Time-Dependent Failure Models

Probability models used to describe

non-constant failure rate processes

Overview

Weibull Distribution
Probability Functions
Characteristics of Weibull Distribution
MTTF, Variance, Median and Mode
Conditional Reliability

Normal Distribution
Probability Functions
Mean, Variance and Moments
Central Limit Theorem

Lognormal Distribution
Probability Functions
Mean, Variance

Weibull Distribution
Weibull distribution can attain many shapes
for various values of shape parameter .
It can model a great variety of data and life
characteristics,
including
constant,
increasing, and decreasing failure rates
(CFR, IFR, and DFR).
Therefore, it is one of the most widely used
lifetime
distributions
in
reliability
engineering.

Two-Parameter Weibull Distribution:

= shape parameter, or slope parameter
= scale parameter, or characteristic life parameter

Probability Functions
CFR, IFR, DFR

Hazard Function

h(t )

, t 0, 0, 0

Reliability Function

t
R (t ) exp

Cumulative Distribution Function

F (t ) 1 exp

Probability Density Function

f (t )

t
exp

3

Characteristics of Weibull
Distribution
Characteristic Effects of Shape (Slope)
Parameter,
Effects of on Weibull Hazard Function
(CFR, IFR, DFR depending on the value of )
Effects of on Weibull PDF
Effects of on Weibull Reliability Function

Characteristic Effects of Scale Parameter,

Effects of on
Weibull Hazard Function

h(t )

, t 0, 0, 0

Effects of on
Weibull Hazard Function
All three life stages of the bathtub curve can
be modeled with the Weibull distribution
and varying values of :

< 1: DFR

= 1: CFR (Exponential distribution)

> 1: IFR
1< < 2, the h(t) curve is concave
the failure rate increases at a
decreasing rate as t increases
= 2, a straight line relationship
between h(t) and t
> 2, the h(t) curve is convex
the failure rate increases at an
increasing rate as t increases

Effects of on
Weibull Hazard Function
Value
0<<1
=1

Property
DFR (pdf is similar to
Exponential Dist.)
CFR (pdf is
Exponential Dist.)

Application
Model infant
mortality
Model
useful life

1< <2 IFR (hazard function is

concave)
=2

LFR (hazard function

is linearly increasing)
(pdf is Rayleigh
distribution)

>1
>2

IFR (hazard function is

convex)

34

IFR (pdf approaches

normal, symmetrical)

Model
wearout

Effects of on Weibull PDF

f (t )

t
exp

8

Effects of on Weibull PDF

0 < < 1:

As t 0, f(t) .

As t , f(t) 0.

f(t) decreases monotonically as t increases

f(t) is convex

The mode is non-existent.

= 1:

exponential distribution

> 1:

f(t) = 0 at t = 0

f(t) increases as t approaches to the mode and

decreases thereafter

1< < 2.6, f(t) is positively skewed right tail

2.6 < < 3.7, its coefficient of skewness

approaches zero no tail, approximate the normal
pdf,

> 3.7, f(t) is negatively skewed left tail

9

Effects of on
Weibull Reliability Function

R (t ) exp

10

Effects of on
Weibull Reliability Function

R (t ) exp

0 < < 1, R(t) decreases sharply and

monotonically, and is convex.

= 1, R(t) decreases monotonically but less

sharply than for 0 < < 1, and is convex.

> 1, R(t) decreases as t increases. As wearout sets in, the curve goes through an inflection
point and decreases sharply.

11

Scale Parameter,
The scale parameter influences both the
mean and the spread of the Weibull
distribution.
All Weibull reliability functions pass
through the point (, 0.368).

R ( ) exp e 1 0.368

is also called characteristic life.

has the same units as T, such as
hours, miles, cycles, actuations, etc.

12

Effects of on Weibull PDF

f (t )

t
exp

Increasing the value of has the effect of

stretching out the pdf
influences both mean and spread
Since the area under a pdf curve is a constant
value of one, the peak'' of the pdf curve will also
decrease with the increase of

13

Examples (Weibull Dist.)

Example 1
A two-speed synchronized transfer case used in a
large industrial dump truck experiences failures
that seem to be well approximated by a two
parameter Weibull distribution with

18, 000 km and 2.7

(1) Characterized the failure process based on the
values of and .

(2) What is the 10,000 km reliability?

R(10000) e

10000 2.7
)
18000

0.815

(3) What is the 24,000 km reliability?

R(24000) e

24000 2.7
)
18000

0.1137
14

Examples (Weibull Dist.)

Example 1 (Cont.)
A two-speed synchronized transfer case used in a
large industrial dump truck experiences failures
that seem to be well approximated by a two
parameter Weibull distribution with

18, 000 km and 2.7

(4) What is the B10 life? (km at which 10% of the
population will fail, or 90% reliability is desired.)
R( ) 1

100

R(10 ) 0.9 e

t
) 2.7
18000

10 7821.7

(5) If the Weibull slope can be changed to 1.7 by

changing the synchronizer gear tooth design, how
will this affect the above answers?
R(10000) 0.6920 0.815
R(24000) 0.1957 0.1137

15

Examples (Weibull Dist.)

Example 2
The characteristic life for a highly turbocharged
diesel engine in a military application is 1,800
miles with a Weibull slope of 1.97. What is the B10
life?

16

Examples (Weibull Dist.)

Example 3
A device that shows running-in failure pattern, has
threshold time to failure of 150 days and characteristic
life of 100days. What is the probability that this device
will fail before 200 days of running times? What is the
probability that this device will survive for 180days? What
is the age-specific failure rate at 180 days? (Hint: the
shape factor in this case could be one of the following;
1.2 ,0.5, 4.5)

17

MTTF and Variance

1
MTTF 1

2
1

1
1

where ( x) is the gamma function:

( x) y x 1e y dy
0

To obtain the gamma function, use Standard

Table in the textbook, and

( x) ( x 1)( x 1), x 0
( x) ( x 1)!, x integer > 0

18

Median and Mode

Design Life and B Life
R (t R ) e

(t R / )

R ( B ) e

( B / )

t R ( ln R )1/

1 / 100

B ln(1 / 100)

1/

Median
R ( B50 ) 0.5 e

( B50 / )

B50 tmed (ln 2)1/

Mode
tmode

(1 1/ )1/

for 1
for 1

19

Conditional Reliability
Conditional reliability is useful in
describing the reliability of an item
following a burn-in period or after a
warranty period, T0.
For Weibull distribution, the conditional
reliability is

P[T t T0 ]
R (t | T0 )
P[T T0 ]

[( t T0 ) / ]
(T0 / )

e
exp (T0 / ) [(t T0 ) / ]
R(t | T0) = R(t) for all t if and only if R(t) is
exponential.

20

Burn-In

Early failures (infant mortality) common reliability

problem, esp. in electronic equipment usually
caused by manufacturing defects--- quality problems.
Problems are typically common in new products and
may disappear as technology matures.

Ideal build-in Q&R upfront and reduce such

problems but hard to do with complex technology and
pressure to reduce product development cycle time.

To achieve reliability goals and reduce field-failure,

common practice to burn-in components and systems
to screen out units that would fail early esp. important
in safety-critical applications can be viewed as a type
of 100% inspection.

Done at use condition or low-level stress environment

to avoid undue aging of components and systems.

Burn-in is expensive incorporate costs and benefits

and decide on optimal trade-off

21

Examples (Weibull Dist.)

Example 4
Given a Weibull distribution with a characteristic
life of 127,000 hr and a slope of 3.74, find the
mean and standard deviation. Also find the
probability of surviving the mean life.
Solution
MTTF (1

1
) (1.27) 0.9025

3.74
2
2
B (1 ) (1
) (1.53) 0.88757

3.74
MTTF 127000(0.9025) 114617 .5
A (1

) (1

2 2 ( B A2 )

34382.5

22

Examples (Weibull Dist.)

Example 4
For the two-speed synchronized transfer case
problem in Example 1, we have two designs. Find
the mean and standard deviation of the life.
A: 18, 000 km and 2.7
Solution

MTTF 16007.57
6388.294
18, 000 km and 1.7
B:

MTTF 16063.74
9752.76

23

Normal (Gaussian) Distribution

Normal distribution is not a true reliability
distribution since the random variable
ranges from minus infinity to plus infinity.
When the probability that the random
variable takes on negative values is
negligible, the normal distribution can be
used successfully to model fatigue and
wear out phenomena.
It will be used to analyze lognormal
distribution.

24

Probability Density Function

For T ~ N(, 2), the PDF is:

1 (t ) 2
1
f (t )
exp
, t
2
2
2
is the location parameter
shifts the graph left or right on the horizontal axis
2 is the scale parameter
stretches or compresses the graph
1 t
f (t )

1 z2 / 2
e
where z
is the pdf for standard
2
normal distribution (=0, 2=1).

25

CDF and Reliability Function

F(t)

R(t)

Cumulative Distribution Function, F(t)

t
F (t )

where z is the cdf for standard normal

distribution (=0, 2=1).

Reliability Function, R(t)

t
R(t ) 1 F (t ) 1

26

Hazard Function

1 t

f (t )

h(t )

R (t )
t
1

h(t) is an increasing function

the normal distribution can only be
used to model wear-out (IFR)
phenomena.
27

Examples of
Normal Distributions

28

Mean, Variance and Moments

Mean and Variance
E[T ]
Var[T ] E (T ) 2 2

Moments of Normal Distribution

m is odd

E[(T ) ]
m

m ! m
2m / 2 (m / 2)! m is even

29

Central Limit Theorem

If X1, X2, , Xn is a random sample of size n
taken from a population (either finite or
infinite) with mean and finite variance 2,
and if X is the sample mean, as the sample
size n approaches to infinite,

2
X ~ N ,
n

The central limit theorem basically states:

The sampling distribution of the mean
becomes approximately normal regardless
of the distribution of the original population.
The sampling distribution of the mean is
centered at the population mean, .
The standard deviation of the sampling
distribution of the mean approaches
.
/ n
Due to the central limit theorem, the normal
distribution plays a central role in classical
statistics.

30

Examples (Normal)
Example 1
A component has a normal distribution of failure
times with =20,000 cycles and =2,000 cycles.
Find the reliability of the component and the
hazard function at 19,000 cycles.

31

Lognormal Distribution
The lognormal distribution is used
extensively in reliability applications to
model failure times. (The lognormal and
Weibull distributions are the most
commonly used distributions in reliability
applications.)
Like the Weibull distribution, the lognormal
can take a variety of shapes.
The relationship between lognormal
distribution and normal distribution is very
useful in analyzing the lognormal
distribution.

32

Lognormal Distribution
If T ~ LogNor( , 2 ), then Ln(T ) ~ N( , 2 )
with Probability Density Function

1 (ln t ) 2
1
f (t )
exp
, t0

t 2
2

is the mean of ln(T), and

2 is the variance of ln(T), and
0
Lognormal
T

2
exp
2

Mean
Variance
Median
Mode

Normal
Ln(T)

2 2

e 1

e
e

33

Lognormal Distribution
PDF and CDF
f (t )

1 ln t

, t0
t

ln t
F (t )
, t0

Reliability Function
ln t
, t0

R (t ) 1 F (t ) 1

Hazard Function
1 ln t

f (t ) t

, t0
h(t )

R (t )
ln t
1

1 z2 / 2
where z
and z are the
e
2

pdf and cdf for standard normal

distribution (=0, 2=1).

34

Examples of
Lognormal Distribution

35

T=0.5
T=1.5

The lognormal distribution is a distribution skewed to

the right.
The pdf starts at zero, increases to its mode, and
decreases thereafter.
The degree of skewness increases as T increases.
36

Examples (Lognormal)
Example 1
The failure time of a certain component is log2
normally distributed with 5 and 1 .
1. Find the reliability of the component and the
hazard rate for a life of 150 time units.
2. Find the mean and variance of the failure time of
the component.

37