the central line and (3) about 2.5% of Ihe plotted points in an imaginary band between two and thre~ standard deviations on both sides of the central line. The points are located back and forth across the central line in a random manner, with no points beyond the controllim_ its. The natural pattern of the points, or subgroup average values, forms its OWnfrequency distribution. If all the points were stacked up at one end, they would form a normal curve. When a process is in control, only chance causes of variation are present. Small variations in machine performance, operator performance, and material characteristics are expected and are considered to be part of a stable process. When a process is in control, certain practical advantages accrue to the producer and consumer: I. Individual units of the product will be more uniform, or, stated another way, there will be less variation., 2. Because the product is more uniform, fewer samples are needed to judge the quality. Therefore, the cost of inspection can be reduced to a minimum. This advantage is extremely important when 100% conformance to specifications is nOI essential. 3. The process capability, or spread of the process, is easily attained from 60. With a knowledge of the process capability, a number of reliable decisions relative to specifications can be made, such as the product specifications; the amount of rework or scrap when there is insufficient tolerance; and whether to produce the product to tight specifications and permit interchangeability of components or to produce the product to loose specifications and use selective matching of components. 4. The percentage of product that falls within any pair of values can be predicted with the highest degree of assurance. For example, this advantage can be very important when adjusting filling machines to obtain different percentage of items below, between, or above particular values. 5. It permits the customer to use the supplier's data and, therefore, to test only a few subgroups as a check on the supplier's records. The X and R charts are used as statistical evidence of process control. 6. The operator is performing atisfactorily from a quality viewpoint. Further improvement in the process can be achieved only by changing the input factors: materials, equipment, environment, and operators. These changes require action by management. When only chance causes of variation are present, the process is stable and predictable over time, as shown in Figure l8-2l(a). We know that future variation as shown by the dotted curve will be the same unless there has been a change in the process due to an assignable cause.

STATISTICAL

PROCESS CONTROL

507

Prediction

(a) Only chance causes of variation present

Size~

/ 'y/ / / "/' ~--7---:::"'-

/-,

/'

Jf

Prediction

Size ~ Figure 18-21 Stable and Unstable Variation

(b) Assignable causes of variation present

Out-of-Control Process
Figure 18-2 (b) illustrates the effect of assignable causes of variation over time. The unnatural, unstable nature of the variation makes it impossible to predict future variation. The assignable causes must be found and corrected before a natural stable process can continue. The term out of control is usually thought of as being undesirable; however, there are situations where this condition is desirable. It is best to think of the term out of control as a change in the process due to an assignable cause. A process can also be considered out of control even when the points fall inside the 3a limits. This situation, as shown in Figure 18-22, occurs when unnatural runs of variation are present in the process. It is not natural for seven or more consecutive points to be above or below the central line as shown at (a). Another unnatural run occurs at (b), where six points in a row are steadily increasing or decreasing. At (c),'the space is divided into four equal bands of 1.5a. The process is out of control when there are two successive points at 1.5a beyond.4
4

For more information, see A. M. Hurwitz and M. Mather, "A Very Simple Set of Process Control Rules," Qualiry Engineering 5, no. I (1992-1993): 21-29.

CHAPTER 18

VCL ~------LCL Process ~------out of control

Xo LCL

f--------

VCL

--t-~----!'.-~---LCL

(a) Seven consecutive (b) Six consecutive points above points increasing or or below decreasing Figure 18-22 Some Unnatural Runs-Process Out of Control

(c) Two consecutive points in outer quarter

There are some common questions to ask when investigating an out-of-control process: 1. Are there differences in the measurement accuracy of the instruments used? 2. Are there differences in the methods used by different operators? 3. Is the process affected by the environment? If so, have there been any changes? 4. Is the process affected by tool wear? 5. Were any untrained workers involved in the process? 6. Has there been any change in the source of the raw materials? 7. Is the process affected by operator fatigue? 8. Has there been any change in maintenance procedures? 9. Is the equipment being adjusted too frequently? 10. Did samples come from different shifts, operators, or machines? It is advisable to develop a checklist for each process using these common questions as a guide.

ess Capability
Control limits are established as a function of the averages-in other words, controllimits are for averages. Specifications, on the other hand, are the permissible variation in the size of the part and are, therefore, for individual values. The specification or tolerance limits are established by design engineers to meet a particular function. Figure 18-23 shows that the location of the specifications is optional and is not related to any of the other features in the figure. The control limits, process spread (process capability), distribution of averages, and distribution of individual values are interdependent. They are determined by the process, whereas the specificationS: have an optional location. Control charts cannot determine if the process is meeting specifications.

STATISTICAL

PROCESS CONTROL

509

Upper specification (optional location) --------------------

Distribution of averages

--f-T-(--Control
limi ts

I I

'- '-

+ 3<T__

x Y... __

I /"/"

/"

"/

I
Process

USL Distribution of individual values

capability

3<T

L II

_
~
Lower specification (optional location) -------------------LSL

Figure 18-23 Relationship of Limits, Specifications, and Distributions

The true process capability cannot be determined until the X and R charts have achieved the optimal quality improvement without a substantial investment for new equipment or equipment modification. When the process is in statistical control, process capability is equal to 60, where 0 = Ro/d2 and d2 is a factor from Appendix Table A. In the example problem, it is 60 = 6(Ro/d2) = 6(0.079/2.059) = 0.230 It is frequently ~ecessary to obtain the process capability by a quick method rather than by using the X and R charts. This method assumes the process is stable or in statistical control, which mayor may not be the case. The procedure is as follows: 1. Take 25 subgroups of size 4, for a total of 100 measurements. 2. Cr lculate the range, R, for each subgroup. 3. Calculate the average range:

Ii = ~ Rig.

4. Calculate the estimate of the population standard deviation:

where

d2

is obtained from Appendix Table A and is 2.059 for n = 4.

5. The process capability will equal 60.

CHAPTER 18

LSL
Tolerance 20

USL

Capability

10

/
13.09 13.15 13.21 Hole location-mm

Nominal

o
13.27

Figure 18-24 Relationship of Process Capability to Tolerance

Remember that this technique does not give the true process capability and should be used only if circumstances require its use. Also, more than 25 subgroups can be used to improve accuracy. The relationship of process capability and specifications is shown in Figure 18-24. Tolerance is the difference between the upper specification limit (USL) and the lower specification limit (LSL). Process capability and the tolerance are combined to form a capability index, defined as USL - LSL 6cr where USL - LSL = upper specification - lower specification, or tolerance Cp = capability index 6cr = process capability

If the capability index is greater than 1.00, the process is capable of meeting the specifications; if the index is less than 1.00, the process is not capable of meeting the specifications. Because processes are continually shifting back and forth, a Cp value of 1.33 has become a de facto standard, and some organizations are using a 2..00 value. Using the capability index concept, we can measure quality, provided the process is centered. The larger the capability index, the better the quality. We should strive to make the capability index as large as possible. This result is accomplished by having realistic specifications and continual striving to improve the process capability. The capability index does not measure process performance in terms of the nominal or target value. This measure is accomplished using Cpk, which is

STATISTICAL

PROCESS CONTROL

511

C
pk

Min {(USL - X) or (X - LSL)} = ------.--3cr

A Cpk value of 1.00 is the de facto standard, with some organizations using a value of 1.33. Figure 18-25 illustrates Cp and Cpk values for processes that are centered and also off center by lcr.
Case I Cp = (USL - LSL)/6 = 8a/6a = 1.33

6a

LSL
Cp = 1.33
Cpk

USL

LSL

Cp = 1.33 Cpk = 1.00

= 1.33

Case II Cp = (USL - LSL)/6 = 6a/6a = 1.00

6a

6a

LSL

Cp = 1.00
Cpk

Cp = 1.00 Cpk = 0.67

= 1.00

Case III Cp = (USL - LSL)/6 =

4a/fu

= 0.67
6'1

6a

LSL
Cp =0.67 Cpk= 0.67

USL

LSL

1'0 USL
Cp = 0.67 Cpk = 0.33

Figure 18-25 Cp and Cpk Values for Three Different

Situations

51 Z

CHAPTER

18

Comments concerning 1. The

2. 3. Cp Cpk Cp

Cp

and

Cpk

are as follows:

value does not change as the process center changes. when the process is centered.
Cpo

Cpk

is always equal to or less than

4. A

Cpk

value greater than 1.00 indicates the process conforms to specifications.

5. A Cpk value less than 1.00 indicates that the process does not conform to specifications. 6. A Cp value less than 1.00 indicates that the process is not capable. 7. A Cpk value of zero indicates the average is eqUid to one of the specification limits. 8. A negative
Cpk

value indicates that the average is outside the specifications.

Quality professionals will use these eight items to improve the process. For example, if a Cl' value is less than one, then corrective action must occur. Initially 100% inspection is necessary to eliminate noncomformities. One solution would be to increase the tolerance of the specifications. Another would be to work on the process to reduce the standard deviation or variability.

Different Control Charts for Variables

Although most of the quality control activity for variables is concerned with the X and R charts, there are other charts that find application in some situations. These charts are described in Table 18-5.

Control Charts for Attributes

An attribute, as defined in quality, refers to those quality characteristics that c' specifications or do not conform to specifications. There are two types: 1. Where measurements are not possible, for example, as color, missing parts, scratches, and damage. 2. Where measurements can be made but are not made In other words, although the diameter of a hole can be crometer, it may be more convenient to use a "go-no go" forms or does not conform to specifications. .orm to

visually inspected items such because of time, cost, or need. measured with an inside migauge and determine if it con-

Where an attribute does not conform to specifications, various descriptive terms are used. A nonconformity is a departure of a quality characteristic from it. intended level

TABLE 18-5

Different Control Charts for Variables

Type
X

Central Line

Central Limits
UCLj( LCLj(

Comments Use when more sensitivity is desired than R; when

and 5

x
5

=X +
=
X -

A35 A35

UCLs = 845 LCLs = 835

UCLj( LCLj(

n > 10; and when data are

collected automatically. Use when only one observation is possible at a time. Data needn't be normal.

MX

Moving average, and moving range, MR

x
R

= X + Ai? = X - All
oi? = =

UCLR = LCLR

oil
X

X and moving R

x
R

UCLx LCLx UCLR LCLR

+
-

2.660 R 2.660 R

=X
=
= (0) =

3.276 R
R

Use when only one observation is possible at a time and the data are normal. Equations are based on a moving range of two.

Median and Range

UCLMd

MdMd

+ As
-

RMd RMd

LCLMd = MdMd

As

Use when process is in a maintenance mode. Benefits are less arithmetic and simplicity.

= D6 LCLR = Ds
UCLR

RMd RMd

or state that occurs with a severity sufficient to cause an associated product or service not to meet a specification requirement. The definition of a defect is similar, except it is concerned with satisfying intended normal or reasonably foreseeable usage requirements. Defect is appropriate for use when evaluation is in terms of Ilsage, and nonconformity is appropriate for conformance to specifications. The term nonconforming unit is used to describe a unit of product or service containing at least one nonconformity. Defective is analogous to defect and is appropriate for use when a unit of product or service is evaluated in terms of usage rather than conformance to specifications. In this section we are using the terms nonconformity and nonconforming unit. This practice avoids the confusion and misunderstanding that occurs with defect and defective in product-liability lawsuits. Variable control charts are an excellent means for controlling quality and subsequently improving it; however, they do have limitations. One obvious limitation is that these charts cannot be used for quality characteristics that are attributes. The converse is not true, because a variable can be changed to an attribute by stating that it conforms or does not conform to specifications. In other words, nonconformities such as missing parts, incorrect color, and so on, are not measurable, and a variable control chart is not applicable.