You are on page 1of 13

1

Thailand Statistician https://www.statassoc.or.th Contributed paper Some Properties of the Three-Parameter Crack Distribution Phitchaphat Bowonrattanaset[a]*, Kamon Budsaba[a,b] [a] Department of Mathematics and Statistics, Faculty of Science and Technology, Thammasat University, Phathum Thani, 12121, Thailand. [b] Centre of Excellence in Mathematics, CHE, Si Ayutthaya Rd., Bangkok 10400, Thailand. *Author for correspondence, e-mail: jinnyice@hotmail.com

Abstract The purpose of this paper is to propose the main theoretical properties of a new lifetime distribution. We will call this new distribution that the Crack distribution (CR) because it covers physical characteristics of fatigue cracks. CR has three parameters. CR relates to the two-parameter distributions i.e. the Birnbaum-Saunders distribution, the Inverse Gaussian distribution which are two well-known fatigue-lifetime distributions and the two-parameter distribution i.e. the Length Biased Inverse Gaussian distribution. These three distributions are the special cases of CR based on non-classical parametrization. The main theoretical properties i.e. the characteristic function, the moment generating function and the cumulative distribution function on three-parameter CR are established in closed form. The existing conditions of each function are considered to satisfy their definition. _______________________________ Keywords: The Crack distribution, fatigue-lifetime, parametrization.

1.Introduction The Reliability Theory achieved numerous applications in Physics, Engineering, Statistics, Environment and Economics. One of the important notions in the reliability theory is the notion of lifetime distributions. In this paper, we will study the new three-parameter fatigue-lifetime distribution: The Crack distribution. It relate to three known two-parameter distributions i.e. the Birnbaum-Saunders distribution, the Inverse Gaussian distribution, and the Length Biased Inverse Gaussian distribution. The value of their random variables is time before failure. Birnbaum and Saunders [1,2] extended properties of Birnbaum-Saunders distribution. Desmond [3] compared two fatigue-life models i.e. Birnbaum-Saunders distribution and Inverse Gaussian distribution. The Birnbaum-Saunders distribution is used in case of cyclic loading while the Inverse Gaussian distribution is used in case of non-cyclic loading. The practitioners can use the Crack Lifetime distribution to be general form of both cases. Shuster [4] indicated a method to obtain the exact probabilities of the Inverse Gaussian distribution by using Standard Normal tables and Logs tables. Chhikara and Folks [5] gathered many properties of the Inverse Gaussian distribution. Khattree [6] studied about the length biased distribution of Inverse Gaussian distribution and Gamma distribution. Ahmed, Budsaba, Lisawadi, and Volodin [7] proved the new parametrization of the Birnbaum-Saunders distribution and various estimation strategies. Their proposed parameters are important since they fit the physical phenomena of fatigue cracks. The parameters l > 0 and q > 0 correspond to the thickness of the machine element and the nominal treatment pressure on the machine element, respectively. In this paper, we will study the Crack Lifetime distribution in form of their parameters and add the new weighted parameter

p . The engineering example of

Crack random variable is time after a machine element is forced by cyclic or non-cyclic loading until the crack achieves the critical value. After a machine element is forced, a slightly crack may happen but the element still works. When it arrives at the critical point, tolerance exceeds and the element do not properly work. We will first review the probability density function (pdf.) base on the proposed parameterization of the Birnbaum-Saunders distribution, the Inverse Gaussian distribution, the Length Biased Inverse Gaussian distribution, the needed integral formula and the principal square root of a complex number. Then we will show the relevance of four above-mentioned distribution. Next we will prove the characteristic function, the moment generating function including their existing conditions and the cumulative distribution function of three-parameter CR in close form base on the proposed parameters of Ahmed, Budsaba, Lisawadi, and Volodin [7].

2.The supported knowledge 2.1.The Birnbaum-Saunders Distribution Let X have the Birnbaum-Saunders distribution where a > 0 is the shape parameter

b > 0 is the scale parameter and the median


then X has BS a, b distribution. The pdf. of X in form of classical parameters is given by

1 x +b exp 12 2a2 foBS ( x; a, b) = 2a ( 2pb) x3 2 0

b x + - 2 x > 0 , b x , otherwise

variance of X is a 2b2 + 1 1 where mean of X is b + and . 2 4


,1 then relation between X and Z are Let Z : N 0

a2

5a 2

( )

a X =b Z + 2
Z= 1 X a b

2 a Z + 1 2
b X .

The cumulative distribution function (cdf.) of X in form of classical parameters

is given by

1 x FoBS ( x; a, b) = F a b

b . x

The relation between classical parameters a , b and proposed parameters l ,

q in physical interpretation.
l = a= 1 a2 1 l
and

q = a 2b b =l q

and

The pdf. of the X : BS l , q is

fnBS ( x;l , q) 3 1 1 2 2 1 x q q l exp - - l + x = 2q 2p x 2 q 0


2.2.The Inverse Gaussian Distribution

2 q , x > 0 x ,otherwise.

(1)

The pdf. of an Inverse Gaussian distribution random variable X classical parameters is


2 b - 32 ) b( x - m x exp ,x > 0 2 foIG ( x; mb) = 2p , 2mx 0 , otherwise

with the

where m> 0 and b > 0. The parameter m is the mean of the distribution and b is a scale parameter. The relation between classical parameters a , b and proposed parameters l ,

q in physical interpretation.
l = b m
and and

q=

2 m b

m= l q

b = l 2q

The new parametrization of the Inverse Gaussian probability density function with the proposed parameters is

3 l 2 q exp 1 x - l - 2 q fnIG ( x;l , q) = q 2p x 0

2 q , x > 0 x , otherwise .

(2)

2.3.The Length Biased Inverse Gaussian Distribution Definition The length biased pdf. of its original pdf. Let X be a non-negative random variable having an absolutely continuous pdf.

X f ( . Let X have a finite first moment E . Let Y be a non-negative random variable with )
pdf. h where Y is the length biased random variable associated with X . The definition of length biased distribution is given by

()

h ( x) =

xf ( x)

, x > 0 where f ( x) is pdf. of the original distribution. E X

The finite first moment of the Inverse Gaussian distribution is E X = m= l q . Hence, the pdf. of the Length Biased Inverse Gaussian distribution is given by

( )

1 1 2 q exp 1 x - l - 2 q fnLB x;l , q = q 2p x 0

2 q , x > 0 x , otherwise .

(3)

2.4.Definite Integrals of Exponentials of Complicated Arguments and Powers [8]

- 12

exp - r x - q x dx =

p exp - 2 rq r

)
rq

(4)

where Rer > 0, Req 0.

- n- 1 2

exp - r x - q x dx = ( - 1)

p n - 2 e r qn

(5)

where Rer > 0, Req > 0. From (5) when n = 1 ;

- 1 12 -

exp - r x - q x dx = -

p 12 12 exp - 2r q r q

p . exp - 2 r q q

Thus,

- 32

exp - r x - q x dx =

p exp - 2 r q q

(6)

where Rer > 0 ,Req > 0.

2.5.The Principal Square Root of a Complex Number Proposition Given a complex number z = rei w , where r is a non-negative real number and w is a real number: - p < w p , then

Re z is non-negative real number


and

Im z is real number
such that

Im z is

non-negative real number if w is non-negative real number; negative real number if w is negative real number.

3.The Relevance of the pdfs. of the Birnbaum-Saunders distribution, the Inverse Gaussian distribution, the Length Biased Inverse Gaussian distribution, and the Crack distribution.

fnIG x;l , q fnCR x;l , q, p = fnBS x;l , q fnLB x;l , q

( ( (

) ) )

,p = 1 ,p = 1 2 ,p = 0

fnCR ( x;l , q p ) = p , fnIG (x;l , q ) + (1 - p )fnLB x;l , q (

fnBS ( x;l , q) =

1 fnIG ( x;l , q) + fnLB ( x;l , q ) 2

3.1.The Probability Density Function of the Crack Distribution Let a continuous random variable X : CR l , q p , ,

l > 0 correspond to thickness of a machine element q > 0 correspond to nominal treatment pressure of a machine element
(Proving origin of l , q is seen from [7])

0 p 1.
The probability density function of X : CR l , q p ,

is given by

fnCR x;l , q p , 1 = q 2p 0

)
p l 3 1 2 2 q + ( 1- p) q 1 x - l exp - x 2 q x
2 q , x > 0 (7) x , , otherwise

where l > 0, q > 0 and 0 p 1 .

4.The Characteristic Function

, Theorem The characteristic function of X : CR l , q p j


Proof.

is

( t) =

l 1-

1 - p 11- 2q ti

1- 2qti

1- 2 ti . q

( t)

= E itX e

exp( itx) p l q 2p 1

q + ( 1- p) q x x
12

32

12

1 x exp - - l 2 q

2 q dx x

1 q 2p

p l

q + ( 1- p) q x x
- 32

32

1 l 2q dx x exp itx - - 2l + 2q x

el pl q

0 2p l e ( 1- p)

q 2p

2 1 - ti - l q / 2 x dx 2q x l 2q / 2 . 1 - 12 x exp - ti - x dx 2q x

exp

(8)

Let us consider each integral in (8) seperately. From (6)

- 32

exp - r x - q x dx =

p exp - 2 r q q

- 32

exp

2 1 - ti - l q / 2 x dx = 2p exp - l 1- 2q ti 2q x , l 2q

(9)

and from (6) where Rer > 0, Req > 0

, since l > 0 q > 0 ; Rer =

1 l 2q >0 q= ,Re > 0 we obtain t . 2q 2

From (4)

- 12

exp - r x - q x dx =

p exp - 2 rq if Rer > 0, Req 0 r

) (

- 12

exp

2 2qp 1 - ti - l q / 2 x dx = exp - l 1- 2qti 2q x 1- 2q ti

(10)

and from (4) where Rer > 0, Req 0

, since l > 0 q > 0 ; Rer =

1 l 2q >0 q= ,Re > 0 0 we obtain t . 2q 2

Putting (9) and (10) into (8), we get that

( t) =

el pl

exp - l 1- 2q ti l 2q 2p el ( 1- p) 2qp + exp - l 1- 2q ti 1- 2q ti q 2p

2p

l 1-

1 - p 11- 2q ti

1- 2qti

1- 2q . ti

For completing this proof, we will consider the existence of j

( t)

by considering only

1 1- 2q ti
Let

[since e (

l 1-

1- 2q ti

and - p 11

1- 2q are defined for all t ] . ti

1- 2q = a + bi ti

(11) where a,b are real numbers.


2 2

, By squaring both sides of (11), we get a = b + 1 1 ab = - qt and by the principal


, , square root of a complex number, a 0 b . Thus, a 1 b , t .
This implies that

1- 2q = a + bi 0 ti

[Since 1- 2q = a + bi = 0if a = 0andb = 0 ] ti

Therefore,

1 1- 2qti

is defined for all t .

, Now, we can complete this proof, the characteristic function of X : CR l , q p j

is

( t) =

l 1-

1 - p 11- 2q ti

1- 2qti

for all t 1- 2 ti q

which accords to the definition of the characteristic function that t . 5.The Moment Generating Function

, Theorem The moment generating function of X : CR l , q p f X ( t) = e


l 1-

is

1 - p 11- 2q t

1- 2qt

1- 2 t q

and exists if t < Proof.

1 2q .

f X ( t) = E tX e

exp( tx) p l q 2p 1

q + ( 1- p) q x x

32

12

1 x exp - - l 2 q

2 q dx x

1 q 2p

p l

32 12 2 q + ( 1- p) q - 1 x - 2l + l q dx exp tx q x x 2 x

el q

2p
q

p l

q + ( 1- p) q x x
- 32

32

12

exp

1 2q

l 2q / t x x

2 dx

el pl
l

0 2p e ( 1- p)

exp
- 12

1 2q

q 2p

exp

l 2q / 2 t x dx x 2 1 - t - l q / 2 . x dx 2q x

(12)

Let us consider each integral in (12) seperately. From (6)

- 32

exp - r x - q x dx =

p exp - 2 r q q

- 32

exp

2 2p 1 - ti - l q / 2 x dx exp - l 1- 2q ti = 2q x , l 2q

(13)

from (6) where Rer > 0, Req > 0,

1 2 since q > 0, l > 0 then Req = l q 2 > 0 but Rer = - t which Rer > 0 . 2q
We obtain that integral in (14) exists if t <

1 2q .

From (4)

- 12

exp - r x - q x dx =

p exp - 2 rq r

) )
(14)

- 12

exp

1 2q

l 2q / 2 2qp = t x dx exp - l 1- 2qt x 1- 2q t

where Rer > 0, Req 0,


2 since q > 0, l > 0 then Req = l q 2 > 0 0,

but Rer =

1 1 - t which Rer > 0 We get >t . . 2q 2q 1 2q . 1 of the existence of both 2q

We obtain that integral in (14) exists if t <

Putting (13) and (14) into (12) under the condition t < integrals,

10

f X ( t) =

el pl 2p

2p l q
2

exp - l 1 - 2q + t

el ( 1- p) q 2p

2qp exp - l 1 - 2q t 1- 2q t

l 1- 1- 2qt

1 - p 11- 2qt

1- 2q t

We will next find the existing condition of the moment generating function from two points. The condition of the existence of both integrals in (12) is t < f X t will be undefined if the denominator

1 . 2q

()

1- 2qt = 0 and the expression inside

square root 1- 2q < 0 ,hence t < t

1 2q .

, Hence, the moment generating function of X : CR l , q p f X ( t) = e


l 1-

is

1 - p 11- 2q t

1- 2qt

1 1- 2 t and exists if t < q 2q .

It is obvious that it differs from finding the characteristic function. The support of the characteristic function is t while the support of the moment generating function is

t<

1 2q .

6.The Cumulative Distribution Function

, Theorem The cumulative distribution function of X : CR l , q p

is

x F - l FnCR ( x;l , q p) = q , 0
Proof.

q ( 1- 2p) e2l x

- F x + l 1 q

q ,x> 0 x , x 0.

To prove this theorem, we will show that

d FnCR ( x;l , q, p) = fnCR ( x;l , q, p) dx d FnCR ( x;l , q, p) dx


,if x > 0

11

d x F - l dx q

q ( 1- 2p) e2l x

- F x + l 1 q

q x

x = - l f q

q d x - l q x dx q x q d x x +l + ( 1- 2p) e2l + l f x dx q q 1

q x

d F ( x) = f ( x) dx

x +l = 2x 2p q

2 1 x q - l q exp - 2 q x x 2 x x 1- 2p q l - 1 + l q - l + exp 2 x 2 q x 2x 2p q 3 1 2 2 q + ( 1- p) q 1 x - l exp - x 2 q x 2 q x

,f x =

( )

1 2p

x2 2

= p l q 2p 1

It is clear that when ..,

d FnCR ( x;l , q, p) = fnCR ( x;l , q, p) = 0. dx

12

, Hence, the cdf. of X : CR l , q p is


x F - l FnCR ( x;l , q, p) = q 0
7.Conclusion This paper is a starting point of study about the new three-parameter Lifetime distribution [the Crack distribution:CR] which relates to three known twoparameter distributions. We have proved to find the closed forms of the characteristic function, moment generating function including their existing conditions and the cumulative distribution function of CR base on non-classical parametrization. The

q ( 1 - 2p) e2l x

- F x +l 1 q

q ,x > 0 x ,x 0.

existing condition t <

1 of the moment generating function depends on parameter q . 2q

There are still many topics of CR which have not been discevered e.g. the statistical inference, the survival analysis. It will be more contributions if one can join the statistical knowledge to the engineering knowledge. Acknowledgement This research is partially supported by Center of Excellence in Mathematics, the Commission on Higher Education, Thailand. References [1] Birnbaum, Z. W., & Saunders, S. C. A new family of life distributions. J. Applied Probability, 1969; 6: 319-327. [2] Birnbaum, Z. W., & Saunders, S. C. Estimation for a family of life distributions with applications to fatigue. J. Applied Probability, 1969; 6: 328-347. [3] Desmond, A. F. On the relationship between two fatigue-life models. IEEE Reliability, 1986; 35(2): 167-169. [4] Shuster, J. On the inverse Gaussian distribution function. J.American Statistical Association, 1968; 63: 1514-1516. [5] Chhikara, R. S., & Folks, J. L. The inverse Gaussian distribution-theory, methodology, and applications, New York: Marcel Dekker, 2008. [6] Khattree, R. Characterization of inverse-Gaussian and gamma distributions through their length-biased distributions. IEEE Reliability, 1989; 38(5): 610-611. [7] Ahmed S. E., Budsaba K., Lisawadi S., & Volodin, A. I. Parametric estimation for the Birnbaum-Saunders lifetime distribution based on a new parameterization. J. Thailand Statistician, 2008; 6: 213-240.

13

[8] Gradshteyn, I. S., & Ryzhik, I. M. Table of integrals, series, and products (7th ed.). MA: Elsevier Academic Press, 2007.

You might also like