Section 3

Failure Data Analysis Using

Weibull

Objectives
‹ Understand the concepts behind Weibull
Analysis
‹ Understand the parameters associated with
the Weibull distribution
‹ Know when the Weibull distribution can be
used
‹ Given a set of times to failure be able to:
‹Understand the type of failure
‹Forecast future reliability levels
‹ Understand how data from components that
haven’t / didn’t fail can be used
50
 Background

The Weibull Distribution was invented in 1937 by

Waloddi Weibull when comparing mortality rates of
different population groups.

High Probability Steady state Probability Increasing Probability

of death when young of death due to random events of death when past the
“wear out” age

51

Background

He invented a formula that could describe the different

shaped graphs in each of the three zones.
β
T 
−  
η
R (T ) = e 
(2 Parameter Weibull)

R(T) = Reliability at Time T

T = Time considered
η = Characteri stic Life
β = Shape Parameter
e = 2.71828 (the base for natural logs) 52
 The Weibull Distribution

Nolan and Heap “discovered” use for Weibull’s work when they wrote the
paper that was the birth place of Reliability Centered Maintenance.
They found that traditional maintenance policies were based on
assumption that items wore out at some “age” limit.
Whereas their review of 10 years of failures in civil aviation revealed most
occurred according to β<1.
Their work promoted age exploration and modeling behaviour.

53

Failure Behaviour

Prob. of fail
Traditional view of Maintenance

Age
Wear out
Maintain before wear out
54
 Failure Behaviour

Prob. of fail
Trouble is most things don’t wear out

Infant mortality

Age
55

Failure Behaviour

Prob. of fail
Fixed time maintenance = increased
risk at increased cost!

Age

Maint. 56
 Failure Behaviour

Birthplace of RCM was to recognise the

importance of knowing the failure curve
of each failure mode before deciding on
which maintenance task is applicable.

57

Failure Behaviour

Reliability Centered Maintenance recognises

6 different combinations of the Weibull
curves.

A
B
C
D
E
58
F
 The Weibull Distribution

• Failure patterns
Failure
– Type A – Bathtub (4%) patterns
from
– Type B – Worst old (2%)
aviation
– Type C – Slow ageing (5%) industry
– Type D – Best new (7%) study
– Type E – Constant (14%)
– Type F – Worst new (68%)

59

Background

β<1 Weibull understood that

these same characteristics
were displayed in the
β=1 strength of steel and he
was engaged by the U.S.
airforce up until 1975.
β>1

60
 The Weibull Distribution
Why Important?
• One of the most widely used distributions
• Highly flexible
• Best fits many real world applications:
• Weibull distribution represents the life of components and parts
whereas the Exponential distribution represents the life of
assemblies and systems
• Mechanical components: ball bearings, motors, fatigue failure of
some simple structures
• Failures where chemical actions are a predominant mechanism

61

Advantages of Weibull
Analysis

Primary advantage of Weibull is the ability by the

analyst to use few data points to make failure forecasts.
This means that it is possible to use data as the first
few failures emerge and decide appropriate action
before more failure data is generated.
This gives the reliability engineer the tool to model
failure behaviour, make decisions on future
performance levels, determine warranty periods, and
confirm B10 lives.

62
 Advantages of Weibull
Analysis

• Bad data or dirty data can also provide valuable

information
• Indications of mixed failure modes can be
recognised by a relatively inexperienced
practitioner
• Information regarding items that have not failed in
service can also be valuable to making reliability
predictions (suspensions)

63

Advantages of Weibull
Analysis
The Weibull formula provides the Reliability analysts with
the capability to play “what if?”.
• What if time at temperature was significant?
• What if the next replacement interval was 4
years?
• If failures occur in service, a prediction of the
number of failures in the next period of time can
be made.
• What if we use a component with superior
reliability?
64
 The Weibull Distribution
The most commonly used distribution for life studies

Typically viewed on Failure Rate vs. Time plot (Hazard

Function)

Failure
Rate
Eta1 Eta2 Eta3

Beta<1 Beta=1 Beta>1

Gamma 1 Gamma 2 Gamma 3 Time

65

The Weibull Distribution

Implies quality problems or insufficient “burn in”. If

associated with maintenance overhaul may indicate
β<1 procedure problems or poor commissioning
practices.

Random failures or failures independent of time in

β=1 service. Causes typically are random events such
as lightning, human error or variable conditions.

Wear out failures at a definite or predictable end of

life. Typically age related due to service conditions
β>1 such as corrosion, wear or fatigue cracking.

66
The Weibull Distribution
Often this will indicate an early onset of failure of
the rest of the population of components.

1<β<4

Ideally this behaviour occurs outside the design

life i.e. majority of items have reached their wear
out age and its time for replacement.
β>4 If this occurs before the design life it is an
indication of under-design or underestimate of
the stress levels in service.

67

The Weibull Distribution

What will a Weibull for a complex piece of

machinery look like?

β=1 In complex machinery the Weibull’s tend to

cancel each other out.

What use is Weibull analysis for complex

machinery?

68
 The Weibull Distribution

Target a dominant mechanism!

69

Failure Behaviour

Objective

z To know the relationship between failure

behaviour and different maintenance tasks

70
 Beta Values : Wear In & Random
Shapes

Probability of failure
(Failure rate)

β = 0.5

β= 1

Time

71

Beta Values: Wearout Shapes

Probability of failure
(Failure rate)
β=4

β=2

Time

72
 Advantages of Weibull
Analysis

Exercise

Consider a car
• Weibull shapes
• Maintenance actions

73

Advantages of Weibull
Analysis

74
 Advantages of Weibull
Analysis

Weibull Shapes: Tyres Weibull Shapes:

Brake Pads

Maintenance task: Maintenance task:

75

Advantages of Weibull
Analysis

Weibull Shape: Body Weibull Shape: Paint

Work

Maintenance task: Maintenance task:

76
 Advantages of Weibull
Analysis

Weibull Shape: Headlight Weibull Shape: Stoplight

Maintenance Task: Maintenance Task:

77

Advantages of Weibull
Analysis

Weibull Shape: Battery Weibull Shape: Oil

Maintenance Task: Maintenance Task:

78
 Advantages of Weibull
Analysis

Weibull Shape: Interior light

Maintenance Task:

79

Failure Behaviour

Weibull parameters can describe any failure

behaviour over an asset’s lifetime using any
of the 3 zones of the bathtub curve.
Zone 1 Zone 2 Zone 3

Failure
Rate
Eta1 Eta2 Eta3

Beta<1 Beta=1 Beta>1

Gamma 1 Gamma 2 Gamma 3 Time

80
 Failure Behaviour

Consider a constant failure rate (random).

Please close
your training
manual

Say failure rate =1/10yrs

81

Failure Behaviour

MTTF = 1 / λ
What is the MTTF?
MTTF = 1 / (1/10)
MTTF = 10 yrs

Say failure rate λ =1/10yrs

82
 Failure Behaviour

If there is an initial population of 100 light bulbs in

a room, how many will be left when you come back
after 10 years? (non maintainable system)
100

Say failure rate =1/10yrs

83
10 years

Failure Behaviour

Years No Rem λ No Failed Cum failures

1 100 0.1 10
2 90 0.1 9
3 81 0.1 8.2
4 72.9 Answer equals 63.2
0.1 7.4
failures.
5 65.61 0.1 6.7
6 59.049 0.1 6.1
7 53.1441 0.1 5.5
8 47.82969 0.1 5.0
9 43.04672 0.1 4.5
10 38.74205 0.1 4.1

This is an exponential decay pattern given by

the expression Reliability = e-(t/MTTF)

84
 Failure Behaviour

Reliability = e -(t/MTTF)

120

100

80
A loss of 63.2 over 10 years
60

40

20

0
1 3 5 7 9 11 13 15 17 19 21 23

85

Failure Behaviour

For random failures with constant failure rate,

63.2% of items will have failed by the MTTF.
This point is known as the Characteristic life
(in Weibull terms η).
100

63.2%

MTTF
86
 Failure Behaviour

If we know
β the shape of the curve and
η the characteristic life,
we can predict the probability of behaviour over
any future time.

Failure curves can be determined from

z test data
z history
z engineering judgment
87

Advantages of Weibull
Analysis

What use is the Weibull Formula?

In a population of 17,354 wooden power poles,

records
indicate 2,100 have failed or been condemned over
last 10 years. How many do we need to budget for
replacement over next 10 years.

88
 Advantages of Weibull
Analysis

What use is the Weibull Formula?

Based on our practice of overhauling boiler tubes

every 4 years, what’s the probability of getting
through the next 4 years without a failure.

89

Advantages of Weibull
Analysis

What use is the Weibull Formula?

The life of brand x seals has been variable in our

critical feed pumps, how many are likely to fail next
year?

90
 Advantages of Weibull
Analysis

What use is the Weibull Formula?

How many tyres will need to be budgeted for on our

fleet of 38 Euclid rear dump trucks over the next 5
years?

91

Advantages of Weibull
Analysis

What use is the Weibull Formula?

Times to Failure were extracted from a CMMS and

there were two distinct distributions evident.
The first had a beta of 0.5 and eta 75 hours.
The second had a beta of 4 and eta of 26280 hrs.
What could this indicate?

92
 Advantages of Weibull
Analysis

When used?

• At or below the component level

• Only when a single failure mode is expected
• Typically applied in analysing mechanical failures

93

The Weibull Distribution

The Beta β or shape factor gives an indication of

the Physical Behaviour

The characteristic life parameter η gives an

indication of the typical time to failure. For
constant failure rates it is approximately the
same as the MTTF

94
 The Weibull Distribution

Each zone commences at Gamma - γ

Failure
Rate
Eta 1 Eta 2 Eta 3

Beta<1 Beta=1 Beta>1

G am m a 1 G amm a 2 G am m a 3 Tim e

95

The Weibull Distribution

If Gamma 3 indicates the beginning of the

wear out zone, what does Gamma 2 indicate?

96
 The Weibull Distribution

Failure
Rate
Eta1 Eta2 Eta3

Beta<1 Beta=1 Beta>1

Gamma 1 Gamma 2 Gamma 3 Time

97

Definitions

Shape Parameter (beta), β: the slope of the Weibull

curve.

Scale Parameter (Eta), η: the characteristic life or

the point at which 63.2% of the items have failed.

Location parameter (gamma), γ: the failure free

period.

98
 Practical Weibull Analysis

What do we need?
1. The time origin
This can be the installation time, or last overhaul date, provided the overhaul
is assumed to restore full life.

2. A horizontal scale
that is relevant to the failure mechanism, calendar time, operating hours, miles,
no of landings, start/stops etc.

3. Clear Definition of what constitutes failure.

99

Practical Weibull Analysis

How?
1. Enter the data

2. Check how the data fits the distribution (the line)

3. Use the results

100
 Practical Weibull Analysis

Tour Weibull

101

Practical Weibull Analysis

What use is the Weibull Formula?

The life of brand x seals has been variable in our critical
feed pumps, 3600, 2040, 6000, 5760, 3240,
4800,5760,3600,4800,4560

Q1/ What type of failure is it?

Q2/ At what life should I replace them to achieve 87% Reliability.
Q3/ What is the B10 life? B1 Life? Characteristic life?

102
 Practical Weibull Analysis

What if a design change increased the wear life by 3

months.
Q1/ How would you reflect that in Weibull
parameters?

What if the reliability prior to wearing out age

improved from 1 random failure every 10,000 hrs to 1
random failure every 100,000 hrs.
Q2/ How would you reflect that?

103

Practical Weibull Analysis

My Weibull does not start at zero.

Example: corrosion failures don’t start to appear in the

first 12 years life of a structure.

Q What Weibull parameter will you adjust to

represent this?

104
 Dirty Data

My Weibull Graph is not a straight line.

Q1/ What could you do in this situation?

105

Dirty Data

The failure rate graph of my engine is a flat line and

the reliability level is too low. How can I use Weibull
to help me to improve?

106
 Practical Weibull Analysis

Data deficiencies are common. Statisticians always want

more data, however, the reliability analyst wants to
minimise in-service failure data.

The Weibull method is a powerful tool for the Reliability

engineer because he can be effective with small quantities
of data.

As CMMS and failure reporting schemes are used more

commonly- companies will have better data.

107

Practical Weibull Analysis

I have only got one data point on a shovel swing

shaft failure after 5 years of service, which the NDT
inspector says is fatigue.
Q1/How can I predict the reliability of replacements
when I don’t have any failure data?

Use Weibayes analysis & assign beta =4.

Check component life for 90% reliability.
Examine Weibull and Failure rate versus time graph.
Is it what you expected. Try beta =2. Is this more
realistic?
108
 Practical Weibull Analysis

Q1/ In the last example I also have 3 shafts still in

service after 6 years. Can I use this data?

109

Practical Weibull Analysis

Q1/ In the last example I had 3 shafts still in service.

Try after 4 years instead of 6 years
What difference do the suspensions make?

110
 Practical Weibull Analysis
Lets say you find that an engineer returns from
holidays and has records of the previous lives at
replacement. Service conditions have not
changed. Failures occurred at 48180 hrs, 54312
hrs, 63948 hrs and 71832 hrs.

Q1/ Were your first assumptions valid for the

weibull parameters?
Q2/ What replacement age will achieve 90%
reliability?

111

The Weibull Distribution

Q/ I find the Weibull graph confusing?

A/ The failure rate versus time graph is the best way to

visualise failure rate behaviour.
The Weibull graph tells you if the plot is a straight line
and therefore how good a fit it is.
If there are corners in your curve, it is an indication 2 or
more failure modes.

112
 The Weibull Distribution

I understand that for wear out failures that

replacement or overhaul before the wear zone
is a good maintenance strategy.

Q1/ What about when β=0.5?

Q2/ What about when β=1?

113

The Weibull Distribution

Q1/ Is a steep β good? How does it

influence my maintenance decisions?

Q2/ Maintenance people tend to be busy, why spend

hours crunching data?

Q3/ How can I prove the change in maintenance practice

was any good?

114
 The Weibull Distribution

Q1/ β=1 is also known as what type of failure?

Q2/ What type of failure for bearing in clean lubricated service?

Q3/ A group of bearings with inadequate lubrication?

Q4/ A bearing with poor alignment?

Draw the shape of the curves.

115

Revisiting the Objectives

‹ Understand the concepts behind Weibull Analysis

‹ Understand the parameters associated with the Weibull
distribution
‹ Know when the Weibull distribution can be used
‹ Given a set of times to failure be able to:
‹Understand the type of failure
‹Forecast future reliability levels
‹ Understand how data from components that haven’t /
didn’t fail can be used

116
 Summary

Weibull Distributions due to their “flexibility” can be

used in many situations to analyse test data
Weibull parameters
Shape parameter = β
Scale parameter = η
Location parameter = γ

117

Weibull Excel Exercise – 3

Day Course

• Refer to the exercises section at the

back of the manual

118