Introduction to SPC
• Review of normal (Gaussian) distributions.
(VERY IMPORTANT for SPC)
• For a certain population of people, the
average male height is 68 inches, with a
standard deviation of 3 inches.
• What is the probability that the next male
from this population has a height greater
than 70 inches? (what information here is
missing?)
SPC
Normal Review (continued)
• What is the probability that the average height of
the next nine males is greater than 70 inches?
• Central Limit Theorem (see next slide)
• If the mean of the population shifted to 72 inches,
what is the probability that the next male is less
than 70 inches (assume normal distribution)?
• If a sample of nine came from the new population,
what is the probability the average of the sample
will be less than 70 inches?
SPC
Central Limit Theorem:
Example of the Roll of the Dice
0
5
10
15
20
25
30
35
40
1 2 3 4 5 6
0
5
10
15
20
25
30
35
40
1 1 2 2 3 3 4 4 5 5 6
0
5
10
15
20
25
30
123456
Theoretical Frequency Distribution based on 216 rolls...
One Die
µ=3.5, o=1.87
Average of
Two Dice
µ=3.5, o=1.23
Average of
Three Dice
µ=3.5, o=0.99
SPC
Statistical Process Control
Customer Supplier
•Sample output of process and
make inferences about its state
•Demonstrate that the
distribution of process output
is known and unchanging
•Plot and monitor over time
•Use statistical tests to detect shifts
and anomalies and react to them quickly
•Use statistical evidence to guide and
confirm process improvements
SPC
Evolution from Inspection to SPC
Customer Supplier
100%
Insp.
100%
Insp.
SPC.
Customer Supplier
100%
Insp.
Sample
Insp.
Customer Supplier
Sample
Insp.
Customer Supplier
SPC.
SPC.
SPC
Statistical Process Control Topics
• Introduction to Variability
• Control Charts
– General info
– xbar Charts, R Charts
– p Charts, np Charts
– Type 1 and Type 2 Errors
• Process Capability Analysis
SPC
Variation
Measure
Performance
Analyze
and Act
Improve
Capabilities
“The less variation, the
better off we are.”
Common cause variation:
Inherent in the system
Assignable cause variation:
Eventrelated, special
(assignable ~ special)
SPC
Analyze and Act:
React to Assignable Causes
• Note unusual variation diagnosed by using a
common test to evaluate individual data points
• Identify cause by noting what change in the process
occurred at that point in time
• Eliminate cause or build in the cause
• Monitor performance to verify the effect of the ―fix‖
• Generally, assignable causes cause points outside of
control limits!
SPC
Improve Process Capabilities:
Drive Out Common Causes
• Variation is inherent in the system
• Don’t react to individual points (this is tampering)
• Analyze possible factors affecting variation (use
Cause and Effect Diagram, Pareto Analysis)
• Work to reduce variation: Make an improvement,
that is, introduce a special cause
• Monitor performance to verify the effect of the
intended improvement
SPC
Xbar and R charts
Customer Supplier
•Sample output of process  parameter
of interest is continuously variable
•Plot one chart to track sample means and
another one to track sample ranges (variation)
•Use statistical evidence to detect changes
and improve the process: to better position
the mean and to reduce variation
SPC
Underlying Assumptions
• process mean µ and standard deviation o when the process
is in control
• process may go out of control in two possible ways
– mean shifts to µ
1
, with standard deviation unchanged
– standard deviation shifts to o
1
, with mean unchanged
• sample means are normally distributed (when in or out of
control, because either:
• process output measurements on individual units are
normally distributed when in or out of control
• OR Central Limit Theorem applies:
– n > 30 OR
– If distribution unimodal or symmetric, then much smaller n’s
are acceptable to assume normality (n on the order of 4).
SPC
Basic Probabilities Concerning the
Distribution of Sample Means
{ } 0027 . 0 3 3 Prob = ÷ < + > x x x or x o µ o µ
{ } 0006 . 0 4 4 Prob = ÷ < + > x x x or x o µ o µ
n x / o o =
Std. dev. of the sample means:
SPC
Estimation of Mean and Std. Dev.
of the Underlying Process
• use historical data taken from the process when it was
―known‖ to be in control
• usually data is in the form of samples (preferably with
fixed sample size) taken at regular intervals
• process mean µ estimated as the average of the sample
means (the grand mean)
• process standard deviation o estimated by:
• standard deviation of all individual samples
• OR mean of sample range R/d
2
, where
sample range R = max. in sample minus min. in sample
and d
2
= value from lookup table (appendix A7)
SPC
Example: Estimation of Mean and Std.
Dev. of the Underlying Process
Output Sample Sample
Measurements Mean Range
1 2 2 3 4 2 2.6 2
2 2 4 2 1 2 2.2 3
3 1 2 4 1 2 2.0 3
4 2 2 3 4 2 2.6 2
5 2 1 2 3 1 1.8 2
6 1 5 2 2 3 2.6 4
Average: µ=2.3 R=2.7
Sample
Estimate of the process mean = µ = 2.3
Estimate of the process std. dev.:
(1) Combined std. dev. of all 30 points: o = 1.1
OR (2) o = R/d
2
(n=5) = 2.7/2.326 = 1.2
SPC
Determination of Control Limits
For the xbar chart:
 Center Line = grand mean
 Control Limits: Co's usually use
 Can analyze process capability based on the specification limits
For the R chart:
 Center Line = average range =
 Control Limits:
Alternative: Use an Economic Approach:
 Consider the cost impact of outofcontrol detection delay (Type
2 error), false alarm (Type 1 error) and sampling costs
 Difficult to estimate costs
n x / 3 3 o µ o µ ± = ±
R
R D LCL and R D UCL 3 4 = =
SPC
Ex. Two Machines Process Capability
Analysis and xbar and Rcharts
SPC
Xbar vs. R charts
• R charts monitor variability: Is the
variability of the process stable over time?
Do the items come from one distribution?
• Xbar charts monitor centering (once the R
chart is in control): Is the mean stable over
time?
>> Bring the Rchart under control, then look
at the xbar chart
SPC
How to Construct a Control Chart
1. Take samples and measure them.
2. For each subgroup, calculate the sample average
and range.
3. Set trial center line and control limits.
4. Plot the R chart. Remove outofcontrol points
and revise control limits.
5. Plot xbar chart. Remove outofcontrol points
and revise control limits.
6. Implement  sample and plot points at standard
intervals. Monitor the chart.
SPC
Xbar and R chart example:
• Look at handout: R Chart.
• Rbar = sum( R )/num. samples = 87/25 = 3.48.
• UCL = D
4
Rbar = 2.114*3.48 = 7.357.
• LCL = D
3
Rbar = 0
• Review samples, eliminate sample 3.
• Do over! New Rbar = 3.29, UCL = 6.95
• Rbar chart now in control, proceed to Xbar!
SPC
Xbar chart
• Grand mean, Xbar = 500.6/24 = 20.86.
• Control limits = 20.86 +/ A2Rbar = 20.86
+ (.5777)*3.29
• UCL = 20.86 + 1.9 = 22.76
• LCL = 20.86 – 1.9 = 18.96
• Bring Xbar chart under control—eliminate
points 15, 22, 23.
SPC
Conclusion of problem
Redo R chart without samples 15, 22, and 23 (and
3 is out as well).
Rbar = 3.24
Control limits (repeat previous procedure = [0,
6.845].
Grand mean (center line for xbar) = 20.77
Control limits = 20.77 +/ (.5777)(3.24) = [18.90,
22.64]
Our control limits for both charts are now set.
SPC
Type 1 and Type 2 Error
Type 1 Error No Error
No Error Type 2 Error
Alarm No Alarm
InControl
OutofControl
SPC
Common Tests to Determine if the
Process is Out of Control
• One point outside of either control limit
• 2 out of 3 points beyond UCL  2 sigma
• 7 successive points on same side of the central line
• of 11 successive points, at least 10 on the same
side of the central line
• of 20 successive points, at least 16 on the same
side of the central line
SPC
Type 1 Errors for these Tests
Test Probability Type 1 Error
2/3
7/7
10/11
16/20
1/1 2(0.00135)
0.0027
0.00052
(0.5)
7
0.0078
11
) 5 . 0 (
11
11
) 5 . 0 (
10
) 5 . 0 (
10
11


.

\

+


.

\

0.00586
i i
i
÷


.

\

¿
20
) 5 . 0 ( ) 5 . 0 (
20
16
20
0.0059
3
) 0228 . 0 ( ) 9772 . 0 (
2
) 0228 . 0 (
2
3
+


.

\

SPC
Type 2 Error
Suppose µ
1
> µ
Type 2 Error =
{ } 1 x 3 x Prob µ µ o µ = + s
( )   x x o µ o µ / 3 1 ÷ + u =
[This is the probability of a sample average being below
the upper control limit. We have not examined
possibility of being below LCL, why?]
• Power = 1 Type 2 Error. Power increases as …
n increases, as (µ
1
÷µ) increases, and as o decreases.
• Extension to µ
1
< µ is straightforward
SPC
Sensitivity of Type I and Type II Errors
• To (UCLLCL)/o
• To n
• To o
• To µ
1
 µ
SPC
Example of Type 1 and 2 Errors
Suppose: µ = 100
µ
1
= 102
o = 4
n = 9
[assume 3 sigma control limits]
Find Type 2 Error:
SPC
Example of Type 1 and 2 Errors (cont.)
Type 2 Error = Type 1 Error =
Prob{Shift detected in third sample after shift occurred}
=
Average number of samples taken before the shift is detected
=
Prob{no false alarm for first 32 samples, but then
false alarm occurs in 33rd sample}
Average number of samples before a false alarm (ARL)