Introduction to Basic Reliability Maths

Probability and statistics are critical to reliability analysis, forecasting and understanding risk

Basic Reliability Maths 1

 Presentation Contents
•Probability
 Multiplication/Addition/Binomial/Bayes Theorem
 Series/Parallel system calculations

•Statistics
 Distributions and Deviation
 Normal/Lognormal/Exponential/Weibull f(t)

 Cumulative Distribution
 Goodness of Fit
Time

•Functions
 MTBF/MTTF/MTTR/Availability
 Reliability and Hazard Functions R1 R3
 Reliability Block Diagrams
R2 R4

Basic Reliability Maths 2

 Definitions and Conventions
Random – unbiased, equally likely to occur

Independent event – no memory, past events have no effect on future events

Conditional event – first event influences the outcome of second event

Mutually exclusive – first event prevents second event (e.g. coin toss: once head cannot be tail)

Probability lies inclusively between 0 (never happen) and 1 (certain to happen)

Notation:
P(A) – probability of A happening
P(B) – probability of B happening
P(AB) – probability of independent events A and B happening together
P(A+B) – probability of independent events A or B happening
1 – P(A) = P(Ā) – is probability of A not happening
P(A|B) – conditional probability of A occurring given B has occurred

Basic Reliability Maths 3

 Reliability Terms
0 10 20 30 40 50 60 70 80 90 100
1.0000
Failure Pattern

Failure Rate
Chance of Failure in 0.1000
Hazard Rate
next year (if still alive 0.0100
at beginning of year). λ(t) or z(t)
0.0010

Note: Log Scale 0.0001

0.04
Most Probable Age
Failure Probability 0.03 Piles of Dead at Death (84)
(at Age of Death)
Density Function 0.02 (distribution mode)
(Failure Distribution) 0.01
f (t) MTTF ~ 70 yr
0.00

1.0
Reliability R(t) Probability of death
0.8
or Survival Curve
Failure Probability
on or before his 70th
0.6
Pattern birthday
Unreliability 0.4
Probability of living to
Cumulative Failure 0.2
his 90th birthday
Probability F(t) 0.0
Age at Death
Basic Reliability Maths 4 59 - L10 Age 4
Source: Howard Witt, Reliability Engineer, 2008
 Probability Maths
Venn
P(A) – probability of A happening A Diagrams B
P(B) – probability of B happening

P(AB) – probability of independent events A and B happening together

A AB B
P(AB) = P(A) x P(B)

P(A+B) – probability of independent events A or B happening

P(A+B) = P(A) + P(B) – P(AB) A A+B B

Probability of mutually exclusive events – B cannot happen if A has happened

P(A+B) = P(A) + P(B) – (0) because P(AB) = 0
A B
Because B cannot happen together with A, A and B can never occur and so P(AB) = 0

Probability of dependent events – B can only occur if A occurs first

A AB B
P(B|A) = P(AB)
P(A)
Basic Reliability Maths 5
Series Configurations and Arrangement
System
P(A) AB P(B) P(AB) = P(A) x P(B) x … x P(n)
A B n

For the System to work both A

and B must work together

Series Systems
Rsystem= R1 x R2 x R3
= 0.995 x 0.995 x 0.995

= 0.985
R1 R2 R3 i.e. the pump-set will not operate 15 times in 1000 uses

Basic Reliability Maths 6

Parallel Configurations and Arrangements
System

A
A A+B B P(A+B) = P(A) + P(B) – P(AB)

B
P(A+B) = 1 – [{1 - P(A)} x {1 - P(B)} x … x {1 x P(n)}]

For the System to work either A

or B must work

Parallel System (Full active redundancy – all working at the same time)
Rsystem= 1-[(1- R1)x(1- R2)x … (1-Rn)]
= 1 – [(1-0.6) x (1-0.6)] = 1 – [(0.4) x (0.4)]

= 1 – [(0.16)]

= 0.84

Basic Reliability Maths 7

 Redundancy for Unrepaired Items
Redundancy Type

Active Parallel Standby

(replicated and in service) (replicated awaiting failure)

Full Partial Conditional Identical Units Different Units

R1 R1 R1 R1 R1

R2 R2 R2 2/3 R2 R2

2 out of
R3 R3 R3 3 voting R3 R3
‘m’ out of
‘n’ running

Uses Parallel Uses Binomial Uses Bayes Uses Poisson

Arrangement Theorem to Theorem to Distribution
to calculate calculate calculate to calculate
System System System System
Reliability Reliability Reliability Reliability

Basic Reliability Maths 8

 Redundancy and System Reliability Behaviour
Each has individual equations Redundancy Type
Active Parallel Standby
(replicated and in service) (replicated awaiting failure)

Full Partial Conditional Identical Different

Units Units
1 1 Standby operates
independently
System MTBF

System Reliability
On
System Reliability

Unit MTBF
5 e Un

Sta
Units i t

nd
2 Ac
2

by
tiv
Units e
In Active units
Series 1.5 operate together
0 0
Time Time
No of Units

System Failure Rate

System Failure Rate

In Series 1
λ 2 3 4 5 λ One Unit Alone
2 Units y
MTBF of 2 units in fully active e n db
tiv Sta
redundancy is 1-1/2 of single unit Ac
5 Units

0 Time Time
0
These curves apply to
Units of Identical Reliability UNREPAIRED systems where Two Units Arranged in
in Fully Active Redundancy Basic Reliability Maths Differing Redundancy 9
failed lives are not restarted
 Mean Time Between Failure
Only applies to constant failure rate situations

t1 t2 t3 t4
Up

Down

MTBF = (t1 + t2 + t3 + t4) / 4 Repaired systems

MTTF = (t1 + t2 + t3 + t4) / 4 Non-repaired systems e.g. bearings, transistors, drink glass

Basic Reliability Maths 10

 Mean Time To Repair

Up

Down
r1 r2 r3 r4

MDT = (r1 + r2 + r3 + r4) / 4 Mean Downtime of all operating time losses

Also known as MTTR - Mean Time To Repair

Estimated Mean Time = [Shortest + 4 x most likely + (shortest + longest)] / 6 with s.d. = b/6 (approx)
Used to estimate time for a job. Ask experienced people to estimate the shortest, most likely and longest
times for a job. Only used at start when there is no recorded data and then improved as data is collected.

Basic Reliability Maths 11

 Availability
Combines Reliability and Maintainability to indicate System Readiness

A = Uptime’s = Uptime
Uptime’s + Downtime’s Total Time

A = MTBF For constant failure rate situations only

MTBF + MDT

MTBF is Mean Time Between Failure

MDT is Mean DownTime

Basic Reliability Maths 12

 Reliability Block Diagrams
R1 R3 R1 R3

R2 R4 R2 R4

Which arrangement is more reliable if all are R = 0.9?

How much more reliable is it?
Rsystem= 1-[(1- R1)x(1- R2)x …(1-Rn)] Rsystem= R1 x R2 x …Rn

Left Hand Side Reliability Block Diagram:

For each parallel pair Rsyst = 1 – [(1 – 0.9) x (1 – 0.9)] = 0.99
For the resulting series Rsyst = 0.99 x 0.99 = 0.98

Right Hand Side Reliability Block Diagram:

For each series pair Rsyst = 0.9 x 0.9 = 0. 81
For the resulting parallel pair Rsyst = 1 – [(1 – 0.81) x (1 – 0.81)] = 0.964

Percentage Improvement
The LHS arrange is more reliable by 0.98 – 0.964 = 1.66%
0.964
Basic Reliability Maths 13
Combination of Series Parallel

Basic Reliability Maths 14

 Statistical Analysis of Failure Data

The ideal result in statistical analysis is to test the whole population. Then
we know the complete range of outcomes. But that is hardly ever possible.
Instead, we take a sample of the whole population and test the sample. If
the sample is big enough, and it truly represents the population, then we
can infer probable future outcomes for the population from the historic
data applying to the sample, provided all remained equal in the future.

For confidence in using statistical results, there needs:

1. A large enough sample
2. The sample is representative of the population
3. The history of the sample represents the population
4. An indication of how certain the sample reflects the population
5. The future remains the same as the past

Basic Reliability Maths 15

Frequency Distribution

Basic Reliability Maths 16

Cumulative Frequency Distribution

106 plants have 20 or less tomatoes

18 plants have 12 or less tomatoes

Basic Reliability Maths 17

Fitting a Known Equation to the Data

This is the equation for a normal distribution

Basic Reliability Maths 18

 Common Distributions in Reliability Analysis
f(t) z(t) R(t)
β>1
β<1 β>1 β>1
β<1
Weibull β~1
β~1 β~1
β<1
Time Time Time
f(t) K~1 z(t) R(t)
K<1
K<1
K<1
Gamma K~1 K~1
K>1
K>1 K>1
Time Time Time
f(t) z(t) R(t)

Exponential

Time Time Time

f(t) z(t) R(t)

Normal

Time Time Time

Type Probability Distribution Failure Rate Curve Reliability Curve

There are many graphical shapes with formulas for failure rate and reliability. We only need to match the
Basic Reliability Maths 19
parts’ failures to the right curve and we can use the maths for the curve to predict the part’s future!
 Confidence Limits of Sample Data
Population Distribution
Frequency of Occurrence

Sample Distribution

Variable being Measured

Extent of Difference between Mean of Sample and
Population Mean is Gauged by Confidence Limits

E.g. A 95% confidence interval for a mean of 35 ± 2,

means that we are 95% sure that the true mean of the
population lies somewhere between 33 and 37

Basic Reliability Maths 20

 The Reliability of Reliability – Confidence Limits
Reliability should never be used to set operating and maintenance policies.
It is the result of operating and maintenance policies!

• Don’t use Reliability Prediction of the Future… it is not repeatable

• Use Reliability to make comparisons between options

Predicted Failure Rate

Predicted Failure Rate

e.g. We can be 95% confident
that by using site data the Predicted Failure Rate
60% 60% 60%
Achieved Failure Rate will be 90% 90% 90%
within 3.5 times the Predicted 95% 95% 95%
Failure Rate; 90% confident it will
within 2.5 times; 60% confident it
times will be within 1.5

λ Achieved Failure Rate λ λ

90% 1.5λ 2.5λ 3λ
95%
2.5λ 4λ 6λ
3.5λ 5λ 8λ
3.5λ Confidence of Prediction Confidence of Prediction Confidence of Prediction
λActual λSample
using Site Specific Data using Industry Database using Generic Databases
e.g. OREDA e.g. FARADIP.3
Basic Maintainability
Source: Smith, Dr, David J., Reliability, Reliability Maths
and Risk, Seventh Edition, Elsevier, 2005, Section 4.4 21
 Calculating the Reliability of a Part
20

Lifetimes of 200 Switches in Laboratory Test

Number of Failures ‘n’

15

10 20
18
16
14
13 12
5 11 10
9 8 7
6 6 5 5 Etc.
4 4

5 10 15
Number of 1,000 Operations

This is the distribution of switches failing the test. Each bar is the number of switches that failed during each
block of 1,000 operations. The switches were set-up on a laboratory testing device that flicked them on and
off continually and counted the cycles. Laboratory testing is commonly used to test the reliability of new
products that have not yet been used ‘in-the-field’ and have no failure data.
Basic Reliability Maths 22
 Modelling Failure Data on a Spreadsheet

No of items failing during a time period

Failure Rate =
Total number of items remaining in sample

Probability of failure during a time period =

No of items failing during a time period

Total number of items in sample

Reliability = 1 – Probability of Failure

From Failure History

Basic Reliability Maths 23

Rate of Failure from History of Part
Interval Failure Rate = No of items failing during a time period
Number of items remaining at period start
= 16 . = 16 =
20
200 - 20 -18 162 0.099 = 9.9%
Lifetimes of 200 Switches in Laboratory Test
Number of Failures ‘n’
15

10 20
18
16
14
13 12
5 11 10 9 8 7
6 6 5 5 Etc.
4 4

5 10 15
Number of 1,000 Operations

10% This is a relatively constant

random failure rate of 10% of
7.5% parts per 1,000 operations
Failure Rate %

10.7% 11.1% 11.1%

10.4% 10.5% 10.4% 10.1%
5% 10% 10% 9.9% 9.6% 9.8% 10.1% 10.3% 9.7% 10% 10%

2.5%

Etc.

5 10 15
Number of 1,000 Operations
Basic Reliability Maths 24
 Hazard Rate from Failure History of Part
Interval Hazard Rate = Interval Failure Rate
Operations in Interval
= 0.099% = 0.000099
1000
10%

7.5%
Failure Rate %

10.7% 11.1% 11.1%

10.4% 10.5% 10.4% 10.1%
5% 10% 10% 9.9% 9.6% 9.8% 10.1% 10.3% 9.7% 10% 10%

2.5%

Etc.

5 10 15
1 Number of 1,000 Operations

0.75
z(t) Hazard Rate

0.5

0.25

Etc.
Parts fail at rate of 0.0001
(per 1000 operations)
5 10 15
Number of 1,000 Operations
Basic Reliability Maths 25
 Probability of Failure
No of items failing during a time period
Probability of failure during a time period = Total number of items at start (t = 0)
10%
= 16 = 0.08 = 8%
200

Reliability
Probability of Failures %

Failure
7.5%

5% 10%
9%
8%
7%
6.5%
6%
5.5%
2.5% 5%
4.5%
4%
3.5%
3% 3%
2.5% 2.5%
2% 2% 2% 2%
1.5% 1.5% 1.5%
1.25% 1.25%

5 10 15 20 25 30 35

100% Number of 1,000 Operations Probability of Failure During a Time Period

From the curve of the 200 test

Probability of Failure %

75%
switches we can estimate how
50%
many switches will fail by a
93%

72% 75% 78% 80% 82% 84% 86% 88% 89.5% 90% 91.5%
certain number of operations in
69%

25% 41% 47% 52% 57% 62% 66%

actual service
34%
27%
19%
10%

5 10 15 20 25 30 35
Number of 1,000 Operations
Cumulative Probability of Failure
Basic Reliability Maths 26
Basic Reliability Maths 27
 Formula for the Reliability of a Part
10%
f(t) Probability of Failures %

This curve is called the ‘Probability Density

Reliability
Failure
7.5%

Function’ and this ‘exponential shape’ is

5% 10%
9%
the characteristic pattern of constant rate
8%
7%
6.5%
random failure.
6%
5.5%
2.5% 5%
4.5%
4%
3.5%
3% 3%
2.5% 2.5%
2% 2% 2% 2%
1.5% 1.5% 1.5%
1.25% 1.25%

5 10 15 20 25 30 35
Number of 1,000 Operations
100%

Reliability = 1 – Probability of Failure

75% Rt = 1 – f(t)
Rt Reliability %

We now have a way to

50% 90% determine the proportion that
81%
73% 66%
60%
will survive past a certain
25%
54% 48% 43% 39% 35% 31%
number of operations
28% 25% 22% 20% 18% 16% 14% 12% 10.5% 9% 7.5% 6.3% 6%

5 10 15 20 25 30 35
Number of 1,000 Operations

Basic Reliability Maths 28

 Reliability Mathematics in ‘Random Failure’ Zone
For a constant failure rate the probability of a component surviving to a
time (t) is the reliability (R(t)).

R(t) = number surviving at instant ‘t’ P(s) = e (-t / MTBF)

number at start when time t = 0 at ½ MTBF 80% of parts
have not failed
The failure rate can be expressed as z(t) below.

z(t) = number failing per unit time at instant ‘t’

number surviving at instant ‘t’
Thus, the failure rate can be defined as the probability of failure in unit
time of a component that is still working satisfactorily.

Chance of Failing
R(t) = e -f(t) for constant chance of failure End of
Infant
Assuming that the failures occur randomly at a constant chance rate, the Mortality Life
Constant Likelihood of Failure
Mean Time To Failure is given below.

MTTF = 1/ failure frequency Time – Age of Part from New

This can also be written as the probability of success (P(s)), The Maintenance Zones of Component Life
or the chance of zero failures.
With this equation we can predict how
R(t) = P(s) = e (-t / MTBF) where t = time
many parts will survive during the ‘random
This is a formula to work out how many parts remain after a period of failure’ period of operating life
time if they are all failing randomly

Basic Reliability Maths 29