Lean Six Sigma Process Improvement

DMAIC – Measure
(Tools & Methodologies)

LSCM 7344

Deepak Iyengar
 Process Yield of the Traditional Firm
990 pcs 985 pcs
1000 pcs A B C
10 rwk 15 rwk

970 pcs
30 rwk
950 pcs 960 pcs
F E D
50 rwk 40 rwk

980 pcs 20 rwk

Yield (Y) = 990 / 1000
= 99% 990 pcs G
10 rwk
 Six Sigma World

990 pcs 985 pcs

1000 pcs A B C
10 rwk 15 rwk

970 pcs 30 rwk

950 pcs 960 pcs

F E D
50 rwk 40 rwk

980 pcs
Rolled Throughput Yield (RTY) = 20 rwk

(990/1000)x(985/1000)x(970/1000)x G
(950/1000)x(960/1000)x(980/1000)x
990 pcs 10 rwk
(990/1000)= 0.837= 83.7%
The hidden factory increases cost and reduces throughput.
 Data Collection Plan
Bad Data = Bad Decisions = Bad Result

• Gather as many classification identification variables as

possible with the data
– Time, machine, auditor, operator, gauge, lab, material,
target, process change, conditions, shift, etc.
• If data occur in time sequence, record the order of data
capture
• For changes over time, record the before and after the
stabilization period
• Screen & filter data to remove data entry errors
 Establish the Baseline Capabilities for the
CTQs
• Need to arrange the data as per the time that
it has occurred (sequentially).
• Conduct appropriate analysis using
appropriate control charts.
• Use the “run test” to detect patterns if the
data looks like it is within its control limits.
Two types of measurements
• “Variable” measurements
– Something that can be measured. e.g. weight, volume,
time etc.
– More information is available
– Could be costly and time consuming to collect
– x-bar chart and R-chart, X-chart and MR-chart
• “Attribute” measurements
– Characteristics such as good/bad, acceptable/unacceptable
– Could be easier to collect
– Doesn’t give much information
– p-chart, c-chart, np-chart, and u-chart
 Statistical Process Control

• Variations and Control

– Random (natural/normal/expected) variation:
Natural variations in the output of process,
created by countless minor factors
– Assignable (abnormal/unexpected) variation: A
variation whose source can be identified
 Common Causes
Standard Deviation/
Mean Spread
n

x i  x  x 
i
2

x i 1 
n n 1
 Variable Measurements
Average

Change in
mean.

Mean
(a) Shift in Mean
 Variable Measurements
Average

Change in
deviation.

Mean
(b) Change in deviation
 Variable Measurements
Average

Change in
deviation.

Change in
mean.

Mean
(c) Shift in Mean and deviation
 Variable Measurements
• X-bar chart
– If data on σ is known

where

• = Standard deviation of distribution of sample means

• σ = Process standard deviation
• n = sample size
• z = Standard normal deviate
 Variable Measurements
• X-bar chart
– If data on σ is unknown

where

= Average of sample means

= Average of Range
 Variable Measurements
• R-chart

Where

= Mean of Range

D3 & D4 = A factor from Table C.4 page 678.

 Control Charts
for Variables
Special Metal Screw

Sample Sample _
Number 1 2 3 4 R x
1 0.5014 0.5022 0.5009 0.5027 0.0018 0.5018
2 0.5021 0.5041 0.5024 0.5020 0.0021 0.5027
3 0.5018 0.5026 0.5035 0.5023 0.0017 0.5026
4 0.5008 0.5034 0.5024 0.5015 0.0026 0.5020
5 0.5041 0.5056 0.5034 0.5047 0.0022 0.5045
R= 0.0021
=x = 0.5027
 Control Charts
for Variables
Control Charts—Special Metal Screw
x-Charts R = 0.0021 A2 = 0.729
= = x = 0.5027
CL = X
=
UCLx = x + A2R
=
LCLx = x - A2R
UCLx = 0.5027 + 0.729 (0.0021) = 0.5042 in.
LCLx = 0.5027 – 0.729 (0.0021) = 0.5012 in.
 Control Charts
Table 5.1 Control Chart Factors
for Variables
Factor for UCL Factor for Factor
Size of and LCL for LCL for UCL for
Control
Sample
Charts—Special
x-Charts
Metal Screw
R-Charts R-Charts
(n) (A2R
) = 0.0021 (D3) (D4)
X-Charts x== 0.5027
2 1.880 0 3.267
3
4 x=x+A
= 1.023 0 2.575
UCL 2R
0.729 0 2.282
5 = 0.577 0 2.115
LCL6 x = x - A0.483
2R 0 2.004
7 0.419 0.076 1.924
8 0.373 0.136 1.864
9 0.337 0.184 1.816
10 0.308 0.223 1.777
 x-Chart—
Special Metal Screw
 Mean and Range Charts

(process mean is
shifting upward)
Sampling
Distribution

UCL

x-Chart Detects shift

LCL
UCL

Does not
R-chart
detect shift
LCL
 Mean and Range Charts

Sampling
Distribution (process variability is increasing)

UCL

Does not
x-Chart
reveal increase
LCL
UCL

R-chart Reveals increase

LCL
 Control Charts
for Variables
Control Charts—Special Metal Screw
R-Charts R = 0.0021 D4 = 2.282
D3 = 0

CL = R
UCLR = D4R
LCLR = D3R
UCLR = 2.282 (0.0021) = 0.00479 in.
LCLR = 0 (0.0021) = 0 in.
 Control Charts
Table 5.1 Control Chart Factors
for Variables
Factor for UCL Factor for Factor
Size of and LCL for LCL for UCL for
Sample x-Charts R-Charts R-Charts
(n) (A2) (D3) (D4)
2 1.880 0 3.267
3 1.023 0 2.575
4 0.729 0 2.282
5 0.577 0 2.115
6 0.483 0 2.004
7 0.419 0.076 1.924
8 0.373 0.136 1.864
9 0.337 0.184 1.816
10 0.308 0.223 1.777
 Range Chart -
Special Metal Screw
 Variable Measurements
• X chart
__
C.L. = X
__ __ __ __
UCL = X + E2R LCL = X - E2R
Where

_
X= Average of sample means

= Average of Range

E2 = A factor from Table C.4 page 678.

 Variable Measurements
• MR chart
__
C.L. = R __
UCL = D4R__
LCL = D3R
Where

= Average of Range

D3 & D4 = A factor from Table C.4 page 678.

 Attribute Measurements

• p-chart
p = mean of proportion
CL = p of defectives.

UCLp = p + zp

LCLp = p – zp

p = p(1 – p)/n
 Attribute Measurements

• c-chart
c = count of occurrences.
CL = c
UCLp = c + z c

LCLp = c –z c
 Attribute Measurements

• np-chart _
np = counts of defectives.
CL = np
UCL = np + 3 np(1-p)

LCLp = np - 3 np(1-p)
 Control Charts
for Attributes
Hometown Bank

UCLp = p + zp
LCLp = p – zp

p = p(1 – p)/n
 Control Charts
Sample
Number
Wrong
Account for Attributes
Number
1 15 Hometown Bank
2 12
3 19 n = 2500
4 2
UCL = p + z
5
6 p
19
4 p Total defectives
7 24 p= Total observations
LCL = p - z
8
p
7
p
9 10
10 17
11 15
12
p = p(1 - p)/n
3
Total 147
 Control Charts
Sample
Number
Wrong
Account Number for Attributes
1 15
2 12 Hometown Bank
3 19
4 2 n = 2500
5 19
UCLp = p + zp
6 4
7 24 147
8 7 p=
LCLp = p - zp
9 10 12(2500)
10 17
11 15
12 3
p = p(1 - p)/n
Total 147
 Control Charts
Sample
Number
Wrong
for Attributes
Proportion
Account Number Defective
1 15 0.006
2 12 Hometown Bank
0.0048
3 19 0.0076
4 2 0.0008 n = 2500
5 19 0.0076
UCLp = p + zp
6
7
4
24
0.0016
0.0096
8 7 0.0028 p = 0.0049
LCLp = p - zp
9 10 0.004
10 17 0.0068
11 15 0.006
12 3 0.0012
p = p(1 - p)/n
Total 147
 Control Charts
for Attributes
Hometown Bank
n = 2500 p = 0.0049

UCLp = p + zp
LCLp = p – zp

p = p(1 – p)/n
 Control Charts
for Attributes
Hometown Bank
n = 2500 p = 0.0049

UCLp = 0.0091
LCLp = 0.0007

p = 0.0014
 p-Chart
Wrong Account Numbers
 Run Tests

• Run test – a test for randomness

• Any sort of pattern in the data would
suggest a non-random process
• All points are within the control limits - the
process may not be random
Nonrandom Patterns in Control charts

• Trend
• Cycles
• Bias
• Mean shift
• Too much dispersion
 Counting Runs
Counting Above/Below Median Runs (MINITAB : Clustering)

7 Runs

B A A B A B B B A A B

unting Up/Down Runs (MINITAB: TRENDS)

8 Runs

U U D U D U D U U D
 Computing z values

Expected no. of runs

CLUSTERING TRENDS

Std. Dev. of runs

 Process Capability
Nominal
value

Process distribution
Lower Upper
specification specification

20 25 30 Minutes

(a) Process is capable

 Process Capability
Nominal
value

Process distribution
Lower Upper
specification specification

20 25 30 Minutes

(b) Process is not capable

 Process Capability
Lower Upper
Specification Specification

Process variability matches

specifications
Lower Upper
Specification Specification

Process variability well within

Lower Upper
specifications Specification Specification

Process variability exceeds

specifications
 Process Capability
Process Capability Ratio
Upper specification - Lower specification
Cp = 6s

Process Capability Index

= x – Lower specification
Cpk = Minimum of 3s ,
Upper specification – x =
3s
 Process Capability
Lower Upper
Cp =1
Specification Specification Cpk =1

Process variability matches

specifications
Lower Upper
Specification Specification

Cp >1
Cpk >1 Process variability well within
Lower Upper
specifications Specification Specification

Cp <1
<1 Process variability exceeds
Cpk
specifications
 Process Capability
Intensive Care Lab
Upper specification = 30 minutes
Lower specification = 20 minutes
Average service = 26.2 minutes s = 1.35 minutes
 Process Capability
Intensive Care Lab
Upper specification = 30 minutes
Lower specification = 20 minutes
Average service = 26.2 minutes s = 1.35 minutes

Cpk = 0.94 Cp = 1.23

Process Process
Capability Capability
Index Ratio
Validity of the Measurement System (Accuracy)

• Bias : Difference between average measured

value and a reference value (standard)
• Stability : Total variation in the measurements
obtained with a measurement system on the
same parts when measuring a single
characteristic over an extended period of time
• Linearity : Difference in bias values through
the expected operating range of the gage
Validity of the Measurement System (Precision)
• Repeatability : Variations in measurements obtained
by one measurement instrument when used several
times by one appraiser, while measuring the identical
characteristic on the same part
• Reproducibility : Variations in measurements made
by different appraisers using the same measuring
instrument when measuring the identical
characteristic on the same part
 Bias
• Corrected through calibration

Bias = Average – Reference Value

Bias % = (|Bias| / tolerance) x 100

 Stability

• Use control charts (X-chart MR- chart or X-bar

chart R-Chart to check for the presence of
assignable causes)
 Repeatability & Reproducibility
• Construct a X-bar chart and a R-chart

R-chart should be under control and X-bar chart

should be out of control. If the averages were
within the control limits, it would mean that
part-to-part variation was less than the
variation due to gage repeatability error.
 R & R variation

• If R&R variation < 10% of observed total

variation, measurement system is acceptable
• If 10% <= R&R variation < 30% of total
variation, measurement system is borderline
acceptable
• If R&R variation >= 30% of total variation,
measurement system is unacceptable