ENG2030M ENGINEERING STATISTICS

3. Modelling Random Behaviour

(Part 2)

1
 Outline
• Random variables
• Distribution functions
• Gaussian / Normal distribution
• Properties
• Standard normal distribution
• Discussion of engineering applications examples
• Linear combinations of random variables
• Probability Plots
• Discrete Random Variables
• Probability Mass function
• Binomial Distribution
• Normal Approximation to binomial
• Engineering applications

2
 Review of Tutorial Case Studies
2. Concrete strength: The design specification requires a minimum strength of
concrete of 60 MPa. Two suppliers A and B claim that they can reliably deliver to
this specification and quoted the mean and standard deviation of their
processes:
Supplier A: mA = 64.5MPa; sA = 2.7MPa; Supplier B: mB = 62.5MPa; sB = 1MPa.

However, supplier B
manufacturing process
has lower variability,
and can be expected B
to deliver far fewer
below spec units.
It should therefore be Initially many engineers
preferred to Supplier were tempted to select
A. A Supplier A given the higher
mean strength of concrete.

Spec
3
 Review of Tutorial Case Studies
3. Manufacturing of a plastic Pin (for the FFD): The design specification for the diameter
of the pin used in the Transit Van fuel filled door design is F5.5 ± 0.1 mm (i.e. the lower
specification limit is LSL = 5.4 mm and the upper specification limit is USL = 5.6 mm).
Injection moulding process parameters - m = 5.543mm, s = 0.0908mm.

26.5%
above USL

USL

LSL-R USL-R

5.7% Revised specification limits to ensure reject rate

below LSL of below 1%

This is not a good strategy as it will result in large

variation in functional performance (opening effort)
LSL 4
 The Normal Distribution
• We can use the normal distribution to establish design
specifications and process targets and tolerances

This is the basis for process

design

For a “capable” process the

specification must include (be
wider than) the natural tolerance

Sometimes called “natural

The 6s interval centred around the process mean
tolerance” of the process.
includes 99.73% of the process values
5
 Linear Combinations of Normal
Distributions
• For the Fuel Filler Door the variable of interest to the Customer
is “Effort to open / close”. This is given by the interference
between the Pin and the Stopper Gap, so we need to look at the
manufacturing variation of both these dimensions.
Let’s look at the
Tolerance Chain
Pin
Pin Diameter

Stopper Gap

Interference
Stopper
What is the
distribution of the
Interference?
6
 Linear Combinations of Independent
Random Variables Normally Distributed

Sometimes called the RSS (Root Sum of

In simple terms s  s12  s22  ...  sn2 Squares) rule for pooled standard
deviation
7
 Solution to the Tolerancing Problem
Numeric Example
mP, sP
Pin Diameter Pin: mP = 5.543mm, sP = 0.0908mm
mS, sS Stopper: mS = 5.25mm, sS = 0.12mm
Stopper Gap
Interference:
Interference
mi, si mi  5.543 5.25  0.293mm
si  sp2  s2S  0.1505mm
Distribution of the Interference

mi = mP - mS

si2= sP2+ sS2

At least 2.5% doors
won’t stay closed!

8
 Project Management Problem
mA, sA
Task A

mB, sB
Task B

Project Duration
mP, sP

Task B cannot start until Task A is completed

Question: What is the total project

duration?

mP = mA + mB
On this basis we can calculate the risk of not
sP2= sA2+ sB2 completing the project in time, estimate the
associated financial penalties, etc…
9
 Modelling Random Behaviour
• The statistical task:

Prediction of
Data Our model
Choose Predict future variation
a good
model

• Two questions:
1. what values to use for m and s?
2. is the Gaussian/Normal model a reasonable one?

10
 Normal Probability Plot
• To answer Question No 2 we can look for evidence of symmetry, lack of
Q 2. Is the skewness and excess kurtosis;
Normal model • A more powerful graphical test is to construct a Normal Probability Plot;
a reasonable
• The normal probability plot compares the distribution in the actual data
one?
sample actual data with a “perfect Normal sample” of the same size;
• For the DC Alternator data the “Perfect Normal sample” uses the
standardized “normal scores” (z-values) for n = 25 (25 measurements).

4%
4%
2%

11
-2.05 -1.55 -1.28
 Normal Probability Plot
The DC alternator data – as run chart and dot plot

17

16

15

14

Measurement no.
13
0 5 10 15 20 25
12
 Normal plot for the DC Alternator Data
17
If our model is right we should see
an approximate straight line

16
Actual sample

15

14

13
-2.5 -1.5 -0.5 0.5 1.5 2.5
Perfect Normal sample

Standardized Normal scores

13
 Interpretation of Normal Plot
What if the points
don’t line up well? Assume there was
This would appear
a very large value
as a point off the
in the sample
18 trend in the
normal plot

17
Actual sample

16
We should
remove the
15 outlier from the
data set and redo
Root causes for the normal plot
14 outliers should always with n-1 values
be investigated!
13
-2.5 -1.5 -0.5 0.5 1.5 2.5
Normal score 14
 Interpretation of Normal Plot
400

350 A normal plot with a clear curve

300 indicates that the normal model
Actual sample

might not be appropriate.

250

200
We can try a
response
150 transformation

100
For example we can take the
50 natural logarithm of x.

0
-2.5 -1.5 -0.5 0.5 1.5 2.5
Normal score
15
 Interpretation of Normal Plot
17
Normal plot after
taking logs
16
Actual sample

A logarithmic
15
transformation
of the response
often improves
14 normality.

13
-2.5 -1.5 -0.5 0.5 1.5 2.5
Normal score

16
 Example
• Normal Plot for the impurity in the Digozo Blue pigment
data (see Computer Tutorial Week 4 – Additional Exercise
– in Blackboard)

Sample No Impurity z (normal scores)

1 19 -2.33
Normal Plot for Impurity
2 37 -1.88 1200

3 42 -1.64
1000
4 56 -1.48
5 66 -1.34
800
6 71 -1.23
7 76 -1.13
600
8 83 -1.04
9 83 -0.95 400
10 103 -0.88
200
(50 data points in total)
0
-3.00 -2.00 -1.00 0.00 1.00 2.00 3.00

17
 Example
• Normal Plot for natural logarithm of impurity in the
Digozo Blue pigment data

Sample No Impurity Ln(impurity) z (normal scores) Normal Plot for ln(impurity)

1 19 2.94 -2.33 8.00
2 37 3.61 -1.88
3 42 3.74 -1.64 7.00
4 56 4.03 -1.48
5 66 4.19 -1.34 6.00
6 71 4.26 -1.23
5.00
7 76 4.33 -1.13
8 83 4.42 -1.04
4.00
9 83 4.42 -0.95
10 103 4.63 -0.88 3.00

(50 data points in total) 2.00

1.00

0.00
-3.00 -2.00 -1.00 0.00 1.00 2.00 3.00

18
 Normal Probability Plots Using Software
• Note that some software (and textbooks) plot the normal
scores on the Y axis, and the actual sample data on the X
axis – this should not affect the analysis.

Minitab
Normal plot for
the DC
Alternator
Voltage data

19
Characteristics of Non-Gaussian Distributions
Skewed
distributions:
e.g. life data /
durability

Negative skewness Positive skewness

(e.g. fatigue failures) (e.g. bearings failure)
skewness – 0
kurtosis - 3

Excess kurtosis
e.g. stock markets
Positive excess kurtosis Negative excess kurtosis
(e.g. high market volatility – many (e.g. currency exchange – many
medium changes, few extreme changes) small changes, more than
20
expected extreme changes)
 Outline
• Random variables
• Distribution functions
• Gaussian / Normal distribution
• Properties
• Standard normal distribution
• Discussion of engineering applications examples
• Linear combinations of random variables
• Probability Plots
• Discrete Random Variables
• Probability Mass function
• Binomial Distribution
• Normal Approximation to binomial
• Engineering applications

21
 Discrete Random Variables
• In the Fuel Filler Door example we have asked
customers for their rating of opening effort of
various prototypes on a scale from 1 to 10.
Histogram of relative
frequencies of customer
ratings for Prototype P1*
Attribute / Discrete Data
35%

30%

25%

20%

15%

10% P(X=4)=0.3 (30%)

5%

0%
1 2 3 4 5 6 7 8 9 10

22
*Data available from Tutorial Sheet 1 – Additional Exercises
 Discrete Random Variables

35%

30%

25%

20%

15%

10% P(X=4)=0.3 (30%)

5%

0%
1 2 3 4 5 6 7 8 9 10

23
 Discrete Random Variables

100%

90%

80%
Cumulative
70%
relative
60%
frequencies plot
of customer 50%

ratings for 40%

Prototype P1 30%

20%

10%

0%
1 2 3 4 5 6 7 8 9 10 24
FFD customer rating data
Discrete Random Variables

25
 Discrete Random Variables - Example
FFD customer rating data – Prototype P1 Relative
(Bins) Frequency
Frequency
35% Score (xi) fi
f(xi)
30% 1 1 0.02
25% 2 2 0.04
20% 3 7 0.14
4 15 0.3
15%
5 8 0.16
10% 6 3 0.06
5% 7 5 0.1
0% 8 6 0.12
1 2 3 4 5 6 7 8 9 10 9 2 0.04
10 1 0.02
10
X xi  fxi   1  0.02  2  0.04  3  0.14  ... 10  0.02  5.1
i1
10

2
s 
10
 x 2i  f x i   X 2
 4.29
Note that this is equivalent
to calculating the
 xi  fi
i1
i1 arithmetic mean in the X
normal way .
n
s  2.07
26
 Binomial Distribution
• Assume that the requirement specification for a DC
Alternator is 14.3V; manufactured alternators with
voltage below this value are rejected.
17

16.5
What is the

Accepted
16
probability
15.5
that in the
Voltage [V]

15 next batch of
14.5
14.3 25 DC
alternators

Rejected
14

13.5 there will be

13 only 3 rejects?
1
2
3
4
5
6
7
8
9

15

20

25
10
11
12
13
14

16
17
18
19

21
22
23
24
Sample No

27
 Binomial Experiment

We don’t know the true

The inspection of a DC Alternator can be
value of p, but our
regarded as a binomial experiment
estimate of p based on
the last 25 samples is
p = 20/25 = 0.8
28
 Binomial Distribution

Example
Based on the last 25
What is the samples our estimate
probability that of the probability of a
in the next reject is
batch of 25 DC p = 5/25 = 0.2
2.36%
alternators
there will be  25 
just 1 reject? f(1)     0.21  1  0.2251  0.0236
1 29
 Binomial Distribution Example
• Assuming p = 0.2 (i.e. 1 in 5 is below spec – 20%)
What is the
probability that x f(x) n x
0 0.38% f(x)     p  1  pnx
the next batch 1 2.36% x
of 25 DC 2 7.08%
3 13.58%
alternators 4 18.67% 25%

there will be 5 19.60%

6 16.33% 20%
0,1,2,3,4, 5, … 7 11.08%
rejects? 8 6.23% 15%

f(x)
9 2.94%
10 1.18% 10%
11 0.40%
12 0.12% 5%
13 0.03%
14 0.01% 0%
15 0.00% 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

30
 Binomial Distribution Example
• Assuming p = 0.2 (i.e. 1 in 5 is below spec – 20%)
x f(x) Cumulative
What is the 0 0.38% 0.38% 100.0%
1 2.36% 2.7%
probability that 2 7.08% 9.8%
90.0%

the next batch 3 13.58% 23.4% 80.0%

4 18.67% 42.1% 70.0%

of 25 DC 5 19.60% 61.7% 60.0%
alternators 6 16.33% 78.0% 50.0%
7 11.08% 89.1%
there will be 5 8 6.23% 95.3% 40.0%

or less rejects? 9 2.94% 98.3% 30.0%

10 1.18% 99.4% 20.0%
11 0.40% 99.8%
10.0%
12 0.12% 100.0%
13 0.03% 100.0% 0.0%
14 0.01% 100.0% 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

15 0.00% 100.0%
Binomial Cumulative Probability Plot for Alternator Data
What is the probability
that the next batch of 25
DC alternators there will P(X>5p=0.2) = 1 – P(X  5 p=0.2) = 38.3%
be more than 5 rejects? 31
 Binomial Distribution Example
• If we reduce the probability of failure to 10% or 1%, the chance (or
risk) of getting more than 2 defects in a batch of 25 decreases
dramatically.
(0.2%)100%
90%

80%

70%

60%
(46.3%) p=0.2
50%
p=0.1
40% p=0.01

30%

20%

(90.2%)10%
0%
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
32
DC Alternator Data
 Binomial Distribution Parameters

Example 25%

20%
• For the DC alternators if
p=0.2: 15%

f(x)
m  n  p  25  0.2  5
10%

5%

s2  n  p  1  p  25  0.2  0.8  4 0%
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

x
33
 Normal Approximation to Binomial
This approximation
works if np > 5 and
n(1-p) > 5

25%
Example P(X2):
20%
2  25  0.2
Z  1.5
15%
25  0.2  1  0.2
f(x)

10%
Using the Z tables:
5% F(Z) = 0.0668 (6.68%)
0%
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Using the binomial cdf P(X2) = 9.8%
34
DC Alternator Data x
 Normal Approximation to Binomial
For example we should calculate
In cases like this a correction
P(X2.5) rather than P(X2)
can be applied to reflect the
fact that the normal 2.5  25  0.2
distribution is defined over the Z  1.25
whole domain, and not only 25  0.2  1  0.2 
discrete values like 2 and 3.
Using the Z tables:
F(-1.25) = 0.105 (10.5%)

25%
Example P(X2):
20%
2  25  0.2
Z  1.5
15%
25  0.2  1  0.2
f(x)

10%
Using the Z tables:
5% F(Z) = 0.0668 (6.68%)
0%
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Using the binomial cdf P(X2) = 9.8%
35
DC Alternator Data x
Summary of Session – learning objectives

• Understand basic probability concepts associated with continuous and

discrete random variables;
• Understand probability distribution functions and probability mass
function and calculation of mean and variance for both continuous and
discrete random variables;
• Properties of the Gaussian (normal) probability distribution function;
• Standardize normal random variables and use of tables for the cumulative
distribution function of a standard normal distribution to calculate
probabilities;
• Linear combinations of normally distributed random variables (the RSS
rule) and practical applications;
• Construction and use of normal probability plot to test normality;
• Properties of binomial distribution, calculation of mean, variance and
probabilities for a binomial random variable.

36