MSA STUDY training presentation for manufacturing

MSA study analysis training for variable data and attribute data

1. 1Subhodeep Krishna Deb
Subhodeep Krishna Deb
Module 2

2. 2Subhodeep Krishna Deb
MSA definition
MSA Parameters
MSA Tools & Methods
MSA Plan & Execution
Interaction points

3. 3Subhodeep Krishna Deb
3
For Variables
1.Bias study
2.Linearity study
3.Stability study
4.R & R (Range Method)
5.GRR (X bar -R Method)
6.ANOVA
MSA Methods
For Attributes
•Signal Detection
•Cross Tabulation (Kappa)
•Visual inspection study
•Gauge Performance Curve

4. 4Subhodeep Krishna Deb
4
Steps of performing MSA
Select 3 appraisers. Arrange 10 samples.
1
st
appraiser to take
readings of all 10
samples.
2
nd
appraiser then
takes reading of 10
samples.
3
rd
appraiser next
takes readings of 10
samples.
Follow steps 3 –5, 3
times; total 90
reading to populate
the data table.
Calculate EV and AV.
Calculate GRR wrt
Total variation (TV)
and Tolerance (Tol)
both.
Plot Average graph
and Range graph

5. 5Subhodeep Krishna Deb
5
Steps for calculating MSA resultsUSL: LSL: Tol: 0.0000
1 2 3 4 5 6 7 8 9 10
1
2
3
Average#DIV/0!#DIV/0!#DIV/0!#DIV/0! #DIV/0!#DIV/0!#DIV/0!#DIV/0!#DIV/0!#DIV/0! X
a= #DIV/0!
Range 0.0000 0.0000 0.0000 0.0000 0.00000.0000 0.00000.00000.00000.0000 R
a= 0.0000
1
2
3
Average #DIV/0!#DIV/0!#DIV/0!#DIV/0! #DIV/0!#DIV/0!#DIV/0!#DIV/0!#DIV/0!#DIV/0! X
b = #DIV/0!
Range 0.0000 0.0000 0.0000 0.0000 0.00000.0000 0.00000.00000.00000.0000 R
b= 0.0000
1
2
3
Average #DIV/0!#DIV/0!#DIV/0!#DIV/0! #DIV/0!#DIV/0!#DIV/0!#DIV/0!#DIV/0!#DIV/0! X
c= #DIV/0!
Range 0.0000 0.0000 0.0000 0.0000 0.00000.0000 0.00000.00000.00000.0000 R
c= 0.0000
Specification: Gauge ID: Date:
APPRAISER TRIAL
PART
AVERAGE
Appraiser 1
Appraiser 2
Appraiser 3 X
p= #DIV/0!
R
p= #DIV/0!
#DIV/0!#DIV/0!#DIV/0! #DIV/0!#DIV/0!#DIV/0!#DIV/0!#DIV/0!#DIV/0!
Average of Range =(R
a + R
b + R
c / No. of Appraiser) R =
Part Average #DIV/0!
0.0000
Range of Average =(Max. X - Min. X) X
DIFF = #DIV/0!
Upper Control Lim it for Range Chart UCLr = R x D4 0.0000

6. 6Subhodeep Krishna DebFrom Data
Sheet R = 0.0083 Xdiff = 0.0024 Rp= 0.1434
Repeatability (Euipment Variation) % Repeatability
EV = R x K1
Trials K
1 % EV = 100 [ EV / TV ]= 10.8434
= 0.0049 2.0000 0.8862
3.0000 0.5908 % EV = 100x[EV/(TOL/6)]= 7.3850
K
1 = 1/d
2* (at m = no of trials, g = no of part x no of appraisers = greater than 15)
Reproducibility (Appraiser Variation) % Reproducibility
AV = [( Xdiff x K2 )
2
- (EV
2
/ nr)] % AV = 100 [ AV / TV]= 1.9303
= 0.0009Change value of K2 as per % AV = 100x[AV/(TOL/6)]= 1.315
your case
n = No of Parts =10.0000 2.0000 3.0000
r = No of Trials =3.0000 0.7071 0.5231
K
2 = 1/d
2* (at m = no of appraisers, g = 1)
Repeatability & Reproducibility (R & R) % Repeatability & Reproducibility (R & R)
GR & R =(EV
2
+ AV
2
) % R & R = 100 [ R&R / TV ] =11.0139
=0.0050 % R & R = 100 x GR&R/(TOL/6) = 7.50
Part Variation ( PV ) % Part Variation ( PV )
Parts K3
PV =Rp x K3
2.0000 0.7071 % PV = 100 [ PV / TV ] =99.3916
3.0000 0.5231
=0.0451 4.0000 0.4467 % PV = 100 x PV/(TOL/6) = 67.69
5.0000 0.4030
Total Variation ( TV ) 6.0000 0.3742 % TV = 100 x TV/(TOL/6) = 68.11
7.0000 0.3534
TV =GRR
2
+ PV
2
8.0000 0.3375 Number of distinct Data Categories
9.0000 0.3249 ndc = 1.41 [ PV / R&R ]
=0.0454 10.0000 0.3146 = 12.7241 Data Categories
Appraisers
K
2
Steps for calculating MSA results

7. 7Subhodeep Krishna Deb
1.More than 50% points in Average graph should be outsidecontrol limits.
2.Averagegraph reflects the Measurement Capability of the Measurement
System.
3.Rangegraph reflects the Measurement Consistency of the Measurement
System.
4.All points in rangegraph should remain withincontrol limit.
5.NDC reflects no of discrete categories permissible in Measurement System.
Steps for interpreting MSA results

8. 8Subhodeep Krishna Deb1. AVERAGE CHART
X Bar 1 2 3 4 5 6 7 8 9 10
D Sarkar 40.84340.84040.84740.85340.85040.86340.87040.86040.85040.880
S K Sinha 40.84040.84040.84040.85040.85040.87040.87040.86040.86040.870 X
p= X dbar
D Rakshit40.85040.83740.84040.85040.85040.86740.87040.86040.85040.880 40.855
UCL 40.86040.86040.86040.86040.86040.86040.86040.86040.86040.860
LCL 40.85040.85040.85040.85040.85040.85040.85040.85040.85040.850
A2 = 1.023 for n=3, 1.88 for n=2 (n=number of appraisers) R bar
Range of average0.010 0.0030.007 0.003 0.0000.007 0.000 0.0000.010 0.010 0.005
40.810
40.820
40.830
40.840
40.850
40.860
40.870
40.880
40.890
1 2 3 4 5 6 7 8 9 10
Average Chart D Sarkar S K Sinha D Rakshit UCL LCL
Steps for interpreting MSA results

9. 9Subhodeep Krishna Deb
GRR –Average Chart Interpretations
Condition Interpretation Action
Less than 50%
of readings are
out of control
limits
Measurement system is not
adequate enough to capture
process variation, or
Parts does not represent expected
process variation
Improve discrimination of the
measurement system, or
Select parts representing
entire process variation
In this example, 22 out of 30 points are
outside the control limit
Since this is more than half of total
points, the conclusion is that the
measurement system is adequate to
detect part-to-part variations
More than 50% points outside control limit indicates that MS variation is much
smaller as compared to part variations, hence MS is capable of detecting part-to-
part variations.

10. 10Subhodeep Krishna Deb2. RANGE CHART
Range 1 2 3 4 5 6 7 8 9 10
D Sarkar 0.010 0.0000.010 0.010 0.0000.010 0.000 0.0000.000 0.000
S K Sinha 0.000 0.0000.000 0.000 0.0000.000 0.000 0.0000.000 0.000
D Rakshit 0.000 0.0100.000 0.000 0.0000.010 0.000 0.0000.000 0.000
UCL 0.005 0.0050.005 0.005 0.0050.005 0.005 0.0050.005 0.005
0.000
0.002
0.004
0.006
0.008
0.010
0.012
1 2 3 4 5 6 7 8 9 10
Range Chart D Sarkar S K Sinha D Rakshit UCL
Steps for interpreting MSA results
Purpose of range chart is to identify whether measurement process
is under control (free from special cause)

11. 11Subhodeep Krishna Deb
Condition Interpretation Action
1 or more point of
any appraiser out of
UCLr
There was a special
cause while taking
reading
Remove these readings and take
same reading again from same part
& appraiser and recalculate
GRR –Range Chart Interpretations
More than 1 point of
only one appraiser
out of UCLr
His method is
different from others
Remove these readings. Train
appraiser on method of
measurement & take readings again
1 or more than one
point of all appraisers
out of UCLr
Measurement System
is sensitive to
appraisers skill
Check why measurement is so
sensitive stop further studies before
taking action on sensitivity
In one part all
appraisers points are
out of UCLr
Part is deformed or
Damaged
Remove all the reading for particular
part and recalculate or replace the
part with new part & take readings

12. 12Subhodeep Krishna Deb
GRR-Acceptance Guideline
% R&R Value Decision
< 10% of TOL or TVGauge is Capable
10 –30% of TOL or TVAcceptable subject to analysis & justification w.r.t.
application, cost of repair & criticality
> 30% of TOL or TVMeasurement system need improvement / corrective
action
Number of distinct data categories should also be checked when doing SPC,
ndc = 1.41 [PV / R&R] > 5 (best = 10)
(this means R&R should always be less that 28% of PV)
< 2, inadequate to provide data for study
= 2, it is equivalent to a go/nogo gauge

13. 13Subhodeep Krishna Deb
Product Control
Measurement is used for
deciding product
acceptance/rejection
TOL/6
Process Control
Measurement is used to
find variation in parts
(variation in process,
identifying special cause,
SPC application)
TV
or
Process Standard
Deviation (If process
variation is known)
GRR % through TV or TOL/6

14. 14Subhodeep Krishna Deb
When Repeatability > Reproducibility
Instrument needs maintenance
Redesign gage for more rigidity
Improve clamping or location of gauging
Excessive within-part variation
Identify the right cause & solution
When Reproducibility > Repeatability
Appraisers needs training on better way of using the gauge
Needs better operational definition
Incremental divisions on instrument are not readable
Need fixture to provide consistency in gauge use.

15. 15Subhodeep Krishna Deb
Relation of R & R with Cp, Cpk
Actual Cp/Cpk is always more than Observed Cp/Cpk

16. 16Subhodeep Krishna Deb
Data
Qualitative
Quantitative
Attribute
Variable
MSA Methods
For Attributes
When MSA to be done for a Gauge
which is used for inspection of a
variable parameter
Snap Gauge, Limit Gauge, Any
special gauge, etc
Signal Detection
method
When MSA to be done for
Inspection method of a parameter
which cannot be measured
Dent, Crack, Fouling, etc
Visual inspection
method
Kappa method

17. 17Subhodeep Krishna Deb
17
For Variables
1.Bias study
2.Linearity study
3.Stability study
4.R & R (Range Method)
5.GRR (X bar -R Method)
6.ANOVA
MSA Methods
For Attributes
•Signal Detection
•Cross Tabulation (Kappa)
•Visual inspection study
•Gauge Performance Curve

18. 18Subhodeep Krishna Deb
18
Steps of performing MSA
Select 3 appraisers.
Take 50 parts randomly
from the process covering
the entire variation (ensure
at least 20% of the parts
are defective).
Measure all the samples
with correct measuring
instrument to collect the
reference data for each
sample.
Conduct trial runs as per
nomenclature of 3
appraisers; perform 3 trials
for each operator.
Note the readings in table.
Populate the table with
450 data in total and
perform calculations as per
below method.

19. 19Subhodeep Krishna Deb
LSL USL
III IIIIII
I.Bad product is always
rejected
II.Gray area, Some time
good is called bad and
bad is called good
III.Good product is always
accepted
Reducing variation is goal of SPC
Reducing variation
is goal of MSA
•Signal Detection is used for both Repeatability & Reproducibility
Steps for calculating MSA results

20. 20Subhodeep Krishna Deb
Steps for calculating MSA results
1.After collection of all the data assign a code for each row i.e. considering 9
observations of each row(3 of each appraiser) as per below nomenclature:
2.Arrange the reference value & codes in descending orders
3.Calculate top width d1 (distance between last part accepted by all the
appraisers to the first part rejected by all (for all specifications)
4.Calculate top width d2 (distance between last part accepted by all the
appraisers to the first part rejected by all (for all specifications)
5.Calculate average width AW = d1+d2/2
6.Calculate % R & R = (AW/Tolerance)*100 +When the row has only 1
-When the row has only 0
xWhen the row has both 1 and 0

21. 21Subhodeep Krishna Deb
Reference value
Varies from : 0.400
to 0.600
+When the row has only 1
-When the row has only 0
XWhen the row has both 1 & 0
Tolerance 0.1
Top width = d1 = 0.566152 -
0.542704
0.023448
Bottom width = d2 = 0.470832 -
0.446697
0.024135
Average width = AW = (d1+d2)/20.023792
% R&R = (AW/Tolerance)X100 23.79%
Decision Remarks
0 called 0 Right Decision,
Effective MS1 called 1
0 called 1Miss Alarm, Consumer’s risk, b
1 called 0False Alarm, Producer’s risk, a
< 10 %Acceptable
< 30 %Conditionally Acceptable
> 30 %Needs improvement
Steps for interpreting MSA results

22. 22Subhodeep Krishna Deb
22
For Variables
1.Bias study
2.Linearity study
3.Stability study
4.R & R (Range Method)
5.GRR (X bar -R Method)
6.ANOVA
MSA Methods
For Attributes
•Signal Detection
•Cross Tabulation (Kappa)
•Visual inspection study
•Gauge Performance Curve

23. 23Subhodeep Krishna Deb
23
Steps of performing MSA
1.Take 50 parts randomly from the process covering the entire variation (ensure
at least 20% of the parts are defective
2.Get measurements on the parts by the operators as 1 (for OK) and 0 (for NOT
OK) decisions
3.Similar to X bar R method perform three trial runs for each set of samples for
each operator. Total 450 data to populate the table.
4.Get these parts measured / decided by a MASTER (an experienced) person for
the results to be used as REFERENCE
5.Compareeach trial of each inspector with the another inspector for their
decision
6.Complete the cross tabulation table, shown in next slide
7.Calculate Kappafor
-A vs B, A vs C, B vs C
-A vs Ref, B vs Ref, C vs Ref

24. 24Subhodeep Krishna Deb
There are 34 times where
A-1 = 1 and B-1 = 1
(that is, of the 50 parts
checked there were 34
matches by A andB on their
FIRST check)
Steps for calculating MSA results

25. 25Subhodeep Krishna Deb
There are 32 times where
A-2 = 1 and B-2 = 1
(that is, of the 50 parts checked
there were 32 matches by A and
B on their SECOND check)
Steps for calculating MSA results

26. 26Subhodeep Krishna Deb
Total : where A-x = 1 and B-x = 1
= 34+32+31 = 97
There are 31 times where
A-3 = 1 and B-3 = 1
(that is, of the 50 parts checked
there were 31 matches by A and
B on their THIRD check)
Steps for calculating MSA results

27. 27Subhodeep Krishna Deb
Count & Expected Count
A*B Cross Tabulation
B
Total
0 1
A
0
Count 44650
Expected Count15.734.350
1
Count 397100
Expected Count31.368.7100
Total
Count 47103150
Expected Count47103150
Expected Count =
(Column Total x Row Total )
/ Grand Total
For A=1 & B=1
Column Total= 103
Row Total = 100
Grand Total= 150
Hence,
Expected count=
(103 x 100)/150= 68.7
Steps for calculating MSA results

28. 28Subhodeep Krishna Deb
A*B Cross Tabulation
B
Total
0 1
A
0
Count 446 50
Expected Count15.734.350
1
Count 397100
Expected Count31.368.7100
Total
Count 47103150
Expected Count47103150
Where
p
o= sum of observed proportions in the
diagonal cells (left to right direction)
p
e= sum of expected proportions in the
diagonal cells (left to right direction)
Steps for calculating MSA results

29. 31Subhodeep Krishna Deb
EffectivenessDecision
More than 90 %Acceptable for the appraiser
More than 80%Marginally acceptable for the appraiser
Less than 80 %Unacceptable for the appraiser-Need improvement
Miss Rate (Consumer’s Risk) Max 2 %
False Alarm rate (Producer’s Risk) Max 5 %
Steps for interpreting MSA results

30. 32Subhodeep Krishna Deb
32
For Variables
1.Bias study
2.Linearity study
3.Stability study
4.R & R (Range Method)
5.GRR (X bar -R Method)
6.ANOVA
MSA Methods
For Attributes
•Signal Detection
•Cross Tabulation (Kappa)
•Visual inspection study
•Gauge Performance Curve

31. 33Subhodeep Krishna Deb
•100% subjective inspection is not 100% effective
•200% inspection is less effective than 100% (no ownership, conflict, multiply
individual effectiveness)
•Rate of improvement noticed will be less than actual improvement for subjective
inspections
•There are more chances of mismatch in acceptance criteria between customer &
supplier
Limitations:

32. 34Subhodeep Krishna Deb
1.Collect min 20 samples covering good, bad (include marginal one which is part of process)
2.Decide the reference value-It should be inline with internal / external customer
requirement. Team should come with common consensus on reference value
3.Identify the parts with numbers
4.Ask a operator who is regularly checking these product to separate good and bad parts
5.Record his decision about every part as good and bad
6.Repeat step 4 and 5 with 2-3 operators for at 2-3 times
7.Calculate Effectiveness of inspection, miss & false alarm
8.Decide whether measurement system is accepted
Steps of performing MSA

33. 35Subhodeep Krishna Deb
Srl
APPRAISER:A APPRAISER:B APPRAISER:C
Trials
1
Trials
2
Trials
3
Trails
1
Trails
2
Trails
3
Trials
1
Trials
2
Trials
3
1 G G G G G G G G G
2 G G B G G G G G G
3 G B G G G B G G G
4 B B B B B B B B B
5 B B B B B B B B B
6 G G G G G G G G G
7 G G G G G G G G G
8 G B G G B B G G B
9 B B B B B B B B B
10 G G G G G G G G G
11 G G G G G G G G G
12 B G G G G G G G G
13 B B B B B B G G B
14 B B B B B B B G B
15 B B B B B B B B B
16 G G G G G G G G G
17 G G G G G G G G G
18 B B B B B B B B B
19 G G G G G G G G G
20 G G G G G G G G G
G=Good; B=Bad
Reference
G B
Appraiser
G
Correct
Decision
Miss
Alarm
B
False
Alarm
Correct
Decision
Steps for calculating MSA results

34. 36Subhodeep Krishna Deb
Ref
G B
A
G 33 2
B 3 22
Number of samples (N)=20 Number of Good samples (NG)=12
Number of trials (R)=3 Number of BAD samples (NB)=8
Srl DESCRIPTION A B C
1Number of miss alarm (Nm) 02 01 05
2Number of false alarm (Nf) 03 01 02
3Effectiveness of inspection = No of good decisions / Total55/60=0.92 58/60=0.97 53/60=0.88
4
Probability of miss P(MISS) = No. miss / No. of opportunities
= Nm / (NBxR)
02/(8x3)=0.08 01/(8x3)=0.0405/(8x3)=0.20
5
Probability of false alarm P(FA) = No of false alarm / No of
Opportunities for false alarm = Nf/ (NGxR)
03/(12x3)=0.08 01/(12x3)=0.0302/(12x3)=0.06
Ref
G B
B
G 35 1
B 1 23
Ref
G B
C
G 34 5
B 2 19
Miss Rate (Consumer’s Risk) Max 2 %
False Alarm rate (Producer’s Risk) Max 5 %
Steps for calculating MSA results

35. 37Subhodeep Krishna Deb
Effectiveness (E)
> 0.9: accepted
0.8-0.9 : conditionally accepted
< 0.8: unacceptable
Probability of False alarm P (FA)
< 0.05 : Accepted
0.05 -0.1: Conditionally accepted
> 0.1: Unacceptable
Probability of miss P (MISS)
< 0.02 : Accepted
0.02 -0.05: Conditionally accepted
> 0.05 : Unacceptable
Note: This the thumb rule.
Organization & customer has
To decide who much risk is
Acceptable considering the
Importance of the parameter
Steps for interpreting MSA results

36. 38Subhodeep Krishna Deb
MSA definition
MSA Parameters
MSA Tools & Methods
MSA Plan & Execution
Interaction points

37. 39Subhodeep Krishna Deb
Steps to be followed to IMPLEMENT MSA
Plan
•Prepare complete gauge list.
•Categorize all gauges to Major Gauge Groups which need to be covered in given time line.
•Refer control plan of each process to identify importance and criticality of each gauge. Select gauge as per criticality.
•Considering point 2 & 3 prepare MSA Plan.
•Plan not more than 2 MSA per day.
Perform
•100% gauges of the Gauge –list need not be covered.
•Perform MSA as per MSA plan.
Present
•Analyze and Conclude study with interpretation.
•For MSAs out of acceptable limit, take necessary action and again perform MSA.
•Part with lowest tolerance to be taken for MSA.

38. 40Subhodeep Krishna Deb
Points to be confirmed before starting MSA
1.No of Parts X No of Appraiser should be minimum 15
2.Appraisers must be the users of the measurement system
3.Parts to be numbered from 1 to n (10) so that numbers are not visible to
appraisers
4.Gauge should be calibrated
5.Parts should be clean and dent free
6.Measurement should be in random order
7.All parts should be retained after study till completion of study
8.Observer should have a ref copy of the MSA readings
9.10 samples should represent the maximum process variation; follow Systematic
sampling for selection.

39. 41Subhodeep Krishna Deb
For related assistance & service:
[email protected]