MSA & SPC

Measurement System Analysis

PRECISION and ACCURACY
Low Precision, High Precision, High Precision,
Low Accuracy Low Accuracy High Accuracy

.. . ..
...
… ….
….
… ….
..
.. ….
.. . ..
PRECISION : Degree of freedom from
random errors.

ACCURACY : refers to extent to which a

reading could be wrong. The higher the
Accuracy, the lower the Bias and vice-versa.

High precision does not imply high accuracy.

PROPERTIES OF A GOOD
MEASUREMENT SYSTEM :

 Measurement system must be in a state of

statistical control.

 That
is, variation present must be only due to
common causes.
 Properties of a good measurement system
(contd.):

 Variability
of measurement system must be small
compared to variability of manufacturing process.

 Variability
of measurement system must be small
when compared to specification limits.
Objective of MSA
To obtain information related to amount and type
of variation associated with measurement system
when the system is put into use.
Scope
To cover all (types of) inspection, measuring &
test equipment referenced in all Control Plans.
PPAP 4th Edition requires MSA for all new
Instruments & Gauges.
BIAS
is the difference between the observed
average of measurements and the reference
value.

Word “Accuracy” means exactness and hence not

recommended to be used to refer to “Bias”.
Ref. Value BIAS Observed
average
STABILITY (Drift)

is the total variation in the measurements

obtained with a measurement system on
the same master or part when measuring
a single characteristic over an extended
time period.
Stability = Variation in measurement
averages over a period of time.
Observed
average

Stability
Time 1
Time 2
LINEARITY

is the difference in Bias values through

the expected operating range of the
Gauge
Linearity = Difference between Bias values over the
operating range Observed
average
Ref. Value

Lower Range Upper Range

Linearity : indicated by slope of Observed Average Values

Bias

No Bias line

Ref. Value
REPEATABILITY
(also known as Equipment Variation)
is the variation in measurements obtained
with one measurement instrument when
used several times by same appraiser while
measuring the identical characteristic on
the same part.
Repeatability = Variation in measurements with one
Gauge, one Appraiser, Same Part, Same Characteristic.

Same Gauge,
Same Part,
Same Characteristic,
Same Appraiser

Repeatability
REPRODUCIBILITY
(also known as Appraiser Variation)
is the variation in the average of measure-
ments made by different appraisers using
the same measurement instrument when
measuring the same characteristic on the
same part.
 Same Gauge,
Reproducibility Same Part,
Same Characteristic,
but Different
Appraisers
Appraiser C

Appraiser A Appraiser B
PART VARIATION (PV)

Part to part variations found for the same

characteristic.
Normally seen on average control charts.
The subgroup averages reflect the part-to-part
differences.
RESOLUTION

The capability of Measurement System

to detect and faithfully indicate even
small changes in the measured characteristic.
DATA CATEGORIES :
Because of economic and physical limitations,
the Measurement System will not perceive all
parts of a process distribution as having
separate values, but group them together -
as Data Categories.
DATA CATEGORIES :

Data Categorization is a result of :

a) Equipment Apparent Resolution
b) Measurement System variation
(Effective resolution)
UNACCEPTABLE DISCRIMINATION :

The discrimination is unacceptable if it can

not detect Part to Part variations , and

Unacceptable for control if it can not detect

Special Causes variation.
Phases of MSA :

Phase I : Understand measurement process, and

confirm if it satisfies our requirements.

Discover environmental factors having

significant influence on Measurement
System.
Phases of MSA :

Phase II : Verify whether a Measurement System,

once deemed acceptable, continues to
have appropriate statistical properties.
MSA comprise evaluation of 5 types of
variation present in a measurement system
in a statistical manner :
1. Bias
2. Linearity
3. Stability
4. Repeatability
5. Reproducibility
Phase I covers 1 to 5 and Phase II covers 4
& 5.
BIAS :

A Part of reference value 6.00 is checked 15 times

by one appraiser, who obtained following values:
5.80, 5.70, 5.90, 5.90, 6.00, 6.10, 6.00, 6.10, 6.40,
6.30, 6.00, 6.10, 6.20, 5.60, 6.00.

Average = 6.0067. Hence Bias = 0.0067.

- Is this an acceptable measurement bias ?
Hypothesis Testing :
A Statistical Hypothesis is an assertion about a
population, normally about a numeric value.
For example, a hypothesis might state that the
mean life of a product is 30, written as H:
m =30.
Suppose we take 8 samples, and test their life
to get X = 27.6 and s = 1.86.
-

There are 2 types of errors in hypothesis testing:

 Errors in hypothesis testing:

- Reject the hypothesis when it is true (called the

type I error). Its Probability is called the Level
of Significance, denoted by letter a .

- Accept the hypothesis when it is false (called

the type II error). Its Probability is denoted
by (1 - a ).
Suppose we have decided that the Type I error
should not exceed 5 % (Alpha 5%).
Now let us take 2 properties of a distribution.
A) The distribution of sample averages is approximately
normal even if the individual measurements going into
the averages is not normal.
B) Sample averages vary less than the individual
measurements according to formula :
__
s X (known as Standard Error of Mean) = s / v (n)
-
Hypothesis (H0): Measured value represents Ref. Value.
-
Standard Deviation = R / d2 = (Max Xi – Min Xi) / d2*
= (6.4 – 5.6) / 3.553 = 0.22514

Where d2* is the modified d2 constant (for no. of samples < 25).
__
Standard Error of Mean = 0.22514 / v 15 = 0.05813.

Calculated ‘t’ statistic = (X - H0) / s b

-

= (6.0067 – 6.0000) / (0.05813)

= 0.1152.
- Degrees of freedom v as a function of d2* for
No. of samples :1 and sample size:15 = 10.8.

- ‘t’ statistic from Tables for v = 10.8 at confidence

level of 1- a/2 (0.975) = (t v.1-a/2) = 2.206.

- Confidence Interval = Bias +/- (d2 x s b x t v.1-a/2 / d2* )

= 0.0067 +/- (3.47193 x 0.05813 x 2.206 / 3.553)
= + 0.1319 and – 0.1185.

Since ‘0’ falls between the confidence intervals, (also calculated ‘t’
statistic falls within theoretical ‘t’ Statistic value), we can accept the
hypothesis, that is the measured value represents the ref. value.
Since the calculated ‘t’ statistic is inside the acceptable
value, accept the hypothesis (i.e. the Bias is acceptable).

When the Bias is unacceptable, look for following causes:

1. Error in Master or Reference Value
2. Worn instrument
3. Instrument made to wrong dimension
4. Instrument measuring wrong characteristic
5. Instrument not calibrated properly
6. Instrument used incorrectly by appraiser
7. Instrument correction algorithm incorrect
Linearity:
An instrument is used not only for measuring one value,
but many values at various locations. Ex. A Vernier
Caliper is used for checking dimensions 5.80, 10.60,
20.75, 32.45, 50.75, etc.
Select 5 or more parts whose measurements, due to
process variation, cover the entire operating range of
the Gauge.
Have each part measured by layout inspection to
establish its reference value.
Have each part measured 10 or more times by one of the
operators who normally use the gauge.
Linearity:
- Calculate the part bias for each measurement and the
bias average for each part.
- Plot the individual biases and bias averages with
respect to reference value on a linear graph.
- Calculate and plot the best fit line and confidence band
of the line using following equation:
y-i = a xi + b,
where xi is reference value, y-i is bias average,
a is slope and b is intercept.
Linearity: For measurement system to be acceptable,
Bias = 0 line must entirely lie within the
confidence bands indicated below.
+1
Regression line
0

-1

Bias

Ref. Values
Linearity:
When linearity is unacceptable, do following:
1. Re-Calibration to achieve zero bias throughout the
operating range,
2. If re-calibration does not give zero bias, it can still be
used for product / process control, as long as the
measurement system remains stable, but should not be
used for analysis.
3. Customer approval may be required.
Stability:
1. Obtain a sample. If feasible, establish its reference
value.
2. On a periodic basis, measure the sample 3 – 5 times.
Frequency to consider significant factors that could
affect stability (ex. Temperature, Humidity, etc.).
__ __
3. Plot standard X / R or X / s Charts.
4. Establish Control Limits.
5. Identify out-of-control conditions and correct.
6. The standard deviation of the measurements can be
used as an approximation of the measurement
system’s repeatability.
Repeatability & Reproducibility:

The Gauge R & R are often the major contributors

of measurement variation.

The evaluation of variations contributed by

repeatability and reproducibility is called Gauge R
and R Study.
 When is a Gauge R & R study required :

 Estimating process variation and acceptability

of production process for new / modified part,
 Decidingacceptability of a new type of
measurement system,

 Following
changes in control plans affecting
measurement systems.
Key factors for effective Gauge R & R study:

1. Sample parts must be selected from process

which represent its entire operating range.

2. The appraisers chosen should be selected from

those who normally operate the instrument.
3. The measurements should be made in a random
order. All possible bias in measurements to be
avoided.

4. The readings should be rounded to the nearest

1/2 value if possible.

5. The instrument must have a discrimination that

allows at least one tenth of the expected process
variation to be measured.
 Methods of Gauge R & R Study
 Range Method
 Average and Range Method
 ANOVA Method
A popular method of R and R variation
analysis is Average and Range method -
typically done through 2-3 appraisers and
10 randomly chosen parts, 2 – 3 trials.
 Relation of R & R, PV and TV

R&R TV
AV R&R

PV
EV
Acceptance criteria for Gauge R & R :
a) Graphical data agreement required.
b) Gauge R&R value:
- Below 10 % = Acceptable
- 10 - 30 % = May be acceptable
- Above 30 % = Not acceptable. Identify
causes, correct, restudy.
When R & R % is high, number of data
categories will be lower, and vice versa.
IMPORTANCE OF DATA CATEGORIES :
Data Category = 1

 Acceptable for control only when process

variation is small compared to specs.
 Unacceptable for analysis.
IMPORTANCE OF DATA CATEGORIES :
Data categories = 2 to 4
 Suitable for controls with semi-variable
techniques (say p-chart).
 Provide only coarse estimates.
 Hence not acceptable for estimating
process indices and parameters.
IMPORTANCE OF DATA CATEGORIES :
Data Categories = 5 and above

 Acceptable for control as well analysis.

 Can be used with variable control charts
(Ex. X Bar-R Chart).
 Recommended.
Actions on R&R values:

If Repeatability is large compared to Reproducibility,

look for following causes:

1. The Instrument needs maintenance

2. The gauge is probably not rigid and need redesigning
3. The clamping or location for gauging need to be
improved
4. There is excessive with-in-part variation.
Actions on R&R values:

If Reproducibility is large compared to Repeatability,

look for following causes:
1. The Appraiser(s) may require more training
2. Calibrations on the gauge / instrument dial not clear
3. A fixture of some sort may be required to help the\
appraisers to use the gauge more consistently.
Attribute Gauge study:
2 – 3 appraisers and 50 parts, 2 – 3 trials.
Measurement system acceptance criteria:
Effectiveness Miss Rate False Alarm Decision
> / = 95 % =0% </=5% Acceptable
> / = 90 % < / = 2 % < / = 10 % Marginal *
< 90 % >2% > 10 % Unacceptable.
* Do not use for Special Characteristics.
Evaluate above values appraiser-wise.