4.
Probability Distribution Models
RE 30003: Introduction to Quality Engineering
Instructor: Dr. Indranil Hazra, Asst. Prof. @ School of Quality and Reliability, IIT Kharagpur i i i i i ii i i i i i ii i i i i i ii i i i i i
sample of n items without replacement from a lot of N
1. Discrete Distribution Models items of which D are nonconforming or defective.
Binomial Distribution
Poisson Distribution
Consider a process that consists of a sequence of n inde-
Consider the Poisson random variable X be the number
pendent trials. When the outcome of each trial is either
of events occurring in a fixed interval of time or space,
a success or a failure, the trials are called Bernoulli trials.
given a constant average rate of occurrence λ > 0.
» If the probability of success on any trial is p , then
» The Poisson PMF can be written as
the number of successes X in n Bernoulli trials has
e −λλx
the binomial distribution with parameters n ≥ 0 P (X = x) = , x = 0, 1, 2, · · ·
x!
and 0 < p < 1. Fig: Lognormal PDFs with θ = 0 and selected values of ω.
» The mean and variance of the distribution are
» The binomial PMF can be written as Exponential Distribution
µ = λ, σ2 = λ
P (X = x) = nC x p x (1 − p)n−x , x = 0, 1, 2, · · · , n The exponential PDF is defined by a single parameter
» The mean and variance of the distribution are Poisson distribution λ > 0 – the rate parameter.
is useful for model-
µ = np, σ2 = np(1 − p)
ing the number of » The exponential PDF can be defined as
Binomial distribution is used to estimate the probabil-
defects on a per unit f (x) = λe −λx , x ≥ 0
(volume, time, etc.)
ity of a given number of defective items in a batch, where » The exponential CDF can be written as
p represents the fraction of defective or nonconforming
basis, i.e., over a fixed ∫x
items in the population, and X represents the number of
period of time or F (x) = λe −λx d x = 1 − e −λx
space, in a process. 0
Fig: Poisson PMFs for different λ values.
nonconforming items in a random sample of n items. » The mean and variance of the distribution are
2. Continuous Distribution Models µ = 1/λ, σ2 = 1/λ2
Normal Distribution Gamma Distribution
The normal distribution is the most important distribu-
The gamma distribution is defined by the shape param-
tion in statistical quality control. It is defined by the loca-
eter r > 0 and the scale parameter λ > 0.
tion parameter (mean) µ and the scale parameter (stan-
Fig: Binomial PMFs for different n and p values. dard deviation) σ > 0. » The gamma PDF can be defined as
Negative Binomial Distribution λ(λx)r −1e −λx
» If X is a normal random variable, then the PDF of X f (x) = , x ≥0
Consider a sequence of independent trials, each with Γ(r )
can be defined as ∫∞ r −1 −t
probability of success p , and let X denote the trial on 1 x−µ 2 where Γ(r ) = 0 t e d t is the gamma function.
− 12 ( σ )
which the r th success occurs. Then, X represents a neg- f (x) = p e , −∞ < x < ∞
σ 2π » The gamma CDF can be written as
ative binomial random variable. ∫x
» The normal CDF
∫
is defined as ( ) λ(λx)r −1e −λx 1
1x
x −µ F (x) = dx = I (λx, r )
» The negative binomial PMF can be written as F (x) = p e
x−µ 2
− 21 ( σ )
dx = Φ 0 Γ(r ) Γ(r )
x−1 r x−r −∞ σ 2π σ where I (λx, r ) is the incomplete gamma function.
P (X = x) = C r −1 p (1 − p) , x = r, r + 1, · · ·
» The mean and variance of the distribution are
» The mean and variance of the distribution are
r 2 r (1 − p) µ = r /λ, σ2 = r /λ2
µ= , σ =
p p2
Negative Binomial distribution models the number of
trials needed to achieve a specified number of successes.
It is used to analyze failure rates and assess process sta- Fig: Normal PDF with µ and σ parameters.
bility over repeated tests.
A normal random variable with mean µ = 0 and stan-
Geometric Distribution dard deviation σ = 1 is known as the standard normal Fig: Exponential PDFs with selected Fig: Gamma PDFs with λ = 1 and
A special case of the negative binomial distribution is if variate, commonly denoted as Z . values of λ. selected values of r .
r = 1, in which case we have the geometric distribution. Note: If r = 1, the gamma distribution reduces to the
» The standard normal PDF can be written as exponential distribution with parameter λ!
» The geometric PMF can be written as 1 − z2
f (z) = p e 2 , −∞ < z < ∞ Weibull Distribution
P (X = x) = p(1 − p)x−1, x = 1, 2, · · · 2π
The Weibull distribution is defined by the scale parame-
» The standard normal
∫ CDF can be written as
» The mean and variance of the distribution are z
1 − z2 ter θ > 0 and the shape parameter β > 0.
1 2 (1 − p) F (z) = p e 2 d z = Φ(z)
µ= , σ = −∞ 2π
p p2 » The Weibull PDF can be defined as
( )β−1
Note: If {X i }ni=1 are normally and independently dis- β x −(x/θ)β
Geometric distribution is useful in determining the f (x) = e , x ≥0
tributed random variables with parameters {µi }ni=1 and θ θ
probability of first occurrence of a defect on the n th trial. ∑n
{σi }i =1, then Y = i =1 a i X i will also follow normal distri-
n » The Weibull CDF can be written as
It helps estimate the number of trials required to en- ∑n ∑n ∫x ( )β−1
bution with µ y = i =1 ai µi and σ y = i =1 ai2σ2i .
2 β x −(x/θ)β −(x/θ)β
counter the first defect in a production line, making it F (x) = e dx = 1−e
0 θ θ
valuable for analyzing rare events in sequential testing. Lognormal Distribution
» The mean and variance of the distribution are
Hypergeometric Distribution Variables in a system sometimes follow an exponential ( )
relationship X = e Y . If the exponent Y is a random vari- µ = θΓ 1 + 1/β
Suppose that there is a finite population consisting of N [ ( ) { ( )} ]
2 2 2
able, then X = e Y is also a random variable. A special case σ = θ Γ 1 + 2/β − Γ 1 + 1/β
items. D (D ≤ N ) of these items fall into a class of inter-
est. A random sample of n items is selected from N with- occurs when Y follows the normal distribution. In that
out replacement, where X number of items fall into the case, the distribution of X is called the lognormal distri-
class D . Then X is a hypergeometric random variable. bution defined by the location parameter θ and the scale
parameter ω.
» The hypergeometric probability mass function
(PMF) can be written as » The lognormal PDF can be written as
N −D 1 − 12 ( ln x−θ
2
ω ) , 0< x <∞
D
Cx C n−x f (x) = p e
P (X = x) = NC
, x = 0, 1, · · · , min(n, D) xω 2π
n
» The lognormal CDF can be written as
» The mean and variance of the(distribution
)( are
) ∫x ( )
nD nD D N −n 1 − 21 ( ln x−θ )
2 ln x − θ
µ= , 2
σ = 1− F (x) = p e ω dx = Φ
N N n N −1 0 xω 2π ω
Fig: Weibull PDFs with θ = 1 and selected values of β.
» The mean and variance of the distribution are
In quality sampling, hypergeometric distribution is the θ+ω2/2 2 2θ+ω2 ω2 Note: If β = 1 is fixed, the Weibull distribution turns into
µ=e , σ =e (e − 1)
appropriate probability model for selecting a random an exponential distribution with λ = 1/θ .