Introduction to Weibull Analysis

BY BILL KEETER, ARMS RELIABILITY ENGINEERS, USA, LLC

Weibull Analysis of failures is an important tool in reliability improvement. During this two hour short course attendees will be introduced to basic Weibull
concepts and learn the practical application of Weibull techniques in determining maintenance strategies. Attendees will learn how to use Weibull to identify
failure mechanisms that are present in their facilities and which maintenance strategies are appropriate for dealing with the failure mechanisms.

HISTORY
Weibull analysis of failure data is an important Reliability Engineering
tool. The Weibull distribution was invented by Swedish engineer Waloddi
Weibull in the 1930s to describe life data. It is the most popular statistical
model for life data and is used in other applications such as weather
forecasting. Weibull has the advantage of using very small sample sizes to
make engineering judgments. A reasonable expectation of future behavior <1
can be obtained using engineering knowledge and a single data point.

Weibull Parameters
Eta ( ) – The characteristic life, or the point at which 63.2% of items will
have failed with the same failure mode. Figure 2. Infant Failure Curve.
Beta ( ) – The slope or shape characteristic of the failure curve. Beta is
Random Failures
used to help determine what sort of maintenance activities are appropriate
Random failures are characterized by =1. Random failures occur due to
for a given failure mode.
random events in the life of the system. These failures are typical of
Gamma ( ) – Describes the point at which the Weibull curve changes shape
maintenance errors, natural effects, and mixed failure modes.
characteristics.

The Bathtub Curve

The bathtub curve is a combination of all the phases of life data in a
single curve. It is typically used to represent the way the lives of people
progress. If we make it past infancy we usually die because of random =1
events such as diseases, automobile accidents, and natural disasters. Later
in life we begin to wear out and die due to old age.

<1 >1
1 2 3
Figure 3. Random (Exponential) Failure Curve.

Early Wear Out Failures

Early wear out failures are characterized by 1< <4. These failures may
require replacements at very low ß lives if their failure creates significant
hazards to equipment or people.

Old Age Wear Out Failures

Old age wear out failures are characterized by >4. Old age wear out
within the design life of the system is a concern. The value of steep s is
Figure 1. Lifetime (Bathtub) Failure Curve. that they are very predictable and lend themselves well to timed
replacements.
Weibull Shapes
Equipment failure mechanisms can be represented by various regions or
combinations of regions of the bathtub curve. Failures can be infant,
random, wear out, or some combination of the three regions.

Infant Failures
Infant failures are characterized by <1. They are typical of quality
>4
issues such as poor maintenance and operating procedures or varying
replacement parts quality.

Figure 4. Wear Out Failure Curve.

RELIABILITY WORLD 2005 CONFERENCE PROCEEDINGS 289

Failure Prediction with Limited Data Varying Life Distributions
Weibull can be used to make failure predictions with a single data point The exponential distribution is very useful for describing many life
if we have some engineering knowledge about the failed part and the failure distributions, but it is not effective for analyzing every set of life data. Dr.
mode. We simply plot the failure point on Weibull paper and draw a line with Waloddi Weibull invented the Weibull distribution that is widely used to
a slope equal to what we suspect the failure characteristics would be. This predict life behavior from limited life data sets.
method of prediction is called Weibayes.
Mathematically the distribution is expressed as:
The Reliability Equation Equation 2. Reliability of Varying Lifetime Behavior
R(t) = e-(t/ )^

The Exponential Distribution

Reliability is the probability that a given piece of equipment will operate Where:
without failure for a given period of time under stated conditions. In essence t = Mission Time
the definition tells us that reliability is determined by equipment design, = Characteristic Life
equipment operation, and mission time. = Weibull shape Characteristic

Mathematically reliability for a given failure mode with exponential life Once we know eta and beta we can make predictions about failure rates,
distribution is: cumulative failures, and failure risks.

Equation 1. Reliability for Random Lifetime Behavior. The cumulative probability of failure at time t, the Cumulative Distribution
R(t) = e- t
Function (CDF) is expressed by:
Equation 3. Cumulative Probability of Failure.
= Failure Rate F(t) = 1 – R(t) = 1 - e-(t/ )^

t = Mission Time
The probability density at time t, the probability density function (pdf) is
The failure rate ( ) = 1/MTTR or 1/MTBF. The units for the failure rate and expressed by:
the mission time must be the same. Equation 4. Probability Density Function.
f(t) = ( /t)(t/ ) e-(t/ )^

A pump has an MTBF of 36 months. What is its reliability for the following
mission times? The failure rate at time t is expressed as:
Equation 5. Failure Rate.
h(t) = ( /t)(t/ )
Mission Time Reliability
Problems
One Month 1. What is the characteristic life ( ) for a pump that has a reliability of .95
for a mission time of 20,000 hours and a Weibull Shape Parameter (ß) of
5000 Hours 2.5?
2. What is the reliability of a fan for a mission time of 17,000 hours if the
10 Years fan has an eta of 8,700 hours and a beta of 4.0?
3. What is the cumulative probability that a group of pumps with the failure
characteristics in problem one will fail by 35,000 hours?

290 RELIABILITY WORLD 2005 CONFERENCE PROCEEDINGS