Chapter 8.
Process and Measurement
System Capability Analysis
Process Capability
Natural tolerance limits are defined as follows:
Uses of p
process capability
p y
Reasons for Poor Process Capability
Process may y have g good
potential capability
Data collection steps:
p
Example 7-1, Glass Container Data
Probability Plotting
Mean and
standard
deviation could be
estimated from
the plot
p
• The distribution may not be normal; other types of probability
plots can be useful in determining the appropriate distribution.
Process Capability Ratios
Example 5-1: USL = 1.00 microns, LSL = 2.00 microns
R
σ= = 0.1398
d2
Practical interpretation
PCR = proportion of tolerance interval used by process
For the hard bake process:
One-Sided PCR
• Cp does not take
process centering into
account
• It is a measure of
potential capability,
p p y, not
actual capability
Measure of Actual Capability
Cpk: one-sided PCR for specification limit nearest to process average
Normality
y and Process Capability
p y Ratios
• The assumption of normality is critical to the
interpretation
p of these ratios.
• For non-normal data, options are:
1. Transform non-normal data to normal.
2
2. Extend the usual definitions of PCRs to handle non
non-normal
normal
data.
3. Modify the definitions of PCRs for general families of
d st but o s
distributions.
• Confidence intervals are an important way to express the information
in a PCR
Confidence interval
is wide because the
sample
p size is small
Confidence interval
is wide because the
samplel size
i iis smallll
Process Performance Index:
Use only when the process is not in control.
If the process is normally distributed and in control:
R
Pp = C p and Ppk = C pk because σ ≈
ˆ ˆ ˆ ˆ ˆ
d2
Process Capability Analysis Using Control Chart
• Specifications are not
needed to estimate
parameters.
Since LSL = 200
Process Capability Analysis Using Designed
Experiments
Gauge and Measurement System Capability
Simple model:
y: total observed measurement
x: true value of measurement
ε: measurement error
x and ε are independent. x ∼ N ( µ , σ P2 ) and ε ∼ N ( 0, σ Gauge
2
)
k = 5.15 → number of standard deviation between 95% tolerance interval
Containing 99% of normal population.
k = 6 → number of standard deviations between natural tolerance limit of
normal population.
p 7-7, USL = 60 and LSL = 5. With k =6
For Example
P/T ≤ 0.1 is taken as appropriate.
Estimating Variances
Because we estimate σˆ Gauge = ( 0.887 ) = 0.79 :
2 2
→
Signal to Noise Ration (SNR)
SNR: number of distinct levels that can be reliably obtained from
measurements SNR should be ≥ 5
measurements.
For Example 7-7: ρ M = 0.0786 and ρˆ P = 1 − ρˆ M = 0.9214.
The gauge (with estimated SNR < 5) is not capable
according to SNR criterion.
Discrimination Ratio
DR > 4 for capable
p g
gauges.
g
For example, 7-7:
The gauge is capable according to DR criterion.
Gauge R&R Studies
• Usually conducted with a factorial experiment (p: number of
randomly selected parts, o: number of randomly selected
operators, n: number of times each operator measures each part)
( )
Pi ∼ N 0, σ P2 : effects of parts
O ∼ N ( 0, σ ) : effects of operators
j
2
O
( PO ) ∼ N ( 0, σ ) : joint effects of parts and operators
ij
2
PO
ε ∼ N ( 0, σ ) : effects of operators
ijk
2
• This is a two-factor factorial experiment.
• ANOVA methods are used to conduct the R&R analysis.
• Negative estimates of a variance component would
lead to filling a reduced model
model, such as
as, for example:
σ2: repeatability variance component
→
→
For the example:
With LSL = 18 and USL = 58:
This gauge is not capable since P/T > 0.1.
Linear Combinations
( )
For x1 , x2 ,… , xn , assume xi ∼ N µi , σ i2 and independent from each other
other.
Let y = a1 x1 + a2 x2 + ... + an xn .
⎛ n n
2⎞
Then y ∼ N ⎜ ∑ ai µi , ∑ ai σ i ⎟
2
⎝ i =1 i =1 ⎠
Nonlinear Combinations
g ( • ) : nonlinear
li i off x1 , x2 ,… , xn
function
f
µi : nominal (i.e., average) dimension for xi (for i = 1, 2,...n)
Taylor series expansion of g ( • ) :
→
R is higher order (2 or higher) remainder of the expansion.
R→0
→
Assume I and R are centered at the nominal values.
α = 0.0027: fraction of values falling outside the natural tolerance limits.
S
Specification
f limits are equal to natural tolerance limits.
(
Assume I = 25 ± 1 Amp or 24 ≤ I ≤ 26 and I ∼ N 25, σ I2 . )
Zα / 2 = Z 0.00135 = 3.0
30
26 − 25
→ = 3.0 → σ I = 0.33
σI
(
Assume R = 4 ± 0.06 ohms or 3.94 ≤ I ≤ 4.06 and I ∼ N 4, σ R2 . )
4.06 − 4.00
→ = 3.0 → σ R = 0.02
σR
Assuming V is approximately normal:
→
Estimating Natural Tolerance Limits
For normal distribution with unknown mean and variance:
• Difference
Diff between
b t tolerance
t l limits
li it and
d confidence
fid limits
li it
• Nonparametric tolerance limits can also be calculated.
