Professional Documents
Culture Documents
A Practical Guide
3rd Edition
Solutions Manual
Page 1 of 229
Contents
Page 2 of 229
Solutions to Chapter 2
Exercises
2.1 Simplify the following Boolean functions
a) ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
(𝐴 ∩ 𝐵 ∪ 𝐶) ∩ 𝐵̅
= (𝐴 ∩ 𝐵 ∪ 𝐶) ∪ 𝐵
= 𝐴𝐵 ∪ 𝐵 ∪ 𝐶
= 𝑩∪𝑪
b) (𝐴 ∪ 𝐵) ∩ (𝐴̅ ∪ 𝐵̅ ∩ 𝐴̅)
= (𝐴 ∪ 𝐵) ∩ (𝐴̅)
= (𝐴𝐴̅ ∪ 𝐵𝐴̅)
̅
=𝑩∩𝑨
c) ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
𝐴 ∩ 𝐵 ∩ 𝐵 ∩ 𝐶 ∩ 𝐵̅
= ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
(𝐴 ∩ 𝐵 ∩ 𝐵 ∩ 𝐶) ∩ 𝐵̅
= (𝐴̅ ∪ 𝐵̅ ∪ 𝐵̅ ∪ 𝐶̅ ) ∩ 𝐵̅
= (𝐴̅ ∪ 𝐵̅ ∪ 𝐶̅ ) ∩ 𝐵̅
= 𝐴̅ ∩ 𝐵̅ ∪ 𝐵̅ ∩ 𝐵̅ ∪ 𝐶̅ ∩ 𝐵̅
= 𝐴̅𝐵̅ ∪ 𝐵̅𝐵̅ ∪ 𝐶̅ 𝐵̅
= 𝐴̅𝐵̅ ∪ 𝐵̅ ∪ 𝐶̅ 𝐵̅
̅
=𝑩
= 𝐴 ∩ 𝐵 ∩ (𝐶̅ ∩ (𝐶 ∩ 𝐴̅) ∩ 𝐵)
= 𝐴𝐵𝐶̅ 𝐶 𝐴̅𝐵
=ϕ
Page 3 of 229
2.3 Simplify the following Boolean expressions
a) ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
(𝐴 ∩ 𝐵 ∪ 𝐶) ∩ 𝐵̅
= (𝐴 ∩ 𝐵 ∪ 𝐶) ∪ 𝐵
= 𝐴𝐵 ∪ 𝐵 ∪ 𝐶
=𝑩∪𝑪
= 𝐵𝐴̅ ∪ 𝐵̅𝐴̅
= 𝐴̅(𝐵 ∪ 𝐵̅)
̅
=𝑨
𝐺 = (𝐴 ∪ 𝐵 ∪ 𝐶) ∩ (𝐴̅ ∪ 𝐵 ∪ 𝐶) ∩ 𝐶̅
𝐺 = (𝐴 ∪ 𝐵 ∪ 𝐶) ∩ (𝐴̅𝐶̅ ∪ 𝐵𝐶̅ )
̅
𝑮 = 𝑩𝑪
𝐏𝐫(𝑮) = 𝟎. 𝟗 × 𝟎. 𝟏 = 𝟎. 𝟎𝟗
Page 4 of 229
2.5 Simplify the following Boolean equations
a) ̅̅̅̅̅̅̅̅̅̅̅̅̅
(𝐴 ∪ 𝐵 ∪ 𝐶) ∩ (𝐴 ∩ 𝐵̅ ∩ 𝐶̅ ) ∩ 𝐶̅
= (𝐴 ∪ 𝐵 ∪ 𝐶) ∩ (𝐴̅ ∪ 𝐵 ∪ 𝐶) ∩ 𝐶̅
= (𝐴 ∪ 𝐵 ∪ 𝐶) ∩ (𝐴̅𝐶̅ ∪ 𝐵𝐶̅ )
̅
= 𝑩𝑪
b) (𝐴 ∪ 𝐵) ∩ 𝐵̅
= 𝐴𝐵̅ ∪ 𝐵𝐵̅
̅
= 𝑨𝑩
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
(𝐀 ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
∪ (𝐁 ∩ 𝐂)) ∩ (𝐁 ∪ (𝐃 ∩ 𝐀))
= 𝐴̅𝐵̅𝐵̅𝐷
̅ ∪ 𝐴̅𝐶̅ 𝐵̅𝐷
̅ ∪ 𝐴̅𝐵̅𝐵̅𝐴̅ ∪ 𝐴̅𝐶̅ 𝐵̅𝐴̅
= 𝐴̅𝐵̅𝐷
̅ ∪ 𝐴̅𝐶̅ 𝐵̅𝐷
̅ ∪ 𝐴̅𝐵̅ ∪ 𝐴̅𝐶̅ 𝐵̅𝐴̅
̅𝑩
= 𝑨 ̅
2.7 Use both Equations (2.25) and (2.29) to find the reliability Pr(s). Which equation is preferred for
numerical solution?
Pr(𝑠) = Pr(𝐸1 ∪ 𝐸2 ∪ 𝐸3 )
Pr(𝐸1 ) = 0.8; Pr(𝐸2 ) = 0.9; Pr(𝐸3 ) = 0.95
Pr(𝐸1 ∪ 𝐸2 ∪ 𝐸3 )
= [Pr(𝐸1 ) + Pr(𝐸2 ) + Pr(𝐸3 )] − [Pr(𝐸1 ∩ 𝐸2 ) + Pr(𝐸1 ∩ 𝐸3 ) + Pr(𝐸2 ∩ 𝐸3 )] + Pr(𝐸1 ∩ 𝐸2 ∩ 𝐸3 )
Page 5 of 229
= [0.8 + 0.9 + 0.95] − [0.8 × 0.9 + 0.8 × 0.95 + 0.9 × 0.95] + 0.8 × 0.9 × 0.95
= 𝟎. 𝟗𝟗𝟗
Pr(𝐸1 ∪ 𝐸2 ∪ 𝐸3 )
2.8 A stockpile of 40 relays contain 8 defective relays. If five relays are selected at random and the
number of defective relays is known to be greater than two, what is the probability that exactly four
relays are defective?
𝐷 𝑁−𝐷
( )( )
Pr(𝑋 = 𝑥) = 𝑥 𝑛 − 𝑥
𝑁
( )
𝑛
8 40 − 8
( )( )
Pr(𝑋 = 4) = 4 5 − 4 = 0.003404
40
( )
5
Similarly:
Pr(𝑋 = 3) = 0.0422
Pr(𝑋 = 5) = 8.51 × 10−5
0.003404
Pr(𝑋 = 4 |𝑋 > 2) = = 𝟎. 𝟎𝟕𝟒𝟒𝟖𝟖
0.0422 + 0.003404 + 8.51 × 10−5
Page 6 of 229
2.9 Given that P = 0.006 is the probability of an engine failure on a flight between two cities, find the
probability of:
1000
a) Pr(𝑋 = 0) = ( ) 0.0060 × 0.9941000 = 𝟎. 𝟎𝟎𝟐𝟒𝟑
0
1000
b) Pr(𝑋 ≥ 1) = 1 − ( ) 0.0060 × 0.9941000 = 𝟎. 𝟗𝟗𝟕𝟓𝟕
0
1000 1000
c) Pr(𝑋 ≥ 2) = 1 − ( ) 0.0061 × 0.994999 − ( ) 0.0060 × 0.9941000 = 𝟎. 𝟗𝟖𝟐𝟖𝟕
1 0
60 𝑒 −6
b) Pr(𝑋 ≥ 1) = 1 − 0!
= 𝟎. 𝟗𝟗𝟕𝟓𝟐
61 𝑒 −6 60 𝑒 −6
c) Pr(𝑋 ≥ 2) = 1 − 1!
− 0!
= 𝟎. 𝟗𝟖𝟐𝟔𝟓
2.10 A random sample of 10 resistors is to be tested. From past experience, it is known that the probability
of a given resistor being defective is 0.08. Let X be the r.v. for number of defective resistors.
a) What kind of distribution function would be recommended for modeling the r.v.?
b) According to the distribution function in (a), what is the probability that in the sample of 10
resistors, there are more than 1 defective resistors in the sample?
Page 7 of 229
Let X be the r.v. of being given a defective resistor
10 10
Pr(𝑋 > 1) = 1 − ( ) 0.081 × 0.929 − ( ) 0.080 × 0.9210 = 𝟎. 𝟏𝟖𝟕𝟗
1 0
2.11 How many different license plates can be made if each consists of three numbers and three letters,
and no number or letter can appear more than once on a single plate?
This problem is a question of determining the number of possible permutations. Assuming that the
digits 0 through to 9 can be used and that all 26 letters of the alphabet can be used, then the solution
is given by:
2.12 The consumption of maneuvering jet fuel in a satellite is known to be normally distributed with a
mean of 10,000 hours and a standard deviation of 1,000 hours. What is the probability of being able
to maneuver the satellite for the duration of a 1-year mission?
Let T be the r.v. for the consumption life of the satellite jet fuel
Pr(𝑇 = 𝑡) ~𝑁𝑜𝑟𝑚(𝜇 = 10,000; 𝜎 = 1,000)
2.13 Suppose a process produces electronic components, 20% of which are defective. Find the distribution
of x, the number of defective components in a sample size of five. Given that the sample contains at
least three defective components, find the probability that four components are defective.
This is a conditional probability with the Binomial distribution with parameters N = 5 and p = 0.2
5
( ) 0.24 × 0.81
Pr(𝑋 = 4 |𝑋 ≥ 3) = 4
5 5 5
( ) 0.23 × 0.82 + ( ) 0.24 × 0.81 + ( ) 0.25 × 0.80
3 4 5
Page 8 of 229
0.0064
Pr(𝑋 = 4 |𝑋 ≥ 3) = = 𝟎. 𝟏𝟏𝟎𝟓
0.0512 + 0.0064 + 0.00032
2.14 If the heights of 300 students are normally distributed, with a mean of 68 inches and standard
deviation of 3 inches, how many students have:
a) heights of more than 70 inches?
𝑋−𝜇 70−68 2
Pr(𝑋 > 70) = Pr (𝑍 > 𝜎
) = Pr (𝑍 > 3
) = Pr (𝑍 > 3) = 𝟎. 𝟐𝟓𝟐𝟓
Therefore, the number of students that are greater than 70 inches in height is given by:
300 × 0.2525 = 𝟕𝟔
2.15 Assume that for a certain type of resistor, 1% are bad when purchased. What is the probability that
a circuit with 10 resistors has exactly 1 bad resistor?
𝜇 𝑥 𝑒 −𝜇
Pr(𝑋 = 𝑥) = 𝑤ℎ𝑒𝑟𝑒 𝜇 = 𝜆𝑡
𝑥!
Page 9 of 229
2.17 A guard works between 5 p.m. and 12 midnight; he sleeps an average of 1 hour before 9 p.m., and
1.5 hours between 9 and 12. An inspector finds him asleep, what is the probability that this happens
before 9 p.m.?
𝑃 𝑟(𝐸|𝐴𝑗 ) Pr(𝐴𝑗 )
𝑃 𝑟(𝐴𝑗 |𝐸) = 𝑛
∑𝑗=1 𝑃 𝑟(𝐸|𝐴𝑗 ) Pr(𝐴𝑗 )
𝑃 𝑟(𝐸|𝐴1 ) Pr(𝐴1 ) 1⁄ × 4⁄
𝑃 𝑟(𝐴1 |𝐸) = = 4 7 = 𝟎. 𝟒
𝑃 𝑟(𝐸|𝐴1 ) Pr(𝐴1 ) + 𝑃 𝑟(𝐸|𝐴2 ) Pr(𝐴2 ) 1⁄ × 4⁄ + 1.5⁄ × 3⁄
4 7 3 7
2.18 The number of system breakdowns occurring with a constant rate in a given length of time has a
mean value of two breakdowns. What is the probability that in the same length of time, two
breakdowns will occur?
𝜇 𝑥 𝑒 −𝜇
Pr(𝑋 = 𝑥) = 𝑤ℎ𝑒𝑟𝑒 𝜇 = 𝜆𝑡
𝑥!
𝐺𝑖𝑣𝑒𝑛 𝜇 = 2
22 𝑒 −2
Pr(𝑋 = 2) = = 𝟎. 𝟐𝟕𝟏
2!
2.19 An electronic assembly consists of two subsystems, A and B. Each assembly is given one
preliminary checkout test. Records on 100 preliminary checkout tests show that subsystem A failed
10 times. Subsystem B alone failed 15 times. Both subsystems A and B failed together five times.
Page 10 of 229
b) What is the probability that A alone fails.
Pr(𝐵) = 0.2
Pr(𝐴 ∩ 𝐵) = 0.05
𝑃 𝑟(𝐴 ∩ 𝐵) 0.05
𝑎) 𝑃 𝑟(𝐴|𝐵) = = = 𝟎. 𝟐𝟓
Pr(𝐵) 0.2
2.20 A presidential election poll shows one candidate leading with 60% of the vote. If the poll is taken
from 200 random voters throughout the U.S., what is the probability that the candidate will get less
than 50% of the votes in the election? (Assume the 200 voters sampled are true representatives of
the voting profile.)
Using a Binomial approximation for a Normal Distribution we find the solution as follows:
100 − 120
𝑃 𝑟(𝑋 < 100) = Pr (𝑍 < ) = Pr(𝑍 < −2.887) = 𝟎. 𝟎𝟎𝟏𝟗𝟒𝟔
√48
2.21 A newspaper article reports that a New York medical team has introduced a new male contraceptive
method. The effectiveness of this method was tested using a number of couples over a period of 5
years. The following statistics are obtained:
a) Estimate the mean probability of an unwanted pregnancy per use. What is the standard deviation
of the estimate?
Page 11 of 229
Year Times Employed Unwanted Pregnancies (X) Pr(X) (xi-x_bar)^2
1 8200 19 0.00232 6.05E-07
2 10100 18 0.00178 5.90E-08
3 2120 1 0.00047 1.14E-06
4 6120 9 0.00147 4.71E-09
5 18130 30 0.00165 1.33E-08
𝑛
1 1
𝑥̅ = ∑ 𝑥𝑖 = (0.00232 + 0.00178 + ⋯ + 0.00165) = 𝟎. 𝟎𝟎𝟏𝟓𝟒
𝑛 5
𝑖=1
𝑛
1 1
𝑆2 = ∑(𝑥𝑖 − 𝑥̅ )2 = (6.05 × 10−7 + 5.9 × 10−8 + ⋯ + 1.33 × 10−8 ) = 𝟒. 𝟓𝟓 × 𝟏𝟎−𝟕
(𝑛 − 1) 4
𝑖=1
b) What are the 95% upper and lower confidence limits of the mean and standard deviation?
𝑆 𝑆
[𝑥̅ − 𝑡𝛼/2 ≤ 𝜇 ≤ 𝑥̅ + 𝑡𝛼/2 ] = 0.95
√𝑛 √𝑛
2.22 Suppose the lengths of the individual links of a chain distribute themselves with a uniform
distribution, shown below.
f(x)
Page 12 of 229
a) What is the height of the rectangle?
b) Find the cumulative PDF for the above distribution. Make a sketch of the distribution and
label the axes.
0 𝑥 < 𝑚 − 𝑎⁄2
𝑥 − (𝑚 − 𝑎⁄2)
𝐹(𝑥) = 𝑚 − 𝑎⁄2 ≤ 𝑥 ≤ 𝑚 + 𝑎⁄2
𝑎
{ 1 𝑥 ≥ 𝑚 + 𝑎⁄2
c) If numerous chains are made from two such links hooked together, what is the PDF of two-link
chains?
Page 13 of 229
𝐹(𝑥) = ∬ 𝑓(𝑠)𝑓(𝑡)𝑑𝑡𝑑𝑠
𝑓𝑜𝑟 𝑥 ≤ 2𝑚
1 1 𝑥 − 2𝑚 + 𝑎 1
𝐹(𝑥) = 𝑏ℎ = (𝑥 − 2𝑚 + 𝑎) ( ) = 2 (𝑥 − 2𝑚 + 𝑎)2
2 2 𝑎2 2𝑎
𝑑 𝑑 1
𝑓(𝑥) = [𝐹(𝑥)] = [ (𝑥 − 2𝑚 + 𝑎)2 ]
𝑑𝑥 𝑑𝑥 2𝑎2
(𝒙 − 𝟐𝒎 + 𝒂)
𝒇(𝒙) =
𝒂𝟐
𝑓𝑜𝑟 2𝑚 ≤ 𝑥 ≤ 2𝑚 + 𝑎
1 1 2𝑚 + 𝑎 − 𝑥 1
𝐹(𝑥) = 1 − 𝑏ℎ = 1 − (2𝑚 + 𝑎 − 𝑥) ( 2
) = 1 − 2 (2𝑚 + 𝑎 − 𝑥)2
2 2 𝑎 2𝑎
𝑑 𝑑 1
𝑓(𝑥) = [𝐹(𝑥)] = [1 − 2 (2𝑚 + 𝑎 − 𝑥)2 ]
𝑑𝑥 𝑑𝑥 2𝑎
(𝟐𝒎 + 𝒂 − 𝒙)
𝒇(𝒙) =
𝒂𝟐
d) Consider a 100-link chain. What is the probability that the length of the chain will be less than
100.5 m if a = 0.1m?
Using the Central Limit Theorem, this problem will approximation a normal; distribution with
2
𝜇 = 100𝑚 & 𝜎 2 = 100𝑎 ⁄12
100.5𝑚 − 100𝑚
Pr(𝑋 < 100.5𝑚) = Pr (𝑍 < ) = Pr(𝑍 < 1.732) = 𝟎. 𝟗𝟓𝟖𝟑
10𝑎⁄
√12
Page 14 of 229
1. f ( x, y ) 0 x, y
2.
f ( x, y )dxdy 1
2 1
0 0
1 2
2
x 2 y 2 yx3 2
y2 y y3 y2
0 4 6 dy 0 4 6 dy 12 12 1
0 0
(i) Find Pr(x > y) the limits for this problem will be 0<x<1 & 0<y<x
1 x
0 0
x 1
1
xy3 y 2 x 2 1
x4 x4 x5 x5 5
0 6 4 dx 0 6 4
dx
30 20
0
60
0
(ii) Pr( y > x) the limits for this problem will be 0<x<1 & x<y<2
1 2
0 x
2
1
xy 3 y 2 x 2 1
8x 4 x 2 x 4 x 4
0 6 4 dx 0 6 4 6 4 dx
x
1
8x 2 4 x 3 x 5 x 5 55
12 12 30 20 0 60
(iii) Pr(x = y) does not exist as this is a continuous RV and x=y would not form a volume
under the integral. Therefore, Pr(x = y) = 0
Page 15 of 229
2.24 A company is studying the feasibility of buying an elevator for a building under construction. One
proposal is a 10-passenger elevator that, on average, would arrive in the lobby once per minute. The
company rejects this proposal because it expects an average of five passengers per minute to use the
elevator.
a) Support the proposal by calculating the probability that in any given minute, the elevator does
not show up, and 10 or more passengers arrive.
There are two Poisson Distributions in this problem; one for the elevator and one for the
passengers.
𝜇𝑥 𝑒 −𝜇 1
Elevator - Pr(𝑋 = 𝑥) = 𝑥!
𝑤ℎ𝑒𝑟𝑒 𝜇 = 𝜆𝑡 = 𝑚𝑖𝑛 × 1 min = 1
10 𝑒 −1
Pr(𝑋 = 0) = = 𝟎. 𝟑𝟔𝟗
0!
𝜇𝑛 𝑒 −𝜇 5
Passengers - Pr(𝑁 = 𝑛) = 𝑛!
𝑤ℎ𝑒𝑟𝑒 𝜇 = 𝜆𝑡 = 𝑚𝑖𝑛 × 1 min = 5
9
𝜇𝑛 𝑒 −𝜇
Pr(𝑁 ≥ 10) = 1 − ∑ = 𝟎. 𝟎𝟑𝟐
𝑛!
1
Therefore, the probability that an elevator does not show up and 10 or more passengers arrive
is given by:
𝑃𝑟(𝑋 = 0 ∩ 𝑁 ≥ 10) = 0.369 × 0.032 = 𝟎. 𝟎𝟏𝟏𝟕
b) Determine the probability that the elevator arrives only once in a 5-minute period.
𝜇𝑥 𝑒 −𝜇 1
Elevator - Pr(𝑋 = 𝑥) = 𝑥!
𝑤ℎ𝑒𝑟𝑒 𝜇 = 𝜆𝑡 = 𝑚𝑖𝑛 × 5 min = 5
51 𝑒 −5
Pr(𝑋 = 1) = = 𝟎. 𝟎𝟑𝟑𝟕
1!
2.25 The frequency distribution of time to establish the root causes of a failure by a group of experts is
observed and given below.
Page 16 of 229
Test whether a normal distribution with known = 10 is an appropriate model for these data.
In order to test whether the normal distribution is an adequate fit in this case, we use a Chi Squared
Goodness-of-Fit test where
b) H1: the r.v (X) does not follow a Normal Distribution with 10
The Chi Squared Statistic is calculated for each row of the data given above using the following
process. The complete solution is shown in the table below.
𝑛
1 1
𝑥̅ = ∑ 𝑥𝑖 = (350 + 1080 + 2450 + 2240 + 1080) = 𝟕𝟐
𝑛 100
𝑖=1
45 − 72 55 − 72
Pr(45 < 𝑋 < 55) = Pr ( <𝑍< ) = Pr(< −2.7 < 𝑍 < −1.7) = 𝟎. 𝟎𝟒𝟏𝟏
10 10
𝐹𝑟𝑒𝑞𝑒𝑥𝑝 (45 < 𝑋 < 55) = N × Pr(45 < 𝑋 < 55) = 100 × 0.0411 = 𝟒. 𝟏𝟏
2
(𝐹𝑟𝑒𝑞𝑜𝑏𝑠 − 𝐹𝑟𝑒𝑞𝑒𝑥𝑝 ) (7 − 4.11)2
Chisq Stat = = = 𝟐. 𝟎𝟑𝟐
𝐹𝑟𝑒𝑞𝑒𝑥𝑝 4.11
Time Range (hrs) Mid Range Obs Freq F*Mid Pr(x) Ex(F) Chisq Stat
45 55 50 7 350 0.0411 4.110 2.0324
55 65 60 18 1080 0.1974 19.740 0.1533
65 75 70 35 2450 0.3759 37.595 0.1791
75 85 80 28 2240 0.2853 28.529 0.0098
85 95 90 12 1080 0.0861 8.608 1.3370
Total 100 7200 W 3.7116
Sample Mean 72 R 7.815
Sigma (Known) 10
2 (𝑘
𝑅 ≥ 𝜒1−𝛼 − 𝑚 − 1) where k = # of intervals & m = number of unknown parameters
Given that W < R (ie: 3.712 < 7.815) we do not reject the null hypothesis at the 5% significance level.
Page 17 of 229
2.26 A random number generator yields the following sample of 50 digits:
Digit 0 1 2 3 4 5 6 7 8 9
Frequency 4 8 8 4 10 3 2 2 4 5
Is there any reason to doubt the digits are uniformly distributed? (Use the Chi-square goodness-of-fit
test.)
In order to test whether the uniform distribution is an adequate fit in this case, we use a Chi Squared
Goodness-of-Fit test where
a) Ho: the r.v (X) follows a Uniform Distribution over the range of data; or
The Chi Squared Statistic is calculated for each data set given above using the following process.
The complete solution is shown in the table below.
1
Pr(𝑋) = = 0.1 Pr(X)
10
9
h = 0.1
X
0 9
2
(𝐹𝑟𝑒𝑞𝑜𝑏𝑠 − 𝐹𝑟𝑒𝑞𝑒𝑥𝑝 ) (4 − 5)2
Chisq Stat = = = 𝟎. 𝟐
𝐹𝑟𝑒𝑞𝑒𝑥𝑝 5
Page 18 of 229
Digit Freq Pr(x) Ex(F) Chisq Stat
0 4 0.10 5.0 0.20
1 8 0.1 5.0 1.80
2 8 0.1 5.0 1.80
3 4 0.1 5.0 0.20
4 10 0.1 5.0 5.00
5 3 0.1 5.0 0.80
6 2 0.1 5.0 1.80
7 2 0.1 5.0 1.80
8 4 0.1 5.0 0.20
9 5 0.1 5.0 0.00
Total 50 W 13.60
R 16.92
2 (𝑘
𝑅 ≥ 𝜒1−𝛼 − 𝑚 − 1) where k = # of intervals & m = number of unknown parameters
Given that W < R (ie: 13.6 < 16.92) we do not reject the null hypothesis at the 5% significance level.
2.27 A set of 40 high-efficiency pumps is tested, all of the pumps fail (F = 40) after 400 pump-hours (T =
400). It is believed that the time to failure of the pumps follows an exponential distribution. Using
the following table and the goodness-of-fit method, determine if the exponential distribution is a good
choice.
In order to test whether the Exponential distribution is an adequate fit in this case, we use a Chi
Squared Goodness-of-Fit test where
Page 19 of 229
The Chi Squared Statistic is calculated for each data set given above using the following process.
The complete solution is shown in the table below.
𝑛 40
𝜆= = = 𝟎. 𝟏 𝒉𝒓−𝟏
𝑇 400
2
(𝐹𝑟𝑒𝑞𝑜𝑏𝑠 − 𝐹𝑟𝑒𝑞𝑒𝑥𝑝 ) (6 − 7.25)2
Chisq Stat = = = 𝟎. 𝟐𝟏𝟓𝟖
𝐹𝑟𝑒𝑞𝑒𝑥𝑝 7.25
2 (𝑘
𝑅 ≥ 𝜒1−𝛼 − 𝑚 − 1) where k = # of intervals & m = number of unknown parameters
Given that W < R (ie: 1.193 < 9.49) we do not reject the null hypothesis at the 5% significance level.
2.28 Use Eqn 2.74 and calculate mean and variance of a Weibull distribution.
Page 20 of 229
Making a change of variable, we get:
1
𝑡 𝛽 𝑑𝑢 𝛽 𝑡 𝛽−1
𝑢 = (𝛼) => 𝑡 = 𝛼𝑢𝛽 𝑑𝑡
= 𝛼 (𝛼)
∞
1 𝛽 𝑡 𝛽−1 −𝑢 𝛼 𝛼 𝛽−1
𝜇 = 𝐸(𝑡) = ∫ 𝛼𝑢𝛽 ( ) 𝑒 ( ) 𝑑𝑢
𝛼 𝛼 𝛽 𝑡
0
Page 21 of 229
Page 22 of 229
2.29 Consider the following repair times
Given Data
Time Range (hrs) Freq
0 4 17
4 24 41
24 72 12
72 300 7
300 5400 9
Use the Chi-square goodness-of-fit test to determine the adequacy of a lognormal distribution with
known parameters 𝜇𝑡 = 2.986 & 𝜎𝑡 = 1.837
In order to test whether the Lognormal distribution is an adequate fit in this case, we use a Chi
Squared Goodness-of-Fit test where
a) Ho: the r.v (T) follows a Lognormal distribution with parameters 𝜇𝑡 = 2.986 & 𝜎𝑡 = 1.837; or
The Chi Squared Statistic is calculated for each data set given above using the following process.
The complete solution is shown in the table below.
2
(𝐹𝑟𝑒𝑞𝑜𝑏𝑠 − 𝐹𝑟𝑒𝑞𝑒𝑥𝑝 ) (17 − 16.5)2
Chisq Stat = = = 𝟎. 𝟎𝟏
𝐹𝑟𝑒𝑞𝑒𝑥𝑝 16.5
Page 23 of 229
2 (𝑘
𝑅 ≥ 𝜒1−𝛼 − 𝑚 − 1) where k = # of intervals & m = number of unknown parameters
2 (5
𝑅 ≥ 𝜒1−𝛼 − 0 − 1)
2 (4)
𝑅 ≥ 𝜒1−𝛼
Given that W > R (ie: 11.98 > 9.49) we reject the null hypothesis at the 5% significance level.
Given that W < R (ie: 11.98 < 13.28) we do not reject the null hypothesis at the 1% significance level.
2.30 Consider the following time to failure data with the ranked value of t i. Test the hypothesis that the
data fit a normal distribution. (Use the Kolmogorov test for this purpose.)
b. A K-S goodness to fit test was used where the rejection region (R) was defined as 𝑅 ≥
𝐷𝑛 (𝛼) with a 5% significance level (ie: 0.05 )
Therefore Dn ( ) was found to be 0.409 from tables of Critical value for the Kolmogonov-Smirnov
statistic and the Maximum K-S statistic for the data set was calculated as shown in the table below to be
Max K-S = 0.2294
Page 24 of 229
𝑛
1 1
𝑥̅ = ∑ 𝑥𝑖 = (10.3 + 12.4 + 13.7 + ⋯ + 16.1) = 𝟏𝟒
𝑛 10
𝑖=1
𝑛
1 1
𝑆2 = ∑(𝑥𝑖 − 𝑥̅ )2 = [(10.3 − 14)2 + (12.4 − 14)2 + ⋯ + (16.1 − 14)2 ] = 𝟐. 𝟖𝟒
(𝑛 − 1) 9
𝑖=1
{ 1 𝑡(𝑛) ≤ 𝑡 < ∞
As the Max (K-S) Value < R, (ie: 0.2294 < 0.409), we do not reject the Ho: Hypothesis that a Normal
Distribution is an appropriate model for these data at the 95% confidence limit.
2.31 If a device has a cycle-to-failure, t, which follows an exponential distribution with = 0.003
failures/cycle.
b) If the device is used in a space experiment and is known to have survived for 300 cycles, what is
the probability that it will fail sometimes after 1000 cycles?
Page 25 of 229
2.32 A highly stressed machine part’s time to failure is normally distributed with 90% of the failures
occurring between 200 and 270 hrs of use (ie: 5% below 200 hrs and 5% above 270 hrs)
𝑇 − 𝑥̅ 200 − 235
𝑃𝑟 (𝑍 < ) = 𝑃𝑟 (𝑍 < )
𝜎 𝜎
b) Determine the part’s life if no more than 1% replacement probability can be tolerated.
𝑇0.01 − 235
−2.326 = => 𝑇0.01 = −2.326 × 21.28 + 235 = 𝟏𝟖𝟓. 𝟓 𝒉𝒓𝒔
21.28
c) Compute the probability that the part functions for an additional 50 hrs, if the part has been
operating for 200 hrs without failure.
250 − 235
1 − Pr(𝑇 < 250 ) 1 − Pr (𝑍 < 21.28 )
= = 𝟎. 𝟐𝟓𝟑𝟏
1 − Pr(𝑇 < 200 ) 1 − Pr (𝑍 < 200 − 235 )
21.28
Page 26 of 229
2.33 Fifty compressors were stress tested for 200 days and the following failure times grouped were
regarded. Perform a χ2 goodness-of-fit test at the 5% significance level to determine the adequacy
of an exponential distribution
0 < T < 50 9
50 < T < 75 11
Firstly, the mean and standard deviation of the sample needs to be generated:
𝑛
1 1
𝑥̅ = ∑ 𝑥𝑖 = × (9 × 25 + 11 × 62.5 + 9 × 87.5 + 12 × 125 + 9 × 175) = 95.5
𝑛 50
𝑖=1
𝑛
2
1 1
𝜎 = ∑(𝑥𝑖 − 𝑥̅ )2 = × [9 × (25 − 95.5)2 + 11 × (62.5 − 95.5)2 … ]
𝑛−1 49
𝑖=1
∴ 𝜎 = 50.429
Then find Pr(T) for each of the intervals (shown in table below) using the normal distribution table.
𝑒𝑖 = 𝑛 × 𝑝𝑖
Page 27 of 229
This information is shown in the following table:
Interval
#
Lower Upper Failures Pr (T) e_i Chi^2
0 50 9 0.1543 1.3890 41.7051
50 75 11 0.1587 1.7460 49.0491
75 100 9 0.1934 1.7403 30.2837
100 150 12 0.3245 3.8944 16.8701
150 200 9 0.1208 1.0871 57.5981
Total 50 9.8568 195.5061
To find R from the chi squared table, we need the number of intervals (k) and the number of
unknowns (m). In this problem, k = 5, and m = 2.
2 (𝑘 2 (2)
∴ 𝑅 = 𝜒1−𝛼 − 𝑚 − 1) = 𝜒0.95 = 5.99
In this case, clearly W (195.5061) > R (5.99). Therefore we reject the normal distribution as a
good fit for this data.
2.34 Perform a 𝜒 2 goodness-of-fit test at the 10% significance level on the following grouped data to fit
a uniform PDF of time to failure between two limits of a = 0 and b = 1000 hrs.
In order to test whether the uniform distribution is an adequate fit in this case, we use a Chi Squared
Goodness-of-Fit test where
The Chi Squared Statistic is calculated for each data set given above using the following process.
The complete solution is shown in the table below.
1
Pr(𝑋) = 1000 = 0.001 Pr(X)
9
h = 0.001
X Page 28 of 229
0 1000
𝐹𝑟𝑒𝑞𝑒𝑥𝑝 (0 < 𝑋 < 200) = N × Pr(0 < 𝑋 < 200) = 106 × 0.001 × (200 − 0) = 𝟐𝟏. 𝟐
2
(𝐹𝑟𝑒𝑞𝑜𝑏𝑠 − 𝐹𝑟𝑒𝑞𝑒𝑥𝑝 ) (16 − 21.2)2
Chisq Stat = = = 𝟏. 𝟐𝟖
𝐹𝑟𝑒𝑞𝑒𝑥𝑝 21.2
Interval Freq Pr(x) Ex(F) Chisq Stat
0 200 16 0.001 21.200 1.28
200 400 18 0.001 21.200 0.48
400 600 25 0.001 21.200 0.68
600 800 23 0.001 21.200 0.15
800 1000 24 0.001 21.200 0.37
Total 106 W 2.96
R 7.78
2 (𝑘
𝑅 ≥ 𝜒1−𝛼 − 𝑚 − 1) where k = # of intervals & m = number of unknown parameters
Given that W < R (ie: 2.96 < 7.78) we do not reject the null hypothesis at the 10% significance level.
2.35 A turbine in a generator can cause a great deal of damage if it fails critically. The manager of the
plant can choose to either replace the old turbine with a new one, or to leave it in place.
The state of the turbine is categorized as either ‘good,’ ‘acceptable’ or ‘poor.’ The probability of
critical failure per quarter depends on the state of the turbine:
The technical department has made a model for the degradation of the turbine. In this model it is
assumed that, if the state is ’good’ or ‘acceptable,’ then the at the end of the quarter it stays the same
with probability 0.95 or degrades with probability 0.05 (ie: ‘good’ becomes ‘acceptable’ and
‘acceptable’ becomes ‘poor’). If the state is ‘poor’ then it stays ‘poor.’
Determine the probability that a turbine which is ‘good’ becomes ‘good,’ ‘acceptable’ and ‘poor’ in
the next three quarters and does not fail.
Page 29 of 229
2.36 Assume there are, on the average, 3 random intermittent interruptions in a network service per
week. What is the probability of at least one interruption in a two-week period?
Solution:
This problem can be solved using the Poisson distribution
𝜇 𝑥 𝑒 −𝜇
Pr(𝑋 = 𝑥) =
𝑥!
And
𝜇 = 𝜆𝑡
Given:
𝜆=3
𝑡=2
𝜇=6
Find:
Pr(𝑋 ≥ 1)
Therefore:
60 𝑒 −6
Pr(𝑋 ≥ 1) = 1 − Pr(𝑋 = 0) = 1 − = 1 − 𝑒 −6 = 1 − 0.002479 = 0.9975
0!
Page 30 of 229
2.37 The life of a light bulb is log-normally distributed with mean of 12,000 hr. and standard deviation
of 2000 hr. Compute the probability that:
a. A light bulb will last 12,000 hrs
b. A light that has already worked for 10,000 hrs, will last 12,000 hrs.
Solution:
a. Given:
μ𝑦 = 12,000
σ𝑦 = 2,000
μ𝑦 12000
μ𝑡 = ln = ln = 9.3790
σ2𝑦 20002
√1 + ( )
√1 + ( 2 ) [ 120002 ]
[ μ𝑦 ]
σ2𝑦 20002
σ𝑡 = √ln (1 + ) = √ln (1 + ) = 0.1655
μ2𝑦 120002
Find:
Pr(𝑇 > 12,000)
ln(𝑇) − μ𝑡 ln(12000) − 9.3790
𝑧= = = 0.082763
σ𝑡 0.1655
b. Find:
Pr(𝑇 = 12,000) × Pr(T > 10,000|T = 12,000)
Pr(𝑇 = 12,000|𝑇 > 10,000) =
Pr(𝑇 > 10000)
Note, logically:
Pr(𝑇 > 10,000|𝑇 = 12,000) = 1
Find:
Pr(𝑇 > 10,000)
ln(𝑇) − μ𝑡 ln(10000) − 9.3790
𝑧= = = −1.0187
σ𝑡 0.1655
Page 31 of 229
Solutions to Chapter 3
Exercises
3.1 For a gamma distribution with the scale parameter of 400, and the shape parameter of 3.8, determine
Pr(x < 200).
3.2 Time to failure of a relay follows a Weibull distribution with = 10 years, = 0.5. Find the following:
b) The MTTF
1 1
𝑀𝑇𝑇𝐹 = 𝛼Г (1 + ) = 10Г (1 + ) = 𝟐𝟎 𝒚𝒓𝒔
𝛽 3.8
𝑡 𝑡
1 𝑡
𝑅(𝑡) = 𝑒𝑥𝑝 [− ∫ ℎ(𝑡)𝑑𝑡] = 𝑒𝑥𝑝 [− ∫ 𝑑𝑡] = 𝑒𝑥𝑝[−2√𝑡]
√𝑡 0
0 0
𝑅(𝑡) = 𝒆𝒙𝒑[−𝟐√𝒕]
c) MTTF
∞ ∞
Page 32 of 229
𝑑𝑢 1
Let 𝑢 = √𝑡 => 𝑑𝑡
=2
√𝑡
Therefore:
∞
−2𝑢 −2𝑢
1 ∞ −2√𝑡
1 ∞ 𝟏
𝑀𝑇𝑇𝐹 = ∫ 𝑒 ∙ 2𝑢 ∙ 𝑑𝑢 = [−𝑒 (𝑢 + )] = [−𝑒 (√𝑡 + )] =
2 0 2 0 𝟐
0
d) Variance
𝜎 2 = 𝐸(𝑡 2 ) − [𝐸(𝑡)]2
∞ ∞ ∞
2) 2 2
1 3
𝐸(𝑡 = ∫ 𝑡 𝑓(𝑡)𝑑𝑡 = ∫ 𝑡 𝑒 −2√𝑡 𝑑𝑡 = ∫ 𝑡 2 𝑒 −2√𝑡 𝑑𝑡
√𝑡
0 0 0
Therefore:
∞ ∞
2) 3 −2𝑢
𝐸(𝑡 =∫𝑢 𝑒 ∙ 2𝑢 ∙ 𝑑𝑢 = 2 ∫ 𝑢4 𝑒 −2𝑢 ∙ 𝑑𝑢
0 0
3 ∞ 𝟑
𝐸(𝑡 2 ) = [−𝑒 −2𝑢 (𝑢4 + 2𝑢3 + 3𝑢2 + 3𝑢 + )] =
2 0 𝟐
3 1 2 𝟓
𝜎2 = −( ) =
2 2 𝟒
3.4 Assume that 100 components are placed on test for 1,000 hours. From previous testing, we believe that
the hazard rate is constant, and the MTTF = 500 hours. Estimate the number of components that will
fail in the time interval of 100 to 200 hours. How many components will fail if it is known that 15
components failed in T < 100 hours?
1 1
𝑀𝑇𝑇𝐹 = 500 = => 𝜆 = = 0.002 ℎ𝑟1
𝜆 500
Therefore, 15 components are expected to fail in the 100-200 hour time interval.
Knowing that 15 units failed before T=100 does not change the previous result as the exponential
distribution is memory less.
Page 33 of 229
3.5 Assume that t, the random variable that denotes life in hours of a specified component, has a cumulative
density function (cdf) of
100
1 t 100
Ft t
0 t 100
Determine the following
a) PDF f(t)
100
𝑓(𝑡) = { 𝑡 2 𝑡 ≥ 100
0 𝑡 < 100
100
𝑅(𝑡) = 1 − 𝐹(𝑡) = 1 − { 1 − 𝑡 𝑡 ≥ 100
0 𝑡 < 100
100
𝑅(𝑡) = { 𝑡 𝑡 ≥ 100
1 𝑡 < 100
c) MTTF (Using a practical upper limit of 1 million hrs to avoid trivial solution)
∞ 1𝑒6
100
𝑀𝑇𝑇𝐹 = ∫ 𝑅(𝑡)𝑑𝑡 = ∫ 𝑑𝑡 = [100ln(𝑡)]1𝑒6
1𝑒2 = 𝟗𝟐𝟏𝒉𝒓𝒔
𝑡
0 100
3.6 Show whether a uniform distribution represents an increasing failure rate, decreasing failure rate, or
constant failure rate.
The uniformly distributed density and cumulative functions are shown below.
Page 34 of 229
From these figure we get:
𝑏 𝑏
1 𝑏−𝑡
𝑅(𝑡) = ∫ 𝑓(𝑡)𝑑𝑡 = ∫ 𝑑𝑡 = ( )
𝑏−𝑎 𝑏−𝑎
𝑡 𝑡
𝑡 𝑡
1 𝑡−𝑎
𝐹(𝑡) = ∫ 𝑓(𝑡)𝑑𝑡 = ∫ 𝑑𝑡 = ( )
𝑏−𝑎 𝑏−𝑎
𝑎 𝑎
𝑓(𝑡) 1 𝑏−𝑡 1
𝜆(𝑡) = =( )÷( )=( )
𝑅(𝑡) 𝑏−𝑎 𝑏−𝑎 𝑏−𝑡
The following plot of the hazard rate indicates that it is an increasing failure rate function.
𝑓(𝑡) 𝑓(𝑡)
ℎ(𝑡) = = ∞
𝑅(𝑡) ∫ 𝑓(𝑡)𝑑𝑡
𝑡
2𝑡 𝑡2 2𝑡 𝑡2
2 exp (− 2 ) 2 exp (− 2 )
𝛼 𝛼 𝛼 𝛼
ℎ(𝑡) = =
∞ 2𝑡 𝑡2 𝑡2
∫𝑡 2 exp (− 2 ) 𝑑𝑡 exp (− 2 )
𝛼 𝛼 𝛼
2𝑡
ℎ(𝑡) =
𝛼2
Page 35 of 229
b) Find the Reliability function R(t)
2𝑡 𝑡2
𝑓(𝑡) 𝛼 2 exp (− 𝛼 2 ) 𝑡2
𝑅(𝑡) = = = exp (− 2 )
𝜆(𝑡) 2𝑡 𝛼
𝛼 2
𝑡2
𝑅(𝑡) = exp (− )
𝛼2
ax 2 1
c) Find the MTTF. Notice: 0 e
2 a
∞ ∞
𝑡2 1
𝑀𝑇𝑇𝐹 = ∫ 𝑅(𝑡)𝑑𝑡 = ∫ exp (− 2 ) 𝑑𝑡 = √𝜋𝛼 2
𝛼 2
0 0
𝛼
𝑀𝑇𝑇𝐹 = √𝜋
2
3.8 Due to the aging process, the failure rate of a non-repairable item is increasing according to (t) =
t-1. Assume that the value of and are estimated as ˆ 1.62 and ˆ 1.2 10 5 hour. Determine
the probability that the item will fail sometime between 100 and 200 hours. Assume an operation
beginning immediately after the onset of aging.
𝑡 𝑡
Page 36 of 229
3.9 Suppose r.v. X has the exponential PDF f(x) = exp[-x], for x > 0, and f(x) = 0, for x 0. Find Pr(x
> a + b | x > a) given a, b > 0.
3.10 The following time to failure data are found when 158 transformer units are put under test. Use a non-
parametric method to estimate f(t), h(t), and R(t) of the transformers. No failures are observed prior
to 1,750 hours.
Time Interval Freq
0 2250 17
2250 2750 54
2750 3250 27
3250 3750 17
3750 4250 19
4250 5000 24
Using Equations 3.54 through to 3.56 we create the following table and plots for the data set given
above:
Time Interval Freq Ns(ti) h(ti) R(ti) f(ti)
0 2250 17 158 4.78E-05 1.00E+00 4.78E-05
2250 2750 54 141 7.66E-04 8.92E-01 6.84E-04
2750 3250 27 87 6.21E-04 5.51E-01 3.42E-04
3250 3750 17 60 5.67E-04 3.80E-01 2.15E-04
3750 4250 19 43 8.84E-04 2.72E-01 2.41E-04
4250 4750 24 24 2.00E-03 1.52E-01 3.04E-04
Page 37 of 229
3.11 A test was run on 10 electric motors under high temperature. The test was run for 60 hours, during
which six motors failed. The failures occurred at the following times: 37.5, 46.0, 48.0, 51.5, 53.0, and
54.5 hours. We don't know whether an exponential distribution or a Weibull distribution model is
better for representing these data. Use the plotting method as the main tool to discuss the
appropriateness of these two models.
Solution
Using Equations 3.51 through to 3.53 we create the following table and plots for the data set given
above:
From the Weibull plot shown below, β was calculated to be greater than 7 therefore; an exponential
distribution does not appear to be a good fit for this data set.
Also, looking at the data set as represented in the hazard function plot shown below, we can see that
the hazard rate increased significantly in the last 20 hours of operation. This also supports the finding
that an exponential model is not an appropriate model for the given data set as it appears to be an IFR
data set.
Page 38 of 229
3.12 A test of 25 integrated circuits over 500 hours yields the following data:
Plot the PDF, hazard rate, and reliability function for each interval of these integrated circuits using a
non-parametric method.
Solution
Using Equations 3.54 through to 3.56 we create the following table and plots for the data set given
above:
0.004000
0.010000 1.000000
0.003500
0.002500
R(t)
f(t)
0.006000 0.600000
0.002000
0.001000
0.002000 0.200000
0.000500
0.000000 0.000000 0.000000
100 200 300 400 500 100 200 300 400 500 100 200 300 400 500
Tim e (hrs) Tim e (hrs) Tim e (hrs)
The hazard rate appears to be fairly constant until approximately 350 hours have elapsed and then it appears
to increase quite rapidly. This may be indicative of the wear out region beginning at that time or some other
failure mode due to an external failure mechanism.
More information would be required to reduce the uncertainty in the data presented and draw more accurate
conclusions on the failure rate.
Page 39 of 229
3.13 Total test time of a device is 50,000 hours. The test is terminated after the first failure. If the PDF of
the device time-to-failure is known to be exponentially distributed, what is the probability that the
estimated failure rate is not greater than 4.610-5 per hour.
𝑇 = 50000
𝑟=1
𝜆𝑢𝑝𝑝𝑒𝑟 ≤ 4.6 × 10−5 ℎ𝑟 −1
We now need to determine the one sided upper confidence limit for a Type II – Failure Terminated
Test.
From Table 3.2; we get:
2
𝜒1−𝛼 (2𝑟)
𝑃𝑟 [𝜆𝑢𝑝𝑝𝑒𝑟 ≤ ]=1−𝛼
2𝑇
2
𝜒1−𝛼 (2𝑟)
𝜆𝑢𝑝𝑝𝑒𝑟 ≤
2𝑇
2
2𝑇𝜆𝑢𝑝𝑝𝑒𝑟 ≤ 𝜒1−𝛼 (2𝑟)
2
2 × 50000 × 4.6 × 10−5 ≤ 𝜒1−𝛼 (2 × 1)
2
4.6 ≤ 𝜒1−𝛼 (2)
This equation can now be solved numerically or by looking at the Chi Squared table at Appendix A:
Table A.3
Using the tables, we look at row 2 (due to the fact we have 2 degrees of freedom).
Move along the columns until we find a value greater than or equal to 4.6
2 2
From there, read the 𝜒1−𝛼 value at the top of the column which in this case is 𝜒0.90
Therefore, there is a 90% chance that the failure rate 𝜆𝑢𝑝𝑝𝑒𝑟 ≤ 4.6 × 10−5 ℎ𝑟 −1
3.14 A manufacturer uses exponential distribution to model number "cycle-to-failure" of its products. In
this case, r.v. T in the exponential PDF represents the number of cycles to failure = 0.003 f/cycle.
b) If a component survives for 300 cycles, what is the probability that it will fail sometime after
500 cycles? Accordingly, if 1,000 components have survived 300 cycles, how many would
one expect to fail after 500 cycles?
Page 40 of 229
Pr(𝑇 > 500 ) 𝑒 −500×0.003
Pr(𝑇 > 500 |𝑇 > 300) = = = 𝟎. 𝟓𝟒𝟗
Pr(𝑇 > 300 ) 𝑒 −300×0.003
3.15 The shaft diameters in a sample of 25 shafts are measured. The sample mean of diameter is 0.102
centimeter, with a standard deviation of 0.005 centimeter. What is the upper 95% confidence limit on
the mean diameter of all shafts produced by this process, assuming the distribution of shaft diameters
is normal?
Using equation 2.135, we find the 95% confidence intervals for the mean as follows:
𝑆 𝑆
[𝑥̅ − 𝑡1−𝛼/2 ≤ 𝜇 ≤ 𝑥̅ + 𝑡1−𝛼/2 ] = 0.95
√𝑛 √𝑛
0.005 0.005
[0.102 − 𝑡 0.05 ≤ 𝜇 ≤ 0.102 + 𝑡 0.05 ]
1− √25 1− √25
2 2
3.16 The sample mean life of 10 car batteries is of 102.5 months, with the standard deviation of 9.45
months. What are the 80% confidence limits for the mean and standard deviation of a PDF that
represents these batteries?
Assuming a Normal life distribution we find the 80% confidence intervals for the mean using equation
2.135, as follows:
𝑆 𝑆
[𝑥̅ − 𝑡1−𝛼/2 ≤ 𝜇 ≤ 𝑥̅ + 𝑡1−𝛼/2 ] = 0.80
√𝑛 √𝑛
9.45 9.45
[102.5 − 𝑡1−0.2 ≤ 𝜇 ≤ 102.5 + 𝑡
1−
0.2 ]
2 √10 2 √10
Using equation 2.136, we find the 80% confidence intervals for the standard deviation as follows:
(𝑛 − 1)𝑆 2 (𝑛 − 1)𝑆 2
[√ 2 ≤𝜎≤√ 2 ] = 0.80
𝜒1−𝑎⁄2 (𝑛 − 1) 𝜒𝑎⁄2 (𝑛 − 1)
Page 41 of 229
3.17 The breaking strength X of 5 specimens of a rope of 1/4 inch diameter are 660, 460, 540, 580, and 550
lbs. Estimate the following:
a) The mean breaking strength by a 95% confidence level assuming normally distributed strength.
𝑛
1 1
Sample Mean 𝑥̅ = ∑ 𝑋𝑖 = (660 + 460 + 540 + 580 + 550 ) = 558
𝑛 5
𝑖=1
𝑛
1 2
Sample Variance 𝑆 = ∑(𝑥𝑖 − 𝑥̅ )2 = 5520
𝑛−1
𝑖=1
The confidence interval for the true mean (μ) is found using Eqn 2.135 as follows:
𝑆 𝑆
[𝑥̅ − 𝑡1−𝛼 ≤ 𝜇 ≤ 𝑥̅ + 𝑡1−𝛼 ] = 0.95
2 √𝑛 2 √𝑛
72.25 72.25
[558 − 𝑡0.975 ≤ 𝜇 ≤ 558 + 𝑡0.975 ]
√5 √5
b) The point estimate of strength value at which only 5% of such specimens would be expected to
break if (μ) is assumed to be an unbiased estimate of the true mean, and (𝑆 2 ) is assumed to be
the true standard deviation. (Assume x is normally distributed.)
If the r.v of breaking strength (X) is assumed to be normally distributed with these parameters,
than the 5th percentile is calculated by finding the cumulative failure function value for 5%:
𝑋0.05 − 558
𝑍0.05 = 𝛷𝑆−1 (0.05) = −1.645 =
72.25
The confidence interval for the true Standard Deviation (σ) is found using Eqn 2.136 as follows:
(𝑛 − 1)𝑆 2 (𝑛 − 1)𝑆 2
[√ 2 ≤𝜎≤√ 2 ] = 0.9
𝜒1−𝑎⁄2 (𝑛 − 1) 𝜒𝑎⁄2 (𝑛 − 1)
Page 42 of 229
Time (hrs) 0.4 1 2 5 8 12 25
Number of Failures 1 3 5 15 20 30 50
The trend line shown in the plot above is given by the following expression.
1
𝑙𝑛 [𝑙𝑛 ( )] = 𝛽 ln(𝑡) − 𝛽ln(𝛼)
𝑅(𝑡)
Using this expression and the trend line value given in the plot above
𝑦 = 1.564𝑥 − 3.803 the shape parameter 𝛽 is given by:
𝛽 = 1.564
1
𝑙𝑛 [𝑙𝑛 ( )] = 𝛽 ln(𝑡) − 𝛽ln(𝛼)
𝑅(𝑡)
Using this expression and the trend line value given in the plot above
𝑦 = 1.564𝑥 − 3.803 the scale parameter 𝛼 is given by:
Page 43 of 229
𝛼 = 11.38
The exponential and lognormal distributions can also be used to model this data set as the shape
parameter is close to unity. However, as shown in the plots below; neither of these distributions
provides as good a fit as the Weibull Distribution.
3.19 Seven pumps have failure times (in months) of 15.1, 10.7, 8.8, 11.3, 12.6, 14.4, and 8.7. (Assume an
exponential distribution.)
𝑇 81.6
̂ =
𝑀𝑇𝑇𝐹 = = 11.7 𝑚𝑜𝑛𝑡ℎ𝑠
𝑟 7
12
𝑅(𝑇 = 12) = 𝑒 −𝜆𝑡 = 𝑒 −11.7 = 0.357
This is a Type II Failure Terminated case. Therefore, 95% two-sided interval of is found using
Table 3.2 as follows:
2 2
𝜒𝛼/2 (2𝑟) 𝜒1−𝛼/2 (2𝑟)
𝑃𝑟 [ <𝜆< ]=1−𝛼
2𝑇 2𝑇
Page 44 of 229
2 2
𝜒0.025 (2 × 7) 𝜒0.975 (2 × 7)
<𝜆<
2 × 81.6 2 × 81.6
3.20 The average life of a certain type of small motor is 10 years, with a standard deviation of 2 years. The
manufacturer replaces free of charge all motors that fail while under warranty. If the manufacturer is
willing to replace only 3% of the motors that fail, what warranty period should be offered? Assume
the time to failure of the motors follows a normal distribution.
Solution
We want to find the time associated with the 3 percentile cumulative failure given the r.v for life (X)
of the small motors is normal distributed with 𝜇 = 10 and 𝜎 = 2.
Therefore, in order for the manufacturer to avoid replacing more than 3% of motor failures, the warranty
period should be set at ≤ 6.24 yrs.
3.21 A manufacturer claims that certain machine parts will have a mean diameter of 4 cm, with a standard
deviation of 0.01 mm. The diameters of five parts are measured and found to be (in mm): 39.98, 40.01,
39.96, 40.03, and 40.02. Would you accept this claim with a 90% confidence level?
𝑛
1 1
Sample Mean 𝑥̅ = ∑ 𝑋𝑖 = (39.98 + 40.01 + 39.96 + 40.03 + 40.02) = 40 𝑚𝑚
𝑛 5
𝑖=1
𝑛
1
Sample Standard Deviation (𝑆) = 𝑠𝑞𝑟𝑡 [ ∑(𝑥𝑖 − 𝑥̅ )2 ] = 0.0292 𝑚𝑚
𝑛−1
𝑖=1
The confidence interval for the true mean (μ) is found using Eqn 2.135 as follows:
𝑆 𝑆
[𝑥̅ − 𝑡1−𝛼 ≤ 𝜇 ≤ 𝑥̅ + 𝑡1−𝛼 ] = 0.9
2 √𝑛 2 √𝑛
0.0292 0.0292
[40 − 𝑡0.95 ≤ 𝜇 ≤ 40 + 𝑡0.95 ]
√5 √5
The confidence interval for the true Standard Deviation (σ) is found using Eqn 2.136 as follows:
(𝑛 − 1)𝑆 2 (𝑛 − 1)𝑆 2
[√ 2 ≤𝜎≤√ 2 ] = 0.9
𝜒1−𝑎⁄2 (𝑛 − 1) 𝜒𝑎⁄2 (𝑛 − 1)
Page 45 of 229
(4)8.5 × 10−4 (4)8.5 × 10−4
[√ 2 (4) ≤ 𝜎 ≤ √ 2 (4) ]
𝜒0.95 𝜒0.05
Therefore, the claim that the true mean is 4cm is supported at the 90% confidence interval however,
the claim that the true standard deviation is 0.01mm is not supported at the 90% confidence interval.
3.22 You are to design a life test experiment to estimate the failure rate of a new device. Your boss asks
you to make sure that the 80% upper and lower limits of the estimate interval (two-sided) do not differ
by more than a factor of 2. Due to cost constraints, the components will be tested until they fail.
Determine how many components should be put on test.
Using a Type II Failure Terminated approach, we can find the number of components (𝑟) required as
follows:
2
𝜒1− 𝛼 (2𝑟)
2 2
𝜆𝑢𝑝𝑝𝑒𝑟 2𝑇 𝜒1−𝛼/2 (2𝑟)
≤2 => ≤ 2 => ≤2
𝜆𝑙𝑜𝑤𝑒𝑟 𝜒𝛼2 (2𝑟) 2
𝜒𝛼/2 (2𝑟)
2
2𝑇
3.23 For an experiment, 25 relays are allowed to run until the first failure occurred at t = 15 hours. At this
point, the experimenters decide to continue the test for another 5 hours. No failures occur during this
extended period, and the test is terminated. Using the 90% confidence level, determine the following:
𝑇 495
̂ =
𝑀𝑇𝑇𝐹 = = 495 ℎ𝑟𝑠
𝑟 1
Using a Type I Time Terminated Test approach, we can find the two sided interval for MTTF as
follows:
Page 46 of 229
2𝑇 2𝑇
𝑃𝑟 [ < 𝑀𝑇𝑇𝐹 < ] =1−𝛼
𝜒 2 𝛼 (2𝑟 + 2) 2
𝜒𝛼/2 (2𝑟)
1−
2
2 × 495 2 × 495
2 < 𝑀𝑇𝑇𝐹 < 2
𝜒0.95 (4) 𝜒0.05 (2)
Using a Type I Time Terminated Test approach, we can find the two sided interval for Reliability
as follows:
2
𝜒1− 𝛼 (2𝑟 + 2) 2
𝜒𝛼/2 (2𝑟)
2
𝑃𝑟 [𝑒𝑥𝑝 [− 𝑡𝑜 ] < 𝑅(𝑡𝑜 ) < 𝑒𝑥𝑝 [− 𝑡𝑜 ]] = 0.90
2𝑇 2𝑇
2 2
𝜒0.95 (4) 𝜒0.1 (2)
𝑒𝑥𝑝 [− × 25] < 𝑅(25) < 𝑒𝑥𝑝 [− × 25]
2 × 495 2 × 495
3.24 A locomotive control system fails 15 times out of the 96 times it is activated to function. Determine
the following:
a) A point estimate for failure probability of the system
In this situation, control system failures can be modeled using the binomial distribution where
the probability of failure is constant. Therefore, the best estimate for p in this case is found
using Equation 3.97 as follows:
15
𝑝̂ = 96 = 0.156
b) 95% two-sided confidence intervals for the probability of failure. (Assume that after each
failure, the system is repaired and put back in an as-good-as-new state.)
The 95% confidence interval for the estimated probability can be found using the Clopper-
Pearson procedure using Equations 3.98 & 3.99 as follows:
Page 47 of 229
𝑝𝑙𝑜𝑤𝑒𝑟 = 0.09
𝟎. 𝟎𝟗 ≤ 𝒑 ≤ 𝟎. 𝟐𝟒𝟓
3.25 A sample of 10 measurements of a sphere diameter gives a mean of 4.38 inches, with a standard
deviation of 0.06 inch. Find the 99% confidence limits of the actual mean and standard deviation.
The confidence interval for the true mean (μ) is found using Eqn 2.135 as follows:
𝑆 𝑆
[𝑥̅ − 𝑡1−𝛼 ≤ 𝜇 ≤ 𝑥̅ + 𝑡1−𝛼 ] = 0.99
2 √𝑛 2 √𝑛
0.06 0.06
[4.38 − 𝑡0.995 ≤ 𝜇 ≤ 4.38 + 𝑡0.995 ]
√10 √10
𝟒. 𝟑𝟏𝟖 𝒊𝒏 ≤ 𝝁 ≤ 𝟒. 𝟒𝟒𝟐 𝒊𝒏
The confidence interval for the true Standard Deviation (σ) is found using Eqn 2.136 as follows:
(𝑛 − 1)𝑆 2 (𝑛 − 1)𝑆 2
[√ 2 ≤𝜎≤√ 2 ] = 0.9
𝜒1−𝑎⁄2 (𝑛 − 1) 𝜒𝑎⁄2 (𝑛 − 1)
Page 48 of 229
3.26 A sample of 7 measurements is taken from a study of an industrial process, which is assumed to follow
a normal distribution. For this sample, the 95% confidence error on estimating the mean () is 2.2.
What sample size should be taken if we want the 99% confidence error to be 1.5, assuming the same
sample variance?
For this sample, the 95% confidence error was calculated to be 2.2. From this information, we can
calculate the sample standard deviation (S) as follows:
𝑆 2.2√7
Confidence Error = |𝑡1−𝛼/2 | = 2.2 Therefore 𝑆 = = 2.379
√𝑛 𝑡0.975
If we want to determine the sample size such that, the 99% confidence error is ≤ 1.5, we can do this
by iteratively increasing 𝑛 and the DOF of the 𝑇 𝑑𝑖𝑠𝑡 (𝑛 − 1) until we achieve a confidence error ≤
1.5
Applying this approach, the sample size (𝑛) required to achieve this level of confidence error was
calculated to be 𝑛 ≥ 21.
𝑆
99% Confidence Error = |𝑡1−𝛼 | ≤ 1.5
2 √𝑛
2.379
𝑡0.995 = 1.48
√21
3.27 Suppose the generic failure rate of a component corresponding to an exponential time to failure model
is 𝑔 = 10−3 ℎ𝑟 −1 with a standard deviation of 𝑔 /2 . Assume that ten components are closely
observed for 1500 hours and one failure is observed. Using the Bayesian method, calculate the mean
and variance of from the posterior distribution. Calculate the 90 percent lower confidence limit.
𝐿(𝐷𝑎𝑡𝑎|𝜆)𝜋0 (𝜆)
𝜋(𝜆|𝐷𝑎𝑡𝑎) =
∫ 𝐿(𝐷𝑎𝑡𝑎|𝜆)𝜋0 (𝜆)𝑑𝜆
Prior:
Assume that the prior failure rate is Normally Distributed as follows:
𝜋0 (𝜆)~𝑁(𝜆̂ = 0.001; 𝜎̂ = 0.0005)
Likelihood:
The observed evidence can be modeled by a Poisson distribution as follows.
(𝜆𝑇)𝑥 𝑒 −𝜆𝑇
𝐿(𝐷𝑎𝑡𝑎|𝜆) = = (1500𝜆)1 𝑒 −1500𝜆
𝑥!
Posterior:
The updated posterior distribution is therefore given by:
Page 49 of 229
F=@(x) 1500*x.*(exp(-x.*1500).*normPDF(x,0.001,0.0005));
denom = quadgk(F,0.00005,0.0035)
denom = 0.2881
The 90% Confidence Interval for the failure rate is given as:
𝟎. 𝟎𝟎𝟎𝟑𝟒 ≤ 𝝀 ≤ 𝟎. 𝟎𝟎𝟏𝟔𝟖
3.28 In the reactor safety study, the failure rate of a diesel generator can be described as having a lognormal
distribution with the upper and lower 90% bounds of 310-2 and 310-4 respectively. If a given nuclear
plant experiences 2 failures in 8,760 hours of operation, determine the upper and lower 90% bounds
given this plant experience. (Consider the reactor safety study values as prior information.)
𝐿(𝐷𝑎𝑡𝑎|𝜆)𝜋0 (𝜆)
𝜋(𝜆|𝐷𝑎𝑡𝑎) =
∫ 𝐿(𝐷𝑎𝑡𝑎|𝜆)𝜋0 (𝜆)𝑑𝜆
Prior:
Assume that the prior failure rate is Normally Distributed as follows:
𝜋0 (𝜆)~𝐿𝑁(𝜆̂𝑡 = −5.809; 𝜎̂𝑡 = 1.399)
Likelihood:
The observed evidence can be modeled by a Poisson distribution as follows.
(𝜆𝑇)𝑥 𝑒 −𝜆𝑇 (8760𝜆)2 𝑒 −8760𝜆
𝐿(𝐷𝑎𝑡𝑎|𝜆) = =
𝑥! 2!
Posterior:
The updated posterior distribution is therefore given by:
Page 50 of 229
p05=quadgk(F2,0.00003,0.0001197)
p05 = 0.0500
p95=quadgk(F2,0.00003,0.0007175)
p95 = 0.9500
The 90% Confidence Interval for the failure rate is given as:
𝟎. 𝟎𝟎𝟎𝟏𝟐 ≤ 𝝀 ≤ 𝟎. 𝟎𝟎𝟎𝟕𝟐
3.29 Five measurements of the breaking strength of a computer board were recorded as 0.28, 0.30, 0.27,
0.33, 0.31 Kgf. Find the point estimate and the 99% confidence intervals for the actual mean breaking
strength assuming the breaking strength is distributed exponentially.
This is a Type II (Failure Terminated Test) where Kgf is used instead of time.
Two sided 99% Confidence Interval for the actual Mean Breaking Stress (MBS)
3.30 The number of days in a 50 day period during which x failures of an assembly line is recorded as
follows. Use a Chi-square goodness of fit test to determine whether a Poisson distribution is a good
fit to these data. Perform the test at a 5% significance level.
In order to test whether the Poisson distribution is an adequate fit in this case, we use a Chi Squared
Goodness-of-Fit test where
c) Ho: the r.v (X) follows a Poisson distribution; or
The Chi Squared Statistic is calculated for each data set given above using the following process.
The complete solution is shown in the table below.
∑ 𝑥 ∙ 𝐹𝑟𝑒𝑞 (𝑥) 45
𝜇= = = 0.9
𝑇𝑜𝑡𝑎𝑙 𝐹𝑎𝑖𝑙𝑢𝑟𝑒𝑠 50
Page 51 of 229
(0.9)0 𝑒 −0.9
Pr(𝑋 = 0) = = 𝟎. 𝟒𝟎𝟔𝟔
0!
2
(𝐹𝑟𝑒𝑞𝑜𝑏𝑠 − 𝐹𝑟𝑒𝑞𝑒𝑥𝑝 ) (21 − 20.3)2
Chisq Stat = = = 𝟎. 𝟎𝟐𝟐
𝐹𝑟𝑒𝑞𝑒𝑥𝑝 20.3
2 (𝑘
𝑅 ≥ 𝜒1−𝛼 − 𝑚 − 1) where k = # of intervals & m = number of unknown parameters
Given that W < R (ie: 0.68 < 7.82) we do not reject the null hypothesis at the 5% significance level.
3.31 Fifty identical units of a manufactured product are tested for 300 hours, only one failure is observed
(the failed unit is replaced with a good one).
𝑟 1
𝜆̂ = = = 6.67 × 10−5 ℎ𝑟 −1
𝑇 50 × 300
b) Find the 90% confidence interval (two-sided) for the actual failure rate.
2
2
𝜒𝛼/2 (2𝑟) 𝜒1− 𝛼 (2𝑟 + 2)
2
𝑃𝑟 [ <𝜆< ] = 0.90
2𝑇 2𝑇
2 2
𝜒0.05 (2) 𝜒0.95 (4)
<𝜆<
30000 30000
Page 52 of 229
3.32 A mechanical life test of 18 circuit breakers of a new design was run to estimate the percentage failed
by 10,000 cycle of operation. Breakers were inspected on a schedule, and it is known that failures
occurred between certain inspections as shown:
Using linear regression of the trend line shown above, the percentage of failures by 10,000
cycles is found to be 5.24%
The trend line shown in the plot above is given by the following expression.
1
𝑙𝑛 [𝑙𝑛 ( )] = 𝛽 ln(𝐶𝑦𝑐) − 𝛽ln(𝛼)
𝑅(𝐶𝑦𝑐)
Using this expression and the trend line value given in the plot above
𝑦 = 2.3854𝑥 − 8.4138 the shape (𝛽) and scale (𝛼) parameters are given by:
Page 53 of 229
3.33 Fifty-eight fans in service are supposed to have an exponential life distribution with an MTTF of
28,700 hours. Assuming that a failed fan is replaced with a new one that does not fail; predict the
number of such fans that will fail in 2000 hours.
1 1
𝜆̂ = = = 3.48 × 10−5 ℎ𝑟 −1
𝑀𝑇𝑇𝐹 28700
̂
𝐹(𝑡) = 1 − 𝑒 −𝜆𝑡
−5
𝐹(2000) = 1 − 𝑒 −2000×3.48×10 = 0.0673
3.34 A manufacturer tests 125 high performance contacts and finds that 3 are defective.
In this situation, contact failures can be modeled using the binomially distributed where the
probability of failure is constant.
3
Therefore, the best estimate for p in this case 𝑝̂ = 125 = 0.024
b) What is the 90% confidence interval for the estimated probability in (a)?
The 90% confidence interval for the estimated probability can be found using the Clopper- Pearson
procedure as follows:
𝟎. 𝟎𝟎𝟔𝟔 ≤ 𝒑 ≤ 𝟎. 𝟎𝟔𝟎𝟗
Page 54 of 229
3.35 If the time-to-failure PDF of a component follows a linear model as follows,
c t 0 t 10,000
f t
0 otherwise
Determine
a) Reliability function.
Knowing that the total area under the PDF must equal unity, we can determine “c”
∞ 10000
10000
𝑐𝑡 2
[ ] = 1 => 𝒄 = 𝟐 × 𝟏𝟎−𝟖
2 0
𝑡
𝑐𝑡 2
𝑅(𝑡) = 1 − 𝑐 ∫ 𝑡𝑑𝑡 = 1 −
2
0
1 𝑡≤0
{ 0 𝑡 > 10000
0 𝑡≤0
𝑓(𝑡) 2 × 10−8 𝑡
ℎ(𝑡) = = 0 ≤ 𝑡 ≤ 10000
𝑅(𝑡) 1 − 1 × 10−8 𝑡 2
{ 0 𝑡 > 10000
Page 55 of 229
3.36 The cycle-to-failure T for a certain kind of component has the instantaneous failure rate (t) = 2.510-
t 0 per cycles. Find the MCTF (mean-cycle-to-failure), and the reliability of this component at
5 2
100 cycles.
𝑡 𝑡 𝑡
−5 2
2.5 × 10−5 𝑡 3
𝑅(𝑡) = 𝑒𝑥𝑝 [− ∫ ℎ(𝑡)𝑑𝑡] = 𝑒𝑥𝑝 [− ∫(2.5 × 10 𝑡 )𝑑𝑡] = 𝑒𝑥𝑝 [− ]
3 0
0 0
2.5 × 10−5 𝑡 3
𝑅(𝑡) = 𝑒𝑥𝑝 [− ]
3
𝑴𝑪𝑻𝑭 = 𝟒𝟒
3.37 The following data were collected by Frank Proschan in 1983. Operating hours to first failure of an
engine cooling part in 13 aircrafts are:
Aircraft 1 2 3 4 5 6 7 8 9 10 11 12 13
Hours 194 413 90 74 55 23 97 50 359 50 130 487 102
a) Would these data support an increasing failure rate, decreasing failure rate or constant failure rate
assumption?
The following table and Time-on-Test plot show that the data has a fairly constant failure rate.
The following non-parametric hazard function shows great variability in the failure rate as a
function of time. Additional testing would be required to reduce the uncertainty surrounding the
question as to whether the data is IFR, DFR or constant.
3.38 The following times to failure in hours were observed in an experiment where 14 units were tested
until eight of them have failed: 80, 310, 350, 470, 650, 900, 1100, and 1530
Assuming that the units have a constant failure rate, calculate a point estimate of the failure rate. Also
calculate a 95% one-sided confidence interval of the failure rate.
𝑟 8
𝜆̂ = = = 5.49 × 10−4 ℎ𝑟 −1
𝑇 14570
2 2
𝜒1−𝛼 (2𝑟) 𝜒0.95 (16)
𝜆𝑢𝑝𝑝𝑒𝑟 = = = 𝟗. 𝟎𝟐 × 𝟏𝟎−𝟒 𝒉𝒓−𝟏
2𝑇 29140
Page 57 of 229
3.39 A life test of 10 small motors with a newly designed insulator has been performed. The following
data are obtained:
Motor No. 1 2 3 4 5 6 7 8 9 10
Failure Time
1175 1200 1400 1450 1580 1870 1930 2120 2180 2430
(hr)
The trend line shown in the plot above is given by the following expression.
1
𝑙𝑛 [𝑙𝑛 ( )] = 𝛽 ln(𝑡) − 𝛽ln(𝛼)
𝑅(𝑡)
Using this expression and the trend line value given in the plot above
𝑦 = 4.3482𝑡 − 32.834 the shape (𝛽) and scale (𝛼) parameters are given by:
𝜷 = 𝟒. 𝟑𝟓 & 𝜶 = 𝟏𝟗𝟎𝟑
𝑡 𝛽
−( )
𝑅(𝑡) = 𝑒 𝑎
4380 4.35
−( )
𝑅(6 𝑚𝑜𝑛𝑡ℎ𝑠) = 𝑅(4380 ℎ𝑟𝑠) = 𝑒 1903 =0
Page 58 of 229
3.40 Use the data in problem 3.39 to perform a total-time-on-test plot.
The Time-on-Test plot shown above details that the data exhibits an IFR over the time period shown
3.41 A company redesigns one of its compressors and wants to estimate reliability of the new product.
Using its past experience, the company believes that the reliability of the new compressor will be
higher than 0.5 (for a given mission time). The company's testing of one new compressor showed that
the product successfully achieved its mission.
a) Assuming a uniform prior distribution for the above reliability estimate, find the posterior
estimate of reliability based on the test data.
1 1
𝜋0 (𝑅) = = =2
𝑏 − 𝑎 1 − 0.5
1 1
2𝑅 2
𝐸[𝑅]𝑃𝑟𝑖𝑜𝑟 = ∫ 2𝑅 𝑑𝑅 = [ ] = 𝟎. 𝟕𝟓
0.5 2 0.5
Page 59 of 229
Applying Bayes Theorem, with one test case and zero failures, we get the following posterior:
1
𝐿(𝐷𝑎𝑡𝑎|𝑅)𝜋0 (𝑅) ( ) 𝑅1 (1 − 𝑅)0 × 2 𝑅
𝜋(𝑅|𝐷𝑎𝑡𝑎) = = 1 =
∫ 𝐿(𝐷𝑎𝑡𝑎|𝑅)𝜋0 (𝑅)𝑑𝑅 2 ∫1 (1) 𝑅1 (1 − 𝑅)0 𝑑𝑅 ∫ 𝑅𝑑𝑅
1
0.5 1 0.5
𝑅 𝑅
𝜋(𝑅|𝐷𝑎𝑡𝑎) = 1 = = 2.67𝑅
𝑅2 0.375
[2]
0.5
1 1
2.67𝑅3
𝐸[𝑅]𝑂𝑛𝑒 𝑆𝑢𝑐𝑐𝑒𝑠𝑠𝑓𝑢𝑙 𝑇𝑒𝑠𝑡 = ∫ 𝑅. 2.67𝑅 𝑑𝑅 = [ ] = 𝟎. 𝟕𝟕𝟖
0.5 3 0.5
b) If the company conducted another test, which resulted in another mission success, what would
be the new estimate of the product reliability?
Assuming another successful test was performed, we get the following posterior for R:
2
2 × ( ) 𝑅 2 (1 − 𝑅)0 𝑅2 𝑅2 𝑅2
𝜋(𝑅|𝐷𝑎𝑡𝑎) = 2 = = = = 3.43𝑅 2
1 2 1 2 3 1 0.29167
2 0 𝑅
2 ∫0.5 ( ) 𝑅 (1 − 𝑅) 𝑑𝑅 ∫0.5 𝑅 𝑑𝑅 [ ]
2 3 0.5
1 1
2
3.43𝑅4
𝐸[𝑅] 𝑇𝑤𝑜 𝑆𝑢𝑐𝑐𝑒𝑠𝑠𝑓𝑢𝑙 𝑇𝑒𝑠𝑡𝑠 = ∫ 𝑅3.43𝑅 𝑑𝑅 = [ ] = 𝟎. 𝟖𝟎𝟒
0.5 4 0.5
3.42 The specification calls for power transistor to have a reliability of 0.95 at 2000h. Five hundred
transistors are placed on test for 2000h with 15 failures observed. The failed units are substituted upon
failure. Has this specification been met? What is the chance that the specification has not been met?
𝑟 15
𝜆̂ = = = 1.5 × 10−5 ℎ𝑟 −1
𝑇 500 × 2000
−5 ×2000
𝑅(2000) = 𝑒 −1.5×10 = 𝟎. 𝟗𝟕
Find the probability that the specification has not been met?
Need to find significance value (𝛼) using the following relationship for the one sided confidence limits
of a Type I – Time Terminated Test
Page 60 of 229
2
𝜒1−𝛼 (2𝑟 + 2)
𝑃𝑟 [𝑒𝑥𝑝 [− 𝑡𝑜 ] < 𝑅(𝑡𝑜 ) < 1] = 1 − 𝛼
2𝑇
2
𝜒1−𝛼 (32)
𝑒𝑥𝑝 [− × 2000] < 𝑅(2000) < 1
2000000
Solving this expression for (𝛼), we get 𝑅(2000) ≥ 0.95 when 𝛼 ≤ 0.017
Therefore; there is a 1.7% chance that the specification has not been met.
3.43 The following failure times, in days, were obtained by placing 30 units on a 100 day life test: 8, 12,
22, 51,73 and 85 days.
a) Perform a Weibull plot and show that a two parameter Weibull PDF is a good fit. What are the
estimates of the PDF parameters?
i Time (days) F(t)% ln(time) ln(ln(1/R(t)))
1 8 2.07 2.08 -3.869
2 12 5.37 2.48 -2.897
3 22 8.68 3.09 -2.399
4 51 11.98 3.93 -2.058
5 73 15.29 4.29 -1.796
6 85 18.60 4.44 -1.581
n = 30
The trend line shown in the plot above is given by the following expression.
1
𝑙𝑛 [𝑙𝑛 ( )] = 𝛽 ln(𝑡) − 𝛽ln(𝛼)
𝑅(𝑡)
Using this expression and the trend line value given in the plot above
𝑦 = 4.3482𝑡 − 32.834 the shape (𝛽) and scale (𝛼) parameters are given by:
Two methods are used to show the hazard rate. Firstly, the nonparametric plot below shows that
the hazard rate is fairly constant throughout the time range with a value of approximately 0.003
Page 61 of 229
Secondly; the MLE failure rate is given by:
𝑟 6
𝜆̂ = = = 2.26 × 10−3 ℎ𝑟 −1
𝑇 251 + 24 × 100
The results from the plotting method in part (a) and the nonparametric methods discussed above
are fairly consistent with one another.
3.44 The time to failure of a solid state power unit has a hazard function in the form of
1
ℎ(𝑡) = 1.34 × 10−4 𝑡 2 𝑓𝑜𝑟 𝑡 ≥ 0
𝑡 𝑡 3 𝑡
1 2.68 × 10−4 𝑡 2
𝑅(𝑡) = 𝑒𝑥𝑝 [− ∫ ℎ(𝑡)𝑑𝑡] = 𝑒𝑥𝑝 [− ∫(1.34 × 10−4 𝑡 2 )𝑑𝑡] = 𝑒𝑥𝑝 [− ]
3
0 0 0
3
2.68 × 10−4 𝑡 2
𝑅(𝑡) = 𝑒𝑥𝑝 [− ]
3
Page 62 of 229
Solving for 𝑡; we get 𝑡 ≤ 23.3 ℎ𝑟𝑠 in order to achieve 𝑅(𝑡) ≥ 0.99
c) Compute MTTF.
∞ ∞ 3
2.68 × 10−4 𝑡 2
𝑀𝑇𝑇𝐹 = ∫ 𝑅(𝑡)𝑑𝑡 = ∫ 𝑒𝑥𝑝 [− ] 𝑑𝑡
3
0 0
This integral was solved via numerical integration using Matlab as follows:
F=@(x) exp(-(1/3*2.68e-4).*(x.^(1.5)));
MTTF=quadgk(F,0,+inf)
MTTF = 451.7
d) If the unit has operated for 50 hrs, what is the probability that the unit will operate for another 50
hrs.
e) What is the mean residual time to failure at 𝑡 = 50ℎ𝑟𝑠? Compare it with the results shown at part
c.
∞ ∞ 3
1 1 2.68 × 10−4 𝑡 2
𝑀𝑅𝐿(50ℎ𝑟𝑠) = ∫ 𝑅(𝑡)𝑑𝑡 = ∫ 𝑒𝑥𝑝 [− ] 𝑑𝑡
𝑅(50) 0.969 3
50 50
This integral was solved via numerical integration using Matlab as follows:
F=@(x) exp(-(1/3*2.68e-4).*(x.^(1.5)));
MRL=quadgk(F,50,+inf)/F(50)
Therefore expected life is now 415.3 + 50 = 465.3 hrs which is higher than that found in part c.
Page 63 of 229
3.45 The following censored data reflect failure time in months of a new laser printer. Censored times result
from the removal of the printer due to upgrades. Determine the reliability of the printer over its 2yr
warranty period.
This problem was solved using both the Kaplan-Meier and Rank Increment Methods for comparison
with the assumption of a Weibull underlying distribution
The trend lines shown in the plot above are given by the following expression.
1
𝑙𝑛 [𝑙𝑛 ( )] = 𝛽 ln(𝑡) − 𝛽ln(𝛼)
𝑅(𝑡)
Using this expression and the trend lines given in the plot above; the shape (𝛽) and scale (𝛼)
parameters are given by:
Page 64 of 229
3.46 Consider an r.v. with cdf F(t) = t/2, 0 < t ≤ 2. Do the following:
a) Derive expressions for the corresponding time-to-failure PDF, hazard function, and reliability
function.
dF (t ) f (t )
f (t ) R(t ) 1 F (t ) h(t )
dt R(t )
Therefore
d (t / 2) 1 t 1 1 1
f (t ) R(t ) 1 h(t )
dt 2 2
2 1 t
2
(2 t )
b) Compute Pr(0.1 < T ≤ 0.2|T > 0.1).
3.47 A life test on silicon photodiode detectors in which 28 detectors were tested at very high stress
condition is performed. Failures were found after the inspections at 2500 (1 failure), 3000 (1 failure),
3500 (2 failures), 3600 (1 failure), 3700 (1 failure), and 3800 (1 failure). The other 21 detectors had
not failed after 3800 h of operation. Use these data to estimate the life distribution of such photodiode
detectors running at the test conditions:
a) Compute and plot a nonparametric estimate of the cdf for time to failure at the test conditions.
This problem was solved using Equations 3.52 through to 3.54 where 𝑛 = 28. The table and plot
below shows the results:
Page 65 of 229
b. Compute hazard rate.
The hazard rate values are shown in the table above. The following plot provides a graphical
representation of the results.
Page 66 of 229
3.48 Data on minutes to breakdown for an insulating fluid were obtained for 11 tests at 30 kV. After 100
min, there were 7 breakdowns (failures) at the following times (in minutes): 7.74, 17.05, 20.46, 21.02,
22.66, 43.40, and 47.30. The other four units had not failed.
b. What do you think about the suggestion of using an exponential distribution to model the data?
From the plot above, we estimate the parameters of the Weibull model to be:
𝛽 = 1.78 & 𝛼 = 30
Since Beta is relatively close to 1; we could use an exponential model for this data set.
c. Assuming an exponential distribution, obtain the maximum likelihood point estimate of failure rate.
𝑟 7
𝜆̂ = = = 1.21 × 10−2 ℎ𝑟 −1
𝑇 179.63 + 4 × 100
2 (2 × 7) 2 (2 × 7 + 2)
𝜒0.025 𝜒0.0975
<𝜆<
1159.3 1159.3
e. Compute the MLE of the approximate reliability at 95%, one-sided confidence interval of reliability
at t = 50 min
2
𝜒1−𝛼 (2𝑟 + 2)
𝑃𝑟 [𝑒𝑥𝑝 [− 𝑡𝑜 ] < 𝑅(𝑡𝑜 ) < 1] = 0.95
2𝑇
Page 67 of 229
2
𝜒0.95 (16)
𝑒𝑥𝑝 [− × 50] < 𝑅(50) < 1
1159.3
3.49 A manufacturer of clocks claims that a certain model will last 5yrs on the average with a standadrd
deviaton of 1.2 yrs.A random sample of 6 such clocks lasted 6, 5.5, 4, 5.2, 5, and 4.3 yrs. Discuss the
adequacy of such a claim. Use any method to support your statistical or probabalistic conclusion.
The problem was solved using the Kolmogorov-Smirnov Test at the 5% significance level with the
following assumptions:
Ho: the r.v (T) follows a Normal Distribution with known parameters 𝜇 = 5𝑦𝑟 & 𝜎 = 1.2𝑦𝑟
H1: the r.v (T) does not follow a Normal Distribution
The rejection region (R) was defined as 𝑅 ≥ 𝐷𝑛 (𝛼) with a 5% significance level (ie: 𝛼 =
0.05)
n = 6 (the number of failure times recorded)
Therefore Dn ( ) was found to be 0.519 from tables of Critical value for the Kolmogonov-Smirnov
statistic and the Maximum K-S statistic for the data set was calculated as shown in the table below to
be Max K-S = 0.2023
As the Max (K-S) Value < R, (ie: 0.2023 < 0.519), we do not reject the Ho: Hypothesis that a Normal
Distribution with known parameters 𝜇 = 5𝑦𝑟 & 𝜎 = 1.2𝑦𝑟 is an appropriate model for these data at
the 95% confidence limit.
Page 68 of 229
3.50 A sample of 100 specimens of a Titanium alloy were subjected to a strain-life fatigue test to determine
crack initiation. The test was run up to a limit of 100,000 cycles. The observed times of crack initiation
(in units of 1000 of cycles) were 18, 32, 39, 53, 59, 68, 77, 78 and 93. No cracks had initiated in any
of the other 91 specimens.
b) Plot the data on Weibull paper and use the plot to obtain an estimate of Weibull shape and scale
parameters. Comment on the adequacy of the Weibull distribution.
From the plot above, we estimate the parameters of the Weibull model to be:
𝛽 = 1.94 & 𝛼 = 317
Page 69 of 229
3.51 The reliability of a turbine blade can be represented by the following expression:
𝑡 2
𝑅(𝑡) = (1 − ) 0 ≤ 𝑡 ≤ 𝑡0
𝑡0
𝑡 2
𝐹(𝑡) = 1 − 𝑅(𝑡) = 1 − (1 − )
𝑡0
𝑑 𝑑 𝑡 2 2 𝑡
𝑓(𝑡) = (𝐹(𝑡)) = [1 − (1 − ) ] = (1 − )
𝑑𝑡 𝑑𝑡 𝑡0 𝑡0 𝑡0
𝑓(𝑡) 2 𝑡 1 2
ℎ(𝑡) = = (1 − ) × 2 = 𝑡
𝑅(𝑡) 𝑡0 𝑡0 𝑡 𝑡0 (1 − 𝑡 )
(1 − )
𝑡0 0
This integral was solved via numerical integration using Matlab as follows:
F=@(x) (1-x./3000).^2;
MTTF=quadgk(F,0,3000)
MTTF = 1000
3.52 The following censored data reflects failure time in months of a new computer scanner. Determine the
reliability of this scanner over its 2 yr warranty period.
This problem was solved using both the Kaplan-Meier and Rank Increment Methods for comparison
with the assumption of a Weibull underlying distribution
Page 70 of 229
The trend lines shown in the plot above are given by the following expression.
1
𝑙𝑛 [𝑙𝑛 ( )] = 𝛽 ln(𝑡) − 𝛽ln(𝛼)
𝑅(𝑡)
Using this expression and the trend lines given in the plot above; the shape (𝛽) and scale (𝛼)
parameters are given by:
𝑡 𝛽 𝑡 𝛽
−( ) −( )
𝑅(24𝑚𝑡ℎ𝑠)𝐾−𝑀 = 𝑒 𝛼 = 0.60 𝑅(24𝑚𝑡ℎ𝑠)𝑅−𝐼𝑛𝑐𝑟 = 𝑒 𝛼 = 0.66
3.53 Use Rank Increment rank statistic to produce a Weibull plot using the following multiple censored
data. Which part of the bathtub curve are these data most likely coming from?
Test No. 1 2 3 4 5 6 7 8 9 10
Time to failure 309+ 386 180 167+ 122 229 104+ 217+ 168 138
“+”=censoring.
Page 71 of 229
The trend line shown in the plot above is given by the following expression.
1
𝑙𝑛 [𝑙𝑛 ( )] = 𝛽 ln(𝑡) − 𝛽ln(𝛼)
𝑅(𝑡)
Using this expression and the trend line given in the plot above; the shape (𝛽) and scale (𝛼) parameters
are given by:
Given that 𝛽 is much greater than 1; this indicates that this data set is most likely from the wear out
region of the bathtub plot.
3.54 The failure time (in hour) of a component is lognormal distributed with parameters μt = 4 and σt = 0.9.
1 0.9 2
MTTF exp t t2 exp 4 81.86 hours
2 2
b. When should the component be replaced, if the minimum required reliability for the component is
0.95?
In order for the component to have a minimum Reliability of 0.95, the cumulative failure function
F(t) must be ≤ 0.05; therefore, we must calculate the F(0.05). Using Matlab with μt = 4 & σt = 0.9,
we get:
F(0.05)=logninv(0.05,4,0.9)
F(0.05) = 12.42 hrs
Therefore, in order to remain at least 95% Reliable, the components should be replaced at:
T ≤ 12.42 hrs
c. What is the value of the hazard function at this time (i.e., for the time calculated in part (b))?
Page 72 of 229
1
h(12.42hrs)
f (12.42)
1
1
exp ln(12.42) 42
R(12.42) 0.95 0.9 12.42 2 0.9
2
0.0092
h(12.42hrs) 0.0097
0.95
d. What is the mean residual life at this time if the component has already survived to this time (i.e:
for the time calculated in part (b))?
∞ ∞
1 1
𝑀𝑅𝐿(12.42 ℎ𝑟𝑠) = ∫ 𝑅(𝑡)𝑑𝑡 = ∫ [1 − 𝐹(𝑡)]𝑑𝑡
𝑅(12.42) 0.95
12.42 12.42
This integral was solved via numerical integration using Matlab as follows:
F=@(x) (1-logncdf(x,4,0.9));
MRL=quadgk(F,12.42,+inf)/F(12.42)
MRL = 73.3
This means, that a component that has survived to T = 12.42 hrs is expected to live for another 73.3
hrs bringing its total life to 85.7 hrs.
The total value of T + MRL (12.42 + 73.3 = 85.7 hrs) is greater than the original MTTF (81.86 hrs)
which is what we would expect.
3.55 On a surface of a highly stressed motor shaft there are 5 crack initiation points observed after the shaft
is tested at 6000rpm for 5hrs.
a) Calculate the point estimate of a crack initiation per cycle (per turn of the shaft)
𝑟 7
𝜆̂ = = = 2.78 × 10−6 𝑐𝑦𝑐𝑙𝑒𝑠 −1
𝑇𝑐𝑦𝑐𝑙𝑒𝑠 6000 × 60 × 5
b) Determine the 90% two-sided confidence interval of the probability of a crack initiation per cycle.
Firstly, we find the 90% confidence of the Reliability using the expression for a Type 1 – Time
Terminated Test where (t is in cycles) we get:
2 2
𝜒1−𝛼/2 (2𝑟 + 2) 𝜒𝛼/2 (2𝑟)
𝑃𝑟 [𝑒𝑥𝑝 [− 𝑡𝑜 ] < 𝑅(𝑡𝑜 ) < 𝑒𝑥𝑝 [− 𝑡𝑜 ]] = 0.90
2𝑇 2𝑇
2 2
𝜒0.95 (12) 𝜒0.05 (10)
𝑒𝑥𝑝 [− 6
× 1] < 𝑅(1) < 𝑒𝑥𝑝 [− × 1]
3.6 × 10 3.6 × 106
Page 73 of 229
1.09 × 10−6 < 𝐹(1) < 5.84 × 10−6
1 1
𝑀𝐶𝐶𝐼 = = = 𝟑. 𝟔 × 𝟏𝟎𝟓 𝒄𝒚𝒄𝒍𝒆𝒔
̂𝜆 2.78 × 10−6
To find probability of no cracks, we need to find the reliability interval for observing no crack in
the time period. Assume 90% confidence intervals.
2 2
𝜒1−𝛼/2 (2𝑟 + 2) 𝜒𝛼/2 (2𝑟)
𝑃𝑟 [𝑒𝑥𝑝 [− 𝑡𝑜 ] < 𝑅(𝑡𝑜 ) < 𝑒𝑥𝑝 [− 𝑡𝑜 ]] = 0.90
2𝑇 2𝑇
2 2
𝜒0.95 (2) 𝜒0.05 (0)
𝑒𝑥𝑝 [− × 1] < 𝑅(1) < 𝑒𝑥𝑝 [− × 1]
2 × 60 × 2000 × 10 2 × 60 × 2000 × 10
3.56 A household appliance is advertised as having more than a 10 yr life. If the following is its time to
failure PDF, determine the reliability of a new appliance at 10 yrs. What is its MTTF when it is new
and what is its MTTF when a 1 yr warranty period ends; assuming it still works?
Reliability at 10 yrs
10
3.57 Assume exponential PDF model with the failure rate λ describes the time to failure of a component.
Suppose we have observed failures of 𝑛 such components in a life test as: 𝑡1 , 𝑡2 , … . . 𝑡𝑛
Page 74 of 229
a) Write the likelihood function representing this set of observed failure data
𝑛
b) Given the failure data observed, write the equation for the posterior PDF of λ, if the prior PDF of λ
can be represnted by a uniform distribution between 𝜆1 𝑎𝑛𝑑 𝜆2
Assuming a uniform prior, the posterior is given by:
c) If 𝑛 = 2; 𝑡1 = 100, 𝑡2 = 150, 𝜆1 = 0.05 𝑎𝑛𝑑 𝜆2 = 0.001 calculate the posterior PDF mean and
95% bounds of λ.
𝜆2 2𝜆 2
[𝐻𝑖𝑛𝑡: ∫ 𝜆2 𝑒 𝑎𝜆 = 𝑒 𝑎𝜆 ( − + )]
𝑎 𝑎2 𝑎3
𝜆2 𝑒 −250𝜆
𝜋(𝜆|𝐷𝑎𝑡𝑎) = 0.05 = 7.83 × 106 𝜆2 𝑒 −250𝜆
∫0.001 𝜆2 𝑒 −250𝜆 𝑑𝜆
F2=@(x) (1/denom)*(x.^2).*exp(-250.*x);
p05=quadgk(F2,0.0001,0.003268)
p05 = 0.0500
p95=quadgk(F2,0.0001,0.02493)
p95 = 0.9500
F3=@(x) (1/denom)*(x.^3).*exp(-250.*x);
mean=quadgk(F3,0.0001,0.5)
mean = 0.0120
Page 75 of 229
3.58 Consider the following reliability test data obtained for a given component whose time to failure
distribution is known to follow the exponential distribution.
Test No. 1 2 3 4 5 6 7 8
Time to failure 823 900+ 1120 2301 3000+ 3000+ 3211 4000+
“+”=censoring.
𝐿(𝐷|𝜆) = 𝜆4 𝑒 −𝜆18355
𝜆̂ = 0.00022
c) Use Kaplan-Meier order statistics to see whether exponential PDF is an appropriate model. What
would be the corresponding estimate of λ?
Page 76 of 229
The following table adn plot show the Kaplan-Meier method applied to the data set above.
Kaplan-Meier Method
Test No. Time to failure j 1-1/j R(tj) Ln(R(t))
0 1
1 823 8 0.875 0.875 -0.134
2 900 7
3 1120 6 0.833 0.729 -0.316
4 2301 5 0.800 0.583 -0.539
5 3000 4
6 3000 3
7 3211 2 0.500 0.292 -1.232
8 4000 1
From the plot above, the estimate of failure rate is calculated to be 𝝀̂ = 𝟎. 𝟎𝟎𝟎𝟒 however, we can
see that the plot is quite scattered and therefore; the MLE method dsicussed above is assumed to
be more accurate with the limtied amount of data provided.
3.59
a) The time to failure of a component follows an exponential distribution, and a right censored life
test of 10 components with replacement yielded three faiures in 8760 hrs. Calculate the point
estimate of the unreliability associated with the output from the system after 1 yr of operation.
𝑟 3
𝜆̂ = = = 3.42 × 10−5 ℎ𝑟𝑠 −1
𝑇 10 × 8760
̂ ̂
𝐹(𝑡) = 1 − 𝑒 −𝜆𝑡 => 𝐹(8760) = 1 − 𝑒 −𝜆8760 = 0.259
Firstly, we find the 90% confidence of the Reliability using the expression for a Type 1 – Time
Terminated Test we get:
2 2
𝜒1−𝛼/2 (2𝑟 + 2) 𝜒𝛼/2 (2𝑟)
𝑃𝑟 [𝑒𝑥𝑝 [− 𝑡𝑜 ] < 𝑅(𝑡𝑜 ) < 𝑒𝑥𝑝 [− 𝑡𝑜 ]] = 0.90
2𝑇 2𝑇
Page 77 of 229
2 2
𝜒0.95 (8) 𝜒0.05 (6)
𝑒𝑥𝑝 [− × 8760] < 𝑅(8760) < 𝑒𝑥𝑝 [− × 8760]
175200 175200
3.60 Calculate the conditional relaibility of a unit at 𝑡 = 1000 ℎ𝑟𝑠 if it has survived up to 𝑡 = 500 ℎ𝑟𝑠.
Assume that the time to failure of the units follow a Weibull distribution with 𝛼 = 1800 ℎ𝑟𝑠 and 𝛽 =
1.8
1000 1.8
𝑒𝑥𝑝 [− (1800) ]
𝑅(1000)
= = 𝟎. 𝟕𝟖𝟏
𝑅(500) 500 1.8
𝑒𝑥𝑝 [− ( ) ]
1800
3.61 What the relaibility function of a component if its hazard rate is expressed as ℎ(𝑡) = 0.01/(𝑡 + 1)
𝑡 𝑡
𝐥𝐧(𝒕 + 𝟏)
𝑹(𝒕) = 𝒆𝒙𝒑 [− ]
𝟏𝟎𝟎
3.62 The following censored data reflect failure time in months of a new laser printer. Censored times result
from the removal of the printer due to upgrades. Determine the reliability of the printer over its 2yr
warranty period. Use the adjusted rank method; the Kaplan-Meier method.
This problem is exactly the same as problem 3.45. See the solution provided above at 3.45
Page 78 of 229
3.63 In a reliability test under normal stress load, 500 components have been exposed to a given stress
profile under constant environmental conditions. After completion of the test, 12 components were
failed. Determine:
a) An estimate of the reliability of this component to endure the given stress profile.
Assuming components can be either in a failed or not failed state only; we therefore have a binomial
case:
𝑟 12
Pr(𝐹𝑎𝑖𝑙) = 𝑝̂ = = = 0.024
𝑁 500
b) The upper 90% confidence limit for the expected number of reliable components under the same
stress profile , if we use 1000 such components in a plant.
The 90% confidence interval for the estimated probability of failure can be found using the
Clopper- Pearson procedure as follows:
Expected number of reliable components under the same stress profile is given by 1000 × 𝑅
106 106
𝑐 𝑐𝑁1−𝛼 1−𝛼
∫ 𝛼 𝑑𝑁 = 1 => [ ] =1 => 𝑐 =
𝑁 1−𝛼 1 106−6𝛼
−1
1
Page 79 of 229
b) If we have gathered the following cycles to failure data, set up the maximum likelihood function
and a point estimate of 𝛼:
From the plot above, the slope gives us 𝛼 directly: 𝜶 = −𝟎. 𝟔𝟖𝟔𝟖
Page 80 of 229
3.65 Life data were gathered in a right censored test that ended in 6 months.
a) Use Kaplan-Meier plotting method and determine whether a Weibull distribution is a proper fit
The following table and plot were created using the grouped Kaplan-Meier method where:
𝐶𝑖
𝐻𝑖" = 𝐻𝑖 − = adjusted # at risk assuming that censored time occur uniformly over the interval
2
𝐹𝑖
= Conditional Probability of a failure in the 𝑖 𝑡ℎ given survivial up to time 𝑡𝑖−1
𝐻𝑖"
𝐹𝑖
𝑝𝑖 = 1 − = Conditional Probability of survival
𝐻𝑖"
𝐹𝑖
𝑅𝑖 = [1 − ] × 𝑅𝑖−1 = Un𝑐onditional Probability of survival
𝐻𝑖"
Grouped Kaplan-Meier
Interval
start finish Fi Ci Hi Hi" pi Ri F(ti)% ln(ti) ln(ln(1/R(t)))
0 0 73 1
1 2 1 5 73 70.5 0.986 0.986 1.42 0.693 -4.248
2 3 3 2 67 66 0.955 0.941 5.90 1.099 -2.800
3 4 5 7 62 58.5 0.915 0.861 13.94 1.386 -1.896
4 5 7 5 50 47.5 0.853 0.734 26.62 1.609 -1.173
5 6 9 4 38 36 0.750 0.550 44.97 1.792 -0.515
6 0 25 25 12.5 1.000 0.550 44.97 -0.515
Page 81 of 229
From the plot above, we can see that the data is a very good fit for a Weibull Plot. This is evident
from the lack of scatter in the data and the 𝑅 2 value of 0.9995.
The trend line shown in the plot above is given by the following expression.
1
𝑙𝑛 [𝑙𝑛 ( )] = 𝛽 ln(𝑡) − 𝛽ln(𝛼)
𝑅(𝑡)
Using this expression and the trend line given in the plot above; the shape (𝛽) and scale (𝛼)
parameters are given by:
Given that 𝛽 is much greater than 1; this indicates that this data set is most likely from the wear out
region of the bathtub plot.
From the plot, we find the time at which 10% of units have failed (𝑡10 )
d) From the plot, estimate the probability that the component will fail in the inteval 5 < 𝑇 < 6
From the plot, we find the difference between the Reliability at 𝑇 = 6 and 𝑇 = 5
1
𝑙𝑛 [𝑙𝑛 ( )] = 3.363 ln(6) − 6.5513 => 𝑅(6) = 0.726
𝑅(6)
1
𝑙𝑛 [𝑙𝑛 ( )] = 3.363 ln(5) − 6.5513 => 𝑅(5) = 0.554
𝑅(5)
Page 82 of 229
3.66 In a mission, 11 of 2012 switches in a plant are defective. Estimate the reliability of a single component
and the associated 90% confidence limit of the true reliability. Use a uniform prior estimate of the
probability of a single failure to be in the range of 0.001-0.05 and find the posterior estimate of the
reliability including the 90% probability bounds. Compare the results with the MLE method.
𝑟 11
Pr(𝐹𝑎𝑖𝑙) = 𝑝̂ = = = 0.00547
𝑁 2012
The 90% confidence interval for the estimated probability of failure can be found using the Clopper-
Pearson procedure as follows:
1 1
𝜋0 (𝑝) = = = 25
𝑏 − 𝑎 0.05 − 0.001
Applying Bayes Theorem for the data given, we get the following posterior:
2012 11
𝐿(𝐷𝑎𝑡𝑎|𝑝)𝜋0 (𝑝) 25 × ( ) 𝑝 (1 − 𝑝)2001
𝜋(𝑝|𝐷𝑎𝑡𝑎) = = 11
∫ 𝐿(𝐷𝑎𝑡𝑎|𝑝)𝜋0 (𝑝)𝑑𝑝 25 ∫0.05 (2012) 𝑝11 (1 − 𝑝)2001 𝑑𝑝
0.001 11
𝑝11 (1 − 𝑝)2001
𝜋(𝑝|𝐷𝑎𝑡𝑎) = 0.05
∫0.001 𝑝11 (1 − 𝑝)2001 𝑑𝑝
These integrals were solved using Matlab as follows:
Page 83 of 229
F=@(x) (x.^11).*(1-x).^2001;
denom = quadgk(F,0.001,0.05)
denom = 9.3174e-033
These values are extremely close to the previous values found using the Clopper- Pearson
procedure.
3.67 Experts estimate the relaibility of a new device at the end of its warranty period as more than 0.9.
Assume that this estimate can be expressed as a noninformative prior (ie: in the form of a uniform
distribution) for the reliability. One has performed tests on 7 such components and has observed one
failure over the same warranty period. What is the posterior estimate of the reliabaility?
1 1
𝜋0 (𝑅) = = = 10
𝑏 − 𝑎 1 − 0.9
1 1
10𝑅 2
𝐸[𝑅]𝑃𝑟𝑖𝑜𝑟 = ∫ 10𝑅 𝑑𝑅 = [ ] = 𝟎. 𝟗𝟓
0.9 2 0.9
Applying Bayes Theorem, with 7 test cases and 1 failure, we get the following posterior:
7
𝐿(𝐷𝑎𝑡𝑎|𝑅)𝜋0 (𝑅) 10 × ( ) 𝑅 6 (1 − 𝑅)1
𝜋(𝑅|𝐷𝑎𝑡𝑎) = = 6
∫ 𝐿(𝐷𝑎𝑡𝑎|𝑅)𝜋0 (𝑅)𝑑𝑅 10 ∫1 (7) 𝑅 6 (1 − 𝑅)1 𝑑𝑅
0.9 6
𝑅 6 (1 − 𝑅)1
𝜋(𝑅|𝐷𝑎𝑡𝑎) = 1
∫0.9 𝑅 6 (1 − 𝑅)1 𝑑𝑅
Page 84 of 229
The integral above was solved using Matlab to determine the percentiles of interest as follows:
F = @(x) (x.^6).*(1-x);
denom = quadgk(F,0.9,1)
denom = 0.0033
F2 = @(x) (1/denom).*(x.^6).*(1-x);
p05 = quadgk(F2,0.9,0.90316)
p05 = 0.0500
p95 = quadgk(F2,0.9,0.98102)
p95 = 0.9500
F3 = @(x) (1/denom).*(x.^7).*(1-x);
mean = quadgk(F3,0.9,1)
mean = 0.9370
The following plot shows the prior and posterior distributions of Reliability plotted simultaneously
for comparison.
Page 85 of 229
3.68 Consider the following reliability test data obtained for a given component whose time-to-failure
distribution is known to follow exponential distribution.
Test No. 1 2 3 4 5 6 7 8
Time-to- 823 900+ 1120 2301 3000+ 3000+ 3211 4000+
failure
+ = Right censored data
a. Write the likelihood function for the component.
b. What is the maximum likelihood estimate of the failure rate?
Solution:
a. Likelihood Function
𝑁𝑓 𝑁𝑠
𝑓
𝐿= ∏ 𝑓(𝑡𝑖 |𝜆) ∏ 𝑅(𝑡𝑗𝑠 |𝜆)
𝑖=1 𝑗=1
𝐿 = ∏ 𝜆 exp(−𝜆𝑡𝑖 ) ∏ exp(−𝜆𝑡𝑗 )
𝑖=1 𝑗=1
𝐿 = 𝜆 exp[−𝜆(𝑡1 + 𝑡2 + 𝑡3 + 𝑡4 + 𝑡5 + 𝑡6 + 𝑡7 + 𝑡8 )] = 𝜆exp(−18355 λ)
b. Solving for the MLE of λ, take the log-likelihood equation and differentiate:
ln 𝐿 = ln(λ) − 18355𝜆
𝑑 ln 𝐿 1
= − 18355 = 0
𝑑𝜆 𝜆
1
∴𝜆= = 5.448 × 10−5
18355
3.69 A sample of 100 of a new model automobile tire is reliability road tested with the life (in miles travelled
before failure) determined for each tire as follows:
Life Intervals (in 103 miles) Number of Failures
40-45 11
45-50 19
50-55 40
55-60 21
60-65 9
Page 86 of 229
Using nonparametric reliability methods estimate:
a. Hazard rate vs. miles traveled.
b. Reliability vs. miles traveled.
Solution
Using the following formulae:
𝑁𝑓 (𝑡𝑖 )
ĥ(𝑡𝑖 ) =
𝑁𝑠 (𝑡𝑖 )Δ𝑡
𝑁𝑠 (𝑡𝑖 )
𝑅̂ (𝑡𝑖 ) =
𝑁
Δ𝑡 = 5000
𝑁 = 100
60-65 9 0 0.0000E+00 0
3.70 Hazard rate of a component is h(𝑡) = 𝑒 −0.5𝑡 , where t is time measured in months.
a. Derive reliability function, R(t), and time-to-failure distribution, f(t)
b. Do you consider this a reliable component? Why?
Solution:
a. The relationship between the Reliability function and the hazard function is:
𝑡
𝑅(𝑡) = exp [− ∫ h(𝜏)𝑑𝜏]
0
𝑡 𝑡
1 1
∫ h(𝜏)𝑑𝜏 = ∫ exp(−0.5𝜏)d 𝜏 = ( exp(−0.5𝑡) − (exp 0)) = 2 − 2 exp(−0.5𝑡)
0 0 −0.5 −0.5
∴ 𝑅(𝑡) = exp[2 − 2 exp(−0.5𝑡)]
Page 87 of 229
3.71 Cycle-to-failure data from a dishwasher testing is shown below.
i Cycle-to-failure
1 7
2 12
3 19
4 29
5 41
6 67
Solution:
Plotted using Excel:
i # to fail 1/R(t_i)
1 7 1.111111
2 12 1.351351
3 19 1.724138
4 29 2.380952
5 41 3.846154
6 67 10
Page 88 of 229
3.72 Suppose the exponential distribution represents the embedded crack sizes (flaws), x, that
randomly occur during manufacturing. That is 𝑓(𝑥) = 𝜆𝑒 −𝜆𝑥 . Assume that a conjugate prior of is a
Gamma PDF with parameters =52, =2.68 (mm). If some pressure vessels have been inspected and
the following crack sizes were observed (1.0, 1.1, 1.08, 1.92, 2.05, 2.09, 2.11, 3.32)
a.) Write the expression representing the posterior PDF of .
b.) Given the posterior PDF of , write the expression for calculating the probability that a
pressure vessel will have a crack of size exceeding 10 mm.
Solution:
a.)
𝐿(𝑥|𝜆)𝑓(𝜆)
𝑓(λ|𝑥) = ∞
∫0 𝐿(𝑥|𝜆)𝑓(𝜆)𝑑𝜆
8
b.) To find the probability, we need to evaluate the posterior to find an estimate for λ, this is now
simple since all the values are known and for a gamma distribution:
𝐸(𝜆) = 𝛼 ′ 𝛽 ′
𝛼 ′ = 52 + 1 = 53
1
β′ = = 0.066476
1
1 + 1.1 + 1.08 + 1.92 + 2.05 + 2.09 + 2.11 + 3.32 + 2.68
∴ 𝐸(λ) = 53 × 0.066476 = 3.5232
∴ 𝑓(𝑥) = 3.5232 exp(−3.5232𝑥)
Page 89 of 229
Using the exponential distribution find:
Pr(𝑥 > 10) = 1 − Pr(𝑥 < 10) = 1 − 𝐹(𝑥 = 10) = 1 − (1 − exp(−3.5232(10))
= 4.9995 × 10−16
3.73 Suppose that we have 10 reliability observations in a test with failure and censored times as
follows: 2; 5; 5; 6+; 7; 12; 14+; 14+; 14+; 14+ (“+” indicates right censored observations)
Use the data and plot the lognormal distribution using the rank adjustment method. Estimate the
parameters of the distribution. Is this an appropriate fit?
Solution:
Using the rank adjustment method:
(𝑛 + 1) − 𝑗𝑡𝑖−1
𝑗𝑡𝑖 = 𝑗𝑡𝑖−1 +
2 + 𝑛 − 𝑖𝑡𝑖
4 6+
7 14+
8 14+
9 14+
10 14+
Page 90 of 229
The lognormal equation when linearized takes the form of:
1 𝜇
Φ−1 [𝐹(𝑡𝑖 )] = ln(𝑡𝑖 ) −
𝜎 𝜎
3.74 Seven pumps have failure times (in months) of: 15.1, 10.7, 8.8, 11.3, 12.6, 14.4, and 8.7. Use the
nonparametric method and estimate the hazard rate, PDF and the reliability function. Are the pumps
experiencing wear-out?
Solution:
Since there are only 7 samples – the small samples approximations should be used:
1
ĥ(𝑡𝑖 ) = , 𝑖 = 1,2, … 𝑛 − 1.
(𝑛 − 𝑖 + 0.625)(𝑡𝑖+1 − 𝑡𝑖 )
𝑛 − 𝑖 + 0.625
𝑅̂ (𝑡𝑖 ) = , 𝑖 = 1,2, … . 𝑛 − 1.
𝑛 + 0.25
1
𝑓̂(𝑡𝑖 ) = , 𝑖 = 1,2, … 𝑛 − 1.
(𝑛 + 0.25)(𝑡𝑖+1 − 𝑡𝑖 )
Page 91 of 229
i t h(t) f(t) R(t)
7 15.1
This broadly resembles a bathtub curve! So it could be argued that the pumps are wearing out. However,
the increase in failure rate is only from a single point so this could just be an outlier and at least a couple
more data points would be required for a conclusive result.
3.75Consider the failure data for cycle to failure of a unit as 8, 33, 15+, 27, 18, 24+ 13+, 12, 37, 29+
25, 30 (+ = rightly censored). What is a point estimate of cycle-to-failure using:
a.) Use probability plotting and the rank adjustment method
b.) Using maximum likelihood estimation
Solution:
a.) Using the rank adjustment method:
(𝑛 + 1) − 𝑗𝑡𝑖−1
𝑗𝑡𝑖 = 𝑗𝑡𝑖−1 +
2 + 𝑛 − 𝑖𝑡𝑖
𝑖 − 0.375
𝐹(𝑡) =
𝑛 + 0.25
Page 92 of 229
Order the data and generate the following table:
i t j F(t) ln(ln(1/R)) ln(t)
3 13+
4 15+
6 24+
9 29+
Therefore:
𝛽 = 2.2509
𝛽 ln(α) = 7.7057
∴ 𝛼 = 30.6735
Page 93 of 229
MTTF for the Weibull is found using:
1+𝛽
𝑀𝑇𝑇𝐹 = 𝛼Γ ( )
𝛽
𝑀𝑇𝑇𝐹 = 27.1686 𝑐𝑦𝑐𝑙𝑒𝑠
𝑛𝑓 𝑠 𝑛
̂ ̂
𝑛𝑓 𝛽̂ 𝛽̂ 𝑡𝑖 𝛽 𝑡𝑖 𝛽̂ 𝑡𝑗 𝛽
+ ∑ ( ) ln ( ) + ∑ ( ) = 0
𝛼̂ 2 𝛼̂ 𝛼̂ 𝛼̂ 𝛼̂
𝑖=1 𝑗=1
This has to be solved numerically and was completed in excel using the SOLVER function.
3.76 Five components are put on a reliability test with the following failure times observed: 98,
116, 2485, 2526, 2920 hours.
a.) If we assume the data come from a lognormal PDF, find the point estimate of the log-mean
and log-standard deviation of the PDF.
b.) What is the 90% two-sided confidence interval of the log-mean?
Solution:
a.) The plotting method can be used to quickly estimate the log-mean and log-standard deviation.
Plotting ln(t) against F(t):
𝑖 − 0.375
𝐹(𝑡) =
𝑛 + 0.25
Using the straight line equation to fit the line of best fit:
1 μ𝑡
Φ−1 [𝐹(𝑡𝑖 )] = ln(𝑡𝑖 ) −
𝜎𝑡 𝜎𝑡
Page 94 of 229
The following table is generated:
i t ln(t) F(t) NormInv(F(t))
Page 95 of 229
3.77 Twenty components are reliability tested with 3 failing and 17 passing.
a.) What is the 90% two-sided confidence interval for reliability?
b.) If we expect, a priori, that reliability should be more than 0.75 (and following a uniform
distribution), write the posterior distribution of the reliability.
Solution:
a.) Using the Clopper-Pearson procedure to find the confidence interval given the data:
𝑥 3
Pr(𝐹𝑎𝑖𝑙) = 𝑝̂ = =
𝑛 20
−1
𝑛−𝑥+1
𝑝𝑙𝑜𝑤𝑒𝑟 = {1 + 𝐹1−𝛼 [2𝑛 − 2𝑥 + 2; 2𝑥]}
𝑥 2
−1 −1
𝑝𝑢𝑝𝑝𝑒𝑟 = {1 + (𝑛 − 𝑥) {(𝑥 + 1)𝐹1−𝛼 [2𝑥 + 2; 2𝑛 − 2𝑥]} }
2
−1 −1
20 − 3 + 1 18
𝑝𝑙𝑜𝑤𝑒𝑟 = {1 + 𝐹 0.1 [2 × 20 − 2 × 3 + 2; 2 × 3]} = {1 + 𝐹0.95 [36; 6]}
3 1−
2 3
∴ 𝑝𝑙𝑜𝑤𝑒𝑟 = 0.0422
−1 −1
𝑝𝑢𝑝𝑝𝑒𝑟 = {1 + (20 − 3) {(3 + 1)𝐹 0.1 [2 × 3 + 2; 2 × 20 − 2 × 3]} }
1−
2
−1 }−1
= {18{4 × 𝐹0.95 [8;34]}
∴ 𝑝𝑢𝑝𝑝𝑒𝑟 = 0.3437
Therefore:
(1 − 𝑝𝑢𝑝𝑝𝑒𝑟 ) ≤ 𝑅 ≤ (1 − 𝑝𝑙𝑜𝑤𝑒𝑟 )
0.6563 ≤ 𝑅 ≤ 0.9578
b.) Given that the prior is a uniform distribution and R > 0.75
1 1
π0 = = =4
𝑏 − 𝑎 1 − 0.75
The likelihood function in this case is a binomial:
20
𝐿(𝐷𝑎𝑡𝑎|𝑅) = ( ) 𝑅17 (1 − 𝑅)3
17
Page 96 of 229
3.78 Using the rank adjustment method of probability plotting, determine the parameters of the
Weibull distribution using the multiple censored data shown in the table below.
1 2 3 4 5 6 7 8 9 10
309+ 380 197 167+ 103+ 229 102 217+ 168 138
Solution:
Using the rank adjustment method:
Where:
(𝑛 + 1) − 𝑗𝑡𝑖−1
𝑗𝑡𝑖 = 𝑗𝑡𝑖−1 +
2 + 𝑛 − 𝑖𝑡𝑖
𝑖 − 0.375
𝐹(𝑡) =
𝑛 + 0.25
2 103+
4 167+
7 217+
9 309+
Page 97 of 229
From the linearized Weibull equation:
1
ln ln [ ] = 𝛽 ln(𝑡) − 𝛽ln(α)
𝑅(𝑡)
Therefore:
𝛽 = 2.4887
𝛽 ln(α) = 13.963
∴ 𝛼 = 273.3124
3.79 Time to failures (in days) of eight components that went under reliability tests are: 143,
164, 188, 190, 192, 206, 209, 213.
a.) Estimate the non-parametric estimate of reliability and hazard rate associated with each
component.
b.) Write the log-likelihood function associated with above data assuming the time to failure
follows a Weibull distribution.
c.) Using part b), write the expressions to determine the point estimates and confidence
intervals for the shape parameter and scale parameter of the Weibull distribution model.
Solution:
a.) Non-parametric estimates
Since we only have 8 samples, use the estimates for small samples:
1
ĥ(𝑡𝑖 ) =
(𝑛 − 𝑖 + 0.625)(𝑡𝑖+1 − 𝑡𝑖 )
𝑛 − 𝑖 + 0.625
𝑅̂ (𝑡𝑖 ) =
𝑛 + 0.25
i t h(t) R(t)
8 213 0.075758
b.) Assuming the data follows a Weibull distribution the log-likelihood function is:
𝑛
𝐿 = ∏ 𝑓(𝑡𝑖 |𝛼, 𝛽)
𝑖=1
Page 98 of 229
𝛽(𝑡)𝛽−1 𝑡 𝛽
𝑓(𝑡) = exp [− ( ) ]
𝛼𝛽 𝛼
𝑛
𝛽(𝑡𝑖 )𝛽−1 𝑡𝑖 𝛽
𝐿=∏ exp [− ( ) ]
𝛼𝛽 𝛼
𝑖=1
𝑛
𝛽(𝑡𝑖 )𝛽−1 𝑡𝑖 𝛽
ln 𝐿 = ∑ ln { exp [− ( ) ]}
𝛼𝛽 𝛼
𝑖=1
𝑛
𝑡𝑖 𝛽
ln 𝐿 = ∑ {ln 𝛽 + ln(𝑡𝑖 )𝛽−1 − ln 𝛼 𝛽 + ln exp [− ( ) ]}
𝛼
𝑖=1
𝑛
𝑡𝑖 𝛽
ln 𝐿 = ∑ {ln 𝛽 + (𝛽 − 1) ln 𝑡𝑖 − 𝛽 ln 𝛼 − ( ) }
𝛼
𝑖=1
𝑛 𝑛
𝑡𝑖 𝛽
ln 𝐿 = ln 𝛽 − 𝛽 ln 𝛼 + ∑(𝛽 − 1) ln 𝑡𝑖 − ∑ ( )
𝛼
𝑖=1 𝑖=1
c.) To find the estimates of α and β the first derivatives of the above equation is taken:
𝜕 ln 𝐿
=0
𝜕𝛼
𝜕 ln 𝐿
=0
𝜕𝛽
These typically have to be solved numerically to find estimates for α and β
Page 99 of 229
Solutions to Chapter 4
Exercises
4.1 Consider the circuit below
X1 X2
X4
X3 X5
Assume the reliability of each circuit R(xi) = exp(-it), and i = 2.010-4 per hour for all i. Find the
following:
a) Minimal path sets
𝑃1 = 𝑋1 𝑋2 ; 𝑃2 = 𝑋3 𝑋5 ; 𝑃3 = 𝑋2 𝑋3 𝑋4
c) MTTF
∞
𝑀𝑇𝑇𝐹𝑆𝑌𝑆 = ∫ 𝑅𝑆𝑌𝑆 (𝑡)𝑑𝑡
0
𝑅𝑠𝑦𝑠 = Pr(𝑃1 ∪ 𝑃2 ∪ 𝑃3 )
𝑅𝑠𝑦𝑠 = 1 − (1 − 𝑅𝑋1 𝑅𝑋2 )(1 − 𝑅𝑋3 𝑅𝑋5 )(1 − 𝑅𝑋2 𝑅𝑋3 𝑅𝑋4 )
𝑅𝑠𝑦𝑠 = 𝑅𝑋1 𝑅𝑋2 + 𝑅𝑋3 𝑅𝑋5 + 𝑅𝑋2 𝑅𝑋3 𝑅𝑋4 − 𝑅𝑋1 𝑅𝑋2 𝑅𝑋3 𝑅𝑋4 − 𝑅𝑋2 𝑅𝑋3 𝑅𝑋4 𝑅𝑋5 − 𝑅𝑋1 𝑅𝑋2 𝑅𝑋3 𝑅𝑋5
+ 𝑅𝑋1 𝑅𝑋2 𝑅𝑋3 𝑅𝑋4 𝑅𝑋5 )
We told that the failure rate for each component is the same 𝑖𝑒: 𝜆𝑖 = 2 ×
−4
10−4 and therefore 𝑅𝑖 = 𝑅 = 𝑒 −2×10 𝑡
1 1 3 1
𝑀𝑇𝑇𝐹𝑆𝑌𝑆 = ( + − + ) = 𝟑𝟗𝟏𝟕 𝒉𝒓𝒔
𝜆 3𝜆 4𝜆 5𝜆
4.2 Consider the circuit below find the following with = 1.010-4
A
X4 X7
X1 X6
X3 X5 X8
X2
Find:
a) Minimal path sets
(X2); (X1, X3); (X1, X4, X7); (X1, X5, X8); (X1, X4, X6, X8)
Consider the simplified system with super component A as shown in the figure above.
Then, we can consider the system reliability in two steps as follows:
𝑅𝐴 = Pr(𝑃1 ∪ 𝑃2 ∪ 𝑃3 )
𝑅𝐴 = 1 − (1 − 𝑅4 𝑅7 )(1 − 𝑅5 𝑅8 )(1 − 𝑅4 𝑅6 𝑅8 )
𝑅𝐴 = 𝑅4 𝑅7 + 𝑅5 𝑅8 + 𝑅4 𝑅6 𝑅8 − 𝑅4 𝑅5 𝑅7 𝑅8 − 𝑅4 𝑅5 𝑅6 𝑅8 − 𝑅4 𝑅6 𝑅7 𝑅8 + 𝑅4 𝑅5 𝑅6 𝑅7 𝑅8 )
We told that the failure rate for each component is the same 𝑖𝑒: 𝜆𝑖 = 1 ×
−4
10−4 and therefore 𝑅𝑖 = 𝑅 = 𝑒 −1×10 𝑡
1 A
2
Page 101 of 229
Now consider the System Reliability with the modified figure shown above
The system path sets with Super component A included are:
𝑃1 = 1, 𝐴; 𝑃2 = 1,3; 𝑃3 = 2
d) Probability of failure at 1,000 hours, using cut sets, to verify results from c)
1 1 1 3 3 8 5 1
𝑀𝑇𝑇𝐹𝑆𝑌𝑆 = ( + + − − + − + ) = 𝟏𝟐, 𝟐𝟕𝟒 𝒉𝒓𝒔
𝜆 2𝜆 3𝜆 4𝜆 5𝜆 6𝜆 7𝜆 8𝜆
1 2
3 4
𝑅𝑠𝑦𝑠 = [𝑅1 𝑅2 + 𝑅3 𝑅4 − 𝑅1 𝑅2 𝑅3 𝑅4 ]
𝑅𝑠𝑦𝑠 (1000) = [𝑒 −(λ1 +λ2 )𝑡 + 𝑒 −(λ3 +λ4 )𝑡 − 𝑒 −(λ1 +λ2 +λ3 +λ4 )𝑡 ]
∞
𝑀𝑇𝑇𝐹𝑆𝑌𝑆 = ∫ (𝑅1 𝑅2 + 𝑅3 𝑅4 − 𝑅1 𝑅2 𝑅3 𝑅4 )𝑑𝑡
0
1 1 1
𝑀𝑇𝑇𝐹𝑆𝑌𝑆 = ( + − ) = 𝟏𝟕, 𝟖𝟐𝟏 𝒉𝒓𝒔
(λ1 + λ2 ) (λ3 + λ4 ) (λ1 + λ2 + λ3 + λ4 )
POINT A V2 POINT B
V1 P1
V3
The purpose of the system is to pump water from point A to point B. The time to failure of all the
valves and the pump can be represented by the exponential distributions with failure rates v and p
respectively.
b) If v = 1.010-3 and p = 2.010-3 per hour, and the system has survived for 10 hours, what is the
probability that it will survive another 10 hours?
4.5 Estimate reliability of the system represented by the following reliability block diagram for a
2,500-hour mission. Assume a failure rate of 1.010-6 per hour for each unit.
A B F I
E H
C D G
This problem is best approached using the Minimal Cut set method as follows:
𝑹𝑺 (𝟐𝟓𝟎𝟎) ≥ 𝟎. 𝟗𝟗𝟗𝟗𝟕
4.6 A containment spray system is used to scrub and cool the atmosphere around a nuclear reactor during
an accident. Develop a fault tree using "No H2O spray'' as the top event.
motor motor
operated operated Spray Nozzle
valve P P valve P
Tank A
(ET1) xxxxx
SV1 SV3 SV6 SN1
pumps
SP1 SV5
CONTAINMENT
P P P
H2 O B
/\/\/\/\/\
The system diagram shown above is redrawn into the following Reliability Block and Fault Tree
diagrams as follows:
ET1 Power
1 2
2
1
SV1 SP1 SV5 SV2 SP2 SV2 SP2 SV5 SV1 SP1
G3 G1
G4 B G2 A
D E B C
𝑇 = G1 ∙ G3
G1 = A + G2 where G2 = BC
G1 = A + BC
G3 = B ∙ G4 where G4 = D + E
G3 = BD + BE
Therefore
𝑇 = (𝐴 + 𝐵𝐶) ∙ (𝐵𝐷 + 𝐵𝐸)
𝑇 = 𝐴𝐵𝐷 + 𝐴𝐵𝐸 + 𝐵𝐶𝐷 + 𝐵𝐶𝐸
The Minimal Cut Sets of the system are therefore;
𝐶1 = 𝐴𝐵𝐷, 𝐶2 = 𝐴𝐵𝐸, 𝐶3 = 𝐵𝐶𝐷, 𝐶4 = 𝐵𝐶𝐸
𝑇̅ = ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
𝐴𝐵𝐷 + 𝐴𝐵𝐸 + 𝐵𝐶𝐷 + 𝐵𝐶𝐸
𝑇̅ = ̅̅̅̅̅̅̅̅̅
(𝐴𝐵𝐷) ∙ ̅̅̅̅̅̅̅̅̅
(𝐴𝐵𝐸) ∙ ̅̅̅̅̅̅̅̅̅
(𝐵𝐶𝐷) ∙ ̅̅̅̅̅̅̅̅
(𝐵𝐶𝐸)
𝑇̅ = (𝐴̅ + 𝐵̅ + 𝐷
̅ ) ∙ (𝐴̅ + 𝐵̅ + 𝐸̅ ) ∙ (𝐵̅ + 𝐶̅ + 𝐷
̅ ) ∙ (𝐵̅ + 𝐶̅ + 𝐸̅ )
𝑇̅ = (𝐴̅ + 𝐵̅) ∙ (𝐵̅ + 𝐶̅ + 𝐷
̅ 𝐸̅ )
𝑇̅ = 𝐵̅ + 𝐴̅𝐶̅ + 𝐷
̅ 𝐸̅
Assume that each cut set is independent and using rare event approximation, the probability of
system failure is given by:
4.8 Consider the pumping system below. System cycles every hour. Ten minutes are required to fill the
tank. Timer is set to open contact 10 minutes after switch is closed. Operator opens switch or the
tank emergency valve if he/she notices an overpressure alarm. Develop a fault tree for this system
with the top event "Tank ruptures''.
Contact
Alarm
Switch
Power
Supply Operator Tank
Pump
Emergency
Timer Coil
Valve
The following Fault Tree provides a graphical representation of the events required to occur that
would lead to a Tank Rupture.
Tank
Tank
Over-Pressurised
Failure
Excessive
Emergency
Pumping
Valve
Fails
Circuitry Control
Failure Failure
Operator Fails to
Switch Fails Contact Fails Act
Open Open Timer Coli
Fails
Poorly Trained
Alarm Failure
Operator
G2 G1
G6 G3 G4 G3 1
1 +
A E B D E G5
A E
𝑇 = G1 ∙ G2
G1 = G3 + G4
G3 = E ∙ (A + E) = EA + EE = E
G4 = B + D
G1 = B + D + E
G2 = G3 ∙ G6
G6 = A ∙ E
G2 = E ∙ A ∙ E = AE
Therefore
𝑇 = AE ∙ (B + D + E) = AEB + AED + AEE = AE
𝑇 = AE
To find the Path Sets, we can find the Boolean compliment of the Fault Tree (ie: Success Tree)
𝑇̅ = ̅̅̅̅
𝐴𝐸
𝑇̅ = 𝐴̅ + 𝐸̅
The Minimal Path Sets of the system are therefore;
𝑃1 = 𝐴̅, 𝑃2 = 𝐸̅
4.10 Compare Design 1 and Design 2 below
A B
Design 2
C D
a) Assume that the components are non-repairable. Which is the better design?
Design 1: Consider this as two super blocks in series Left (A+B) and Right (C+D)
𝑅𝐷𝑒𝑠𝑖𝑔𝑛1 = 𝑅𝑅 . 𝑅𝐿 = (1 − 𝐹𝑅 ). (1 − 𝐹𝐿 ) = (1 − 𝐹𝐴 𝐹𝐵 ). (1 − 𝐹𝐶 𝐹𝐷 )
𝑅𝐷1 = [𝑅𝐴 𝑅𝐶 + 𝑅𝐴 𝑅𝐷 − 𝑅𝐴 𝑅𝐶 𝑅𝐷 + 𝑅𝐵 𝑅𝐶 + 𝑅𝐵 𝑅𝐷 − 𝑅𝐵 𝑅𝐶 𝑅𝐷 − 𝑅𝐴 𝑅𝐵 𝑅𝐶 − 𝑅𝐴 𝑅𝐵 𝑅𝐷
+ 𝑅𝐴 𝑅𝐵 𝑅𝐶 𝑅𝐷 ]
𝑅𝐷1 = [4𝑒 −(λ𝐴 +λ𝐶 )𝑡 − 2𝑒 −(λ𝐴 +2λ𝐶 )𝑡 − 2𝑒 −(2λ𝐴 +λ𝐶)𝑡 + 𝑒 −(2λ𝐴 +2λ𝐶 )𝑡 ]
Design 2: Consider this as two super blocks in parallel Upper (A+B) and Lower (C+D)
𝑅𝐷𝑒𝑠𝑖𝑔𝑛2 = (1 − 𝐹𝑈 . 𝐹𝐿 ) = (1 − (1 − 𝑅𝑈 )(1 − 𝑅𝐿 ))
𝑅𝐷2 = [𝑅𝑈 + 𝑅𝐿 − 𝑅𝑈 𝑅𝐿 ]
𝑅𝐷2 = [𝑅𝐴 𝑅𝐵 + 𝑅𝐶 𝑅𝐷 − 𝑅𝐴 𝑅𝐵 𝑅𝐶 𝑅𝐷 ]
Therefore; for a given time Design 2 is the better design. The following Matlab plot shows the
reliability of both designs as a function of time highlighting the fact that Design 2 is indeed a
better design.
In this problem, we want to determine the time for each design such the reliability will be
equal to or greater than 0.99. Using Matlab to solve, we get the following results:
4.11 Consider the following electric circuit for providing emergency light during a blackout.
L AC AC Supply
R Relay
B1 B2
B3 B4
Batteries
In this circuit, the relay is held open as long as AC power is available, and any of the four batteries
is capable of supplying light power. Start with the top event "No Light When Needed".
𝐶4 = 𝐴𝐶 ∙ 𝐵2 ∙ 𝐵3
𝐶5 = 𝐴𝐶 ∙ 𝐵2 ∙ 𝐵4
𝐶6 = 𝑅 ∙ 𝐵1 ∙ 𝐵3
𝐶7 = 𝑅 ∙ 𝐵1 ∙ 𝐵4
𝐶8 = 𝑅 ∙ 𝐵2 ∙ 𝐵3 No AC Power No DC Power
Available Available
𝐶9 = 𝑅 ∙ 𝐵2 ∙ 𝐵4
AC Pwr
Relay
Source
Fail
Fail
B1 B2 B3 B4
Fail Fail Fail Fail
4.12 Consider the following pumping system consisting of three identical parallel pumps, and a valve in
series.
P1
Valve
P2 Output
P3
H2O
The pumps have a constant failure rate of p (per hour) and the valve has a constant failure rate of v
(per hour) (accidental closure).
a) Develop an expression for the reliability of the system using the success tree. (Assume non-
repairable components.)
Valve
𝑅𝑠𝑦𝑠 = 𝑅𝑉 [1 − (1 − 𝑅𝑃1 )(1 − 𝑅𝑃2 )(1 − 𝑅𝑃3 )] Functions
𝑅𝑠𝑦𝑠 = 𝑅𝑉 [1 − (1 − 𝑅𝑃 )3 ]
Pump 1 Pump 2 Pump 3
𝑅𝑠𝑦𝑠 (𝑡) = 𝑒−𝜆𝑉𝑡 [3𝑒−𝜆𝑝 𝑡 − 3𝑒−2𝜆𝑝 𝑡 + 𝑒−3𝜆𝑝 𝑡 ] Functions Functions Functions
1 𝑇 1 𝑇
𝑅𝑎𝑣 = ∫ 𝑅𝑠𝑦𝑠 (𝑡)𝑑𝑡 = ∫ (3𝑒−(𝜆𝑝 +𝜆𝑉)𝑡 − 3𝑒−(2𝜆𝑝 +𝜆𝑉 )𝑡 + 𝑒−(3𝜆𝑝 +𝜆𝑉)𝑡 ) 𝑑𝑡
𝑇 0 𝑇 0
𝑇
1 3𝑒−(𝜆𝑉+𝜆𝑝 )𝑡 3𝑒−2(𝜆𝑉 +𝜆𝑝 )𝑡 𝑒−3(𝜆𝑉 +𝜆𝑝 )𝑡
𝑅𝑎𝑣 = [ − + ]
𝑇 −(𝜆𝑉 + 𝜆𝑝 ) −(2𝜆𝑝 + 𝜆𝑉 ) −(3𝜆𝑝 + 𝜆𝑉 )
0
𝑇
1 −3𝑒−(𝜆𝑉 +𝜆𝑝 )𝑡 3𝑒−2(𝜆𝑉+𝜆𝑝 )𝑡 𝑒−3(𝜆𝑉+𝜆𝑝 )𝑡
𝑅𝑎𝑣 = [ + − ]
𝑇 (𝜆𝑉 + 𝜆𝑝 ) (2𝜆𝑝 + 𝜆𝑉 ) (3𝜆𝑝 + 𝜆𝑉 )
0
c) Repeat questions (a) and (b) for the case that t < 0.1, and then approximate the reliability
functions. Find the average reliability of the system when = 0.001 per hour and T = 10 hours.
Therefore:
𝑅𝑠𝑦𝑠 (𝑡) ≅ (1 − 𝜆𝑡) × [3(1 − 𝜆𝑡) − 3(1 − 𝜆𝑡)2 + (1 − 𝜆𝑡)3 ]
When = 0.001 per hour and T = 10 hours and Assuming that 𝜆𝑉 = 𝜆𝑝 = 0.001, then:
𝑅𝑎𝑣 = 0.505
4.13 In the following system, which uses active redundancy, what is the probability that there will be no
failures in the first year of operation? Assume constant failure rates given below.
= 200×10-6 hr
= 10×10 -6 hr = 200×10-6 hr
= 200×10-6 hr
𝑅𝑠𝑦𝑠 = 𝑅1 [1 − (1 − 𝑅2 )3 ]
This problem can be solved using the Binomial distribution where probability of failure is given
by:
𝐹(𝑡) = 1 − 𝑒 −𝜆𝑡
−4
𝐹(1000) = 1 − 𝑒 −1000×2×10 = 0.1813
2
30
𝑅𝑠𝑦𝑠 (1000) = ∑ ( ) 0.1813𝑖 (1 − 0.1813)30−𝑖
𝑖
𝑖=0
30 30 30
𝑅𝑠𝑦𝑠 (1000) = ( ) 0.18130 (0.8187)30 + ( ) 0.18131 (0.8187)29 + ( ) 0.18132 (0.8187)28
0 1 2
6
1 7
5
3
2 8
4
a) Find all minimal path sets.
A minimal path set is a path set containing the least amount of branches such that no node is
traversed more than once in a path set:
c) Assuming each component has a reliability of 0.90 for a given mission time, compute the system
reliability over mission time.
The system reliability can be calculated using Path Sets or Cut Sets; we will show both methods
for comparison:
Using Path Sets
𝑅𝑠𝑦𝑠 = Pr(𝑃1 ∪ 𝑃2 ∪ 𝑃3 … .∪ 𝑃7 )
𝑅𝑠𝑦𝑠 = 1 − [(1 − 𝑅1 𝑅6 𝑅7 )(1 − 𝑅1 𝑅5 𝑅8 ) … . . (1 − 𝑅2 𝑅4 𝑅5 𝑅6 𝑅7 )]
𝑅𝑠𝑦𝑠 = 1 − [(1 − R3 )3 × (1 − R4 )3 × (1 − R5 )]
𝑅𝑠𝑦𝑠 = 1 − [(1 − 0.93 )3 × (1 − 0.94 )3 × (1 − 0.95 )]
𝑹𝒔𝒚𝒔 = 𝟎. 𝟗𝟗𝟗𝟕
4.16 In the following fault tree, find all minimal cut sets and path sets. Assuming all component
failure probabilities is 0.01, find the top event probability.
TOP
+ +
C D
A B + B
A C
Pr(𝑇) = 𝑃𝑟(𝐶1 + 𝐶2 + 𝐶3 )
Pr(𝑇) = 𝑃𝑟(𝐶1 ) + Pr(𝐶2 ) + Pr(𝐶3 ) − Pr(𝐶1 𝐶2 ) − Pr(𝐶1 𝐶3 ) − Pr(𝐶2 𝐶3 ) + Pr(𝐶1 𝐶2 𝐶3 )
Pr(𝑇) = 𝑃𝑟(𝐴𝐵) + Pr(𝐶𝐷) + Pr(𝐵𝐶) − Pr(𝐴𝐵𝐶𝐷) − Pr(𝐴𝐵𝐶) − Pr(𝐵𝐶𝐷) + Pr(𝐴𝐵𝐶𝐷)
Pr(𝑇) = 3 × 0.012 − 2 × 0.013
𝐏𝐫(𝑻) = 𝟐. 𝟗𝟖 × 𝟏𝟎−𝟒
To find the Path Sets, we can find the compliment of the Fault Tree (ie: Success Tree):
𝑆 = ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
𝐴𝐵 + 𝐶𝐷 + 𝐵𝐶
𝑆 = (𝐴̅ + 𝐵̅) ∙ (𝐶̅ + 𝐷 ̅ ) ∙ (𝐵̅ + 𝐶̅ )
𝑆 = (𝐴̅𝐶̅ + 𝐴̅𝐷 ̅ + 𝐵̅𝐶̅ + 𝐵̅𝐷 ̅ ) ∙ (𝐵̅ + 𝐶̅ )
̅ ̅ ̅
𝑆 = 𝐶 𝐵 + 𝐴𝐷 𝐵 + 𝐵𝐶 𝐵 + 𝐵̅𝐷
(𝐴 ̅ ̅ ̅ ̅ ̅ ̅ ̅ 𝐵̅ + 𝐴̅𝐶̅ 𝐶̅ + 𝐴̅𝐷
̅ 𝐶̅ + 𝐵̅𝐶̅ 𝐶̅ + 𝐵̅𝐷
̅ 𝐶̅ )
̅ ̅ ̅ ̅
𝑆 = (𝐵𝐶 + 𝐵𝐷 + 𝐴𝐶 ) ̅ ̅
Y
Sequence 1 = XY
X
I Sequence 2 = XY
Sequence 3 = X
where sequence 1 is a success and sequences 2 and 3 are failures. The cut sets of system X and Y
are X A B A C D and Y B D E A
𝑋 = ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
𝐴𝐵 + 𝐴𝐶 + 𝐷
𝑋 = (𝐴̅ + 𝐵̅ ) ∙ (𝐴̅ + 𝐶̅ )𝐷
̅
𝑋 = (𝐴̅𝐴̅ + 𝐴̅𝐶̅ + 𝐴̅𝐵̅ + 𝐵̅ 𝐶̅ ) ∙ 𝐷
̅
𝑋 = (𝐴̅ + 𝐵̅ 𝐶̅ ) ∙ 𝐷
̅
𝑋 = 𝐴̅𝐷
̅ + 𝐵̅ 𝐶̅ 𝐷
̅
4.18 In a cement production factory, a system such as the one shown below is used to provide cooling
water to the outside of the furnace. Develop an MLD for this system.
System bounds: S12, S11, S10, S9, S8, S7, S6, S5, S4, S3, S2, S1
Top event: Cooling from legs 1 and 2
Not-allowed events: Passive failures and external failures
Assumptions: Only one of the pumps or legs is sufficient to
provide the necessary cooling. Only one of the
tanks is sufficient as a source.
Tank S11
Valves Leg 2
S10 S9 S1 S2 Valves
S7 S6 S4
S8 S3
S5
Plant
Compute PC
Process-Control PCC
Compute
Common
Bus (CB)
Control CI-1 Control
Instrument - 1 CI-2
Instrument - 2
Field Field
Sensor 1 T1 T2
Sensor 2
Product
Orifice - 1 Control Orifice - 2 Control
(O-1) Valve - 1 (O-2) Valve - 2
(CV-1) (CV-2)
T = (T1 + CI-1 + PC + PCC +CB + CV-1 + O-1) ∙ (T2 + CI-2 + PC + PCC +CB + CV-2 + O-2)
𝑇 = (𝑇1 ∙ 𝑇2) + (𝑇1 ∙ CI-2) + (𝑇1 ∙ CV-2 ) + (𝑇1 ∙ O-2) + (CI-1 ∙ 𝑇2) + (CI-1 ∙ CI-2)
+ (CI-1 ∙ CV-2) + (CI-1 ∙ O-2) + (CV-1 ∙ T2 ) + (CV-1 ∙ CI-2 ) + (CV-1 ∙ CV-2 )
+ (CV-1 ∙ O-2 ) + (O-1 ∙ T2 ) + (O-1 ∙ CI-2 ) + (O-1 ∙ CV-2 ) + (O-1 ∙ O-2 ) + 𝑃𝐶
+ 𝑃𝐶𝐶 + 𝐶𝐵
c) Find the probability of the top event (Using Rare Event Approximation).
Pr(𝑇) = Pr(𝑇1) ∙ Pr(𝑇2) + Pr(𝑇1) ∙ Pr(CI-2) + Pr(𝑇1) ∙ Pr(CV-2) + ⋯ + Pr(𝑃𝐶) + Pr(𝑃𝐶𝐶) + Pr(𝐶𝐵)
Components 𝑃𝐶, 𝑃𝐶𝐶 𝑎𝑛𝑑 𝐶𝐵 are critical to the design as they are single component Cut Sets
a
I Switch
b c
Assume that components a, b and c are identical with a constant failure rate of 110-3 (per hour) and
a constant standby failure rate of 110-5 (per hour). Probability that the switch fails to operate if
component "a" fails is 510-2. Calculate the reliability of this system at t = 200 hours. (Note that
either "a" or "b and c" is required for system operation.)
Solution
𝑡
𝑅𝑠𝑦𝑠 (𝑡) = 𝑅𝐴 (𝑡) + ∫ 𝑓𝐴 (𝑡1). 𝑅𝑆𝑆 (𝑡1). 𝑅′𝐵 (𝑡). 𝑅′𝐶 (𝑡). 𝑅𝐵 (𝑡 − 𝑡1). 𝑅𝐶 (𝑡 − 𝑡1)𝑑𝑡1
0
𝑡
′ ′
𝑅𝑠𝑦𝑠 (𝑡) = 𝑒 −𝜆𝐴 𝑡 + ∫ 𝜆𝐴 𝑒 −𝜆𝐴 𝑡1 0.95 . 𝑒 −𝜆 𝐵 𝑡1 . 𝑒 −𝜆 𝐶 𝑡1 . 𝑒 −𝜆𝐵 (𝑡−𝑡1) . 𝑒 −𝜆𝐶(𝑡−𝑡1) 𝑑𝑡1
0
𝑡
′ ′
𝑅𝑠𝑦𝑠 (𝑡) = 𝑒 −𝜆𝐴 𝑡 + 𝜆𝐴 × 0.95𝑒 −(𝜆𝐵+𝜆𝐶)𝑡 ∫ 𝑒 −𝜆𝐴 𝑡1 𝑒 −(𝜆 𝐵 +𝜆 𝐶 )𝑡1 𝑒 (𝜆𝐵+𝜆𝐶)𝑡1 𝑑𝑡1
0
𝑡
′ ′
𝑅𝑠𝑦𝑠 (𝑡) = 𝑒 −𝜆𝐴 𝑡 + 𝜆𝐴 × 0.95𝑒 −(𝜆𝐵+𝜆𝐶)𝑡 ∫ 𝑒 −(𝜆𝐴 +𝜆 𝐵 +𝜆 𝐶−𝜆𝐵−𝜆𝐶)𝑡1 𝑑𝑡1
0
′ ′ 𝑡
−𝜆𝐴 𝑡 −(𝜆𝐵 +𝜆𝐶 )𝑡
−𝑒 −(𝜆𝐴 +𝜆 𝐵 +𝜆 𝐶−𝜆𝐵−𝜆𝐶)𝑡1
𝑅𝑠𝑦𝑠 (𝑡) = 𝑒 + 𝜆𝐴 × 0.95𝑒 [ ]
(𝜆𝐴 + 𝜆′ 𝐵 + 𝜆′ 𝐶 − 𝜆𝐵 − 𝜆𝐶 )
0
𝜆𝐴 × 0.95𝑒 −(𝜆𝐵+𝜆𝐶)𝑡 ′ ′
𝑅𝑠𝑦𝑠 (𝑡) = 𝑒 −𝜆𝐴 𝑡 + (1 − 𝑒 −(𝜆𝐴 +𝜆 𝐵 +𝜆 𝐶 −𝜆𝐵−𝜆𝐶)𝑡 )
(𝜆𝐴 + 𝜆′ 𝐵 + 𝜆′ 𝐶 − 𝜆𝐵 − 𝜆𝐶 )
a e
c d a
b d
Use the following component reliability values to determine the probability that critical data will not
be saved following a loss of power: Ra = 0.99, Rb = 0.98, Rc = 0.999, Rd = 0.998, Re = 0.99.
The following Event Sequence Diagram shows the sequence of events possible for the system.
𝑆𝑒𝑞 3 = 𝐴̅𝐵𝐶
𝑆𝑒𝑞 3 = (𝑎̅ + 𝑏̅) ∙ (𝑐 + 𝑑 ) ∙ (𝑎 + 𝑑𝑒 )
𝑆𝑒𝑞 3 = 𝑎̅𝑑𝑒 + 𝑏̅𝑐𝑎 + 𝑏̅𝑑𝑎 + 𝑏̅𝑑𝑒
𝑆𝑒𝑞 4 = 𝐴
𝑆𝑒𝑞 4 = 𝑎𝑏
System Failure
𝐹𝑠𝑦𝑠 = 𝑆𝑒𝑞 3 + 𝑆𝑒𝑞 4
𝐹𝑠𝑦𝑠 = 𝑎̅𝑑𝑒 + 𝑏̅𝑐𝑎 + 𝑏̅𝑑𝑎 + 𝑏̅𝑑𝑒 + 𝑎𝑏
𝐹𝑠𝑦𝑠 = 0.99 × 0.002 × 0.01 + 0.98 × 0.001 × 0.01 + 0.98 × 0.002 × 0.01 + 0.98 × 0.002
× 0.01 + 0.01 × 0.02
𝑭𝒔𝒚𝒔 = 𝟐. 𝟔𝟖𝟖 × 𝟏𝟎−𝟒
∞ ∞
2 1
𝑀𝑇𝑇𝐹𝑆𝑌𝑆 = ∫ 𝑅𝑆𝑌𝑆 (𝑡)𝑑𝑡 = ∫ (2𝑒 −𝜆𝑡 − 𝑒 −2𝜆𝑡 )𝑑𝑡 = −
0 0 𝜆 2𝜆
𝑖𝑓 𝜆𝑖 = 1 × 10−2
2 1
𝑀𝑇𝑇𝐹𝑆𝑌𝑆 = − = 150 ℎ𝑟𝑠
0.01 2 × 0.01
Either the MTTF quoted as 1000hrs has been overestimated or the failure rate from field data is
incorrect. Either way, there is a great deal of inconsistency in the component level and system level
failure rates quoted in the problem.
B G
Input C H Output
F
D I
Develop a fault tree for this system with the top event of "No Output from the System". Calculate the
reliability of this system for a 100 hour mission. Assume MTTF of 600 hours for all components.
Therefore System success (S) is assured is at least one of the path sets occurs.
𝑆 = 𝐵̅𝐺̅ + 𝐵̅𝐸̅ 𝐻
̅ + 𝐶̅ 𝐻
̅+𝐷
̅𝐼̅ + 𝐷
̅ 𝐹̅ 𝐻
̅
No Output from
System
B G B E H C H D I D F H
System failure can be found be taking the complement of this expression as follows:
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
𝑆̅ = 𝐵
̅ 𝐺̅ + 𝐵̅𝐸̅ 𝐻
̅ + 𝐶̅ 𝐻̅+𝐷 ̅𝐼̅ + 𝐷̅ 𝐹̅ 𝐻
̅
𝑆̅ = (𝐵 + 𝐺) ∙ (𝐵 + 𝐸 + 𝐻) ∙ (𝐶 + 𝐻) ∙ (𝐷 + 𝐼) ∙ (𝐷 + 𝐹 + 𝐻)
𝑆̅ = (𝐵𝐵 + 𝐵𝐸 + 𝐵𝐻 + 𝐺𝐵 + 𝐺𝐸 + 𝐺𝐻) ∙ (𝐶𝐷 + 𝐶𝐼 + 𝐻𝐷 + 𝐻𝐼) ∙ (𝐷 + 𝐹 + 𝐻)
𝑆̅ = (𝐵 + 𝐺𝐸 + 𝐺𝐻) ∙ (𝐶𝐷 + 𝐶𝐼𝐹 + 𝐻𝐷 + 𝐻𝐼)
This expression cannot be reduced any further and therefore, this provides us with the system Cut Sets:
𝐶1 = 𝐵𝐶𝐷; 𝐶2 = 𝐵𝐶𝐼𝐹 ; 𝐶3 = 𝐵𝐻𝐷; 𝐶4 = 𝐵𝐻𝐼; 𝐶5 = 𝐺𝐸𝐶𝐷; 𝐶6 = 𝐺𝐸𝐶𝐼𝐹; 𝐶7 = 𝐺𝐻𝐷; 𝐶8 = 𝐺𝐻𝐼;
Using the Path Sets of the system, we can find the System Reliability as follows:
𝑅𝑠𝑦𝑠 (𝑡) = Pr(P1 ∪ P2 ∪ P3 ∪ P4 ∪ P5 )
𝑅𝑠𝑦𝑠 (𝑡) = 1 − (1 − 𝑅𝐵 𝑅𝐺 )(1 − 𝑅𝐵 𝑅𝐸 𝑅𝐻 )(1 − 𝑅𝐶 𝑅𝐻 )(1 − 𝑅𝐷 𝑅𝐼 )(1 − 𝑅𝐷 𝑅𝐹 𝑅𝐻 )
As each component is now considered identical, the expression for System Reliability can be
simplified as follows:
𝑅𝑠𝑦𝑠 (𝑡) = 3𝑅2 + 2𝑅 3 − 7𝑅 4 − 𝑅 5 + 8𝑅6 − 5𝑅 7 + 𝑅 8
Two standby subsystems (A and B) and (C and D) are placed in parallel. Unit E should be in series with
the two standby subsystems. Assume time to failure of A, B, C, D and E are exponentially distributed and
switching is perfect for the standby subsystems.
As a designer, if you are asked to select Unit E such that its contribution to the failure of the whole system
is less than 10% of the total system failure probability for a mission time of T = 1000 hrs, what should be
the minimum acceptable constant failure rate of Unit E?
Solution
As all component failure rates are equal, the reliability of subsystem (A and B) is the same as the
reliability of subsystem (C and D).
Therefore
𝑡
𝑅𝑠𝑦𝑠 (𝑡) = [𝑅𝐴𝐵 (𝑡) + 𝑅𝐶𝐷 (𝑡) − 𝑅𝐴𝐵 (𝑡)𝑅𝐶𝐷 (𝑡)] × 𝑅𝐸 (𝑡)
2
𝑅𝑠𝑦𝑠 (𝑡) = [2 × (1 + 0.0001𝑡)𝑒 −0.0001𝑡 − ((1 + 0.0001𝑡)𝑒 −0.0001𝑡 ) ] × 𝑒 −𝜆𝐸 𝑡
𝑅 (1000)
Solving this expression for 𝜆𝐸 such that 𝑅 𝐸 ≤ 0.1; we get 𝝀𝑬 ≤ 𝟐. 𝟒𝟒 × 𝟏𝟎−𝟗 𝒉𝒓−𝟏
𝑠𝑦𝑠 (1000)
B E
A D G
C F H
System Output
Occurs
F G
A C A B D A B E A C D E
The following truth table shows the combination of events that will lead to failure (1) or success (0)
of the system. For simplicity, the table does not show components A or H as the failure of either of
these components will lead to failure of the system.
The truth table above shows all path set combinations for the system whereas, the path sets discussed
previously are minimal path sets.
A X C
IN
Components A, B, C, D and X are identical and have a Weibull time to failure distribution model with
shape parameter β = 1.6
b) Using the results from part a), determine the scale parameter if we desire that the total probability of
system failure not to exceed 10−2 over a ten year period.
𝑡 𝛽 𝑡 𝛽 𝑡 𝛽 𝑡 𝛽
−( ) −( ) −2( ) −4( )
𝑅𝑠𝑦𝑠 (𝑡) = 𝑒 𝛼 + (1 − 𝑒 𝛼 ) × {2𝑒 𝛼 −𝑒 𝛼 }
Solving this expression for 𝛼 such that 𝑅𝑠𝑦𝑠 (87600) ≥ 0.99; we get 𝜶 ≥ 𝟐𝟖𝟐𝟐𝟑𝟑 𝒉𝒓𝒔
d) Use the minimal cut sets to verify that the probability of failure of this system meets the 10−2 over
a ten year period criterion. Explain the reason for any discrepancies.
𝑅𝑠𝑦𝑠 = 1 − Pr(𝐶1 ∪ 𝐶2 ∪ 𝐶3 ∪ 𝐶4 )
𝑅𝑠𝑦𝑠 ≥ 1 − 4 × (1 − 𝑅𝑠𝑦𝑠 )3
3
87600 1.6
−( )
𝑅𝑠𝑦𝑠 ≥ 1 − 4 × (1 − 𝑒 282233 )
𝑅𝑠𝑦𝑠 ≥ 0.9884
This discrepancy is due to the use of rare event approximation and equates to an error of 0.161 %.
However, this error is conservative by nature as it underestimates the system reliability.
A C
B D
Components A and B are share load, but C and D are parallel. Components A and B are identical and follow
an exponential time t failure distribution with the full load failure rate being twice that of the half load
failure rate. If 𝜆𝐶 = 0.001 ℎ𝑟 −1 𝑎𝑛𝑑 𝜆𝐷 = 0.002 ℎ𝑟 −1 , determine the MTTF of this system if the total
system reliability should be at least 0.99 for a mission length of 10 hrs.
For components (A and B), the reliability of the shared load configuration is given by Equation 4.28 as
follows:
𝑡
2𝜆ℎ 𝑒 −𝜆𝑓𝑡
𝑅𝐴𝐵 = 𝑒 −2𝜆ℎ 𝑡
+ (1 − 𝑒 −𝑡(2𝜆ℎ −𝜆𝑓) )
(2𝜆ℎ − 𝜆𝑓 )
𝑅𝐶𝐷 = 𝑅𝐶 + 𝑅𝐷 − 𝑅𝐶 𝑅𝐷
2𝜆ℎ 𝑒 −𝜆𝑓𝑡
𝑅𝑆𝑌𝑆 = {𝑒 −2𝜆ℎ 𝑡 + (1 − 𝑒 −𝑡(2𝜆ℎ −𝜆𝑓) )} × {𝑒 −0.001𝑡 + 𝑒 −0.002𝑡 − 𝑒 −0.003𝑡 }
(2𝜆ℎ − 𝜆𝑓 )
−2𝜆ℎ ×10
2𝜆ℎ 𝑒 −2𝜆ℎ ×10
𝑅𝑆𝑌𝑆 = {𝑒 + (1 − 𝑒 −10×(2𝜆ℎ −2𝜆ℎ ) )} × {𝑒 −0.001×10 + 𝑒 −0.002×10 − 𝑒 −0.003×10 }
(2𝜆ℎ − 2𝜆ℎ )
For a System reliability of 0.99 at T= 10 hrs, we can now solve for 𝝀𝒉 ≤ 𝟒. 𝟗𝟐𝟔𝟔 × 𝟏𝟎−𝟒 𝒉𝒓−𝟏
∞ ∞
Therefore, in this case the MTTF of the system is found to be 587.75 hours
A C
B D
If Pr(A) = Pr(B) = Pr(C) = Pr(D) = Pr(E)=0.05 is the failure probability of components of this system
for a given mission, determine system failure probability using any two methods and discuss which
one is more accurate and why?
𝑅𝑠𝑦𝑠 = 1 − Pr(𝐶1 ∪ 𝐶2 ∪ 𝐶3 ∪ 𝐶4 )
𝑹𝒔𝒚𝒔 ≥ 𝟎. 𝟗𝟗𝟒𝟕𝟓
This discrepancy is due to the use of rare event approximation and equates to an error of 0.43%. This
error is conservative by nature as it underestimates the system reliability. Using Path Sets gives a more
accurate result
Out
In S
C
X
Subsystem X is made of share load components B & C (with identical constant failure rates 𝜆𝑓 & 𝜆ℎ
for full and half load operation respectively). Upon failure of component A, subsystem X is used by
switching via a perfect switch S. Assume no standby failure rate. Assume a constant failure rate for
component A (ie: 𝜆𝐴 )
𝑡
We are told that the switch is perfect and that no standby failure rate. Therefore, we get:
If component A fails at time (t1), subsystem X is called upon. The reliability of subsystem X is given
by:
𝑡
𝑡
−𝜆𝑓 (𝑡−𝑡1)
𝑅𝑋 (𝑡 − 𝑡1) = 𝑒 + 𝜆𝑓 ∫ 𝑒 −𝜆𝑓(𝑡2−𝑡1) ∙ 𝑒 −𝜆𝑓 (𝑡−𝑡2) 𝑑𝑡2
𝑡1
Substituting the reliability expression for subsystem X back into the system reliability expression we
get:
𝑡
𝑡
−𝜆𝑓 𝑡
𝑅𝑠𝑦𝑠 (𝑡) = 𝑒−𝜆𝐴𝑡 + 𝜆𝐴 𝑒 ∫ 𝑒−𝜆𝐴𝑡1+𝜆𝑓𝑡1 ∙ {1 + 𝜆𝑓 (𝑡 − 𝑡1)} 𝑑𝑡1
0
𝑡 𝑡 𝑡
−𝜆𝑓 𝑡 −𝑡1(𝜆𝐴−𝜆𝑓 ) −𝑡1(𝜆𝐴−𝜆𝑓 ) −𝑡1(𝜆𝐴−𝜆𝑓 )
𝑅𝑠𝑦𝑠 (𝑡) = 𝑒−𝜆𝐴𝑡 + 𝜆𝐴 𝑒 [∫ 𝑒 𝑑𝑡1 + ∫ 𝜆𝑓 𝑡𝑒 𝑑𝑡1 − ∫ 𝜆𝑓 𝑡1𝑒 𝑑𝑡1]
0 0 0
𝟏 − 𝒆−𝒕(𝝀𝑨−𝝀𝒇) 𝟏 − 𝒆−𝒕(𝝀𝑨−𝝀𝒇) 𝟏 𝒕 𝟏
𝑹𝒔𝒚𝒔 (𝒕) = 𝒆−𝝀𝑨𝒕 + 𝝀𝑨 𝒆−𝝀𝒇𝒕 [( ) + 𝝀𝒇 𝒕 ( ) − 𝝀𝒇 { 𝟐
−( + ) 𝒆−𝒕(𝝀𝑨−𝝀𝒇) }]
(𝝀𝑨 − 𝝀𝒇 ) (𝝀𝑨 − 𝝀𝒇 ) (𝝀𝑨 − 𝝀𝒇 ) (𝝀𝑨 − 𝝀𝒇 ) (𝝀𝑨 − 𝝀𝒇 )𝟐
A D
C F
B E
Determine the MTTF of this system if all the components are identical with a failure rate of 𝜆𝑖 =
0.005/𝑚𝑜𝑛𝑡ℎ
A D
A D
F F
B E B E
T1
T2
G1 G2 2 out of 3
E Gate
Mutually
Exclusive OR
Gate 2/3
A B C D F G H
Cut Sets
T1 T2
AC EFG
AD EFH
BC EGH
BD
Path Sets
T1 T2
A.B E
C.D F.G
F.H
G.H
T1
T2
G1 G2 2 Out of 3
E AND Gate
A B C D F G H
Component A B C D E F G H
Failure Rate 0.08 0.02 0.07 0.05 0.01 0.05 0.05 0.05
Pr(𝑇1) = 0.08 × 0.07 + 0.08 × 0.05 − 0.08 × 0.07 × 0.05 + 𝐵𝐶 + 𝐵𝐷 − 𝐵𝐶𝐷 − 𝐴𝐵𝐶 − 𝐴𝐵𝐷 + 𝐴𝐵𝐶𝐷
𝐏𝐫(𝑻𝟏) = 𝟎. 𝟎𝟏𝟏𝟓
𝐏𝐫(𝑻𝟐) = 𝟎. 𝟎𝟎𝟎𝟎𝟕𝟐𝟓
4.33 Consider the standby system composed of three identical units. It is known that failure probability of
each block follows a normal distribution. Note that the following censored failure data has been obtained
in a life test involving 10 of the same blocks. Assume perfect switching and no failure in standby mode.
Use Kaplan-Meier plotting method to address the following questions.
Test No. 1 2 3 4 5 6 7 8 9 10
Time to Failure 298 492 692 850 980 1200+ 1200+ 1200+ 1200+ 1200+
Using the Kaplan-Meier plotting process, we can estimate the parameters of the individual
components assuming a Normal Distribution as follows:
Kaplan-Meier Method
i Time (hrs) j 1-1/j R(tj) F(tj)
0 1
1 298 10 0.900 0.900 10.000
2 492 9 0.889 0.800 20.000
3 692 8 0.875 0.700 30.000
4 850 7 0.857 0.600 40.000
5 980 6 0.833 0.500 50.000
6 1200 5
7 1200 4
8 1200 3
9 1200 2
10 1200 1
Using the trend line shown in the pot above, we can estimate the mean and standard deviation of an
individual component as follows:
𝜇𝑖 = 𝑡50 = 𝟏𝟎𝟏𝟎 𝒉𝒓𝒔
A joint distribution of 3 components each having identical Normal distributions, will also be a
Normal distribution with the following parameters.
𝑀𝑇𝑇𝐹𝑠𝑦𝑠 = 𝜇𝑠𝑦𝑠 = 𝜇1 + 𝜇2 + 𝜇3
4.34 The reliability block diagram of a system is shown in the following figure with component reliability
noted in each block. What is the reliability of the system?
0.90 0.78
0.98 0.90
0.99 0.80
This problem can be solved by pivoting around the central component (Y) in the case where Y is first
assumed to be reliable and then where Y is assumed to be unreliable as follows:
𝑹𝒔𝒚𝒔 = 𝟎. 𝟗𝟑𝟒𝟑
a) Write the reliability function R(t) of the following shared load system. Assume that each unit is
identical and can provide 1/3 load, 2/3 load or full load. With the flowing constant failure rates
𝜆1/3 𝑙𝑜𝑎𝑑 = 10−4 ℎ𝑟 −1 ; 𝜆2/3 𝑙𝑜𝑎𝑑 = 10−3 ℎ𝑟 −1 ; 𝜆𝑓𝑢𝑙𝑙 𝑙𝑜𝑎𝑑 = 10−2 ℎ𝑟 −1
Without knowing what the failure times (t1) and (t2) are, we cannot estimate the reliability of the
system at 10,000 hrs.
A B
Switch
f(t) g(t)
C D
4.37 Draw a success tree for the system shown below (with “system output available” as the top event).
System Output
A Available
B F
E H
C G A D E G B F H
C G
Solution:
This problem is solved by setting up four weibull distributions (one for each belt) and assuming
that each belt is independent.
𝑡 𝛽
𝐹(𝑡) = 1 − exp [− ( ) ]
𝛼
For the 4 belts:
𝑡1 1.34
𝐹(𝑡1 ) = 1 − exp [− ( ) ]
2500
𝑡2 1.34
𝐹(𝑡2 ) = 1 − exp [− ( ) ]
3400
𝑡3 1.34
)
𝐹(𝑡3 = 1 − exp [− ( ) ]
8000
𝑡4 1.34
𝐹(𝑡4 ) = 1 − exp [− ( ) ]
6100
𝐹(𝑡1 ) = 0.008585
𝐹(𝑡2 ) = 0.005694
𝐹(𝑡3 ) = 0.001813
𝐹(𝑡4 ) = 0.002606
Solution:
a) For 2 parallel units:
𝑅(𝑡) = 1 − [𝐹1 (𝑡)𝐹2 (𝑡)] = 1 − [(1 − exp(−𝜆1 𝑡))(1 − exp(−λ2 𝑡))]
At t = 100
𝑅(100) = 1 − [(1 − exp(−0.0034 × 100))(1 − exp(−0.0105 × 100))]
𝑅(100) = 1 − [0.28823 × 0.650062] = 0.8126
b) For a standby system with perfect switching and no standby failure (Assuming unit 1 is active
and unit 2 is the standby unit)
𝑡
𝑅(𝑡) = 𝑅1 (𝑡) + ∫ 𝑓1 (𝑡1 ) 𝑅2 (𝑡 − 𝑡1 )𝑑𝑡1
0
𝑡
𝑅(𝑡) = exp(−𝜆1 𝑡) + ∫ 𝜆1 exp(−𝜆1 𝑡1 ) exp[−𝜆2 (𝑡 − 𝑡1 )] 𝑑𝑡1
0
𝑡
𝑅(𝑡) = exp(−λ1 𝑡) + 𝜆1 ∫ exp(−𝜆1 𝑡1 − 𝜆2 𝑡 + 𝜆2 𝑡1 ) 𝑑𝑡1
0
𝑡
𝑅(𝑡) = exp(−λ1 𝑡) + 𝜆1 exp(−λ2 𝑡) ∫ exp[−t1 (λ1 − λ2 )] 𝑑𝑡1
0
𝜆1 exp(−λ2 𝑡)
𝑅(𝑡) = exp(−λ1 𝑡) + {1 − exp[−t(λ1 − λ2 )]}
λ1 − λ2
0.0034 exp(−0.0105 × 100 )
𝑅(100) = exp(−0.0034 × 100) +
0.0034 − 0.0105
× {1 − exp[−100(0.0034 − 0.0105)]}
𝑅(100) = 0.8850
Solution:
a) Event tree for this problem:
𝐵̅ ∙ 𝐴̅ S
𝐵̅ ∙ 𝐴 F
𝐵 ∙ 𝐶̅ ∙ 𝐴̅ S
𝐵 ∙ 𝐶̅ ∙ 𝐴 F
𝐵∙𝐶 F
Failure 2:
𝐵 ∙ 𝐶̅ ∙ 𝐴 = (𝑑 + 𝑒 ∙ 𝑓) ∙ (𝑐 ̅̅̅̅̅̅̅̅̅̅̅̅̅
+ 𝑑 ∙ 𝑓) ∙ (𝑎 ∙ 𝑏 + 𝑎 ∙ 𝑐)
̅̅̅̅ = 𝑐̅ ∙ (𝑑̅ + 𝑓 ̅) = 𝑐̅𝑑̅ + 𝑐̅𝑓 ̅
𝐶̅ = 𝑐̅ ∙ 𝑑𝑓
𝐵 ∙ 𝐶̅ ∙ 𝐴 = (𝑑 + 𝑒𝑓) ∙ (𝑐̅𝑑̅ + 𝑐̅𝑓 )̅ ∙ (𝑎𝑏 + 𝑎𝑐)
= (𝑑𝑐̅𝑑̅ + 𝑑𝑐̅𝑓 ̅ + 𝑒𝑓𝑐̅𝑑̅ + 𝑒𝑓𝑐̅𝑓 ̅) ∙ (𝑎𝑏 + 𝑎𝑐)
= (𝑑𝑐̅𝑓 ̅ + 𝑒𝑓𝑐̅𝑑̅) ∙ (𝑎𝑏 + 𝑎𝑐)
= 𝑎𝑏𝑐̅𝑑̅𝑓 ̅ + 𝑎𝑐𝑐̅𝑑̅𝑓 ̅ + 𝑎𝑏𝑐̅𝑑̅𝑒𝑓 + 𝑎𝑐𝑐̅𝑑̅𝑒𝑓
= 𝑎𝑏𝑐̅𝑑̅𝑓 ̅ + 𝑎𝑏𝑐̅𝑑̅𝑒𝑓
Failure 3:
𝐵 ∙ 𝐶 = (𝑑 + 𝑒 ∙ 𝑓) ∙ (𝑐 + 𝑑 ∙ 𝑓) = 𝑑𝑐 + 𝑑𝑑𝑓 + 𝑐𝑒𝑓 + 𝑑𝑒𝑓𝑓
= 𝑑𝑐 + 𝑑𝑓 + 𝑐𝑒𝑓 + 𝑑𝑒𝑓 = 𝑑𝑐 + 𝑑𝑓 + 𝑐𝑒𝑓
Solution:
a) This is equivalent to writing the path sets of the system. Therefore by inspection of the block
diagram the minimal path sets are:
𝐴𝐵𝐷, 𝐴𝐵𝐸, 𝐴𝐶𝐸
Therefore
𝑅𝑠𝑦𝑠 = Pr(𝐴𝐵𝐷) + Pr(𝐴𝐵𝐸) + Pr(𝐴𝐶𝐸) − Pr(𝐴𝐵𝐷𝐸) − Pr(𝐴𝐵𝐶𝐷𝐸) − Pr(𝐴𝐵𝐶𝐸)
𝑅𝑠𝑦𝑠 = 𝑅(𝐴𝐵𝐷) + 𝑅(𝐴𝐵𝐸) + 𝑅(𝐴𝐶𝐸) − 𝑅(𝐴𝐵𝐷𝐸) − 𝑅(𝐴𝐵𝐶𝐷𝐸) − 𝑅(𝐴𝐵𝐶𝐸)
F1
F2
F3
Solution:
For state S
𝑆 = 𝑋̅𝑌̅ = ̅̅̅̅̅̅̅̅̅̅̅̅̅̅
(𝐴𝐵 + 𝐵𝐶) ∙ ̅̅̅̅̅̅̅̅̅̅̅̅
(𝐵𝐷 + 𝐸)
̅̅̅̅ ∙ 𝐵𝐶
𝑋̅ = 𝐴𝐵 ̅̅̅̅ = (𝐴̅ + 𝐵̅) ∙ (𝐵̅ + 𝐶̅ ) = 𝐴̅𝐵̅ + 𝐴̅𝐶̅ + 𝐵̅𝐵̅ + 𝐵̅𝐶̅ = 𝐴̅𝐵̅ + 𝐴̅𝐶̅ + 𝐵̅ + 𝐵̅𝐶̅
= 𝐵̅ + 𝐴̅𝐶̅
𝑌 = 𝐵𝐷 ∙ 𝐸 = (𝐵̅ + 𝐷
̅ ̅̅̅̅ ̅ ̅ ) ∙ 𝐸̅ = 𝐵̅𝐸̅ + 𝐷 ̅ 𝐸̅
̅ ̅
𝑆 = (𝐵̅ + 𝐴𝐶 ) ∙ (𝐵̅𝐸̅ + 𝐷 ̅ 𝐸̅ ) = 𝐵̅𝐵̅𝐸̅ + 𝐵̅𝐷 ̅ 𝐸̅ + 𝐴̅𝐶̅ 𝐵̅𝐸̅ + 𝐴̅𝐶̅ 𝐷
̅ 𝐸̅
= 𝐵̅𝐸̅ + 𝐵̅𝐷 ̅ 𝐸̅ + 𝐴̅𝐶̅ 𝐵̅𝐸̅ + 𝐴̅𝐶̅ 𝐷
̅ 𝐸̅
= 𝐵̅𝐸̅ + 𝐴̅𝐶̅ 𝐷 ̅ 𝐸̅
= Pr(𝐵𝐸 ) + Pr(𝐴̅𝐶̅ 𝐷
̅ ̅ ̅ 𝐸̅) − Pr(𝐴̅𝐵̅𝐶̅ 𝐷 ̅ 𝐸̅ )
= 0.97804 + 0.891688 − (0.87385)
= 0.9959
Note: rare event approximation could NOT be used for the above instance because the
majority of the terms were probabilities of success which are large and the additional term
was needed to obtain a realistic answer.
For state F1
𝐹1 = 𝑋̅𝑌 = (𝐵̅ + 𝐴̅𝐶̅ ) ∙ (𝐵𝐷 + 𝐸) = (𝐵̅𝐵𝐷 + 𝐵̅𝐸 + 𝐵̅𝐵𝐷 + 𝐴̅𝐶̅ 𝐸 )
= 𝐵̅𝐸 + 𝐴̅𝐶̅ 𝐸 = Pr(𝐵̅𝐸) + Pr(𝐴̅𝐶̅ 𝐸 ) − Pr(𝐴̅𝐶̅ 𝐵̅𝐸)
= 0.00196 + 0.001805 − (0.0017689)
= 1.996 × 10−3
For state F2
𝐹2 = 𝑋𝑌̅ = (𝐴𝐵 + 𝐵𝐶) ∙ (𝐵̅𝐸̅ + 𝐷 ̅ 𝐸̅ ) = 𝐴𝐵𝐵̅ 𝐸̅ + 𝐴𝐵𝐷
̅ 𝐸̅ + 𝐵𝐶𝐵̅𝐸̅ + 𝐵𝐶𝐷
̅ 𝐸̅
̅ ̅
= 𝐴𝐵𝐷 𝐸 + 𝐵𝐶𝐷 𝐸 ̅ ̅
≈ Pr(𝐴𝐵𝐷 ̅ 𝐸̅ ) + Pr(𝐵𝐶𝐷̅ 𝐸̅ )
= 0.0098892 + 0.0098892 = 1.976 × 10−3
Note: in this case, rare event approximation was used since the minus term is orders of
magnitude smaller.
4.43 Data show that punctures in a car tire occur at a constant rate of one-per-10,000 miles. In a car with
the same 4 tires operating, what is the probability that at least one tire puncture occurs after 15,000
miles.
Solution:
An exponential distribution should be used since a constant failure rate is stated.
1
𝜆=
10000
Since there are 4 tyres on the same car and any failure will stop the system, this is a series system
with 4 identical sub-systems:
∴ Pr(t < 15000)sys = R(t)4 = 0.22314 = 0.002479
Time Interval 1 2 3 4 5 6
No. of fires 6 8 16 6 11 11
Do you believe the fire incidents are time-dependent? Prove your answer.
Using the Laplace test to determine if a trend exists, we get the following results:
𝑛
1 1 (6 + 14 + 30 + 36 + 47 + 58)
𝑛 ∑ 𝑡𝑖 − 2 𝑡𝑜 − 0.5 × 58
𝑈= 𝑖=1
= 6 = 𝟎. 𝟒𝟏𝟒𝟓
1 1
𝑡𝑜 √ 58√
12(𝑛) 12(6)
As |𝑈| > 0.3, we can see a slightly positive trend in rate of fire incident; however more evidence
would be required before a conclusive observation could be made.
𝑛 6
𝛽̂ = 𝑛 = = 𝟏. 𝟏𝟗𝟏𝟑
𝑡𝑜 58 58 58 58 58
∑ 𝑙𝑛 𝑙𝑛 + 𝑙𝑛 + 𝑙𝑛 + 𝑙𝑛 + 𝑙𝑛
𝑖=1 𝑡𝑖 6 14 30 36 47
The β value of 1.19 also indicates a slightly increasing rate of occurrence but again, more evidence
would be required before a conclusive observation could be made as this is very close to a value of 1
which would indicate a Homogeneous Poisson Process with no trend.
𝑛 6
𝑅𝑂𝐶𝑂𝐹 = 𝜆̂ = ̂
= = 𝟒. 𝟕𝟓𝟖 × 𝟏𝟎−𝟐 𝒉𝒓−𝟏
𝑡𝑜 𝛽 581.1913
Main
Transformer
Main MTTF = 104 hr
MTTF = 106 hr Generator MTTR = 10 hrs
MTTR = 48hrs
𝐿𝐵 = 𝐿𝐵1 ∙ 𝐿𝐵2
𝑀𝐺𝐶 = 𝑀𝐺 + 𝑀𝑇
𝐺𝐶1 = 𝐺1 + 𝐵1 ∙ 𝐵𝐶1
𝐺𝐶2 = 𝐺2 + 𝐵2 ∙ 𝐵𝐶2
𝜆 𝜆 1 1
𝑞(𝑡) = − exp[−(𝜆 + 𝜇)𝑡] 𝑤ℎ𝑒𝑟𝑒 𝜇= & 𝜆=
𝜆+𝜇 𝜆+𝜇 𝑀𝑇𝑇𝑅 𝑀𝑇𝑇𝐹
1 1
𝜇= & 𝜆=
48 106
10−6 10−6 1
𝑞𝐿𝐵 (24) = − exp [− (10−6 + ) 24] = 𝟏. 𝟖𝟖𝟗 × 𝟏𝟎−𝟓
1 1 48
10−6 + 10−6 +
48 48
1 1
𝜇= & 𝜆=
48 106
10−6 10−6 1
𝑞𝑀𝑇 (24) = − exp [− (10−6 + ) 24] = 𝟏. 𝟖𝟖𝟗 × 𝟏𝟎−𝟓
1 1 48
10−6 + 10−6 +
48 48
1 1
𝜇= & 𝜆=
10 104
10−4 10−4 1
𝑞𝑀𝐺 (24) = − exp [− (10−4 + ) 24] = 𝟗. 𝟎𝟖𝟔 × 𝟏𝟎−𝟒
1 1 10
10−4 + 10−4 +
10 10
1 1
𝜇= & 𝜆=
10 105
10−5 10−5 1
𝑞𝐵𝐶 (24) = − exp [− (10−5 + ) 24] = 𝟗. 𝟎𝟗𝟐 × 𝟏𝟎−𝟓
1 1 10
10−5 + 10−5 +
10 10
Batteries (𝑩𝟏 + 𝑩𝟐 ):
1 1
𝜇= & 𝜆=
4 103
10−3 10−3 1
𝑞𝐵 (24) = − exp [− (10−3 + ) 24] = 𝟑. 𝟗𝟕𝟒 × 𝟏𝟎−𝟑
1 1 4
10−3 + 10−3 +
4 4
1 1
𝜇= & 𝜆=
20 102
10−2 10−2 1
𝑞𝐺 (24) = − exp [− (10−2 + ) 24] = 𝟏. 𝟐𝟕𝟐 × 𝟏𝟎−𝟏
1 1 20
10−2 + 10−2 +
20 20
System Unavailability
𝑞𝑃 (24) = [(9.086 × 10−4 + 1.889 × 10−5 ) ∙ (1.272 × 10−1 + 3.974 × 10−3 × 9.092 × 10−5 )2 ]
+ (1.889 × 10−5 )2
5.3 An operating system is repaired each time it has a failure and is put back into service as soon as possible
(monitored system). During the first 10,000 hours of service, it fails five times and is out of service
for repair during the following times:
Hours
1000 - 1050
3960 - 4000
4510 - 4540
6130 - 6170
8520 - 8560
Using the Laplace test to determine if a trend exists for a Time Terminated data series, we get the
following results:
As |𝑈| << 0.3, we can assume that there is no trend in the data to support an increasing ROCOF.
𝑛 5
𝛽̂ = 𝑛 = = 𝟏. 𝟎𝟗𝟑
𝑡 9800 9800 9800 9800 9800
∑ 𝑙𝑛 𝑡𝑜 𝑙𝑛 1000 + 𝑙𝑛 3660 + 𝑙𝑛
4510
+ 𝑙𝑛 6130 + 𝑙𝑛
8520
𝑖=1 𝑖
The β value of 1.09 also indicates a lack of trend or time dependent variation in the rate of
occurrence of failures. We can therefore assume a HPP with constant ROCOF.
𝑛 5
𝑅𝑂𝐶𝑂𝐹 = 𝜆̂ = ̂
= = 𝟐. 𝟏𝟔𝟕 × 𝟏𝟎−𝟒 𝒉𝒓−𝟏
𝑡𝑜 𝛽 98001.093
Repairs
As |𝑈| ≫ 0.3, we can assume that there is a trend in the data to support an increasing repair rate.
𝑛 5
𝛽̂ = 𝑛−1 = = 𝟏. 𝟕𝟏𝟑
𝑡 200 200 200 200
∑ 𝑙𝑛 𝑡𝑜 𝑙𝑛 + 𝑙𝑛 90 + 𝑙𝑛 120 + 𝑙𝑛 160
𝑖
50
𝑖=1
The β value of 1.71 also indicates a trend or time dependent variation in the rate of occurrence of
repairs. We can therefore assume a NHPP for the repair rate.
𝑛 5
𝜇̂ = ̂
= = 𝟓. 𝟕𝟐 × 𝟏𝟎−𝟒 𝒉𝒓−𝟏
𝑡𝑜 𝛽 2001.713
𝛽 −𝜆𝑡 𝛽 ] −4 ×1001093 ]
𝑅(𝑡 + 𝑡1 |𝑡) = 𝑒 −[𝜆(𝑡+𝑡1) => 𝑅(100) = 𝑒 −[2.167×10 = 𝟎. 𝟗𝟔𝟕𝟑
𝜇 1 5 1 5
𝑎(𝑡) = 𝑤ℎ𝑒𝑟𝑒 𝜇 = = = 0.025 𝜆= = = 0.00051
𝜆+𝜇 𝑀𝑇𝑇𝑅 200 𝑀𝑇𝐵𝐹 9800
0.025
𝑎(𝑡) = = 𝟎. 𝟗𝟖
0.00051 + 0.025
c) If the system has been operating for 10 hours without a failure, what is the probability that it will
continue to operate for the next 10 hours without a failure?
𝛽 −𝑡 𝛽 ]
Pr(𝑋 ≥ 20|𝑋 > 10) = 𝑅(𝑡 + 𝑡1 |𝑡) = 𝑒 −𝜆[(𝑡+𝑡1 )
−4 [201.093 −101.093 ]
𝑅(20|10) = 𝑒 −2.167×10 = 𝟎. 𝟗𝟗𝟕
Assuming repair times are log-normally distributed, we can find the confidence interval as
follows:
𝑛
1 𝑙𝑛50 + 𝑙𝑛40 + 𝑙𝑛30 + 𝑙𝑛40 + 𝑙𝑛40
𝑙𝑛𝜏̅ = ∑ 𝑙𝑛𝜏𝑖 = = 3.68
𝑛 5
𝑖=1
2
∑𝑛𝑖=1(𝑙𝑛𝜏𝑖 − 𝑙𝑛𝜏̅)2
𝑆 = = 0.033
𝑛−1
𝑆 × 𝑡1−𝛼/2 𝑆 × 𝑡1−𝛼/2
𝑃𝑟 (𝑙𝑛𝜏̅ − ≤ 𝑙𝑛𝜏 ≤ 𝑙𝑛𝜏̅ + ) = 1 − 𝛼 = 0.80
√𝑛 √𝑛
Repair No. 1 2 3 4 5
Cycle-to-failure
5010 6730 4031 3972 4197
(Inter-arrival of cycles)
Solutions:
Using the Laplace test to determine if a trend exists for a Failure Terminated data series, we get the
following results:
𝑛−1
1 1
𝑛 − 1 𝑖=1 𝑡𝑖 − 2 𝑡𝑜
∑
𝑈=
1
𝑡𝑜 √
12(𝑛 − 1)
As |U| > 0.3, their does appear to be a slight trend in the data provided and therefore, we cannot
assume the process follows a HPP.
As U is positive, it shows that there is a slightly increasing trend in the data which means the inter-
arrival times of failures is decreasing (ie: IFR).
𝑛 5
𝛽̂ = 𝑛−1 = = 𝟏. 𝟕𝟑𝟐
𝑡𝑜 23940 23940 23940 23940
∑ 𝑙𝑛 𝑙𝑛 + 𝑙𝑛 11740 + 𝑙𝑛 + 𝑙𝑛 19743
𝑡𝑖 5010 15771
𝑖=1
The β value of 1.73 also indicates a trend or time dependent variation in the rate of occurrence of
failures. We can therefore assume a NHPP for the failure rate.
𝑛 5
𝑅𝑂𝐶𝑂𝐹 = 𝜆̂ = ̂
= = 𝟏. 𝟑𝟎𝟎𝟖 × 𝟏𝟎−𝟕 𝑐𝑦𝑐𝑙𝑒 −1
𝑡𝑜 𝛽 239401.732
The reliability of the system 1000 cycles after the 5th repair is given by R(24940|23940)
−7 [249401732 −239401.732 ]
𝑅(24940|23940) = 𝑒 −1.3008×10 = 𝟎. 𝟔𝟗𝟑
A B
I E O
C D
Assume that all components are identical and are repaired immediately after each of them experiences
a failure. The rate of occurrence of the failure for each component is = 0.001 (per hour), and mean-
time-to-repair is 15 hours.
In A C E Out
Given the above information, calculate unavailability of the system assuming that all components are
independent.
1 𝑇𝑅 𝑇𝑡
𝑞𝑎𝑣𝑒𝑟𝑎𝑔𝑒|𝐵𝐶𝐷 = 𝜆𝑇𝑜 + 𝑓𝑟 +
2 𝑇 𝑇
1 15 1
𝑞𝑎𝑣𝑒𝑟𝑎𝑔𝑒|𝐵𝐶𝐷 = 0.00001 × 720 + 0.25 × + = 0.010154
2 724.75 724.75
System Unavailability:
𝑞𝑠𝑦𝑠 = Pr(𝐶1 + 𝐶2 + 𝐶3 ) = 2 × 9.999 × 10−5 + 0.0101543 = 𝟐. 𝟎𝟏 × 𝟏𝟎−𝟒
System Failure
G1 G2 G3
A B C E
A D D B
Suppose all components have the same availability value. Further, consider data obtained for one
component in this set as follows:
Failure Times from when Component was new (hrs) 2823 3984 7232 9812 12314
Repair Duration (hrs) 17 11 28 14 32
As |U| < 0.3, their does not appear to be a strong trend in the data provided and therefore, we can
assume the failure process follows a Homogenous Poisson Process (HPP).
𝑛 5
𝛽̂ = 𝑛 = = 𝟏. 𝟑𝟏
𝑡 13500 13500 13500 13500 13500
∑ 𝑙𝑛 𝑡𝑜 𝑙𝑛
2823
+ 𝑙𝑛
3984
+ 𝑙𝑛
7232
+ 𝑙𝑛
9812
+ 𝑙𝑛
12314
𝑖=1 𝑖
The β value of 1.31 suggests that the trend is not very strong as it is close to a value of 1 which
would indicate a constant failure rate.
𝑛 5
𝑅𝑂𝐶𝑂𝐹 = 𝜆̂ = ̂
= = 𝟏. 𝟗𝟔𝟔 × 𝟏𝟎−𝟓 ℎ𝑟 −1
𝑡𝑜 𝛽 135001.309
Repairs
(17 + 28 + 56 + 70 + 102)
− 0.5 × 102
𝑈= 5 = −𝟎. 𝟕𝟕𝟓
1
102√
12(5)
As |U| > 0.3, their does appear to be a trend in the data provided and therefore, we cannot assume
the repair process follows a Homogenous Poisson Process (HPP).
𝑛 5
𝛽̂ = 𝑛 = = 𝟏. 𝟐𝟑
𝑡 102 102 102 102
∑ 𝑙𝑛 𝑡𝑜 𝑙𝑛 17 + 𝑙𝑛 28 + 𝑙𝑛
56
+ 𝑙𝑛 70
𝑖=1 𝑖
The β value of 1.23 suggests that the trend is not very strong as it is close to a value of 1 which
would indicate a constant repair rate.
𝑛 5
𝜇̂ = ̂
= = 𝟏. 𝟔𝟖𝟐 × 𝟏𝟎−𝟐 ℎ𝑟 −1
𝑡𝑜 𝛽 1021.23
13500
𝑀𝑇𝐵𝐹 = = 2700 ℎ𝑟𝑠
5
1 1
𝜆̂ = = = 0.00037 ℎ𝑟 −1
𝑀𝑇𝐵𝐹 2700
102
𝑀𝑇𝑇𝑅 = = 20.4 ℎ𝑟𝑠
5
1 1
𝜇̂ = = = 0.04902 ℎ𝑟 −1
𝑀𝑇𝑇𝑅 20.4
𝑀𝑇𝐵𝐹 2700
𝑎 = lim 𝑎(𝑡) = = = 𝟎. 𝟗𝟗𝟐𝟓
𝑡→∞ 𝑀𝑇𝐵𝐹 + 𝑀𝑇𝑇𝑅 2700 + 20.4
d) Determine system unavailability assuming all components have the same availability as
calculated at part c).
e) Calculate 90% confidence limits for ROCOF using data in part a). (Assume no trend).
𝜒𝛼2 (2𝑟) 2
𝜒1− 𝛼 (2𝑟 + 2)
2 2
𝑃𝑟 ( ≤𝜆≤ ) = 1 − 𝛼 = 0.90
2𝑇 2𝑇
5.8 For the system represented by the Fault Tree below, if all the components are periodically tested,
determine the average unavailability of the system.
System Failure
B C
All components have a test interval of 1 month and average test duration of 1 hr.
-1 -1
Component Failure Rate ( hr ) Repair Rate ( hr ) Frequency of Repair
A 0.001 0.1 0.1
B 0.0001 0.01 0.2
C 0.0001 0.01 0.3
𝐹𝑠𝑦𝑠 = (𝐴 ∙ (𝐵 + 𝐶))
𝐹𝑠𝑦𝑠 = 𝐴𝐵 + 𝐴𝐶
𝑞𝑠𝑦𝑠 = 𝑞𝑎 (𝑞𝑏 + 𝑞𝑐 )
1 𝑇𝑅𝑖 𝑇𝑡
𝑞𝑖 = 𝜆̅𝑇𝑜𝑖 + 𝑓𝑟𝑖 +
2 𝑇 𝑇
Given Information
𝑇 = 1 𝑚𝑡ℎ = 30 × 24 ℎ𝑟𝑠 = 𝟕𝟐𝟎 ℎ𝑟𝑠
𝑇𝑡 = 1 ℎ𝑟
The following table provides a summary of the individual component unavailability values:
5.9 A repairable unit had the following failure characteristics in the past 5 years. (Assume the unit’s life is
currently at the beginning of its sixth year).
Number of Failure***
Year Required Maintenance
1 n
2 0.9n
3 0.88n
4 0.72n
5 0.88n
Using the Laplace test to determine if a trend exists, we get the following results:
𝑛
1 1 𝑛(1 + 1.9 + 2.78 + 3.5 + 4.38)
∑ 𝑡𝑖 − 𝑡𝑜 − 0.5 × 4.38𝑛
𝑛 𝑖=1 2 5
𝑈= = = 𝟎. 𝟗𝟐𝟑𝟏
1 1
𝑡𝑜 √ 4.38𝑛√
12(𝑛) 12(5)
As |𝑈| > 0.3, we can see a positive trend in rate of number of failures. Therefore, we can assume
that the process is NHPP and that the system is deteriorating over time.
𝑛 5
𝛽̂ = 𝑛−1 = = 𝟏. 𝟔𝟕𝟗
𝑡𝑜 4.38 4.38 4.38 4.38
∑ 𝑙𝑛 𝑙𝑛 1 + 𝑙𝑛 1.9 + 𝑙𝑛 2.78 + 𝑙𝑛
𝑡𝑖 3.5
𝑖=1
b) How many failures (in terms of fraction of n) do we expect to see in the sixth year?
5.10 The following data represent failure times and repair experiences from the life of a component.
Assume 30 days/month
Assuming that the current age of the unit is 30 months:
Failures
𝑛
1 1 (4 + 7 + 15.5 + 20 + 26)
∑ 𝑡𝑖 − 𝑡𝑜 − 0.5 × 30
𝑛 𝑖=1 2 5
𝑈= = = −𝟎. 𝟏𝟐𝟗
1 1
𝑡𝑜 √ 30√
12(𝑛) 12(5)
As |U| < 0.3, their does not appear to be a strong trend in the data provided and therefore, we can
assume the failure process follows a Homogenous Poisson Process (HPP).
𝑛 5
𝛽̂ = 𝑛 = = 𝟏. 𝟎𝟔𝟕
𝑡 30 30 30 30 30
∑ 𝑙𝑛 𝑡𝑜 𝑙𝑛 4 + 𝑙𝑛 7 + 𝑙𝑛
15.5
+ 𝑙𝑛 20 + 𝑙𝑛 26
𝑖=1 𝑖
𝑛 5
𝜆̂ = ̂
= = 𝟏. 𝟏𝟔𝟖 × 𝟏𝟎−𝟒 ℎ𝑟 −1
𝑡𝑜 𝛽 (30 × 30 × 24)1.0686
Repairs:
(3 + 7 + 14 + 17 + 24)
− 0.5 × 24
𝑈= 5 = 𝟎. 𝟑𝟐𝟑
1
24√
12(5)
𝑛 5
𝛽̂ = 𝑛−1 = = 𝟏. 𝟏𝟗𝟏𝟖
𝑡 24 24 24 24
∑ 𝑙𝑛 𝑡𝑜 𝑙𝑛 3 + 𝑙𝑛 7 + 𝑙𝑛 14 + 𝑙𝑛 17
𝑖=1 𝑖
𝑛 5
𝜇̂ = ̂
= = 𝟎. 𝟏𝟏𝟑𝟑 ℎ𝑟 −1
𝑡𝑜 𝛽 241.19178
b) Estimate the instantaneous availability at the present time of 30 months.
0.1133 0.0001168
𝑎(21600) = + exp[−(0.0001168 + 0.1133) × 21600]
0.0001168 + 0.1133 0.0001168 + 0.1133
𝒂(𝟐𝟏𝟔𝟎𝟎) = 𝟎. 𝟗𝟗𝟗
1 5
𝜆̂ = = = 0.000231 ℎ𝑟 −1
𝑀𝑇𝑇𝐹 21600
1 5
𝜇̂ = = = 0.2083 ℎ𝑟 −1
𝑀𝑇𝑇𝑅 24
0.2083 0.000231
𝑎(21600) = + exp[−(0.000231 + 0.2083) × 21600]
0.000231 + 0.2083 0.000231 + 0.2083
𝒂(𝟐𝟏𝟔𝟎𝟎) = 𝟎. 𝟗𝟗𝟖𝟗
c) In a periodically tested system with a constant ROCOF of 10−4 ℎ𝑟 −1 , average constant repair
duration of 10 hrs, average test duration of 5 hrs and average constant frequency of repair of 0.05
per test, determine the best (optimum) inspection interval.
1 1 𝜆 𝜆
𝜇= & 𝜆= 𝑞(𝑡) = − exp[−(𝜆 + 𝜇)𝑡]
10 104 𝜆+𝜇 𝜆+𝜇
1 10 5
0.000999 = 10−4 × 30 × 30 × 24 + 0.05 +
2 𝑇 𝑇
Current
Time (months)
Time
a) Is there practically a trend in ROCOF?
𝑛
1 1
𝑛 ∑ 𝑡𝑖 − 2 𝑡𝑜
𝑖=1
𝑈=
1
𝑡𝑜 √
12(𝑛)
As |U| < 0.3, their does not appear to be a strong trend in the data provided and therefore, we can
assume the failure process follows a Homogenous Poisson Process (HPP).
𝑛 5
𝛽̂ = 𝑛 = = 𝟏. 𝟎𝟖𝟕
𝑡 54 54 54 54 54 54 54
∑ 𝑙𝑛 𝑡𝑜 𝑙𝑛
5.2
+ 𝑙𝑛 10.8 + 𝑙𝑛 18.3 + 𝑙𝑛 28.8 + 𝑙𝑛 33.3 + 𝑙𝑛 44.1 + 𝑙𝑛 49.2
𝑖=1 𝑖
The β value of 1.09 also indicates a constant rate of occurrence which would indicate an
Homogenous Poisson Process (HPP) with no trend.
𝑛 5
𝑅𝑂𝐶𝑂𝐹 = 𝜆̂ = ̂
= = 𝟗. 𝟏𝟔 × 𝟏𝟎−𝟐 𝑚𝑡ℎ−1
𝑡𝑜 𝛽 (54 × 24 × 30)1.087
b) What is the conditional probability of no failure (ie: reliability) in the next 6 months given the
component is operating at the current accumulated component age of 54 months?
c) If the observed average downtime for repair is 14 hrs per repair and repair has no trend, what is the
average conditional unavailability given we are at the age of 54 months and the component is
operating in the next 6 months?
𝜆 𝜆
𝑞(𝑡) = 𝜆+𝜇 − 𝜆+𝜇 exp[−(𝜆 + 𝜇)𝑡]
d) By assuming no trend in ROCOF, how much error would we make in the unconditional average
unavailability?
e) If we decide to adopt a periodically tested policy that reduces the failure rate by a factor of 5 (eg:
using a preventative maintenance program) with an average test duration of 5 min, average repair
duration of 2 hrs and average frequency of repair of 0.01, what should we select as our inspection
interval 𝑇0 in order to maintain the same unavailability value calculated in part c).
1 𝑇𝑅 𝑇𝑡
𝑞̅ = 𝜆̅𝑇𝑜 + 𝑓𝑟 +
2 𝑇 𝑇
5
1 2 60
𝑞̅ = 𝜆̅𝑇𝑜 + 0.01 +
2 𝑇 𝑇
S Available
A B A C
As the two values for A are different, we will assume the components labeled ‘A’ in the diagram are
actually different (ie: 𝐴1 = 0.9 𝑎𝑛𝑑 𝐴2 = 0.98 )
Therefore:
𝑆 = (𝐴1̅ + 𝐵̅)(𝐴̅2 ∙ 𝐶̅ ) = 𝐴1̅ 𝐴̅2 𝐶̅ + 𝐵̅𝐴̅2 𝐶̅
𝐹 = 𝑆̅ = ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
𝐴1̅ 𝐴̅2 𝐶̅ + 𝐵̅𝐴̅2 𝐶̅ = (𝐴1 + 𝐴2 + 𝐶)(𝐵 + 𝐴2 + 𝐶)
𝐹 = (𝐴1 𝐵 + 𝐴1 𝐴2 + 𝐴1 𝐶 + 𝐴2 𝐵 + 𝐴2 𝐴2 + 𝐴2 𝐶 + 𝐶𝐵 + 𝐶𝐴2 + 𝐶𝐶)
𝐹 = 𝐴1 𝐵 + 𝐴2 + 𝐶
In order to find the mutually exclusive cut sets, it is best to use the Combinatorial Truth Table as
follows:
Combin a1 a2 b c Sys Pr(i)
1 0 0 0 0 0
2 0 0 0 1 1 0.008379
3 0 0 1 0 0
4 0 1 0 0 1 0.016929
5 1 0 0 0 0
6 0 0 1 1 1 0.000441
7 0 1 0 1 1 0.000171
8 1 0 0 1 1 0.000931
9 0 1 1 0 1 0.000891
10 1 0 1 0 0
11 1 1 0 0 1 0.001881
12 0 1 1 1 1 0.000009
13 1 0 1 1 1 0.000049
14 1 1 0 1 1 0.000019
15 1 1 1 0 1 0.000099
16 1 1 1 1 1 0.000001
0.0298
̅𝒔𝒚𝒔 = 𝟎. 𝟎𝟐𝟗𝟖
𝒒
If we simply use rare event approximation, we get 𝑞̅ ≤ 0.1 × 0.05 + 0.02 + 0.01 ≤ 𝟎. 𝟎𝟑𝟓
Assuming that the present age of the component is 250 months and a monthly test with duration of 2
hrs has been performed since the unit has been put into operation, determine:
𝑛 5
𝛽̂ = 𝑛 = = 𝟏. 𝟕𝟕𝟓
𝑡 250 250 250 250 250
∑ 𝑙𝑛 𝑡𝑜 𝑙𝑛
73
+ 𝑙𝑛
132
+ 𝑙𝑛
148
+ 𝑙𝑛
193
+ 𝑙𝑛
212
𝑖=1 𝑖
𝑛 5
𝜆̂ = ̂
= = 𝟐. 𝟑𝟔𝟏 × 𝟏𝟎−𝟗 𝒉𝒓−𝟏
𝑡𝑜 𝛽 (250 × 30 × 24)1.775
𝑇 250×30×24
1 1
𝜆̅(𝑡) = ∫ 𝜆(𝑡)𝑑𝑡 = ∫ 𝜆̂𝛽̂𝑡𝛽−1 𝑑𝑡 = 𝜆̂𝑇𝛽−1 = 𝟐. 𝟕𝟕𝟖 × 𝟏𝟎−𝟓 𝒉𝒓−𝟏 = 𝟎. 𝟎𝟐 𝒎𝒕𝒉−𝟏
𝑇 𝑇
0 0
𝑛 5
𝛽̂ = 𝑛−1 = = 𝟏. 𝟒𝟒𝟐
𝑡 1 1 1 1
∑ 𝑙𝑛 𝑜 𝑙𝑛 0.1 + 𝑙𝑛 0.6 + 𝑙𝑛 + 𝑙𝑛 0.8
𝑡𝑖 0.65
𝑖=1
𝑇 1×30×24
1 1
𝜇̅ (𝑡) = ∫ 𝜇(𝑡)𝑑𝑡 = ∫ 𝜇̂ 𝛽̂ 𝑡𝛽−1 𝑑𝑡 = 𝜇̂ 𝑇𝛽−1 = 𝟔. 𝟗𝟒𝟒 × 𝟏𝟎−𝟒 𝒉𝒓−𝟏 = 𝟓 𝒎𝒕𝒉−𝟏
𝑇 𝑇
0 0
1 𝑇𝑅 𝑇𝑡
𝑞̅ = 𝜆̅𝑇𝑜 + 𝑓𝑟 +
2 𝑇 𝑇
2
1 0.02 1 0.2 30 × 24
𝑞̅ = × × 30 + × + = 𝟎. 𝟎𝟏𝟎𝟏𝟕
2 30 5 250 250
5.14 The following historical data for a periodically tested component are available. Duration of repair for
the times the component required repair (in hours) are 7, 8, 10, 9 and 8. Total number of tests is 50.
Overall length of each test is 3.5 hrs. Length of operation 𝑇0 = 1000 ℎ𝑟𝑠
a) Is there any trend in the length of repair? What is the confidence level you can place on the trend
conclusion?
𝑛
1 1 (7 + 15 + 25 + 34 + 42)
𝑛 ∑ 𝑡𝑖 − 2 𝑡𝑜 − 0.5 × 42
𝑈= 𝑖=1
= 5 = 𝟎. 𝟔𝟔𝟒
1 1
𝑡𝑜 √ 42√
12(𝑛) 12(5)
𝑛 5
𝛽̂ = 𝑛−1 = = 𝟏. 𝟒𝟏
𝑡 42 42 42 42
∑ 𝑙𝑛 𝑡𝑜 𝑙𝑛 7 + 𝑙𝑛 + 𝑙𝑛 + 𝑙𝑛 34
𝑖
15 25
𝑖=1
𝑛 5
𝜇̂ = ̂
= = 𝟐. 𝟓𝟗𝟐 × 𝟏𝟎−𝟐 𝒉𝒓−𝟏
𝑡𝑜 𝛽 (42)1.41
Confidence Level
Pr(𝑍 < 0.664) = 0.75
Therefore, the probability of the trend value being lower than 0.664, is found to be approximately
75%. This means there is a lot of uncertainty surrounding the Laplace Test findings and more
evidence would be required before a conclusive assessment can be made regarding the trend.
b) Calculate average unavailability of the component assuming a constant ROCOF and constant
duration of repair.
42
𝜇̅ = = 8.4 ℎ𝑟𝑠
5
5
𝜆̅ = = 0.005 ℎ𝑟 −1
1000
1 𝑇𝑅 𝑇𝑡
𝑞̅ = 𝜆̅𝑇𝑜 + 𝑓𝑟 + 𝑤ℎ𝑒𝑟𝑒 𝑇 = 𝑇𝑜 + 𝑓𝑟 𝑇𝑅 + 𝑇𝑡 = 1000 + 0.84 + 3.5 = 1004.34 ℎ𝑟
2 𝑇 𝑇
1 5 × 1000 8.4 3.5
𝑞̅ = × + 0.1 × + = 𝟎. 𝟎𝟓𝟒𝟑
2 1000 × 50 1004.34 1004.34
5.15 Consider the following observed times of failure in the table below (current time coincides with the
occurrence of last failure)
Failure No. 1 2 3 4 5 6
Inter-arrival of failures (hours) 881 780 870 712 1005 790
𝑛 6
𝛽̂ = 𝑡𝑛 = 𝑡 𝑡 𝑡 𝑡 𝑡 = 1.4448
∑𝑛−1
𝑖=1 ln ( 𝑡 ) ln ( 6 ) + ln ( 6 ) + ln ( 6 ) + ln ( 6 ) + ln ( 6 )
𝑖 𝑡1 𝑡2 𝑡3 𝑡4 𝑡5
𝑛 6
𝜆̂ = ̂ = 1.4448
= 2.6857 × 10−5
𝑡
𝛽 5038
𝑛
b) Write the expression for the MTBF in the future of a repairable component
a) Using the wave height data, what is the point estimate of the parameter, λ (measured in ft-1), of the
exponential distribution?
b) Assume that the occurrence rate of tsunamis follows a homogeneous Poisson process. Using the
inter-arrival of major tsunamis, determine the probability that at least one major tsunami will occur
during the 60-year lifetime of this nuclear plant.
Calculate the point estimate of the period for the inter-arrival times:
5
𝜆𝑡 = = 0.007645 𝑦𝑟 −1
290 + 189 + 97 + 60 + 18
Calculate the point estimate of the period for the inter-arrival times:
Using Eq 5.14 from the textbook:
𝑏 𝑛 𝑏
[∫𝑎 𝜆(𝑡)𝑑𝑡] exp [− ∫𝑎 𝜆(𝑡)𝑑𝑡]
Pr[𝑁(𝑏) − 𝑁(𝑎) = 𝑛] =
𝑛!
[𝜆𝑡 (𝑡2 − 𝑡1 )]𝑛 exp[−𝜆𝑡 (𝑡2 − 𝑡1 )]
Pr[𝑁(𝑡2 ) − 𝑁(𝑡1 ) = 𝑡] =
𝑛!
The probability of at least 1 tidal wave means:
Pr(𝑛 ≥ 1) = 1 − Pr(𝑛 = 0)
0
(𝜆𝑡 (60)) exp(−𝜆𝑡 (60))
Pr(𝑛 = 0) = = exp(−0.007645 × 60) = 0.632095
0!
Pr(𝑛 ≥ 1) = 1 − 0.632095 = 0.367905
c) Calculate the probability that during the 60-year lifetime of the plant, a tsunami will occur and its
height exceeds the designed height of the levee.
𝑆 − 𝐶𝐿
𝑆𝑀 =
√𝑠 2 + 𝐶𝑙 2
𝑆 − 𝐶𝐿
𝑅 = 𝛷[ ] 𝑝𝑒𝑟 𝑐𝑦𝑐𝑙𝑒
√𝑠 2 + 𝐶𝑙 2
c) Express the total number of reversals. N(t) seen by each piston as a function of time.
1
−
𝑆𝑁 𝑞 − 𝐶𝐿
𝑅 = 𝛷[ ]
√𝑠 2 + 𝐶𝑙 2
d) If the shaft is subject to fatigue, express the reliability as a function of time. Metals fatigue,
generally, follow Manson-Coffin Lae: S(N) = SN(−1/q). Assume also that the standard deviation, s,
does not change with N. Also, q is a constant.
1
−
𝑆𝑁 𝑞 − 𝐶𝐿
𝑅 = 𝛷[ ] 𝑋(𝑡)
√𝑠 2 + 𝐶𝑙 2
e) Determine the expected life (50% reliability) of the crank shaft turning at a constant rate, R (RPM).
1
−
𝑆𝑁 𝑞 − 𝐶𝐿
𝑅 = 𝛷[ ] = 0.5
√𝑠 2 + 𝐶𝑙 2
1
−
𝑆𝑁 𝑞 −𝐶𝐿 𝑆 𝑞 𝑆 𝑞
=0 => 𝑁 = (𝐶𝐿) => 𝑡0.5 = 𝑁𝑋(𝑡) = |𝑋(𝑡) (𝐶𝐿) |3.7
√𝑠2 +𝐶𝑙2
No Water Spray
ET1 Power
1 2
1
2
ET1 X V5 Y
𝑅𝑌 = 𝑅𝑉3 𝑅𝑉6 𝑅𝑁1 + 𝑅𝑉4 𝑅𝑉7 𝑅𝑁2 − 𝑅𝑉3 𝑅𝑉6 𝑅𝑁1 𝑅𝑉4 𝑅𝑉7 𝑅𝑁2
𝑅𝑠𝑦𝑠 = 𝑅𝐸𝑇1 𝑅𝑉5 [𝑅𝑉1 𝑅𝑃1 + 𝑅𝑉2 𝑅𝑃2 − 𝑅𝑉1 𝑅𝑃1 𝑅𝑉2 𝑅𝑃2 ][𝑅𝑉3 𝑅𝑉6 𝑅𝑁1 + 𝑅𝑉4 𝑅𝑉7 𝑅𝑁2 − 𝑅𝑉3 𝑅𝑉6 𝑅𝑁1 𝑅𝑉4 𝑅𝑉7 𝑅𝑁2 ]
Using the Birnbaum Importance method, we find the importance of each component as follows:
𝐼𝑖𝐵 = 𝑅𝑠 [𝑅(𝑡)|𝑅𝑖 (𝑡) = 1] − 𝑅𝑠 [𝑅(𝑡)|𝑅𝑖 (𝑡) = 0]
𝐵
𝐼𝐸𝑇1 = 𝑅𝑉5 [𝑅𝑉1 𝑅𝑃1 + 𝑅𝑉2 𝑅𝑃2 − 𝑅𝑉1 𝑅𝑃1 𝑅𝑉2 𝑅𝑃2 ][𝑅𝑉3 𝑅𝑉6 𝑅𝑁1 + 𝑅𝑉4 𝑅𝑉7 𝑅𝑁2 − 𝑅𝑉3 𝑅𝑉6 𝑅𝑁1 𝑅𝑉4 𝑅𝑉7 𝑅𝑁2 ]
𝐹𝑖 [𝑄(𝑡)]
𝐼𝑖𝐹𝑉 (𝑡) =
𝐹𝑠 [𝑄(𝑡)]
Failure Number 1 2 3 4 5 6 7 8 9 10
Cumulative Time on Test (hrs) 12 75 102 141 315 330 342 589 890 1007
Using equation 6.53 and 6.54 we can create a Duane Plot as follows:
𝛽 = 1 − 0.46 = 𝟎. 𝟓𝟒
Given β < 1, we have a reliability growth situation whereby the system is improving with time.
Vendor 1 Vendor 2
Number in Failure Rate Mct Failure Rate Mct
Module System (per 104 hrs) (min) (per 104 hrs) (min)
A 2 45 15 1350 45 20 1800
B 1 90 20 1800 30 15 450
C 2 30 10 600 90 10 1800
240 3750 300 4050
3750
𝑀𝑇𝑇𝑅1 = = 15.63 𝑚𝑖𝑛
240
4050
𝑀𝑇𝑇𝑅2 = = 13.5 𝑚𝑖𝑛
300
A constant failure rate implies an exponential distribution of failure times. Given that an exponential
distribution is memory-less, it is not worth doing any preventative maintenance as the probability of
future failures with or without preventative maintenance is the same.
How much time would you schedule for the balance of the test program, in order to have some
confidence that your controller had met its goal? (Note: Each failure represents a different failure mode
and corrective actions are being taken for each.)
Using MLE methods under the assumption of either a Weibull or Exponential underlying distribution,
we get the following results for the model parameters:
The Weibull assumption provides a higher MLE value (ie: -72.99 > -74.21)
An alternative approach would be to perform a Laplace test to see if a trend exists in the data as follows:
𝑛
1 1
∑ 𝑡𝑖 − 𝑡𝑜
𝑛 𝑖=1 2
𝑈=
1
𝑡𝑜 √
12(𝑛)
𝑛 8
𝛽̂ = 𝑛−1 = = 𝟎. 𝟗𝟓𝟑
𝑡 9000 9000 9000 9000
∑ 𝑙𝑛 𝑡𝑜 𝑙𝑛 1296 + 𝑙𝑛 + 𝑙𝑛 + ⋯ + 𝑙𝑛 8902
𝑖
1582 1855
𝑖=1
𝑛 8
𝜆̂ = ̂
= = 𝟏. 𝟑𝟕 × 𝟏𝟎−𝟑 𝒉𝒓−𝟏
𝑡𝑜 𝛽 (9000)0.9525
Duane Method
Failure Test Hours to Failure t/N log(t) log(t/N)
1 1296 1296.0 7.167 7.167
2 1582 791.0 7.366 6.673
3 1855 618.3 7.526 6.427
4 2310 577.5 7.745 6.359
5 3517 703.4 8.165 6.556
6 5188 864.7 8.554 6.762
7 6792 970.3 8.824 6.878
8 8902 1112.8 9.094 7.015
𝛽̂ = 1 − 0.4133 = 0.5867
𝜆̂ = 10−1.4018 = 0.039646
With the current rate of failures given by the data set, we would require another 26 yrs of failure free
operation to achieve a MTBF goal of 7000 hrs.
Obviously this result is unrealistic and we can conclude that a 7000 hrs MTBF goal is unachievable with
the current system design and time available.
but probability of c and d can be obtained from a Weibull model such as:
𝑡 𝛽
−( )
𝑃𝑟(𝑎) = 1 − 𝑒 𝛼
If 𝜆𝑎 = 𝜆𝑎 = 10−4 ℎ𝑟𝑠 −1 and 𝛼𝑐 = 𝛼𝑑 = 2250 ℎ𝑟𝑠 and 𝛽𝑐 = 𝛽𝑑 = 1.4, Determine the importance of
parameters α, β, and λ in modeling this scenario. Use an importance measure of your choice, and justify
your choice.
Solution:
The complement of the cut set is given by:
𝑆̅ = ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
𝑎𝑏 + 𝑎𝑐 + 𝑑 = (𝑎̅ + 𝑏̅)(𝑎̅ + 𝑐̅)𝑑̅
𝑆̅ = 𝑎̅𝑑̅ + 𝑏̅ 𝑐̅𝑑̅
1000 1.4
𝑅𝑐 = 𝑅𝑑 = exp [− ( ) ] = 0.7252
2250
Using the Birnbaum Importance method, we find the importance of each component as follows:
𝐼𝑖𝐵 = 𝑅𝑠 [𝑅(𝑡)|𝑅𝑖 (𝑡) = 1] − 𝑅𝑠 [𝑅(𝑡)|𝑅𝑖 (𝑡) = 0]
From these calculations, we can see that component D, is almost four times as important as the next
component which is component A (0.9673 versus 0.2493). This makes sense as component D is a
minimal cut set on its own and component A appears in each of the other two minimal cut sets.
𝜕𝑅 𝜕 𝑡 𝛽
−( ) 𝛽𝑡 𝑡 𝛽−1 −( 𝑡 )𝛽
= {𝑒 𝛼 } = 2( ) 𝑒 𝛼
𝜕𝛼 𝜕𝛼 𝛼 𝛼
𝜕𝑅 𝜕 𝑡 𝛽
−( ) 𝑡 𝑡 𝛽 𝑡 𝛽
−( )
= { 𝑒 𝛼 } = 𝑙𝑛 ( ) × ( ) × 𝑒 𝛼
𝜕𝛽 𝜕𝛽 𝛼 𝛼
𝜕𝑅 𝜕 −4
= { 𝑒 −𝜆𝑡 } = −𝑡𝑒 −𝜆𝑡 = −1000𝑒 −1000×1×10 = −𝟗𝟎𝟓
𝜕𝜆 𝜕𝜆
𝜕𝑅 𝜕𝑅 𝜕𝑅
Given that 𝜕𝛼
> 𝜕𝛽
> 𝜕𝜆
, we can conclude that α is more important than β and λ in this case in terms
of its potential to influence the reliability of the system.
Therefore, the overall order of importance in the parameters for the system detailed above is given as
α, β, and then λ.
S = Success
F = Failure
Frequency of initiating event I = 1.5 year−1; components b and d have dependent (common-cause)
failure with a β factor of 0.08.
a) Determine the minimal cut sets in terms of the basic events (I, a, b, c, etc.) for the event tree
scenarios ending in failure (i.e., F).
Solution:
Let z = the probability of a common cause failure between components b and d
Therefore
𝐴 = (𝑎 + 𝑏 + 𝑧)(𝑐 + 𝑑 + 𝑧) = 𝑎𝑐 + 𝑎𝑑 + 𝑎𝑧 + 𝑏𝑐 + 𝑏𝑑 + 𝑏𝑧 + 𝑧𝑐 + 𝑧𝑑 + 𝑧𝑧
𝐴 = 𝑎𝑐 + 𝑎𝑑 + 𝑏𝑐 + 𝑏𝑑 + 𝑧
𝐴̅ = 𝑎̅𝑏̅𝑧̅ + 𝑐̅𝑑̅𝑧̅
Also
𝐵 = 𝑎𝑒 + 𝑐𝑓
Using the following relationships, the availability and unavailability of each component is
determined as follows:
1 1
𝑎𝑣𝑎𝑖𝑙𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = 𝑤ℎ𝑒𝑟𝑒 𝜆 = & 𝜏 = 𝑀𝑇𝑇𝑅
1 + 𝜆𝜏 𝑀𝑇𝐵𝐹
For components b & d; we need to determine the independent and common cause unavailability as
follows:
1 1
𝑎𝑣𝑎𝑖𝑙𝑎𝑏𝑖𝑙𝑖𝑡𝑦𝑏|𝑑 𝑇𝑜𝑡𝑎𝑙 = ( )=( ) = 𝟎. 𝟗𝟗𝟗𝟎𝟕
1 + 𝜆𝜏 1 + (1⁄18,200) × 17
Using this approach, the availability and unavailability of each component is summarized as
follows:
Therefore, the frequency of failure is calculated using the cut sets as follows:
Cutsets Pr(Ci)
1 4.21E-08
2 2.73E-08
3 2.70E-10
4 3.93E-10
5 3.19E-10
6 4.92E-10
7 2.77E-11
8 4.28E-11
Qsys 7.09E-08
1
𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦 𝑜𝑓 𝐹𝑎𝑖𝑙𝑢𝑟𝑒 = 1.5 × × 7.09 × 10−8 = 𝟏. 𝟐𝟏 × 𝟏𝟎−𝟏𝟏 𝒉𝒓−𝟏
8760
c) Calculate the Fussell–Vesely importance measure of all components based on part B above.
𝐹𝑖 [𝑄(𝑡)]
𝐼𝑖𝐹𝑉 (𝑡) =
𝐹𝑠 [𝑄(𝑡)]
Using the unavailability of each cut set, we can find the importance of each component as shown
in the last column of the table below.
The effect on system failure by each component failure is found by including all cut sets which
have the component of interest contained within them which is detailed as follows and summarized
in the adjacent table:
Failed
𝐹𝑎 = 𝐼[ 𝑎̅𝑏̅𝑧̅𝑐𝑓 + 𝑎𝑒𝑐̅𝑑̅𝑧̅ + 𝑎𝑐𝑒 + 𝑎𝑐𝑓 + 𝑎𝑑𝑒 + 𝑎𝑒𝑧] Fi I_fv
Component (i)
𝐹𝑏 = 𝐼[𝑎̅𝑏̅𝑧̅𝑐𝑓 + 𝑎𝑒𝑐̅𝑑̅𝑧̅ + 𝑏𝑐𝑓 + 𝑎𝑒𝑧 + 𝑐𝑓𝑧] a 7.04E-08 0.992
𝐹𝑐 = 𝐼[𝑎̅𝑏̅𝑧̅𝑐𝑓 + 𝑎𝑒𝑐̅𝑑̅𝑧̅ + 𝑎𝑐𝑒 + 𝑎𝑐𝑓 + 𝑏𝑐𝑓 + 𝑐𝑓𝑧] b 6.99E-08 0.986
𝐹𝑑 = 𝐼[𝑎̅𝑏̅𝑧̅𝑐𝑓 + 𝑎𝑒𝑐̅𝑑̅𝑧̅ + 𝑎𝑑𝑒 + 𝑎𝑒𝑧 + 𝑐𝑓𝑧] c 7.06E-08 0.995
𝐹𝑒 = 𝐼[𝑎𝑒𝑐̅𝑑̅𝑧̅ + 𝑎𝑐𝑒 + 𝑎𝑑𝑒 + 𝑎𝑒𝑧] d 6.98E-08 0.984
𝐹𝑓 = 𝐼[𝑎̅𝑏̅𝑧̅𝑐𝑓 + 𝑎𝑐𝑓 + 𝑏𝑐𝑓 + 𝑐𝑓𝑧] e 2.79E-08 0.393
𝐹𝑧 = 𝐼[𝑎𝑒𝑧 + 𝑐𝑓𝑧] f 4.30E-08 0.607
From this table, we can see that component C is the most important in terms of affecting system
reliability. This is intuitive when one considers that component C is present in 6 of the 8 cut sets.
Component A, B & D are the next most important components in the system due there influence
on system unavailability.
μ 1.412 0.945
σ 0.414 0.122
a) What is the Safety Margin and thus the reliability for this new wing?
μ𝑆 − μ𝑙 1.412 − 0.945
𝑆𝑀 = = = 1.0820
√σ𝑆2 + σ2𝑙 √(0.414)2 + (0.122)2
Reliability:
𝑅 = Φ(𝑆𝑀) = Φ(1.082)
From the standard normal table (or using excel function NORM.S.DIST)
𝑅 = 0.8604
b) Assume that the company head wants to up the reliability factor to 0.99. What should be the new
target mean for the strength distribution assuming that standard deviation won’t change?
Setting R=0.99 and using the standard normal table (or excel function NORM.S.INV)
𝑆𝑀 = Φ−1 (𝑅 = 0.99) = 2.3263
μ𝑆 − 0.945
∴ 2.3263 =
√0.4142 + 0.1222
𝜇𝑆 = 2.3263 × √0.4142 + 0.1222 + 0.945 = 1.9491 𝑀𝑃𝑎
Component A B C D E H
Pr(Failure) 0.008 0.008 0.008 0.05 0.007 0.002
𝑇 = 𝐺1 + 𝐺2 + 𝐸 + 𝐻
𝐺1 = 𝐷 ∙ (𝐴𝐵 + 𝐴𝐶 + 𝐵𝐶)
𝐺2 = 𝐴𝐷
𝑇 = 𝐷𝐴𝐵 + 𝐷𝐴𝐶 + 𝐷𝐵𝐶 + 𝐴𝐷 + 𝐸 + 𝐻
Firstly, finding the probability that the system will fail using part a) and rare event approximation:
𝐹𝑇 ≤ 𝐴𝐷 + 𝐷𝐵𝐶 + 𝐸 + 𝐻
𝐹𝑇 ≤ 0.008 × 0.05 + 0.05 × 0.0082 + 0.007 + 0.002 = 0.007603
For component D:
Set Pr(𝐷) = 1
∴ 𝐹𝑇 (𝑇|𝐷 = 1) = 0.008 + 0.0082 + 0.007 + 0.002 = 0.01526
𝐹𝑇 (𝑇|𝐷 = 1) 0.01526
𝐼𝐷𝑅𝐴𝑊 = = = 2.0076
𝐹𝑇 0.007603
For component E:
Set Pr(𝐸) = 1
∴ 𝐹𝑇 (𝑇|𝐸 = 1) = 0.008 × 0.05 + 0.05 × 0.0082 + 1 + 0.002 = 1.0006
𝐹𝑇 (𝑇|𝐸 = 1) 1.000603
𝐼𝐸𝑅𝐴𝑊 = = = 131.60
𝐹𝑇 0.007603
Comparing these, it can be seen that E and H are by far the most important, this is logical based on
the fault tree since E and H are single points of failure at the highest level.
Using one of the statistical uncertainty techniques, obtain the mean and standard deviation of the
equivalent resistor. In what ways is the uncertainty associated with the equivalent resistance different
from the individual resistor? Discuss the results.
1 1 1 𝑅1 𝑅2
= + => 𝑅𝑡𝑜𝑡 =
𝑅𝑡𝑜𝑡 𝑅1 𝑅2 𝑅1 + 𝑅2
25 × 50 1 −2 × 502 −2 × 252
𝑅̅𝑡𝑜𝑡 ≈ + [ × 6.25 + × 25]
25 + 50 2 (25 + 50)3 (25 + 50)3
2 2
502 252
𝑆 2 (𝑅𝑡𝑜𝑡 ) = (2.5)2 ×[ ] + (5)2
× [ ] = 𝟏. 𝟓𝟒𝟑
(25 + 50)2 (25 + 50)2
𝑆(𝑅𝑡𝑜𝑡 ) = √1.543 = 𝟏. 𝟐𝟒 𝜴
The variance of the parallel system is lower than the variance of any individual component. This shows
the benefit of parallel systems in reducing the uncertainty.
7.2 The results of a bootstrap evaluation give 𝜇 = 1 × 10−4 and 𝜎 = 1 × 10−3. Evaluate the number
of pseudo failures (F), in N trials for an equivalent binomial distribution. Estimate the 95% confidence
limits of μ.
This problem is similar to example 7.10 in the text. Following the same process we get:
𝐹 = 𝑁𝜇 = 100 × 1 × 10−4 ≈ 𝟎. 𝟎𝟏
The 95% confidence interval for the estimated mean can be found using the Clopper - Pearson
procedure as follows (Equation 3.116 & 3.117 respectively):
The basic system diagram shown in exercise 4.6 can be represented by the following Reliability Block
Diagram. The modified Fault Tree which considers CCF in the system is also shown below:
No Water Spray
ET1 Power
1 2
Similarly
𝑇2 = 𝑁2 + 𝑉4 + 𝑉7 + 𝑃1𝑃2 + 𝑃2𝑉1 + 𝑃2𝑉5 + 𝑉2𝑃1 + 𝑉1𝑉2 + 𝑉5𝑉2 + 𝐶𝐶𝑉 + 𝐶𝐶𝑃
𝐹𝑠𝑦𝑠 = 𝐸𝑇1 + 𝑃𝑤𝑟 + 𝑁1𝑁2 + 𝑁1𝑉4 + 𝑁1𝑉7 + 𝑁1𝑃2𝑉5 + 𝑁1𝑉2𝑉5 + 𝑉3𝑁2 + 𝑉3𝑉4 + 𝑉3𝑉7
+ 𝑉3𝑃2𝑉5 + 𝑉3𝑉2𝑉5 + 𝑉6𝑁2 + 𝑉6𝑉4 + 𝑉6𝑉7 + 𝑉6𝑃2𝑉5 + 𝑉6𝑉2𝑉5 + 𝑃1𝑉5𝑁2
+ 𝑃1𝑉5𝑉4 + 𝑃1𝑉5𝑉7 + 𝑉1𝑉5𝑁2 + 𝑉1𝑉5𝑉4 + 𝑉1𝑉5𝑉7 + 𝑉1𝑉5𝑃2 + 𝑃1𝑃2 + 𝑃1𝑉2
+ 𝑃2𝑉1 + 𝑉1𝑉2 + 𝐶𝐶𝑉 + 𝐶𝐶𝑃
𝐹𝑠𝑦𝑠 = 𝐸𝑇1 + 𝑃𝑤𝑟 + 𝑁𝑖2 + 2𝑃𝑖 𝑁𝑖 𝑉𝑖 + 4𝑁𝑖 𝑉𝑖 + 2𝑁𝑖 𝑉𝑖2 + 5𝑉𝑖2 + 4𝑉𝑖3 + 5𝑃𝑖 𝑉𝑖2 + 𝑃𝑖2 + 2𝑃𝑖 𝑉𝑖 + 𝐶𝐶𝑉
+ 𝐶𝐶𝑃
Considering rare event approximation, the system failure probability is given by:
𝑄𝑠𝑦𝑠 = 𝑞𝐸𝑇 + 𝑞𝑝𝑤𝑟 + 𝑞𝑁2 + 2(1 − 𝛽)2 𝑞𝑃 𝑞𝑁 𝑞𝑉 + 4(1 − 𝛽)𝑞𝑁 𝑞𝑉 + 2(1 − 𝛽)2 𝑞𝑁 𝑞𝑉2 + 5(1 − 𝛽)2 𝑞𝑉2
+ 4(1 − 𝛽)3 𝑞𝑉3 + 5(1 − 𝛽)3 𝑞𝑃 𝑞𝑉2 + (1 − 𝛽)2 𝑞𝑃2 + 2(1 − 𝛽)2 𝑞𝑃 𝑞𝑉 + 𝛽𝑞𝑃 + 𝛽𝑞𝑉
7.4 A class of components is temperature sensitive in that the components will fail if the temperature is
raised too high. Uncertainty associated with a component’s failure temperature is characterized by a
continuous uniform distribution such as shown below:
If the temperature for a particular component is uncertain but can be characterized by an exponential
distribution with λ = 0.05 per degree Celsius, calculate reliability of this component.
1 1 1
𝑓(𝑇𝑓 ) = = =
𝑏 − 𝑎 150 − 100 50
150 𝑇𝑓
1
𝑅 = 𝑃𝑟(𝑇𝑐 < 𝑇𝑓 ) = ∫ ∙ ∫ 𝜆𝑒 −𝜆𝑇𝑐 𝑑𝑇𝑐 ∙ 𝑑𝑇𝑓
50
100 0
150 150
1 1 𝑒 −𝜆𝑇𝑓
𝑅 = 𝑃𝑟(𝑇𝑐 < 𝑇𝑓 ) = ∫ (1 − 𝑒 −𝜆𝑇𝑓 )𝑑𝑇𝑓 => [𝑇𝑓 + ]
50 50 𝜆 100
100
𝑹 = 𝟎. 𝟗𝟗𝟖
7.5 Consider the cut sets below describing the failure of a simple system: F = AB + BC. The following data
have been found for components A, B, and C:
a) Use the system reduction methods to calculate equivalent number of failures and total test time for
failure of the system.
A C
5 1
𝑌̂𝐴𝐶 = 1 − (1 − ) × (1 − ) = 𝟎. 𝟎𝟎𝟒𝟓
1250 2012
12
𝑌̂𝑠𝑦𝑠 = 𝑌̂𝐴𝐶 × 𝑃̂𝐵 = 0.0045 × = 𝟏. 𝟐𝟓 × 𝟏𝟎−𝟓
4315
𝑚
𝐹𝑖 + 1 5.62 + 1 12 + 1
𝑌̂𝑠𝑦𝑠
′
=∏ =( )( ) = 𝟏. 𝟓𝟗𝟒 × 𝟏𝟎−𝟓
𝑁𝑖 + 1 1250 + 1 4315 + 1
𝑖=1
1 − 𝑌̂𝑠𝑦𝑠
′
1 − 1.5936 × 10−5
𝑁𝑠𝑦𝑠 = = = 𝟐𝟗𝟏, 𝟎𝟕𝟔 𝒉𝒓𝒔
𝑌̂𝑠𝑦𝑠
′
− 𝑌̂𝑠𝑦𝑠 1.5936 × 10−5 − 1.25 × 10−5
b) Given the results of (a), calculate the 90% confidence limits for the unreliability of this system.
The 90% confidence interval for the estimated unreliability of the system can be found using the
Clopper - Pearson procedure as follows (Equation 3.116 & 3.117 respectively):
7.6 The risk R of scenario I may be evaluated by using the expression 𝑅𝑖 = 𝑓𝑖 × 𝐶𝑖 where 𝑓𝑖 is the
frequency and 𝐶𝑖 is the consequence of this scenario. For a given scenario, if the mean frequency is
1 × 10−8 per year with a standard deviation of 1 × 10−8 per year, and mean consequences is 1000
injuries with a standard deviation of 100, determine the mean risk of the scenario and its associated
standard deviation.
Mean Risk
𝑅̅ = 𝑓 ̅ × 𝐶̅ = 1 × 10−8 × 1000 = 𝟏 × 𝟏𝟎−𝟓 𝒊𝒏𝒋𝒖𝒓𝒊𝒆𝒔/𝒚𝒓
𝜕𝑅
= 𝐶 = 𝟏𝟎𝟎𝟎
𝜕𝑓
𝜕𝑅
= 𝑓 = 𝟏 × 𝟏𝟎−𝟖
𝜕𝐶
2
𝑆 2 (𝑅) = (1 × 10−8 )2 × (1000)2 + 1002 × (1 × 10−8 ) = 1.01 × 10−10
B
E
A C
F
D
Z
H J K
If all components are identical and subject to dependent failures (CCF), calculate the failure probability
of the system using the α-Factor Model for parametric CCF probability assessment (use generic data
discussed in Section 7.3). Total failure probability of one unit is 0.001.
Assume that CCF’s are only possible in the parallel redundant components and that each redundant
system will be successful as long as one component operates. The system Fault tree is therefore given
by:
A X Y Z J K
CCBCD CCEF
CCGHI
Bi Ci Di
Ei Fi
Gi Hi Ii
Bi CCCD Ci CCBD Di CCBC
Gi CCHI Hi CCGI Ii CCGH
𝐹𝑆𝑌𝑆 = (𝐴 + 𝑋 + 𝑌)(𝑍 + 𝐽 + 𝐾)
𝐹𝑆𝑌𝑆 = [𝐴 + (𝐵𝑖 𝐶𝑖 𝐷𝑖 + 3𝐵𝑖 𝐶𝐶𝐶𝐷 + 𝐶𝐶𝐵𝐶𝐷 ) + (𝐸𝑖 𝐹𝑖 + 𝐶𝐶𝐸𝐹 )][(𝐺𝑖 𝐻𝑖 𝐼𝑖 + 3𝐺𝑖 𝐶𝐶𝐻𝐼 + 𝐶𝐶𝐺𝐻𝐼 ) + 𝐽
+ 𝐾]
3 2 3
𝑄𝑠𝑦𝑠 = [𝑞𝐴 + 𝑞1−3 + 3𝑞1 𝑞2|3 + 𝑞3|3 + 𝑞1|2 + 𝑞2|2 ][𝑞1−3 + 3𝑞1 𝑞2|3 + 𝑞3|3 + 𝑞𝐽 + 𝑞𝐾 ]
𝑄𝑠𝑦𝑠 = [0.001 + (0.95 × 0.001)3 + 3(0.95 × 0.001)(0.04 × 0.001) + (0.01 × 0.001) + (0.95 × 0.001)2
+ (0.05 × 0.001)][(0.95 × 0.001)3 + 3(0.95 × 0.001)(0.04 × 0.001) + (0.01 × 0.001)
+ 0.001 + 0.001]
𝑡 𝛽
𝑅(𝑡) = 𝑒𝑥𝑝 [− ( ) ]
𝛼
Suppose we are uncertain about the value of β. We have estimated the mean value of β, 𝜇𝛽 = 1.5 with
a coefficient variation of 0.5. Assuming α = 1400 h (with no uncertainty), compute the mean and
coefficient of variation of R(t) for t = 1000 h.
1000 1.5
̅
𝑅 (1000) = 𝑒𝑥𝑝 [− ( ) ] = 𝟎. 𝟓𝟒𝟕
1400
Coefficient of Variation
2
2 (𝑅)
𝜕𝑅 2 𝑡 𝑡 𝛽 𝑡 𝛽
𝑆 = 𝑉𝑎𝑟(𝛽) [ ] = 𝑉𝑎𝑟(𝛽) [− ln ( ) ( ) 𝑒𝑥𝑝 [− ( ) ]]
𝜕𝛽 𝛼 𝛼 𝛼
𝑆(𝑅) 0.0833
𝐶𝑜𝑉𝑅 = = = 𝟎. 𝟏𝟓𝟐𝟑
𝑅̅ 0.547
7.9 Consider the system as represented by the fault tree in Problem 5.8. If each component is replaceable
and can be represented by an exponential time-to-failure model:
Therefore
𝐹̅𝑠𝑦𝑠 = ̅̅̅̅̅̅̅̅̅̅
𝑎𝑏 + 𝑎𝑐 = (𝑎̅ + 𝑏̅)(𝑎̅ + 𝑐̅) = 𝑎̅ + 𝑏̅𝑐̅
𝑅𝑠𝑦𝑠 = 𝑅𝑎 + 𝑅𝑏 𝑅𝑐 − 𝑅𝑎 𝑅𝑏 𝑅𝑐
∞ ∞
1 1 1 1 1 1
𝑀𝑇𝑇𝐹 = + − = −3 + −4
− −3 = 𝟓, 𝟏𝟔𝟕 𝒉𝒓𝒔
𝜆𝑎 𝜆𝑏 + 𝜆𝑐 𝜆𝑎 + 𝜆𝑏 + 𝜆𝑐 10 2 × 10 10 + 2 × 10−4
b) Coefficient of Variation
We are told that each component is exponentially distributed, therefore the mean of each component is
given by 1/λ and the S.D of each component is given by the mean multiplied by the Coefficient of
Variation (CoV). The table below summaries the results calculated for each component given.
The CoV for the MTTF of the system can be found using the following Taylor series approximation to
the MTTF equation:
1 1 1 𝜇𝑏 𝜇𝑐 𝜇𝑎 𝜇 𝑏 𝜇𝑐
𝑀𝑇𝑇𝐹 = 𝑌 = + − = 𝜇𝑎 + −
1⁄ 1⁄ + 1⁄ 1⁄ + 1⁄ + 1⁄ 𝜇𝑐 + 𝜇𝑏 𝜇𝑏 𝜇𝑐 + 𝜇𝑎 𝜇𝑐 + 𝜇𝑎 𝜇𝑏
𝜇𝑎 𝜇𝑏 𝜇𝑐 𝜇𝑎 𝜇𝑏 𝜇𝑐
𝑛
2 (𝑌)
𝜕𝑌 2
𝑆 = ∑ 𝑉𝑎𝑟(𝜇𝑖 ) [ ]
𝜕𝜇𝑖
𝑖=1
2
𝜕𝑌 ( 𝜇𝑏 𝜇𝑐 ) (10000 × 10000)2
=1− 2
= 1 − = 𝟎. 𝟑𝟎𝟔
𝜕 𝜇𝑎 (𝜇𝑏 𝜇𝑐 + 𝜇𝑎 𝜇𝑐 + 𝜇𝑎 𝜇𝑏 ) (10000 × 10000 + 1000 × 10000 + 1000 × 10000)2
2 2
𝜕𝑌 (𝜇𝑐 ) ( 𝜇𝑎 𝜇𝑐 )
= 2 −
= 𝟎. 𝟐𝟒𝟑
𝜕𝜇𝑏 (𝜇 + 𝜇 ) (𝜇 𝜇 + 𝜇 𝜇 + 𝜇 𝜇 )2
𝑏 𝑐 𝑏 𝑐 𝑎 𝑐 𝑎 𝑏
2 2
𝜕𝑌 (𝜇𝑏 ) (𝜇𝑎 𝜇𝑏 )
= 2 − 2 = 𝟎. 𝟐𝟒𝟑
𝜕𝜇𝑐 (𝜇 + 𝜇 ) (𝜇𝑏 𝜇𝑐 + 𝜇𝑎 𝜇𝑐 + 𝜇𝑎 𝜇𝑏 )
𝑏 𝑐
𝑆(𝑌) 349
𝐶𝑜𝑉𝑀𝑇𝑇𝐹 = = = 𝟎. 𝟎𝟔𝟖
𝑀𝑇𝑇𝐹𝑠𝑦𝑠 5167
If 𝑆1 𝑎𝑛𝑑 𝑆2 are standby subsystems with perfect switching, no standby failure and identical
components (i.e., 𝑎1 , 𝑏1 , 𝑎𝑛𝑑 𝑐1 are identical and so are 𝑎2 , 𝑏2 , 𝑎𝑛𝑑 𝑐2 )
a) Determine reliability of the system as a function of time. (Assume constant failure rate: 𝜆𝐴 , 𝜆𝐵 , 𝜆𝐶 ,
𝜆𝐷 and 𝜆𝑎1 = 𝜆𝑏1 = 𝜆𝑐1 = 𝜆1 and 𝜆𝑎2 = 𝜆𝑏2 = 𝜆𝑐2 = 𝜆2 .
Let Components B, C, S1 & S2 be represented by a super block X as shown on the picture above.
b) If the units (i.e., the blocks) are “revealed fault repairable” units with constant failure rate λ and
repair rate μ, determine the availability of the system. (Note that 𝜆𝐴 , 𝜆𝐵 , 𝜆𝐶 , 𝜆𝐷 and 𝜆𝑎1 = 𝜆𝑏1 =
𝜆𝑐1 = 𝜆1 , and 𝜆𝑎2 = 𝜆𝑏2 = 𝜆𝑐2 = 𝜆2 and 𝜇𝐴 , 𝜇𝐵 , 𝜇𝐶 , 𝜇𝐷 and 𝜇𝑎1 = 𝜇𝑏1 = 𝜇𝑐1 = 𝜇1 , and 𝜇𝑎2 =
𝜇𝑏2 = 𝜇𝑐2 = 𝜇2 are constants.)
Using the following expression, the limiting point wise availability for each component can be
found as follows:
𝜇
𝑎(𝑡) =
𝜆+𝜇
𝜇𝐴
𝑎(𝑡)𝐴 = and similarly for components B – D
𝜆𝐴 + 𝜇𝐴
𝜇1
𝑎(𝑡)𝑆1 = and similarly for switch S2
𝜆1 + 𝜇1
Assuming that in the limit as t approaches infinity, the system availability approaches the system
reliability. Therefore, the system availability is given as follows:
𝜇𝐴 𝜇𝐷 𝜇𝐵 𝜇1 3 𝜇𝐶 𝜇2 3
𝐴𝑠𝑦𝑠 (𝑡) = ( )( ) [( )( ) +( )( )
𝜆𝐴 + 𝜇𝐴 𝜆𝐷 + 𝜇𝐷 𝜆𝐵 + 𝜇𝐵 𝜆1 + 𝜇1 𝜆𝐶 + 𝜇𝐶 𝜆2 + 𝜇2
𝜇𝐵 𝜇𝐶 𝜇1 3 𝜇2 3
−( )( )( ) ( ) ]
𝜆𝐵 + 𝜇𝐵 𝜆𝐶 + 𝜇𝐶 𝜆1 + 𝜇1 𝜆2 + 𝜇2
No Flow from D
A D
B S1 C S2
d) Determine minimal cut sets of the system for the fault tree top event.
𝐹𝑠𝑦𝑠 = 𝐴 + 𝐷 + (𝐵 + 𝑆1 )(𝐶 + 𝑆2 )
𝐹𝑠𝑦𝑠 = 𝐴 + 𝐷 + 𝐵𝐶 + 𝐵𝑆2 + 𝐶𝑆1 + 𝑆1 𝑆2
e) If components B and C are identical and subject to CCF, using the cut sets in (d) determine the
probability of failure of the system (use an α-factor model with generic data for dependent failure
analysis).
Let 𝐶𝐶𝐵|𝐶 = the probability of CCF between components B & C. The cut sets then become:
𝐶1 = 𝐴; 𝐶2 = 𝐷; 𝐶3 = 𝐵𝑖 𝐶𝑖 ; 𝐶4 = 𝐵𝑖 𝑆2 ; 𝐶5 = 𝐶𝑖 𝑆1 ; 𝐶6 = 𝑆1 𝑆2 ; 𝐶7 = 𝐶𝐶𝐵|𝐶
Using rare event approximation, the probability of system failure is given by:
Pr(𝐹) = Pr(𝐶1 + 𝐶2 + 𝐶3 + 𝐶4 + 𝐶5 + 𝐶6 + 𝐶7 )
Using the α-factor as discussed on page 349 of the text, we find the probability of system failure
as follows:
In this case, m is the number of components affected by the CCF which equals 2 (B & C).
𝛼𝑡 = 𝛼1 + 2𝛼2
𝑄𝑡 = 𝑄𝐵|𝐶
1 𝛼1 𝛼1
𝑄1 = 𝑄𝑡 = 𝑄
(
2 − 1 𝛼𝑡
) 𝛼1 + 2𝛼2 𝑡
1−1
2 𝛼2 2𝛼2
𝑄2 = 𝑄𝑡 = 𝑄
(
2 − 1 𝛼𝑡
) 𝛼1 + 2𝛼2 𝑡
2−1
From the table on page 349 of the text, we find that the generic alpha values for a system where
two components are affected by CCF as follows:
𝛼1 = 0.95
𝛼2 = 0.05
Therefore:
0.95
𝑄1 = 𝑄 = 0.9048𝑄𝐵|𝐶
0.95 + 2 × 0.05 𝑡
2 × 0.05
𝑄2 = 𝑄 = 0.0952𝑄𝐵|𝐶
0.95 + 2 × 0.05 𝑡
Using the cut sets found previously and rare event approximation, the probability of system failure
is given by:
2
𝑄𝑠𝑦𝑠 = 𝑄𝐴 + 𝑄𝐷 + (0.9048𝑄𝐵|𝐶 ) + (0.9048𝑄𝐵|𝐶 )𝑄𝑆2 + (0.9048𝑄𝐵|𝐶 )𝑄𝑆1 + 𝑄𝑆1 𝑄𝑆2
+ 0.0952𝑄𝐵|𝐶
2
𝑄𝑠𝑦𝑠 = 𝑄𝐴 + 𝑄𝐷 + (0.9048𝑄𝐵|𝐶 ) + (0.9048𝑄𝐵|𝐶 )[𝑄𝑆2 + 𝑄𝑆1 ] + 𝑄𝑆1 𝑄𝑆2 + 0.0952𝑄𝐵|𝐶
a) Is a Weibull distribution a proper fit? Use Kaplan–Meier and Rank-Increment rank statistics to plot
and compare results.
The following table and plot summarize the results of using Kaplan-Meier and Rank Increment
techniques on the data set given assuming a Weibull underlying distribution.
The parameters and MTBF of the Weibull distribution for each of the two techniques used are
summarized in the following table.
b) What is your estimate of the reliability at 2500 h, using each method in (a)?
Using the Weibull parameters found in part (a), the reliability at 2500 hrs was calculated for each
technique and summarized as follows:
c) What is the mean residual life if the unit has worked without failure for 2500 h using each method
in (a)?
∞
1
𝑀𝑅𝐿(2500 ℎ𝑟𝑠) = ∫ 𝑅(𝑡)𝑑𝑡
𝑅(2500)
2500
This integral was solved for each case via numerical integration using Matlab as follows:
Kaplan-Meier
F=@(x) exp(-((x./3534.64797).^0.7555));
MRL=quadgk(F,2500,+inf)/F(2500)
MRL = 5473.5 hrs
Rank Increment
F2=@(x) exp(-((x./3976.02755).^0.8728));
MRL=quadgk(F2,2500,+inf)/F2(2500)
MRL = 4281.8 hrs
MLE
F3=@(x) exp(-((x./3783.3361).^1.063));
MRL=quadgk(F3,2500,+inf)/F3(2500)
MRL = 3487.1 hrs
In order to quantify the uncertainty of each parameter using the MLE method, a Fisher Information
matrix can be developed as follows:
𝑁 𝑆
𝑁 𝑡𝑗 𝑆 𝛽
𝛽𝑡𝛽−1 −(𝑡𝑖)𝛽 −( )
𝑙 = ∏{ 𝛽 𝑒 𝛼 }∙ ∏𝑒 𝛼
𝛼
𝑖=1 𝑗=1
𝑁𝑓 𝑁𝑠 𝛽
𝑡𝑗
𝛽𝑡𝑖 𝛽−1 −(𝑡𝑖)𝛽 −( )
ln(𝑙) = 𝛬 = ∑ 𝑙𝑛 [ 𝛽 𝑒 𝛼 ] + ∑ 𝑙𝑛 [𝑒 𝛼 ]
𝛼
𝑖=1 𝑗=1
𝑁𝑓 𝑁𝑠
𝑡𝑖 𝛽 𝑡𝑗 𝛽
𝛬 = 𝑁𝑓 [𝑙𝑛(𝛽) − 𝛽 ln(𝛼)] + ∑ [(𝛽 − 1)𝑙𝑛(𝑡𝑖 ) − ( ) ] − ∑ ( )
𝛼 𝛼
𝑖=1 𝑗=1
𝑁𝑓 𝑁𝑠
𝜕𝛬 −𝑁𝑓 𝛽 𝛽 𝛽
= + 𝛽+1 [∑(𝑡𝑖 )𝛽 + ∑(𝑡𝑗 ) ]
𝜕𝛼 𝛼 𝛼
𝑖=1 𝑗=1
𝑁𝑓 𝑁𝑠
𝜕 2 𝛬 𝑁𝑓 𝛽 𝛽(𝛽 + 1) 𝛽
2
= 2 − 𝛽+2
[∑(𝑡𝑖 )𝛽 + ∑(𝑡𝑗 ) ]
𝜕𝛼 𝛼 𝛼
𝑖=1 𝑗=1
𝑁𝑓 𝑁𝑠
𝜕𝛬 𝑁𝑓 𝑡𝑖 𝑡𝑖 𝛽 𝑡𝑗 𝑡𝑗 𝛽
= − 𝑁𝑓 ln(𝛼) + ∑ [𝑙𝑛(𝑡𝑖 ) − 𝑙𝑛 ( ) ( ) ] − ∑ 𝑙𝑛 ( ) ( )
𝜕𝛽 𝛽 𝛼 𝛼 𝛼 𝛼
𝑖=1 𝑗=1
𝑁𝑓 𝑁𝑠
𝜕 2 𝛬 −𝑁𝑓 𝑡𝑖 2 𝑡𝑖 𝛽 𝑡𝑗 2 𝑡𝑗 𝛽
= − ∑ [{𝑙𝑛 ( )} ( ) ] − ∑ {𝑙𝑛 ( )} ( )
𝜕𝛽 2 𝛽2 𝛼 𝛼 𝛼 𝛼
𝑖=1 𝑗=1
𝑁𝑓 𝑁𝑠 𝑁𝑓 𝑁𝑠
𝜕2𝛬 −𝑁𝑓 1 𝛽 𝛽 𝑡𝑖 𝑡𝑗 𝛽
= + 𝛽+1 [∑(𝑡𝑖 )𝛽 + ∑(𝑡𝑗 ) ] − 𝛽+1 [∑ 𝑙𝑛 ( ) (𝑡𝑖 )𝛽 + ∑ 𝑙𝑛 ( ) (𝑡𝑗 ) ]
𝜕𝛼𝜕𝛽 𝛼 𝛼 𝛼 𝛼 𝛼
𝑖=1 𝑗=1 𝑖=1 𝑗=1
−𝜕 2 𝛬 −𝜕 2 𝛬
𝜕𝛼 2 𝜕𝛼𝜕𝛽
𝐹=
−𝜕 2 𝛬 −𝜕 2 𝛬
[ 𝜕𝛽𝜕𝛼 𝜕𝛽 2 ]
If we assume a normal distribution for each parameter, we get the following 95% confidence
intervals for each:
𝟎. 𝟑𝟖 < 𝜷 < 𝟏. 𝟕𝟓
7.12 In a structural reliability analysis for a pipeline specimen under cyclic load, the model often used is:
𝑁𝑆 𝑚 = 𝑐
where S = stress amplitude, N = number of cycles to failure, and m and c are constants.
If the mean and variance of applied stress are (𝜇𝑠 𝑎𝑛𝑑 𝜎𝑠2 ), the parameter m is known without any
uncertainty, and the mean and variance of constant c are (𝜇𝑐 𝑎𝑛𝑑 𝜎𝑐2 ).
Determine the mean and variance of N. Under what conditions the uncertainty in N would be higher
than uncertainties of each S and c?
𝑐
𝑁 =
𝑆𝑚
Mean of N
Variance of N
2 (𝑁) 2
𝜕𝑁 2 2
𝜕𝑁 2
𝜎 = 𝜎𝑠 [ ] + 𝜎𝑐 [ ]
𝜕𝑆 𝜕𝑐
𝜕𝑁 −𝑚𝑐
= (𝑚+1)
𝜕𝑆 𝑆
𝜕𝑁 1
= 𝑚
𝜕𝑐 𝑆
2 2
2 (𝑁)
−𝑚𝑐 1
𝜎 = 𝜎𝑠2 [ (𝑚+1) ] + 𝜎𝑐2 [ 𝑚 ]
𝑆 𝑆
7.13 Consider a gas boiler control system shown below. The two gas valves are identical and in series; each
valve has two possible failure modes.
Schematic diagram of the control system for the gas boiler is shown below. A system failure
corresponds to a gas explosion due to unburned gas being ignited.
The three switches are identical and in series; each switch has three possible failure modes.
Stuck closed (probability 𝐹3 = 2 × 10−3 ).
Short circuit (probability 𝐹4 = 2 × 10−3 ).
If a current flows through the circuit, the solenoids are energized and both valves V1 and V2 are open.
The probability of the pilot flame failing 𝐹6 = 10−1 .
The probability of a source of ignition existing 𝐹7 = 1
a) Draw a fault tree associated with the top event of “hazard of a gas explosion.”
Risk of Gas
Explosion
F7
Switch
Pilot Valve
Failure
Failure
F6
F3 F4 F5 F3 F4 F5
F1 F2 F1 F2
F3 F4 F5
c) Recalculate (b) assuming CCFs between the valves. Use the β-factor method with β = 0.05.
a) The required system availability must be at least 0.99. How many generators must be bought?
1 1 0.21 1
𝑎𝑎𝑣|𝑖 ≈ 1 − 𝜆𝑇0 ≈1− × × 𝑦𝑟 ≈ 𝟎. 𝟗𝟒𝟕𝟓
2 2 𝑦𝑟 2
System Availability
When 𝑛 = 2
𝑎𝑠𝑦𝑠 ≈ 1 − (1 − 0.9475)2 ≈ 𝟎. 𝟗𝟗𝟕𝟐
Therefore a minimum of two generators are required to give a system availability of 0.99 or
greater.
However, if power is required all year round (ie: even during test and repair times); than four
generators are required to ensure that two are working whilst two are being tested/repaired.
𝑅𝑠𝑦𝑠 = 𝑅𝑎 + 𝑅𝑏 − 𝑅𝑎 𝑅𝑏
∞ ∞
1 1 1 1 1 1
𝑀𝑇𝑇𝐹 = + − = + − = 𝟕. 𝟏𝟒𝟑 𝒚𝒓𝒔
𝜆𝑎 𝜆𝑏 𝜆𝑎 + 𝜆𝑏 0.21 0.21 0.21 + 0.21
c) Explain how the average availability of a system with three generators is affected if we assume
the probability of generator failures are dependent. Use an α-factor method with generic data.
Assume three generators in parallel with a 2/3 required for system success configuration.
Therefore system unavailability is given by:
𝑄𝑠𝑦𝑠 = 𝑞𝑎|𝑖 𝑞𝑏|𝑖 + 𝑞𝑎|𝑖 𝑞𝑐|𝑖 + 𝑞𝑏|𝑖 𝑞𝑐|𝑖 + 𝑞𝑎𝑏|𝑐𝑐 + 𝑞𝑎𝑐|𝑐𝑐 + 𝑞𝑏𝑐|𝑐𝑐 + 𝑞𝑎𝑏𝑐|𝑐𝑐
Using the α-factor as discussed on page 349 of the text, we find the probability of system failure
as follows:
𝑘 𝛼𝑘 𝑚
𝑄𝑘 = 𝑄𝑡 𝑘 = 1,2, … 𝑚; 𝛼𝑡 = ∑ 𝑘𝛼𝑘
𝑚 − 1 𝛼𝑡 𝑘=1
( )
𝑘−1
In this case, m is the number of components affected by the CCF. From the table on page 349 of
the text, we find that the generic alpha values for a system where three components are affected
by CCF as follows:
𝛼1 = 0.95
𝛼2 = 0.04
𝛼3 = 0.01
𝛼𝑡 = 𝛼1 + 2𝛼2 + 3𝛼3 = 0.95 + 0.08 + 0.03 = 1.06
1 𝛼1 0.95
𝑄1 = 𝑄𝑡 = 𝑄
(
3 − 1 𝛼𝑡
) 1.06 𝑡
1−1
2 𝛼2 0.04
𝑄2 = 𝑄𝑡 = 𝑄
(
3 − 1 𝛼𝑡
) 1.06 𝑡
2−1
3 𝛼2 3 × 0.01
𝑄3 = 𝑄𝑡 = 𝑄𝑡
3 − 1 𝛼𝑡 1.06
( )
3−1
Therefore
2
0.95 0.04 3 × 0.01
𝑄𝑠𝑦𝑠 = 3 ( 𝑄𝑡 ) + 3 ( 𝑄𝑡 ) + ( 𝑄𝑡 )
1.06 1.06 1.06
2
0.95 0.04 3 × 0.01
𝑄𝑠𝑦𝑠 = 3 ( × 0.0525) + 3 ( × 0.0525) + ( × 0.0525)
1.06 1.06 1.06
𝑸𝒔𝒚𝒔 = 𝟎. 𝟎𝟏𝟒𝟎𝟕
We can from the calculations above that a system with three generators and susceptible to common
cause failures will have a greater unavailability than a two generator system without common
cause failures. This highlights the fact that common cause failures are the dominant component of
total system unavailability.
Assume A, B, C, and D are periodically tested (no test staggering) and with the following data:
Determine
𝑇 = 𝐴(𝐶 + 𝐷) ∙ (𝐴 + 𝐵 + 𝐶)
𝑇 = (𝐴𝐶 + 𝐴𝐷) ∙ (𝐴 + 𝐵 + 𝐶)
𝑇 = 𝐴𝐶 + 𝐴𝐷
b) The point-wise average unavailability for one test cycle (assume independence).
Using the following relationship for point wise unavailability, each component’s unavailability
can be determined and used to find the system unavailability per cycle:
1 𝑇𝑟 𝑇𝑡
𝑞𝑖 = 𝜆𝑇0 + 𝑓𝑟 +
2 𝑇 𝑇
𝑞𝑠𝑦𝑠 = 𝑞𝑎 𝑞𝑐 + 𝑞𝑎 𝑞𝑑
Considering common cause failures, the minimal cut set are given as:
Minimal Cuts sets 𝐶1 = 𝐴𝑖 𝐶𝑖 ; 𝐶2 = 𝐴𝑖 𝐷𝑖 ; 𝐶3 = 𝐶𝐶𝑎𝑐 ; 𝐶4 = 𝐶𝐶𝑎𝑑
Assume that the common cause unavailability of components A & C and A & D is the average of
their individual unavailability’s as follows:
𝑞𝑎 + 𝑞𝑐 𝑞𝑎 + 𝑞𝑑
𝑞𝑠𝑦𝑠 = (1 − 𝛽)2 𝑞𝑎 𝑞𝑐 + (1 − 𝛽)2 𝑞𝑎 𝑞𝑑 + 𝛽 ( )+𝛽( )
2 2
0.040149 + 0.184149
𝑞𝑠𝑦𝑠 = 2 × (0.95)2 × 0.040149 × 0.184149 + 2 × 0.05 ( )
2
𝑞𝑠𝑦𝑠 = 𝟎. 𝟎𝟐𝟒𝟓𝟔
This unavailability value is nearly twice the size of the case where common cause failure is not
considered.
𝑇 = 𝐺1 ∙ 𝐺2
𝑇 = (𝐵𝐶 + 𝐵̅𝐶 + 𝐵𝐶̅ ) ∙ (𝐴𝐶)
𝑇 = 𝐵𝐶𝐴𝐶 + 𝐵̅𝐶𝐴𝐶 + 𝐵𝐶̅ 𝐴𝐶
𝑇 = 𝐴𝐵𝐶 + 𝐴𝐵̅𝐶
𝑇 = 𝐴𝐶(𝐵 + 𝐵̅)
𝑇 = 𝐴𝐶
Mean:
𝜇𝐶1 = 𝜇𝑎 + 𝜇𝑐 = 0.001 + 0.0001 = 𝟏. 𝟏 × 𝟏𝟎−𝟑
Standard Deviation:
2
𝜎𝐶1 = 𝜎𝑎2 + 𝜎𝑐2 = 1.01 × 10−6
7.17 What is the MTBF in the future of a repairable component with the following observed times of failure
(current time coincides with the occurrence of last failure)?
𝑛
1 1
𝑛 − 1 𝑖=1 𝑡𝑖 − 2 𝑡𝑜
∑
𝑈=
1
𝑡𝑜 √
12(𝑛 − 1)
As |𝑈| > 0.3, we can see a positive trend in rate of failures; system is deteriorating.
𝑛 6
𝛽̂ = 𝑛−1 = = 𝟏. 𝟓𝟓
𝑡𝑜 4289 4289 4289 4289
∑ 𝑙𝑛 𝑙𝑛 832 + 𝑙𝑛 + 𝑙𝑛 2422 + ⋯ + 𝑙𝑛 3617
𝑡𝑖 1574
𝑖=1
𝑇 𝑇 ̂ ̂
1 1 ̂
𝜆̂(𝑇𝛽 − 4289𝛽 )
𝜆̅(𝑡) = ∫ 𝜆(𝑡)𝑑𝑡 = ∫ 𝜆̂𝛽̂𝑡 𝛽 −1
𝑑𝑡 =
𝑇 − 4289 𝑇 − 4289 𝑇 − 4289
4289 4289
1 𝑇 − 4289
𝑴𝑻𝑩𝑭 = =
̅𝜆 1.396 × 10 × (𝑇 1.55 − 42891.55 )
−5
7.18 Time to failure of incandescent light bulbs can be described accurately with a lognormal distribution.
A test engineer claims that a 10% increase in voltage decreases life by approximately 50%. A
particular brand of 100-W bulb has a median life of 1200 h at 110V.
a) Give an inverse power relationship expression for the life of such light bulbs as a function of
voltage.
1
𝐿(𝑉) =
𝐾𝑉 𝑛
1 1
1200 = … … . 𝑒𝑞𝑛(1) 𝑎𝑛𝑑 600 = … … . 𝑒𝑞𝑛(2)
𝐾110𝑛 𝐾121𝑛
121𝑛 ln(2)
2 = 110𝑛 => 𝑛 = 121 = 𝟕. 𝟐𝟕𝟑
𝑙𝑛( )
110
1
𝐾= = 𝟏. 𝟏𝟖𝟖 × 𝟏𝟎−𝟏𝟖
1200 × 1107.273
Therefore, the complete Inverse Power Relationship in this case is given by:
1
𝐿(𝑉) =
1.188 × 10−18 × 𝑉 7.273
𝐿𝑢𝑠𝑒 𝐾 −1 𝑉𝑢𝑠𝑒
−𝑛
𝑉𝑎𝑐𝑐 𝑛
𝐴𝐹 = = −𝑛 = ( )
𝐿𝑎𝑐𝑐 𝐾 −1 𝑉𝑎𝑐𝑐 𝑉𝑢𝑠𝑒
The results for the given accelerating voltages are summarized below:
V AF
140 5.78
7.19 A scenario is characterized by the following cut sets: S = ab + ac + d. Probability of a and b can be
obtained from an exponential model such as 𝑃𝑟𝑎 (𝑡) = 1 − 𝑒𝑥𝑝(−𝜆𝑎 𝑡), but probability of c and d
can be obtained from a Weibull model such as:
𝑡 𝛽
𝑃𝑟𝑐 (𝑡) = 1 − 𝑒𝑥𝑝 [− ( ) ]
𝛼
If 𝜆𝑎 = 𝜆𝑎 = 10−4 ℎ𝑟𝑠 −1 and 𝛼𝑐 = 𝛼𝑑 = 2250 ℎ𝑟𝑠 and 𝛽𝑐 = 𝛽𝑑 = 1.4 Determine the importance of
parameters α, β, and λ in modeling this scenario. Use an importance measure of your choice, and
justify your choice.
Solution:
The complement of the cut set is given by:
𝑆̅ = ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
𝑎𝑏 + 𝑎𝑐 + 𝑑 = (𝑎̅ + 𝑏̅)(𝑎̅ + 𝑐̅)𝑑̅
𝑆̅ = 𝑎̅𝑑̅ + 𝑏̅ 𝑐̅𝑑̅
1000 1.4
𝑅𝑐 = 𝑅𝑑 = exp [− ( ) ] = 0.7252
2250
Using the Birnbaum Importance method, we find the importance of each component as follows:
𝐼𝑖𝐵 = 𝑅𝑠 [𝑅(𝑡)|𝑅𝑖 (𝑡) = 1] − 𝑅𝑠 [𝑅(𝑡)|𝑅𝑖 (𝑡) = 0]
From these calculations, we can see that component D, is almost four times as important as the next
component which is component A (0.9673 versus 0.2493). This makes sense as component D is a
minimal cut set on its own and component A appears in each of the other two minimal cut sets.
Therefore, for the given system it appears that the Weibull parameters α & β are more important to the
system reliability than λ. To determine the order of parameter importance, we can take the derivatives
of the Weibull Reliability function to compare as follows:
𝜕𝑅 𝜕 𝑡 𝛽
−( ) 𝛽𝑡 𝑡 𝛽−1 −( 𝑡 )𝛽
= {𝑒 𝛼 } = 2( ) 𝑒 𝛼
𝜕𝛼 𝜕𝛼 𝛼 𝛼
𝜕𝑅 𝜕 𝑡 𝛽
−( ) 𝑡 𝑡 𝛽 𝑡 𝛽
−( )
= { 𝑒 𝛼 } = 𝑙𝑛 ( ) × ( ) × 𝑒 𝛼
𝜕𝛽 𝜕𝛽 𝛼 𝛼
𝜕𝑅 𝜕 −4
= { 𝑒 −𝜆𝑡 } = −𝑡𝑒 −𝜆𝑡 = −1000𝑒 −1000×1×10 = −𝟗𝟎𝟓
𝜕𝜆 𝜕𝜆
𝜕𝑅 𝜕𝑅 𝜕𝑅
Given that 𝜕𝛼
> 𝜕𝛽
> 𝜕𝜆
, we can conclude that α is more important than β and λ in this case in terms
of its potential to influence the reliability of the system.
Therefore, the overall order of importance in the parameters for the system detailed above is given as
α, β, and then λ.
d) What is the probability of failure of the system (assuming 2 out of the 3 units need to function for
the system to work)? Solve for both the cases of no-dependent failures and dependent failures and
compare the results.
a) Based on the data in this table plot the risk profile in terms of “annual frequency of exceeding the
given number of fatalities versus number of fatalities.” That is, the so-called “Farmer’s Curve.”
5
9.04E-9
7.41E-9
7.39E-9
6.39E-9
1.89E-9
8.2 If an accident requires occurrence of event I followed by either of events B or C so that a major
consequence happens:
C
Seq 1 = S
B
I Seq 2 = F
B
Seq 3 = F
𝐵̅ = 1 − 𝐵 = 𝟎. 𝟗𝟗𝟕
𝐶̅ = 1 − 𝐶 = 𝟎. 𝟗𝟗𝟑
c) If consequence of each scenario is 100 injuries, what is the total risk of this accident?
Injuries will occur if sequences 2 or 3 eventuate. Therefore, the total risk of this accident is given
by:
C
Seq 1 = S
C
Seq 2 = F
C Seq 3 = S
B
A C
Seq 4 = F
B
Seq 5 = F
b) If Boolean expressions A = x + yz, B = y + t, C = tz, show the minimal cut sets of subsystems A, B,
and C, then find the minimal cut sets of each scenario (sequence) of the event tree.
𝐶𝐵1 = 𝑦; 𝐶𝐵2 = 𝑡;
𝐶𝐶1 = 𝑡𝑧
Subsystem Complements
𝐴̅ = ̅̅̅̅̅̅̅̅̅
𝑥 + 𝑦𝑧 = 𝑥̅ (𝑦̅ + 𝑧̅) = 𝑥̅ 𝑦̅ + 𝑥̅ 𝑧̅
𝐵̅ = ̅̅̅̅̅̅̅
𝑦 + 𝑡 = 𝑦̅𝑡̅
𝐶̅ = 𝑡𝑧
̅ = 𝑡̅ + 𝑧̅
c) If the probability of the components or the subsystems are as follows, then calculate the frequency
(per week) of each scenario: Pr(x) = Pr(y) = 0.1, Pr(z) = Pr(t) = 0.2
2⁄
𝐼= 𝑚𝑜𝑛𝑡ℎ = 0.5 𝑤𝑒𝑒𝑘 −1
4 𝑤𝑒𝑒𝑘𝑠⁄
𝑚𝑜𝑛𝑡ℎ
𝑆𝑒𝑞 4 = 𝟎
The technical department has made a model of the degradation of the turbine. In this model it is assumed
that, if the state is “good” or “acceptable,” then at the end of the quarter it stays the same with probability
0.95 or degrades with probability 0.05 (i.e., good becomes “acceptable” and “acceptable” becomes
“poor”). If the state is “poor,” then it stays “poor.”
a) Determine the probability that a turbine that is “good” becomes “good,” “acceptable,” and “poor”
in the next three quarters, and does not fail.
Prob
Pr(G|G) 0.8573
Pr(A|G) 0.1352
Pr(P|G) 0.0072
Total 0.9997
The cost of repairing the failure is $1.5M (including replacement of parts). The cost of a new turbine
is $5K.
Consider a situation where the old turbine is categorized as “acceptable”; determine the optimal
decision using expected monetary loss (risk).
We are given the probability of failure given the state is acceptable as: Pr(𝐹|𝐴) = 0.001
𝑀𝑜𝑛𝑒𝑡𝑎𝑟𝑦 𝑅𝑖𝑠𝑘 = Pr(𝐹) × 𝐶𝑜𝑠𝑡 = 0.001 × 1.5 × 106 = $𝟏𝟓𝟎𝟎 𝑝𝑒𝑟 𝑞𝑢𝑎𝑟𝑡𝑒𝑟
Therefore, continuing to use the old turbine is less risky than replacement (ie: $1500 < $5000)
What would the decision be if the old turbine was “good” with 10% probability and “acceptable”
with 90% probability?
𝑀𝑜𝑛𝑒𝑡𝑎𝑟𝑦 𝑅𝑖𝑠𝑘 = (0.9 × 0.001 + 0.1 × 0.0001) × 1.5 × 106 = $𝟏𝟑𝟔𝟓 𝑝𝑒𝑟 𝑞𝑢𝑎𝑟𝑡𝑒𝑟
Again, continuing to use the old turbine is less risky than replacement (ie: $1365 < $5000)
It is possible to carry out two sorts of inspection on the old turbine. A visual inspection costs $200,
but does not always give the right diagnosis. The probability of a particular outcome of the
inspection given the actual state of the turbine is given in the table below:
The second sort of inspection uses an x-ray and determines the state of the turbine exactly. The cost
of the x-ray inspection is $800.
Assuming that no information is available on the current state of the turbine, some inspection is
required:
Given that we are told the replacement cost of a turbine is $5k, we only replace it if the state is
found to be poor.
ii. An Acceptable turbine is classified as Poor: This would invoke an unnecessary turbine
replacement costing $5k. The risk of this occurring according to the data in the table above
is 0.1 × $5k = $500. The total risk is $200 + $500 which is still less than the cost of an x-
ray (ie: $700 < $800)
iii. A Poor turbine is classified as Acceptable: This would see continued use of a poor turbine
with possible subsequent failure. The risk of this scenario is 0.1 × $15k + $200 = $1700.
This risk is greater than the cost of an x-ray (ie: $1700 > $800)
a) Determine a Boolean equation representing each of the event tree scenarios in terms of the fault
tree failure events (𝐶1 , 𝐶2 , 𝑎𝑛𝑑 𝐶3 ).
The Boolean Expression and Complement for each Subsystem is given by:
𝐴 = 𝐶1 ∙ 𝐶2
𝐴̅ = 𝐶
̅̅̅̅̅̅̅̅ ̅̅̅ ̅̅̅
1 ∙ 𝐶2 = 𝐶1 + 𝐶2
𝐵 = 𝐶1 + 𝐶3
̅̅̅̅̅̅̅̅̅̅
𝐵̅ = 𝐶 ̅̅̅ ̅̅̅
1 + 𝐶3 = 𝐶1 ∙ 𝐶3
̅̅̅1 ∙ ̅̅̅
𝑆𝑐𝑒𝑛𝑎𝑟𝑖𝑜 2 = 𝐼𝐴𝐵̅ = 𝐼[(𝐶1 ∙ 𝐶2 )(𝐶 𝐶3 )]
𝑆𝑐𝑒𝑛𝑎𝑟𝑖𝑜 2 = 𝐼[∅]
𝑆𝑐𝑒𝑛𝑎𝑟𝑖𝑜 3 = 𝐼𝐵 = 𝐼(𝐶1 + 𝐶3 )
b) If the frequency of the initiating event I is 10−3 year−1, and Pr(𝐶1 ) = 0.001; Pr(𝐶2 ) =
0.008 𝑎𝑛𝑑 Pr(𝐶3 ) = 0.005, calculate the risk (expected injuries per year).
̅̅̅1 ∙ ̅̅̅
𝑆𝑐𝑒𝑛𝑎𝑟𝑖𝑜 1 = 𝐼𝐶 𝐶3 = 0.001 × 0.999 × 0.995 = 𝟎. 𝟎𝟎𝟎𝟗𝟗𝟒 𝒚𝒓−𝟏
𝑆𝑐𝑒𝑛𝑎𝑟𝑖𝑜 2 = 𝐼[∅] = 𝟎
c) Plot the risk-profile curve (Farmer Curve) for this problem. That is to calculate complementary
cumulative probability of exceeding certain number of injuries
3
1.594E-3
6.0E-4
10 100 Injuries