TX427_Frame_C02 Page 9 Wednesday, December 19, 2001 11:42 AM

Maintenance
2 Mathematics
INTRODUCTION
As in the case of other engineering disciplines, mathematics is an indispensable
1
maintenance tool. Mathematics applications in engineering are relatively new. A his-
tory of mathematics is provided in Reference 1.
In maintenance, mathematics find applications in work sampling, inventory con-
trol analysis, failure data analysis, establishing optimum preventive maintenance pol-
icies, maintenance cost analysis, and project management control. Some of the areas
of mathematics used in maintenance include set theory, probability, calculus, differ-
ential equations, Stochastic processes, and Laplace transforms. Even though many
excellent texts are available in areas such as these, this chapter presents essential
mathematical concepts to enable understanding of the material presented in the book.
This should eliminate the need for readers to consult math books.

BOOLEAN ALGEBRA AND PROBABILITY PROPERTIES

Boolean algebra is important in probability theory and is named after George Boole
2
(1813–1864), its originator. Table 2.1 presents selective rules of Boolean algebra.
The capital letters denote arbitrary sets or events and the symbol + denotes the union
of sets or events. The intersection of sets in the table is written without the dot.
Nevertheless, in some other documents it could have been written with the symbol
∩ or with a dot.
3,4
Important properties of probability are as follows:

• The probability of occurrence of event, Y, is always

0 ≤ P(Y ) ≤ 1 (2.1)

where P(Y ) is the probability of occurrence of Y.

• The probability of occurrence and nonoccurrence of Y is given by

P(Y ) + P(Y ) = 1 (2.2)

where Y is the negation of Y and P ( Y ) is the probability of nonoccurrence

of Y.

©2002 CRC Press LLC

TX427_Frame_C02 Page 10 Wednesday, December 19, 2001 11:42 AM

TABLE 2.1
2,3
Commonly Used Boolean Algebra Rules
Rule Description Symbolism

Absorption law Y ( Y + A) = Y
Y + YA = Y
Commutative law AY = YA
A +Y =Y +A
Idempotent law YY = Y
Y +Y =Y
Distributive law A (Y + B) = AY + AB
A + (YB) = (A + Y ) (A + B)
Associative law A (YB) = (AY ) B
(A + B) + Y = A + ( Y + B)

• The probability of the sample space, S, is

P(S ) = 1 (2.3)

• The probability of the negation of the sample space S is

P(S ) = 1 (2.4)

• The probability of an intersection of independent events, Y1, Y2, Y3 …, Yn, is

P ( Y 1 Y 2 Y 3 …Y n ) = P ( Y 1 )P ( Y 2 )P ( Y 3 )…P ( Y n ) (2.5)

where
n = total number of events,
Yi = ith event, for i = 1, 2, 3,…, n,
P(Yi ) = probability of occurrence of event Yi, for i = 1, 2, 3,…, n.

• The probability of the union of n independent events is given by

n

P(Y 1 + Y 2 + Y 3 + … + Y n) = 1 – ∏ (1 – P(Y i)) (2.6)

i=1

• The probability of the union of n mutually exclusive events is expressed by

P(Y 1 + Y 2 + Y 3 + … + Y n) = ∑ P(Y i) (2.7)

i=1

©2002 CRC Press LLC

TX427_Frame_C02 Page 11 Wednesday, December 19, 2001 11:42 AM

Note that for very small values of P(Y1), P(Y2), P(Y3),…,P(Yn), Eq. (2.6) yields almost
the same result to Eq. (2.7).

Example 2.1

Assume that in Eq. (2.6), we have n = 2, P(Y1) = .04, and P(Y2) = .06. Calculate the
probability of the union of independent events Y1 and Y2. Use the same given data
in Eq. (2.7) and comment on the results given by Eqs. (2.6) and (2.7).
For n = 2, Eq. (2.6) yields

P ( Y1 + Y2 ) = 1 – ∏ ( 1 – P ( Y i ) )
i=1

= P ( Y1 ) + P ( Y2 ) – P ( Y1 )P ( Y2 ) (2.8)

Substituting the given values for P(Y1) and P(Y2) into Eq. (2.8), we get

P ( Y1 + Y2 ) = .04 + .06 – ( .04 ) ( .06 )

= .0976

Using the same given data in Eq. (2.7) yields

P ( Y1 + Y2 ) = P ( Y1 ) + P ( Y2 )
= .1000

The above two results are almost identical.

PROBABILITY AND CUMULATIVE DISTRIBUTION

FUNCTION DEFINITIONS
PROBABILITY
5
This is defined by

P ( Y ) = lim ( M/m ) (2.9)

m→∞

where P(Y ) is the probability of occurrence of event Y and M is the total number
of times that Y occurs in the m repeated experiments.

CUMULATIVE DISTRIBUTION FUNCTION

5,6
This is expressed by
t
F (t) = ∫–∞ f ( y ) dy (2.10)

©2002 CRC Press LLC

TX427_Frame_C02 Page 12 Wednesday, December 19, 2001 11:42 AM

where
t = time,
F(t) = cumulative distribution function,
f(y) = probability density function.

By differentiating Eq. (2.10) with respect to t, we get

t
dF ( t ) d ( ∫ –∞ f ( y ) dy )
-------------- = ---------------------------------
- = f (t) (2.11)
dt dt

Setting t = ∞ in Eq. (2.10) yields

∞
F (∞) = ∫–∞ f ( x ) dx = 1 (2.12)

This proves that the total area under the probability density curve is always equal
to unity.

PROBABILITY DISTRIBUTIONS OF CONTINUOUS

RANDOM VARIABLES
Over the years many continuous random variable probability distributions have been
developed. This section presents some of those useful for performing mathematical
7–9
maintenance analysis-related studies.

EXPONENTIAL DISTRIBUTION
This is one of the most widely used probability distributions in engineering, partic-
10
ularly in reliability work. It is relatively easy to handle in conducting analysis. The
distribution probability density function is defined by

–λ t
f (t ) = λe , t ≥ 0, λ > 0 (2.13)

where λ is the distribution parameter.

By substituting Eq. (2.13) into Eq. (2.10) we get the following expression for
the exponential distribution cumulative distribution function:

t –λ y –λ t
F (t) = ∫0 λ e dy = 1 – e (2.14)

Example 2.2
By setting t = ∞ in Eq. (2.14) prove that the value of the cumulative distribution
function is equal to unity.

©2002 CRC Press LLC

TX427_Frame_C02 Page 13 Wednesday, December 19, 2001 11:42 AM

Thus, for t = ∞ Eq. (2.14) becomes

–λ ∞
F (∞) = 1 – e = 1– 0 = 1

The above result proves that values of F(t) for t = ∞ is always equal to unity.

RAYLEIGH DISTRIBUTION
This distribution, developed by John Rayleigh (1842–1919), is used often in reliability
1
engineering and in the theory of sound. Its probability density function is expressed by

f ( t ) =  -----2 te
2 – ( t /α )
2
, t ≥ 0, α > 0 (2.15)
α 

where α is the distribution parameter.

Inserting Eq. (2.15) into Eq. (2.10), we obtain

2
– ( t /α )
F (t) = 1 – e (2.16)

The above equation is the Rayleigh distribution cumulative distribution function.

Example 2.3

Obtain an expression for the probability density function by using Eq. (2.16) in Eq. (2.11).
Substituting Eq. (2.16) into Eq. (2.11) yields

dF ( t )
f ( t ) = -------------- =  -----2 e
2t –( t /α )2
(2.17)
dt  α

Note that Eq. (2.17) is identical to Eq. (2.15). Thus, it proves that by differen-
tiating the cumulative function, F(t), with respect to time, t, yields the probability
density function.

WEIBULL DISTRIBUTION
This distribution was developed by W. Weibull of the Royal Institute of Technology,
11
Stockholm, in the early 1950s. Weibull distribution is useful for representing many
different physical phenomena. Its probability density function is defined by

b
b−1 – ( t /α )
bt e
-, t ≥ 0, b > 0, α > 0
f ( t ) = ------------------------- (2.18)
α
b

where b and α are the shape and scale parameters, respectively.

©2002 CRC Press LLC

TX427_Frame_C02 Page 14 Wednesday, December 19, 2001 11:42 AM

Using Eq. (2.18) in Eq. (2.10), we get

b
– ( t /α )
F (t) = 1 – e (2.19)

Equation (2.19) is also known as Weibull cumulative distribution function.

Example 2.4

Obtain expressions by using Eq. (2.19) for b = 1 and b = 2 and comment on the
resulting equations.
Thus, for b = 1 and b = 2 Eq. (2.19) yields the following expressions, respectively:

– t /α
F (t) = 1 – e (2.20)

and

2
– ( t /α )
F (t) = 1 – e (2.21)

Equations (2.20) and (2.21) are identical to Eqs. (2.14) (i.e., for 1/ α = λ ) and
(2.16), respectively. It means for b = 1 and b = 2 exponential and Rayleigh distri-
butions are the special cases of the Weibull distribution, respectively.

NORMAL DISTRIBUTION
This distribution is sometime called the Gaussian distribution after Carl Friedrich
Gauss (1777–1855), a German mathematician. It is one of the most widely used
statistical distributions. The distribution probability density function is expressed by

(t – µ)
2
1
f ( t ) = -------------- exp – -----------------
- , – ∞ < t < +∞ (2.22)
σ 2π 2σ
2

where µ and σ are the distribution parameters (i.e., mean and standard deviation,
respectively).
Substituting Eq. (2.22) into Eq. (2.10), we obtain

(t – µ)
2
1 t
F ( t ) = -------------- ∫ exp – -----------------
- dy (2.23)
σ 2 π –∞ 2σ
2

4,6,11
The values of Eq. (2.23) are tabulated in various mathematical books.
This distribution was actually discovered by De Moivre as early as in 1733 but
6
due to historical error was attributed to Carl Gauss.

©2002 CRC Press LLC

TX427_Frame_C02 Page 15 Wednesday, December 19, 2001 11:42 AM

GENERAL DISTRIBUTION
This distribution can represent a wide range of physical phenomena, and its proba-
12
bility density function is expressed by

β β
β −1 θt θt
f ( t ) = [ m λ st + (1 – m)βt θ e ] exp [ – m λ t – ( 1 – m ) ( e
s−1 s
– 1 ) ] (2.24)

for 0 ≤ m ≤ 1 and λ , s, β , θ > 0

where λ and θ are the scale parameters, β and s are the shape parameters.
By inserting Eq. (2.24) into Eq. (2.10), we get the following expression for the
cumulative distribution function:

β
θt
F ( t ) = 1 – exp [ – m λ t – ( 1 – m ) ( e
s
– 1)] (2.25)

The following statistical functions are the special cases of the general distribution:

• For m = 1: Weibull
• For m = 1 and s = 2: Rayleigh
• For m = 1 and s = 1: Exponential
• For m = 0 and β = 1: Extreme value
• For s = 1 and β = 1: Makeham
s = 0.5 and β = 1: Bathtub
13
• For

Table 2.2 presents cumulative distribution functions for the distributions discussed
earlier.

TABLE 2.2
Cumulative Distribution Functions for Selective
Distributions
Distribution Name Cumulative Distribution Function (F(t))
β
θt
1 – exp [ – m λ t – (1 – m)(e
s
General – 1) ]
–λ t
Exponential 1–e
2
– (t/ α )
Rayleigh 1–e
b
– (t/ α )
Weibull 1–e
(t – µ )
2
1 t
Normal --------------- ∫ exp – -----------------
- dy
σ 2 π –∞ 2σ 2

©2002 CRC Press LLC

TX427_Frame_C02 Page 16 Wednesday, December 19, 2001 11:42 AM

LAPLACE TRANSFORMS: INITIAL AND FINAL

VALUE THEOREMS
Laplace transforms are useful for solving system of linear differential equations in
mathematical maintenance analysis. These transforms are named for Pierre-Simon
1
Laplace (1749–1827) who died exactly 100 years after the death of Isaac Newton.
14–16
The Laplace transform of the function, f(t), is expressed by

∞
∫0
– st
f (s) = f ( t )e dt (2.26)

where
t = time,
s = Laplace transform variable,
f(s) = Laplace transform of f(t).

Example 2.5
Obtain Laplace transforms of the following two functions:

• f (t) = 1 (2.27)
–λ t
• f (t) = e (2.28)

where λ is a constant.
Using Eq. (2.27) in Eq. (2.26), we get

∞ 1
∫0 1 ⋅ e
– st
f (s) = dt = --- , for s > 0 (2.29)
s

Substituting Eq. (2.28) into Eq. (2.26) yields

∞ – ( s+ λ )t ∞
– λ t – st –e 1
f (s) = ∫0 e e dt = ------------------
s+λ 0
= ------------ , for s > 0
s+λ
(2.30)

Laplace transforms of some selective functions are presented in Table 2.3.

INITIAL AND FINAL VALUE THEOREMS

The initial value theorem is given by

lim f ( t ) (2.31)
t→0

©2002 CRC Press LLC

TX427_Frame_C02 Page 17 Wednesday, December 19, 2001 11:42 AM

TABLE 2.3
Laplace Transforms of Some Common Functions
No. f(t) f(s)
–λ t 1
1 e -----------, for s > – λ
s+λ
1
2 1 --- , for s > 0
s
df (t)
3 ------------- s f (s) − f(0)
dt
t f (s)
4 ∫0 f ( y)d y -----------
s
1
5 t ----2
s
m−1
t 1
6 -------------------- ----m- , for m = 1, 2, 3, …
(m – 1)! s
m−1 λ t
t e 1
7 -------------------- ------------------m- , for m = 1, 2, 3, …
(m – 1)! (s – λ )

6
The Laplace transform of Eq. (2.31) is

lim sf ( s ) (2.32)
s→∞

The final value theorem is expressed by

lim f ( t ) (2.33)
t→∞

6
The Laplace transform of Eq. (2.33) is given by

lim sf ( s ) (2.34)
s→0

Example 2.6

Prove that Eqs. (2.33) and (2.34) are equal. From Table 2.3 and Eq. (2.26) we write

df ( t ) ∞ d f (t)
∫0 e
– st
L ------------ = ------------- dt = sf ( s ) – f ( 0 ) (2.35)
dt dt

where L is the Laplace transform operator.

The limit of

∞ d f (t)
∫0
– st
e ------------- dt
dt

©2002 CRC Press LLC

TX427_Frame_C02 Page 18 Wednesday, December 19, 2001 11:42 AM

as s → 0 is

∞ d f (t) d f (t)
∞
lim ∫ e ∫0
– st
------------- dt = ------------- dt
s→0 0 dt dt
w d f (t)
= lim ∫ ------------- dt
w→∞ 0 dt
= lim [ f ( w ) – f ( 0 ) ]
w→∞

= lim f ( t ) – f ( 0 ) (2.36)
t→∞

The limit of [s f(s) − f(0)] as s → 0 is

lim sf ( s ) – f ( 0 ) (2.37)
s→0

From Eqs. (2.36) and (2.37) we obtain

lim f ( t ) – f ( 0 ) = lim sf ( s ) – f ( 0 ) (2.38)

t→∞ s→0

Equation (2.38) yields

lim f ( t ) = lim sf ( s ) (2.39)

t→∞ s→0

The above equation proves that Eqs. (2.33) and (2.34) are equal.

Example 2.7

Assume that we have

µ λ –( λ + µ )t
f ( t ) = ------------- + ------------- e (2.40)
λ+µ λ+µ

where λ and µ are parameters or constants and t is time.

Prove using Eq. (2.40) that Eqs. (2.33) and (2.34) yield identical results. By
substituting Eq. (2.40) into Eq. (2.33), we get

µ λ –( λ + µ )t µ
lim ------------- + ------------- e = ------------- (2.41)
t→∞ λ + µ λ+µ λ+µ

Taking the Laplace transform of Eq. (2.40) yields

µ 1 λ 1
f ( s ) = ------------- --- + ------------- --------------------------
λ + µ s λ + µ (s + λ + µ)
(s + µ)
= ----------------------------- (2.42)
s(s + λ + µ)

©2002 CRC Press LLC

TX427_Frame_C02 Page 19 Wednesday, December 19, 2001 11:42 AM

Inserting Eq. (2.42) into Eq. (2.34), we get

(s + µ) µ
lim s ----------------------------- = ------------- (2.43)
s→0 s(s + λ + µ) λ+µ

Equations (2.41) and (2.43) prove that Eqs. (2.33) and (2.34) yield identical results.

ALGEBRAIC EQUATIONS
Mathematical maintenance analysis may involve determining roots of algebraic equa-
tions. A root may be described as a value of variable when insertion into the polynomial
equation leads to the value of the equation equal to zero. When all roots of the poly-
11,17,18
nomial equation are found, it is considered solved.

QUADRATIC EQUATION
Although quadratic equations were solved around 2000 BC by Babylonians, in Western
society before the seventeenth century the theory of equations was handicapped by
1
the failure to recognize negative or complex numbers as the roots of equations. A
quadratic equation is defined by

2
Ax + Bx + C = 0 (2.44)

where x is a variable; A, B, and C are the constants.

Solutions to Eq. (2.44) are given below:

1/2
x1 , x2 = ( –B ± D )/2 A (2.45)

where

2
D ≡ B – 4 AC (2.46)

For real A, B, and C the roots can be classified as follows:

• For D > 0: real and unequal

• For D = 0: real and equal
• For D < 0: complex conjugate

If x1 and x2 are the roots of Eq. (2.44) then we have

x 1 x 2 = C/ A (2.47)

and

x 1 + x 2 = – B/ A (2.48)

©2002 CRC Press LLC

TX427_Frame_C02 Page 20 Wednesday, December 19, 2001 11:42 AM

CUBIC EQUATION
Italian mathematicians played an instrumental role in finding the algebraic solution
to cubic equation. In 1545 Girolamo Cardano (1501–1576) published a Latin treatise
1
on algebra at Nuremberg in Germany and included Tartaglia’s solution of the cubic.
Cubic equation is expressed by
3 2
x + B1 x + B2 x + B3 = 0 (2.49)

where x is a variable; B1, B2, and B3 are the constants.

Let
2
L = ( 3B 2 – B 1 )/9 (2.50)

3
M = ( 9B 1 B 2 – 27B 3 – 2B 1 )/54 (2.51)

3 2 1/2 1/3
N = [M + (L + M ) ] (2.52)
and

3 2 1/2 1/3
P = [M – (L + M ) ] (2.53)

The roots of Eq. (2.49) are given below:

B
x 1 = N + P – -----1 (2.54)
3

1 B 1
x 2 = – --- ( N + P ) – -----1 + --- i 3 ( N – P ) (2.55)
2 3 2

1 B 1
x 3 = – --- ( N + P ) – -----1 – --- i 3 ( N – P ) (2.56)
2 3 2

Let
3 2
T = L +M (2.57)
For real B1, B2, and B3 the roots can be classified as follows:
• For T > 0: one real and two complex conjugate
• For T < 0: all real and unequal
• For T = 0: all real and at least two equal

DIFFERENTIAL EQUATIONS

In mathematical maintenance analysis it may be necessary to find solutions to a set

of linear differential equations, particularly when applying the Markov method. Even
though there are various methods for solving differential equations, the Laplace

©2002 CRC Press LLC

TX427_Frame_C02 Page 21 Wednesday, December 19, 2001 11:42 AM

transform approach is probably the most effective technique for solving a set of
linear differential equations.
The following example demonstrates the application of Laplace transforms to
solve a set of linear differential equations.

Example 2.8

Assume that the following two differential equations describe a repairable system:

dP 0 ( t )
- = –λ P0 ( t ) + µ P1 ( t )
--------------- (2.58)
dt

dP 1 ( t )
- = –µ P1 ( t ) + λ P0 ( t )
--------------- (2.59)
dt

where
Pi (t) = probability that the system is in state i at time t, for i = 0 (working
normally), i = 1 (failed),
λ = system failure rate,
µ = system repair rate.

At time t = 0, P0(0) = 1, and P1(0) = 0.

Prove by using Laplace transforms and Eqs. (2.58) and (2.59) that the probability
of the system operating normally, i.e., P0(t), is given by Eq. (2.40).
Taking Laplace transforms of Eqs. (2.58) and (2.59), we get

sP 0 ( s ) – P 0 ( 0 ) = – λ P 0 ( s ) + µ P 1 ( s ) (2.60)

sP 1 ( s ) – P 1 ( 0 ) = – µ P 1 ( s ) + λ P 0 ( s ) (2.61)

where Pi (s) is the Laplace transform of the probability that the system is in state i,
for i = 0,1.
For given initial conditions Eqs. (2.60) and (2.61) become

sP 0 ( s ) – 1 = – λ P 0 ( s ) + µ P 1 ( s ) (2.62)

sP 1 ( s ) = – µ P 1 ( s ) + λ P 0 ( s ) (2.63)

Rearranging Eq. (2.63) yields

λ P0 ( s )
P 1 ( s ) = ---------------- (2.64)
s+µ

©2002 CRC Press LLC

TX427_Frame_C02 Page 22 Wednesday, December 19, 2001 11:42 AM

Substituting Eq. (2.64) into Eq. (2.62), we obtain

(s + µ)
P 0 ( s ) = ----------------------------- (2.65)
s(s + λ + µ)

Taking the inverse Laplace transform of Eq. (2.65) results in

µ λ –( λ + µ )t
P 0 ( t ) = ------------- + ------------- e (2.66)
λ+µ λ+µ

For f(t) = P0(t), Eqs. (2.40) and (2.66) are identical. It means Eq. (2.40) denotes
the probability of the system operating normally when its (i.e., system) failure and
repair rates are given.

PROBLEMS
1. Discuss the following Boolean algebra laws:
• Idempotent law
• Absorption law
2. Give a physical example of mutually exclusive events.
3. What are the independent events?
4. Define the following:
• Probability density function
• Cumulative distribution function
5. Prove that the cumulative distribution function of exponential distribution
is given by

–λ t
F (t) = 1 – e (2.67)

where t is time and λ is the distribution parameter.

6. Write the probability density function of Weibull distribution. What are
the special case distributions of the Weibull distribution?
7. Write the special case statistical functions of the general distribution.
8. Compare general and Weibull distributions.
9. Prove that the Laplace transform of f(t) = t is given by

1
f ( s ) = ----2 (2.68)
s

10. Find the roots of the following equation:

3 2
x + 2x – 5x – 6 = 0 (2.69)

where x is a variable.

©2002 CRC Press LLC

TX427_Frame_C02 Page 23 Wednesday, December 19, 2001 11:42 AM

REFERENCES
1. Eves, H., An Introduction to the History of Mathematics, Holt, Rinehart, and Winston,
New York, 1976.
2. Lipschutz, S., Set Theory and Related Topics, McGraw-Hill, New York, 1964.
3. NUREG-0492, Fault Tree Handbook, Nuclear Regulatory Commission (NRC),
Washington, D.C., 1981.
4. Lipschutz, S., Probability, McGraw-Hill, New York, 1965.
5. Mann, N.R., Schafer, R.E., and Singpurwalla, N.D., Methods for Statistical Analysis
of Reliability and Life Data, John Wiley & Sons, New York, 1974.
6. Shooman, M.L., Probabilistic Reliability: An Engineering Approach, McGraw-Hill,
New York, 1968.
7. Patel, J.K., Kapadia, C.H., and Owen, D.B., Handbook of Statistical Distributions,
Marcel Dekker, New York, 1976.
8. Ireson, W.G., editor, Reliability Handbook, McGraw-Hill, New York, 1966.
9. Dhillon, B.S., Life distributions, IEEE Transac. Reliability, 30, 1981, 457–460.
10. Davis, D.J., An analysis of some failure data, J. Am. Stat. Assoc., 1952, 113–150.
11. Spiegel, M.R., Mathematical Handbook of Formulas and Tables, McGraw-Hill, New
York, 1968.
12. Dhillon, B.S., A hazard rate model, IEEE Transac. Reliability, 29, 1979, 150.
13. Dhillon, B.S., Reliability Engineering in Systems Design and Operation, Van Nostrand
Reinhold Co., New York, 1983.
14. Nixon, F.E., Handbook of Laplace Transformation, Prentice-Hall, Englewood Cliffs,
New Jersey, 1965.
15. Spiegel, M.R., Laplace Transforms, McGraw-Hill, New York, 1965.
16. Oberhettinger, F. and Badii, L., Tables of Laplace Transforms, Springer-Verlag,
Berlin, 1973.
17. Abramowitz, M. and Stegun, I.A., editors, Handbook of Mathematical Functions,
U.S. Government Printing Office, Washington, D.C., 1972.
18. Burington, R.S., Handbook of Mathematical Tables and Formulas, McGraw-Hill,
New York, 1973.

©2002 CRC Press LLC