You are on page 1of 26

2/13/2019

EBB 338 Process Control


Chap 2:Statistical process control

Assoc. Prof. Ir Dr. Pung Swee Yong

(Standardize) Normal Distribution


• Because of its special characteristic of symmetry, it is convenient to visualize the
normal distribution as being divided into units of standard deviation (referred to as z-
values)
• The formula to convert actual values to z-vales:

where :
 = 0,  = 1

• Sometimes, it is also called z-distribution.


• The area under the normalized curve:
-1 < z < +1 = 68.3% of the total area
-2 < z < +2 = 95.4% of the total area
-3 < z < +3 = 99.7% of the total area

1
2/13/2019

Normal Distribution

What are Control Charts?


• Control charts are specialized graphs that provide us with
information in two dimensions:
1. the distribution of the process (average and variance)
2. process trending

UCL: a boundary of 3 above CL.


LCL: a boundary of 3 below CL

2
2/13/2019

Application of Control Charts

• Control charts are used to


i. Baseline process performance
ii. Monitor and control process performance
iii. Evaluate measurement systems
iv. Compare multiple processes
v. Compare processes before and after a change

How?

Process Process

Process Process

3
2/13/2019

How to use Control Charts?


• Sample data (sub-group data) is plotted on a control
chart to monitor process stability.
• In a stable process, samples will be randomly
distributed around the center line of the control chart.
This random arrangement of data reflects the normal
variation that we would expect in any process (common
cause variation).
• When the data pattern is not random, it’s a signal that
a process shift has occurred and the process is
unstable (special cause variation).

How to create Control Charts?


1. Specify the characteristics that you are interested in
control charting.
2. Specify the type of data that will be used:
a. Discrete: counts, proportions, percentages
b. Continuous: all measurement data e.g. length, vol.,
speed, etc.
3. Define the sampling approach:
a. Specify sample size (sub-group)
b. Specify sampling frequency

4
2/13/2019

How to create Control Charts?


4. Select the appropriate control chart.
5. Begin sampling and charting data on a run chart.
6. Plot the data on the control chart when there is a sufficient
number of individual values or subgroups
a. Attribute control charts: 25 or more
b. Variable control charts: 50 or more
7. Monitor the control chart for process stability. If a special
cause signal occurs, take corrective action as required.
8. Upper and lower control limits should be recalculated
whenever a significant change in the process is implemented
that alters the output of process.

Control charts for Discrete (Attribute) data


• Attribute control charts are used with proportion and
count data.
• Typically, attribute control charts are used to display
and track proportions or counts for some
characteristic of interest e.g. proportions of non-
conforming product, the number of customers served
etc.
• Type of attribute control charts

1.p-chart 3. c-chart
2.np-chart 4. u-chart

5
2/13/2019

Control charts for Continuous (variable) data


• Variables control charts are used with continuous
(measurement) data and are normally used to monitor
and control the inputs (x variables) that affect process
performance.
• E.g. length, volume, speed, temperature etc

Control charts for Continuous data: Moving Range


(MR) chart

• Sometimes it is necessary to define rational


subgroups in such a way that each sample can
contain only one value.
• It is impossible to compute a sample range or
sample standard deviation for a sample of size 1 so
some other method is needed to construct control
charts.

6
2/13/2019

Control charts for Continuous data: Moving Range


(MR) chart
• It is used in conjunction with individuals chart to monitor process
variation stability.
• Moving range values are calculated by taking the difference
between successive pairs of measurements
• The characteristics of the MR chart:

The average of the moving ranges

Control charts for Continuous data: Moving Range


(MR) chart

 The difference between data point, xi , and its


predecessor, xi-1 , is calculated as .

 For m individual values, there are m - 1 ranges.

 Next, the arithmetic mean of these values is calculated


as

 If the data are normally distributed with standard


deviation  then the expected value of
=

7
2/13/2019

Control charts for Continuous data: Moving Range


(MR) chart

1. Subtract the second data point from the first data point
and record this value. As an example take a data set of {1,
4, 4, 2, 7, 3}. Subtracting the second data point from the
first gives us: 1-4 = -3.
2. Take the absolute value of the result. Continuing the
example: abs(-3) = 3. Record the result as the first entry in
a list.
3. Repeat step 1 and 2 for the rest of the data points starting
by subtracting the third from the second. Again from the
example data set, {1, 4, 4, 2, 7, 3} : {(1-4), (4-4), (4-2), (2-
7), (7-3)} = {-3, 0, 2, -5, 4} = {3, 0, 2, 5, 4}. This list is the
moving range for your data set.

Control charts for Continuous data: Moving Range (MR)


chart
Example
An MR control chart for the customer service center is shown in Fig.
below. Is the hold time variability stable?

It shows an average for the MR equal to 53.3 sec. The


normal values for MR are expected to fall between 0.0 sec
(LCL) and 174.1 sec (UCL). The one point that is outside the
UCL was caused by malfunction of telephone switching
equipment

8
2/13/2019

Control charts for Continuous data:


individual (X) chart
• Individual chart works best when used with data that
is normally distributed (non-normal data should be
transformed to normal).
• When process is under control, both the sample
statistic (mean, range/std. dev.) should fall within the
control limits.
• When either of them is found to have shifted
significantly, undesirable changes in the process may
be occur.

Control charts for Continuous data:


individual (X) chart
• The characteristics of the individual chart:

: the average of the individual observations

When the calculated value for lower control limit is


negative number, a LCL is not placed on the control
chart.

9
2/13/2019

Control charts for Continuous data: individual chart


Example
The manager of a customer service center is interested in improving
customer satisfaction by reduction telephone on-hold time. He decides to
use an individuals chart to assess hold time stability and prepares the
individuals chart shown in Fig. below.

It shows that the telephone on-hold time is stable. The


average on-hold time is 248.4 sec. The normal range for
hold time is 106.6 sec (LCL) to 390.1 sec (UCL).

Sub-groups
• It is a sample where all items included in the sample
were produced under very similar conditions. Thus,
ONLY common cause of variation within each of the
sample subgroups is included.
• A special cause signal will occur when the variation
between subgroups is significantly greater than
the variation within sub-groups.
• The selection of good rational subgroups is important
in enabling us to distinguish a “signal” (special cause
variation) from the process noise (common cause
variation).
• The size of subgroups is small, i.e. around 2-5 units.

10
2/13/2019

Control charts for Continuous data: Range (R) chart

• It is used to monitor process variation stability.


• R values measure the variation within sample subgroups.
• R-chart is normally used in conjunction with the chart.
• The characteristics of the R chart:

A point plotting outside the control limits is evidence that the


process is out of control.

11
2/13/2019

Control charts for Continuous data: Range (R) chart

Example
A quality engineer for a metal machining company prepared an R chart
(subgroup = 3) to assess process variation stability of electrode diameter
for a product used in the electroplating industry. Comment on her control
chart.

It shows the variability to be stable, since none of the subgroup


ranges fall outside of the control limits. The average range for
electrode diameter is 0.3169 cm and the normal process width
for diameter range is 0.000cm (LCL) to 0.8157 cm (UCL)

s-chart
• This chart can be used in place of the R chart.
• In an S chart, the center line and the 3 upper and
lower control limits are given by

3 upper limit = B 4 s
Center line = s
3 lower limit = B3 s

• The values B3 and B4 depend on the subgroup size.

12
2/13/2019

Example of s-chart

13
2/13/2019

Which is better to use, R or s-chart?

• Both the R and s-chart have the same purpose: to


estimate the process standard deviation and to
determine whether it is in control.
• It seems more natural to estimate the process
standard deviation s than with the range, R. (n < 9)
• In fact, when the population is normal, s is a more
precise estimate of the process standard deviation
than R, because it has smaller uncertainty.
• It follows that the s-chart is a better choice, especially
for larger sample sizes. (n  9)

Control charts for Continuous data: chart when use


together with R chart
• It is used to monitor the stability of the process mean.
• Averages of sample subgroups are plotted on the control chart to
provide a display of subgroup-to-subgroup variation.
• It is typically used in moderate high volume production
environments where it makes sense to use sample subgroups
• The characteristics of the chart:

14
2/13/2019

Control charts for Continuous data: chart when use


together with s chart

15
2/13/2019

Control charts for Continuous data: chart

Example:
After accessing the within subgroup variation using an R chart,
the quality engineer produced an chart to check the sub-group
to sub-group variability.
The chart shows the
process average fall
outside of the control
limits. The average of
the subgroup averages
for electrode diameter
is 9.6681 cm (x) and the
normal range for
average diameter is
between 9.6681 cm
(LCL) and 9.9923 cm
(UCL)

Choosing an appropriate control chart


What type of data do you have?

Variable Attributes

Subgroup size
Defective Defects
P chart C chart
U chart
n>1 n=1 nP chart

I-MR chart

n < 9* n9
X bar chart X bar chart
R chart s chart

16
2/13/2019

How to use Control Charts?


• If a special cause variation occurs, the process should
be studied to determine the cause of the signal.
• If an assignable cause for the problem is identified,
appropriate action should be taken to correct it.
• The following graphs signal a POTENTIAL special cause
signal.
(note: all eight events apply to variable control charts.
Only 1, 4, 7 and 8 are applied to attributed control
charts).

Potential Special cause signal


1. An observation occurring more
than 3 units of std. dev. From the
center line, i.e. any points
occurring outside of the lower or
upper control limits.

2. Two out of three consecutive


observations (all on the same side
of the center line) occurring
more than 2 units of standard
deviation away from the center
line.

17
2/13/2019

Potential Special cause signal


3. Four out of five consecutive
observations (all on the same side
of the center line) occurring
more than 1 unit of standard
deviation away from the center
line.

4. Eight consecutive observations


occurring within 3 unit of
standard deviation from the
center line.

Potential Special cause signal


5. Fifteen consecutive
observations occurring within 1
unit of standard deviation from
the center line.

6. Eight consecutive observations


occurring that are more than 1
unit of standard deviation from
the center line.

18
2/13/2019

Potential Special cause signal


7. Fourteen consecutive
observations that alternate up
and down.

8. Six consecutive observations


that trend downward or upward.

Case study 1
E.g. The QC engineer in charge of a salt packaging
process is concerned about the moisture content in the
packages of salt. The primary concern is that variation in
the ambient humidity in the plant may be causing variation
in the mean moisture content in the packages over time.
Table below shows the data taken over a period of time.
Based on these data, establish appropriate control charts
of moisture content in this salt packaging process.

19
2/13/2019

20
2/13/2019

 At sample number 6 there is a signal since there is a point that


plotted outside of the control limits. This indicates that there is a
special cause operating and that the process variation is not in
control.
 The appropriate action is to determine the nature of the special
cause, and then delete the out-of-control sample and re-compute
the control limits.

Special cause
 A technician neglected to close a vent, causing greater than usual
variation in moisture content during the time period when the
sample was chosen.

Solution: re-training the technician.


Then, delete the data point due to special cause

21
2/13/2019

 With sample # 6 deleted, the value of R is 0.5836.

 With sample # 6 deleted, the value of


X is 2.658 .
 The sample size, n = 5.
 To re-plot X chart, from table A2 = 0.577.
UCL = 2.658 + (0.577)(0.5836) = 2.995;
LCL = 2.658 –(0.577)(0.5836) = 2.321

22
2/13/2019

• There is an out of control signal on the X-bar chart.


• When there is an out of control signal, the process must be stopped and
the special causes remedied.
• Even when a process is in control, the process must be continually
monitored, since new special causes are bound to crop up from time to
time and will need to be detected and corrected.

 Note that while control charts can detect the presence of a special cause,
they cannot determine its nature, nor how to correct it.
 It is necessary for the process engineer to have a good understanding of
the process, so that special causes detected by control charts can be
diagnosed and corrected.

23
2/13/2019

 The production installs a hygrometer to monitor the ambient


humidity and determines that the fluctuations in moisture
content are caused by fluctuations in ambient humidity.

 A dehumidifier is installed to stabilize the ambient humidity.

 After this special cause is remedied, more data are collected


and a new R-chart and X chart are constructed.

 The process is now in a state of statistical control.

24
2/13/2019

Summary of Steps for Control Chart


The steps in using the R (s) chart and X chart are
1. Choose rational subgroups.
2. Compute the R (s) chart.
3. Determine the special causes for any out-of-control
points.
4. Recompute the R (s) chart, omitting samples that
resulted in out-of-control points.
5. Once the R (s) chart indicates a state of control,
compute the X chart.
6. If the X chart indicates that the process is not in
control, identify and correct any special causes.
7. Continue to monitor X and R (s).

 Remove
sample #
6, then re-
calculate
and re-plot
the s-chart

25
2/13/2019

933 911 889 882 903 890 892 908 895 916
897 898 915 913 930 940 912 920 920 890
885 900 905 930 890 895 895 896 922 891
900 905 902 900 890 909 896 894 928 920
879 862 873 871 900 915 902 906 926 915

Max 933 911 915 930 930 940 912 920 928 920
Min 879 862 873 871 890 890 892 894 895 890
Range 54 49 42 59 40 50 20 26 33 30 R-bar 40.3
X-double
X-bar 898.8 895.2 896.8 899.2 902.6 909.8 899.4 904.8 918.2 906.4 bar 903.12

R-chart X bar-chart

1 (b) Describe the approach to be carried


out IF sample #9 is outlier.
(10%)

1. Determine the special causes for any


out-of-control points.
2. Recompute the R (s)/X bar- chart,
omitting samples that resulted in out-
of-control points.

26