Risk Analysis, Vol. 26, No.

1, 2006

DOI: 10.1111/j.1539-6924.2006.00705.x

Risk-Based Integrity and Inspection Modeling (RBIIM)

of Process Components/System
Faisal I. Khan,1 Mahmoud M. Haddara,1 and Subrata K. Bhattacharya1,2

Process plants deal with hazardous (highly flammable and toxic) chemicals at extreme conditions of temperature and pressure. Proper inspection and maintenance of these facilities
is paramount for the maintenance of safe and continuous operation. This article proposes a
risk-based methodology for integrity and inspection modeling (RBIIM) to ensure safe and
fault-free operation of the facility. This methodology uses a gamma distribution to model the
material degradation and a Bayesian updating method to improve the distribution based on
actual inspection results. The method deals with the two cases of perfect and imperfect inspections. The measurement error resulting from imperfect inspections is modeled as a zero-mean,
normally distributed random process. The risk is calculated using the probability of failure and
the consequence is assessed in terms of cost as a function of time. The risk function is used
to determine an optimal inspection and replacement interval. The calculated inspection and
replacement interval is subsequently used in the design of an integrity inspection plan. Two
case studies are presented: the maintenance of an autoclave and the maintenance of a pipeline
segment. For the autoclave, the interval between two successive inspections is found to be
19 years. For the pipeline, the next inspection is due after 5 years from now. Measurements
taken at inspections are used in estimating a new degradation rate that can then be used to
update the failure distribution function.
KEY WORDS: Failure modeling; integrity assessment and evaluation; quantitative risk assessment; riskbased inspection; risk-based maintenance

1. INTRODUCTION

nancial investment, and the environment against

the consequences of failures occurring to these systems, an assessment of the condition of existing infrastructure (major process units, pipelines, etc.) is
necessary. The condition assessment quantifies the
degradation of the material and provides a basis
for the decision-making process regarding preventive
maintenance and/or replacement. With limited maintenance resources, it is essential that the available
funds be spent where they are most effective in reducing potential risks (Concord, 1993; Pandey, 1998).
The traditional methods of assessing the condition of process systems are based on a deterministic
load-resistance methodology in which the integrity
of the components is evaluated by comparing the

Maintaining the integrity of a large number of

aging process components is a subject of prime importance to process companies all over the world. In
Canada, there are more than 250,000 km of pipelines
carrying natural gas, crude oil, and petroleum products (Pandey, 1998). To protect the public, the fiFaculty of Engineering & Applied Science, Memorial University
of Newfoundland, St Johns, NL, A1B 3X5, Canada.
2 Department of Ocean Engineering, Indian Institute of Technology, Chennai, 600 036, India.
Address correspondence to Faisal I. Khan, Faculty of Engineering
& Applied Science, Memorial University of Newfoundland, St
Johns, NL, A1B 3X5, Canada; fkhan@engr.mun.ca.
1

203

C 2006 Society for Risk Analysis

0272-4332/06/0100-0203$22.00/1 

204
current operating conditions with a design-limit state
beyond which the component cannot operate safely.
Load-resistance methods have the disadvantage that
they often yield somewhat conservative results, leading to potentially unnecessary repairs and inspections
that result in an overall increase in maintenance costs.
Use of deterministic methods does not provide information about potential risk that results in the
unrealistic maintenance planning for process plants
(Desjardins, 2002).
Risk is defined as multiplication of the probability of failure and its likely consequence. Risk-based
methods aim at identifying, characterizing, quantifying, and evaluating the likelihood of the loss caused
as a result of the occurrence of a specific event. The
use of risk-based methods for the management of the
process components provides reliable quantification
of potential risks. This provides an alternative strategy for the maintenance of assets instead of the use of
simple ranking (prioritizing) based on reported failure occurrences.
This approach also provides a means for quantitatively establishing future reliability levels for the
components. These levels can be used as a basis for
optimizing reinspection intervals. The uncertainties
associated with the design and operation of process
components have led to an increasing use of riskbased approaches in making decisions regarding asset
integrity management.
In risk-based inspection (RBI) strategy risk is
used as a criterion to prioritize inspection tasks for the
components in a process plant. This provides many
advantages, which include (1) an increase in plant
availability, (2) a decrease in the number of failure
occurrences, (3) a reduction in the level of risk due
to failure, and (4) a reduction in the direct inspection cost of the plant. Operational safety of a process plant increases as a result of decreasing the number of failure occurrences and the potential risk due
to failure. This makes RBI a useful strategy to meet
the rising societal expectations regarding operational
safety of complex onshore and offshore process facilities. These expectations resulted in a number of
regulations that the oil and gas industry has to meet.
The last two decades saw a number of studies that attempted to address the subject. Different methodologies were suggested to meet maintenance planning requirements. These methodologies range from the fully
qualitative to the fully quantitative. A brief summary
of these approaches is presented in the subsequent
section.

Khan, Haddara, and Bhattacharya

2. AVAILABLE RISK-BASED APPROACHES
Since the late 1980s, numerous quantitative, semiquantitative, and qualitative models have been developed to aid plant engineers with the prioritization of
components inspections. In the following sections, we
outline briefly these methods.
2.1. Qualitative Approaches
Qualitative risk index approaches assign subjective scores to the different factors that are thought to
influence the probability and consequences of failure
(Muhlbauer, 1992; Cagno et al., 2000; Dey & Gupta,
2001). These scores are then combined using simple
formulas to give an index representing the level of
risk. Risk index approaches provide a ranking of the
different process components based on the perceived
level of risk estimated. The ranking obtained by using these methods is highly subjective. In addition,
these approaches do not provide any indication of
whether the risk associated with a component is unacceptable and consequently no guidance is provided
regarding whether any risk reduction action is necessary. The index system scoring format suggested
by Muhlbauer (1992) accounts for the use of in-line
inspection tools to locate metal loss corrosion by
awarding up to 8 points out of a maximum of
400 points representing resistance to failure (i.e., 2%).
This underestimates the benefits of high-resolution
pigging, which is known to result in significant reductions to the large percentage of failures that are
attributable to corrosion (2040% of all failures).
Therefore, index systems provide at best an approximate risk-based ranking of process components,
which has serious limitations when being used as a
basis for integrity management decision making.
2.2. Quantitative Approaches
These approaches determine the level of risk
based on direct estimates of the probability and/or
consequences of failure. Current quantitative risk assessment approaches focus on a single aspect of the
consequence associated with failure. Published studies deal with either loss of life risk or economic risk
(Hill, 1992; Concord, 1993; Nessim & Stephens, 1995;
Pandey, 1998; Nessim et al., 2000).
Integration of environmental damage, life safety,
and economic risks has not been addressed adequately. Another limitation of quantitative risk

Risk-Based Integrity and Inspection Modeling

assessment approaches is that they typically base the
failure probability estimates on historical failure rates.
Publicly available databases do not usually allow subdivision of the failure data according to the attributes
of a specific process component and where adequate
subdivision is possible, the amount of data associated with a particular attribute set is very limited because of the rarity of the failures. Failure probabilities
estimated from public data are, therefore, not sufficiently specific to represent a given failure in a specific
process component.
Another approach to calculate failure probability is based on the concept of structural reliability
that includes Markovian models and hot spots on a
component-by-component basis (a number of interesting articles may be seen in Journal of Infrastructure Systems). The effect of the correlation from one
hot spot to another has been investigated by Lotsberg
et al. (1998), Faber and Sorensen (1999), Brown and
May (2000), and Montgomery and Serratella (2002).
Faber et al. (2000, 2003) proposed an informal decision
analysis where the number of considered elements is
reduced in a consistent and systematic way. However,
due to the numerical effort required and the stability of the method such an approach did not prove to
be practical. Straub and Faber (2000) indicated that
an integrated approach to the decision problem that
is suitable for industrial purposes has not yet been
developed.
2.3. Semi-Quantitative Approaches
Semi-quantitative approaches were developed to
provide a practical and easy tool to be used for designing maintenance programs that optimize the use
of resources and in the meantime ensure effective and
efficient asset management. These approaches use
semi-quantitative models for consequence estimation
as well as failure probability calculations. Examples of
these approaches can be found in Khan and Haddara
(2003a, 2003b), Khan and Haddara (2004), and Khan
et al. (2004). It is easily employed in process plants and
to components like pipelines or pressure components.
These approaches provide a tool to ascertain that the
estimated risk of failure satisfies a predetermined acceptance criterion (Khan et al., 2004; Willcocks & Bai,
2000; Dey, 2004).
The risk-based inspection and maintenance approach discussed by Willcocks and Bai (2003) uses
a failure modes effect analysis to identify the failure
modes of system components and their consequences.
It then uses failure patterns and rates to calculate the

205
probability of failure, and it determines the risk to
be used in inspection and maintenance planning. Depending on the level of risk for each mode and pattern
of failure, the required analysis, inspection, maintenance, and repair tasks are selected. For example, a
review of historical failure databases indicates that
the major failure modes in a pipeline are internal corrosion and external impact. Thus, the main efforts (in
terms of design, structural modeling, inspections, etc.)
should be focused on these failure modes. Of course,
this is a simple example of risk-based inspection and
maintenance. In practice, more specific details about
the specific pipeline need to be considered.
Guidelines to help hydrocarbon and process
chemical industries in establishing risk-based inspection programs on fixed equipment and piping were
issued by the American Petroleum Institute (API,
2002). This document describes a recommended
practice for developing risk-based maintenance programs. The document does not preclude the use of any
of the three methodologies mentioned above, but outlines the limitations of each methodology. This document explains the basic elements for developing and
implementing a risk-based inspection program.
2.4. Maintenance Optimization
A large number of articles were published on the
subject of optimizing maintenance through the use
of mathematical models (Montgomery & Serratella,
2000; Khan & Haddara, 2003a, 2003b; Willcocks &
Bai, 2003; Dey, 2004). Most maintenance optimization models are based on lifetime distributions or
Markovian deterioration models. It is often difficult to
collect enough data for estimating the parameters of
a lifetime distribution or the transition probabilities
of a Markov chain. This presents an obstacle in the
way of using these models to design practical maintenance programs. The combined use of the reliability
index methods and the limit state approach may prove
helpful in removing this drawback.
Typically, structural reliability methods are used
to estimate the probability of failure of components
or structures. In these methods, the condition of the
material is described by a state function g. The state
function represents the strength margin that the material has over the applied load. Thus, the limit-state
function, gt is defined as
gt = R L,

(1)

where R is the strength of the material and L is the

applied stress.

206

Khan, Haddara, and Bhattacharya

As degradation of material occurs and operating conditions change, material strength and applied
stress become random functions of time. Thus, the
limit-state function becomes a stochastic process. A
failure model can be developed based on the limitstate function approach and allowing for material
degradation variability, operating conditions variability, and measurement errors. Data collected during inspections provide a means to update the failure model
through the use of Bayess theorem.
Risk-based maintenance takes into consideration
the consequence of failure as well as the probability
of failure. In this article, the consequence is measured
in terms of the cost of failure in addition to the cost
of the inspection/maintenance program.
3. QUANTITATIVE RISK-BASED INTEGRITY
AND INSPECTION MODELING (RBIIM)
RBIIM aims at modeling inspection tasks to
achieve safe operating conditions at minimum cost.
Determining the relationship between inspections
periodicity on one side and the level of safety and
maintenance costs on the other, provides an insight
into the application of the as low as reasonable practicable (ALARP) principle. In order to implement
an integrated RBIIM approach, it is necessary first
to formulate acceptance criteria for each function
of the plant facility. Thereafter, the relationship between each function and the performance of the
various components responsible for achieving this
function is established. To simplify the implementation, the acceptance criteria for each function of
the facility are transformed into acceptance criteria
for the individual components. Subsequently, risk assessment is performed relating the event of component failure to the consequences in terms of monetary
losses. The main objectives of RBIIM are:
1. To identify critical equipment in a process
plant. These are pieces of equipment that have
a high level of risk.
2. To develop an optimum inspection and maintenance strategy for the equipment that will
guarantee the integrity of the plant.
Fig. 1 depicts the overall framework designed to
achieve these objectives. The framework comprises
five stages. These are: identification of equipment
to be analyzed, identification of degradation mechanisms for each component, calculation of the risk
associated with the failure of each component, deter-

mination of an optimal inspection policy (interval) for

each component, and development of a comprehensive policy for the plant integrity assessment.
3.1. Material Degradation Mechanisms
Material degradation is a common cause of process components failure. Material degradation can be
caused by one or more mechanisms. These include:
internal and external thinning due to corrosion; stress
corrosion cracking; brittle fracture; and fatigue due to
vibration. These mechanisms cause material deterioration and thus affect the ability of the component to
withstand the applied load.
Two basic models will be used in this work to
describe material degradation mechanisms: a thinning model and a crack model. The thinning model
is used to describe the reduction in the thickness of
the components material as a result of internal or
external corrosion. The crack model is used to describe the reduction of the load-carrying capacity of
the component as a result of cracks. Cracks may result
from stress corrosion, brittle fracture, or fatigue (see
Kallen, 2002).
3.1.1. Thinning Model
Corrosion wear failure is a complex failure mode
in which a component is affected by a combination of
corrosion and wear. Continuous wear results in material thinning, which may be localized or can spread
over a large surface area. In the former, material
degradation is seen in the form of pitting, while in the
latter, a general thickness loss occurs. Carbon steel or
copper are usually more susceptible to general thinning, while stainless steels and higher alloy materials are usually more prone to localized thinning and
pitting.
The state function in this case, gt , is defined as
 


PD
C t
gt = R St = S 1

, (2)
d
2d
where R is material resistance, St is the applied stress,
S is material strength, C is corrosion rate, P is operating pressure, D is the diameter of the component, d is
the material thickness, and t is a time increment.
Corrosion can occur on the internal or external
surface of the component. The same thinning model
can be used for both cases; however, the corrosion
rate may differ from one case to the other.
The state function gt , is a measure of the ability
of the material to resist failure. Failure will occur only
if the magnitude of the function gt reaches zero.

Risk-Based Integrity and Inspection Modeling

207

Divide the system into independent

(process) components, e.g., vessel, pipeline
segments, reactors, etc.

Select one component at a time and

identify possible degradation mechanisms

Model component
failure using a gamma
stochastic process

Develop a prior for each

degradation mechanism

Fig. 1. Framework of risk-based integrity

and inspection modeling methodology.

Obtain past
inspection
results for the
unit under
investigation

Estimate the
consequence using cost
data for inspection,
failure, and replacement

Determine failure
probability using a
posterior and past
inspection results

Risk calculations

Estimate optimal inspection and

replacement intervals

Have all
components been
considered?

Develop inspection plan for the

integrity of the process system

Equation (2) can be used to define a margin of

safety in terms of material thickness. Assuming a linear degradation model, the incremental wall thickness
loss (x) during an element of time t, is given by

x = Ct = 1

PD
d.
2dS

(3)

The margin of safety in terms of the minimum required material thickness, mt , is then obtained as

mt = d

PD
.
2S

(4)

3.1.2. Stress Corrosion Cracking Model

Stress corrosion cracking (SCC) failure occurs
when the applied stresses on a component in a corrosive environment generate a field of localized surface
cracks, usually along grain boundaries, which renders
the component incapable of performing its function.

208

Khan, Haddara, and Bhattacharya

There are many material/environmental combinations within which SCC can occur. They include caustic cracking, amine cracking, carbonate cracking, sulfide stress cracking, hydrogen-induced cracking, and
polythionic acid cracking, and chloride cracking for
stainless steels (Kallen, 2002).
The SCC state function uses a resistance minus
stress model based on Pariss crack growth law used
in linear elastic fracture mechanics,



PD
gc = KIC Y
+S
A,
(5)
2d
where KIC is the material fracture toughness, Y is a
dimensionless geometric factor, S is residual stress,
and A is the crack depth.
The crack depth, A, is determined by
lcr = Ccr t n ,
lcr
A=
,
Rl/a

(6)

where lcr , Ccr , and Rl/a are the characteristic crack

length, the crack growth rate, and the crack to lengthto-depth (thickness) ratio, respectively. The value
of KIC , material fracture toughness, is taken to be
300 ksi(in)1/2 (10,425 MPa (mm)1/2 ) for stainless steel.
The material fracture toughness for carbon and low
alloy steel may be calculated using the following formula (Kallen, 2002):
KIC


= Minimum 33.2 + 2.806e(0.02(T+100)) , 200 ksi in.
Equation (6) can be used to derive an expression for
the incremental increase in crack depth as

2
x = Ccr t =

Rl/a
KIC
.

PD

Y
+S
2d

(7)

The safety margin for wall thickness, mc , becomes

mc = d Ccr t.

below the transition temperature and as a result are

susceptible to brittle fracture. Furthermore, vibration
may cause components to fail prematurely. Improperly supported piping near vibration sources is prone
to fatigue. In contrast to the other degradation mechanisms discussed above, this mechanism is modeled
using a simple indexing methodology (Kallen, 2002).

3.2. Material Degradation Model

In the previous section, we reviewed the different mechanisms that control material degradation.
The net degradation of the material is the total sum
of degradations that take place as a result of one or
more of the above-mentioned mechanisms. Although
the above-mentioned mechanisms are deterministic,
there is a level of uncertainty associated with some of
their variables. For instance, the strength S, the corrosion rate C, and the operating pressure, P in the
thinning model, and KIC , Y, P, S, and A in the stress
corrosion cracking mechanism, have uncertainty associated with their values. Therefore, these variables
have to be considered random and the material degradation process is expected to be a stochastic process.
Let us define the accumulated material degradation measured at a certain point in time t, to be the
sum of the incremental degradations that occurred
from the start of service till time t. Let us also assume that the incremental material degradations are
independent, exponentially distributed random variables. Then, the cumulative degradation from the start
of service till time t, X(t), is a gamma-distributed
stochastic process with stationary increments. The
gamma-distributed stochastic process has many advantages: it has a nonnegative distribution and it is
easy to manipulate.
A gamma density function with a shape parameter , and a scale parameter , where and > 0, is
given by

(8)

3.1.3. Brittle Fracture and Fatigue

At low temperatures carbon steels may suffer
a brittle fracture at loads significantly below the intended design loads. This is because carbon steels lose
their ductility as the temperature is lowered. The temperature at which ductility is lost is called the transient
temperature. While most materials are designed to
operate well above the transition temperature, some
materials in older plants may still operate near or

f (x) =

(x)1 ex
()

for

x > 0.

(9)

For the thinning model, the cumulative degradation is

assumed to be linear in time. Thus, the shape parameter, , will be replaced by a linear function of time
= o t.
The probability density function describing the
cumulative wall thickness degradation, X(t), is described by Equation (10):

Risk-Based Integrity and Inspection Modeling

f X(t) (x) =

o t
(x)ot1 ex
(ot)

for

x > 0.

209

(10)

This distribution will be used to describe cumulative

material degradation caused by each individual degradation mechanism described above.
Thus, the mean and variance of the cumulative
degradation in the wall thickness, X(t), are given by
0
X(t) = E[X(t)] = t = t,

o
2
(11)
X(t) = Var[X(t)] = 2 t 2 = 2 t.

The parameters of the gamma process, X(t), are calculated as

 2

o =
and = 2 .
(12)

It is common practice to express the standard deviation ( ) in terms of the mean () through the use of
a coefficient of variance (): = . Using this relationship, the gamma density function for corrosion
can be rewritten as
t

1 2
t
2
x
  (x) 2 1 e 2 .
f X(t) (x) =
(13)
t

2
In this expression the average corrosion rate () is the
only uncertain parameter that would vary with time.
Considering this, the above expression can be represented in terms of single-variable Bayesian gamma
conditional function G as


1
1
G x  2 t,

2

t
1 2
t
2
x
  (x) 2 1 e 2 = f X(t) (x).
=
(14)
t

2
For stress corrosion cracking, the cumulative crack
depth is assumed to be a function of tn . Thus, in this
case = ot n and the density function becomes
f X(t) (x) =

o t
n
 (x)ot 1 ex

 o t n

for

x > 0.

(15)

The mean and variance for the cumulative wall thickness degradation become
o
X(t) = E[X(t)] = t n = t n ,

o
2
(16)
X(t) = Var[X(t)] = 2 t n = 2 t n .

The parameters of the gamma process, X(t), are calculated as

 2

and = 2 .
(17)
o =

Substituting for the parameters of the density function, one gets

tn
1 2
tn
2
x
 n  (x) 2 1 e 2
f X(t) (x) =
t

2


1 n 1

.
= Ga x  2 t ,

2


(18)

Similarly, the above equation can be represented in

terms of a single variable conditional gamma function
G as


1
1
G x  2 t n ,

2
tn
1 2
tn
2
x
 n  (x) 2 1 e 2 .
= f X(t) (x) =
t

2


(19)

These gamma density functions are subsequently used

in the inspection updating for the cases of perfect and
imperfect inspections.
3.3. Inspection Updating Modeling
The results of inspection can be effectively used in
updating the prior knowledge of the average degradation rate (). This updating could be best done using
Bayesian updating. Bayess theorem provides a formal and structured approach that can be used to update the prior based on component data gained from
actual inspections. Inspection updating modeling involves two main steps: (1) selection of an appropriate
prior, and (2) Bayesian updating of the prior using
new inspection data. Bayesian updating can be applied for the two cases: perfect or imperfect inspection
data. Bayesian updating is widely discussed in the literature; excellent reviews are given by Kallen (2002)
and Kallen and Noortwijk (2003).
3.3.1. Step ISelection of a Prior
Bayess theorem indicates that a posterior probability can be obtained from an assumed prior probability when modified based on new knowledge

210

Khan, Haddara, and Bhattacharya

3.3.2. Step II(i)Posterior: Perfect
Inspection Updating

(information). A number of probability distributions

have the special property that if used in Bayess theorem as a prior distribution and the new information is used in the form of a random variable with
a certain distribution, the posterior distribution has
the same distribution as the prior. These pairs of
distributions are called conjugate distributions, examples of which are normal-normal-normal, normallognormal-normal, and gamma-exponential-gamma
distributions (see Harr, 1987).
Following the suggestions made by van Noortwijk
and van Gelder (1996), Kallen (2002), and Kallen and
van Noortwijk (2003), the inverted gamma distribution is used here as a prior distribution. The inverted
gamma density is given as
IX(t) (x) =

ba
(a)


 a+1

b
1
exp
x
x

for

The conditional prior density function for the decrease in the thickness of the material, X(t), can be
expressed as
( | x) = 

l(x | ) ()

(21)

l(x | ) () d

where () is the prior density for , and l(x | )

is the likelihood of a measurement x given .
Kallen (2002) used Equation (21) to obtain a
posterior density function for the loss in the material thickness for the perfect inspection case, ( | x),
given by
 

 t
x

( | x) = IX(t)  2 + , 2 + ,
(22)

x 0.
(20)

where and are the parameters of the prior density.

Consider the case when the inspection is repeated
n times, thus, n measurements x1 at time t1 , x2 at
t2 , . . . , xn at tn are carried out. The posterior for the
loss in the material thickness can be written as

An inverted gamma density fits very well as a prior

density function for the material degradation caused
by corrosion and stress corrosion cracking. The density is nonnegative and has a longer tail, which represents uncertainty over higher degradation rates. Fig. 2
depicts the simple discrete prior density and also the
continuous version of this density function. The inverted gamma density function has been fitted to
the discrete prior given in Fig. 2, through the use of
simulation.

( | x1 , . . . , xn )
 

n
n


ti ti1
xi xi1

= IX(t) 
+ ,
+ .
 i=1
2
2
i=1
(23)

80

70

60

Density (%age)

50

Discrete prior (based on expert opinion)

Inverted gama prior

Fig. 2. Discrete prior based on available

information and inverted gamma prior
fitted to these data.

40

30

20

10

0
0.00

0.06

0.11

0.17

0.22

0.28

0.34

0.39

Degradation rate (mm/year)

0.45

0.50

0.56

0.60

0.70

Risk-Based Integrity and Inspection Modeling

211

If one considers that xo = 0 at to = 0, then the posterior

actually depends only on the last inspection, because
the standard deviation ( ) is expressed in terms of the
mean (). Therefore, the above expression reduces to

l(y | ) () d





1
1
f
G y  2 t n ,
()
d
() d.

2
=0



( | x1 , . . . , xn )

 

n
n


tn
xn

= ( | xn ) = IX(t) 
+ ,
+ .
 i=1 2
2
i=1
(24)

Perfect inspections do not usually happen in real life;

it is thus important to model imperfect inspections as
well.
3.3.3. Step II(ii)Posterior: Imperfect
Inspection Model
This model is based on the approach discussed
by Newby and Dagg (2002), and Kallen and van
Noortwijk (2003). Let us consider a stochastic process Y(t) given by
Y(t) = X(t) + ,

(25)

where is the error in the measurement. The error, ,

is considered to be a zero mean normal process with
a standard deviation . Y(t) is the degradation measured (material lost) during an inspection at time t.
A brief description of the posterior density formulation and solution algorithm is presented here;
for more details see, Kallen (2002) and Kallen and
van Noortwijk (2003). The likelihood of a measurement y given the degradation rate is
l(y | ) = fY(t) (y)
= f X(t)+ (x) =

f X(t) (y ) f () d,
(26)

where f X (t) (y ) is the likelihood of the gamma increment, X(t), having parameters and . To solve
the above integral, we replace the integral in Equation (26) by a summation
n




fY(t) (y)
G y i  ot n , p(i );
i=1

where

p(i ) = 1.

(27)

Equation (27) was developed for a nonlinear material

degradation model: a linear model can be obtained by
substituting n = 1.
Using the parameters of the gamma density function, the denominator in Equation (21) becomes

(28)
This integral can also be approximated by the following summation:

l(y | ) () d
0


n
k


i=1

j=1




1 n 1
p( j ) p(i ),
G y  2 t ,

2


(29)
where p(i ) = Pr{ = i } is the discrete density of
. Using these results, the posterior is approximated
as
p(i | y)





k

1 n 1

p(i )
p( j )
G y  2t ,

2
j=1

;
= n



k


1 n 1
p( j )
p(i )
G y  2 t ,

2
i=1
j=1
for i = 1, . . . , n.

(30)
In case of multiple imperfect inspections, we need the
product of the individual likelihoods for the measured
increments

l(y1 , . . . , yk | ) =
lY(tk)Y(tk1 ) (yk yk1 | ).
k

(31)

Using the convolution integral as before, the above

likelihood can be rewritten as
l(y1 , . . . , yk | )

 
=
...
f X(tk)X(tk1 ) (yk yk1 k)

f (1 , . . . , k) d1 , . . . , dk,

(32)

where k = k k1 . Because the s are not independent, the likelihood can be approximated as



l(y1 , . . . , yk | ) = E
f Dk(dk k)
k

N

1 
j
f Dk dk k
N j=1

as

N ,

(33)

where Dk = X(t k ) X(t k1 ) and dk = yk yk1 . In

the last step we used the strong law of large numbers

212

Khan, Haddara, and Bhattacharya

to approximate the expectation of a function through

the use of simulation (see Kallen, 2002).
For each k, we sample k (j = 1, . . . , N) and calcuj
j
j
late k = k k1 . Because the gamma distribution
f Dk = G(x | [t k t k1 ], ) is not defined for x < 0,
j
we need to make sure that the condition (dk k )
0 is satisfied.
l(y1 , . . . , yk | )



N 
 tk tk1 1
 j
1 


.
Ga dk min k , dk 
,
N j=1 k
2
2
(34)
The posterior for multiple imperfect inspections can
now be obtained as
p(i | y)




N 

  t 1
1 

j
Ga dk min k , dk  2 ,
p(i )
 2
N j=1 k


,
= n
N 

  t 1

1 

j
p(i )
Ga dk min k , dk  2 ,
 2
N j=1 k
i=1

(35)
for i = 1, . . . , n and t k = t k t k1 .
We now have a Bayesian updating model that accepts the results of multiple imperfect inspections. The
advantage of using the simulation technique to calculate the posterior density as a function of the degradation rate is that one can use different levels for the
magnitude of the error in each inspection.
3.4. Risk-Based Optimal Inspection
Interval Calculation
Determining an optimal integrity inspection
and/or maintenance action is a problem of optimization under uncertainty. The most comprehensive approach for solving such a problem is through the use
of decision theory, which provides a systematic and
consistent way to evaluate the alternatives and identify an optimal choice (see Nessim & Stephens, 1995;
Nessim et al., 2000; Goyet et al., 2002).
In the previous steps, we have developed a model
to calculate the probability of failure of a process component as a function of time. In order to calculate the
risk, we have to calculate the consequences associated
with that failure. In the present study, we estimate the
consequence in terms of the cost incurred as a result
of failure. Having estimated the risk, we will use it
to decide when to inspect and when to replace the
component. In other words, we intend to determine
the maximum length of time between two consecutive
inspections that would result in a minimum accepted

risk. We will adopt an optimization criterion that has

been suggested by Wagner (1975). This criterion is referred to as the average cost over unbound region.
This criterion is derived using renewal theory and uses
the concept of a components lifecycle. The length of
this cycle spans the time from the start of the service
until a renewal is affected. Renewal consists of either
a preventive or a corrective replacement. The risk factor is calculated as the ratio of the expected costs per
cycle over the expected cycle length. A brief formulation of this criterion is presented below (for details,
see van Noortwijk, 2000, 2003).
We will model the maintenance tasks as a discrete renewal process. In this case, the component
returns to the as-good-as-new condition after each
replacement. A discrete renewal process {N(n), n =
1, 2, 3, . . . } is a nonnegative integer-valued stochastic process that registers the successive renewals in
the time interval (0, n). Let the renewal times T 1 , T 2 ,
T 3 , . . . , be nonnegative, independent, identically distributed, random quantities having a discrete probability function
Pr {Tk = i} = pi ,
and

where i = 1, 2, . . . ,

pi = 1,

(36)

i=1

where pi represents the probability of a renewal at

time i. The cost (consequence) associated with a renewal at time i is denoted by ci , i = 1, 2, 3, . . . . The
risk factors (expected average costs per unit time)
are determined by simply averaging the risk factor
over an unbounded horizon. The expected risk over a
bounded horizon (0, n), denoted by R(, ), is given
by
n



R(, ) =
pi ci + E( (n i)) ,
i=1

for

n = 1, 2, 3, . . . ,

and

(0) 0.

(37)

The cost associated with the occurrence of the event

T i = i is ci plus the additional expected cost during the
interval (i, n), i = 1, . . . , n. Using the discrete renewal
theorem, the risk factor (expected average costs per
unit time) is expressed as
E(cycle risk)
E(cycle length)


ci (, ) pi (, )
E(Ci )
i=1
=
=
,


E(I)
i pi (, )

R(, ) =

i=1

(38)

Risk-Based Integrity and Inspection Modeling

213

where ci and pi are the cost incurred during a unit

of time i and the probability of renewal during this
time, respectively. {m, } represents the replacement level, which is the ratio of the material degradation to the critical safety margin. It is calculated for
each time interval . The random variable I is the cycle length with a cycle cost of cI . Considering replacement cycle as the time period between two renewals,
the numerator of Equation (38) represents expected
cycle risk and the denominator is the expected cycle
length (mean lifetime). Equation (38) can be rewritten as
R(, )


n




j=1


n



j=1

4.1. An Autoclave
Geary (2002) presented a review of risk-based
inspection approaches practiced in the United Kingdom. Four different case studies were performed by
seven different agencies. In the present article, we
will apply the methodology outlined above to the
case study no. 3 of Gearys report: an autoclave. We
will determine the optimal inspection interval for this
component. A brief description of the autoclave is
presented here and the input data used in the study is
given in Table I.
4.1.1. Process Details

ci (, ) pi (, , | P, S . . .) p ( j |
y)

i=1

4. CASE STUDIES

i pi (, , | P, S . . .) p ( j |
y)

i=1

(39)
where ( | y) is the posterior density, (p(j | y)) is the
discrete version of the posterior density over a given
number of the measurements, y = y1 , y2 , . . . , yk .
In order to include the uncertainty over the various material and operating characteristics such as
pressure (P) and material strength (S), we can sample these variables from their respective probability
distribution functions. It is better to include the uncertainty over the degradation rate in the simulation,
instead of using a discrete solution. The advantage of
using a simulation technique is that the degradation
rate, , can be sampled before we go into the loops to
calculate the expected average risk. Also, there will be
no extra loop needed if we assume that other parameters such as the pressure, P, and material strength,
S, are also uncertain. The solution in this case is given
by
E|R(, )|



N




1 
j
j
j
ci (, ) pi , | , P , S . . .
N j=1 i=1


=
,
n




I 
j
j
j
i pi , | , P , S . . .
N j=1 i=1

(40)
where E|R(, )| denotes expected average risk and
is measured as dollar per unit time.

The autoclave reactor is filled with water at 75 C

and pressurized to 10 Barg. The pressure is subsequently released and water is then agitated and a vacuum is applied to evacuate air from the empty space.
Subsequently, an ammonium sulfate solution (as a
catalyst) is introduced along with tetrafluoroethylene
(TFE) (in a gaseous state). TFE is immediately polymerized by the water. The temperature of the exothermic reaction is controlled by cooling the water jacket
(circulating water at 15 C). The process operates in
batches, 3 cycles per day for 48 weeks in a year.
The vessel has been inspected three times and the
results of the inspections are given in Table I. This unit
has been studied using the methodology outlined in
the previous sections and results are briefly discussed
below.
5. RESULTS AND DISCUSSIONS
The prior and posterior density functions for thinning and cracking in the autoclaves material are
shown in Figs. 3 and 4, respectively. It may be observed
from Fig. 3 that the posterior supports the prior. The
posterior for perfect inspections shows a tall peak at
a corrosion rate of 0.185 mm/year, which is comparatively higher than the prior peak. Peaks for the posterior (imperfect inspections) and the prior occur at
almost the same location (0.11 mm/year). This indicates that the results of the inspection (with possible
errors in the measurements) confirm the suitability of
the originally assumed thinning rate.
In case of crack growth, the peak of the posterior (perfect inspections) function occurs at a degradation rate of 0.2 mm/year, which is much higher than
that for the prior given as 0.012 mm/year and the
posterior for the imperfect inspections occurring at

214

Khan, Haddara, and Bhattacharya

Background information

Autoclave dimensions
Material of construction
Active damage mechanism
Inspection history

Other Parameters
Tensile strength
Operating pressure
Corrosion rate
Crack growth rate
Corrosion allowance
n (power function for crack
growth rate)
Crack length to depth ratio
Inspection cost
Preventive replacement cost
Failure cost

Vessel manufactured in 1980

Design temperature 100 C and pressure 25 Barg
Operating temperature 895 C and pressure 10 Barg
5 ft diameter; 11 diameter, 1 in. wall thickness
Stainless steel 321, carbon steel jacket
Stress corrosion cracking
Internal thinning
Three inspections
Material loss (thinning): mean 0.7 mm 0.9 mm 1.5mm,
CoV 0.50
Crack growth: mean: 0.06 mm 0.3 mm 0.6 mm, CoV 0.20
Values
Mean 515 MPa, coefficient of variance (CoV) 0.20
Mean 3.2 MPa, CoV 0.05
Mean 0.1 mm/year; CoV 0.25
Mean 0.01 mm/year; CoV 0.25
4.5 mm
0.50

Table I. Input Details for Autoclave

6
$15,000
$30,000
$3,000,000

a rate of 0.04 mm/year. This indicates that the posterior function for the perfect inspections represents an
idealized condition and does not represent the actual
degradation mechanism in this case. Furthermore, the
prior and posterior functions for the imperfect inspections are fairly close together, which indicates
that they give a better representation of the actual
situation.

Calculations for the life cost were made using the

imperfect inspections assumption and considering all
possible variations in the operating parameters. The
results are shown in Figs. 5 and 6 for the thinning and
cracking mechanisms, respectively. It can be seen from
Fig. 5 that the replacement time is 43 years and that the
failure time is about 212 years. There are two minima
observed in the plot. The first is at 43 years (which

30

25
prior
postprior
post_imp_prior

20

Density

post_imp_prior_CDF

Fig. 3. Prior and posterior density

distribution for corrosion degradation
mechanism.

15

10

0
0

0.1

0.2

0.3

0.4

0.5

Corrosion rate (mm/year)

0.6

0.7

0.8

Risk-Based Integrity and Inspection Modeling

215
1.2

70

60

50
0.8
40
Density

prior
postprior
post_imp_prior
post_imp_prior_CDF

Fig. 4. Prior and posterior density

distribution for crack growth.

30

0.6

0.4
20

Cummulative density function

0.2

10

0
0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0
0.45

Crack growth rate (mm/year)

whereas the failure time is calculated to be 554 years.

This plot has also two minima: one at 85 years and the
other at 535 years. The second minimum is impractical. Now, as the autoclave has already survived for
24 years, the next inspection should be done within
the next 50 years, assuming SCC to be the only degradation mechanism. These results were obtained by
considering the effects of only one degradation mechanism at a time. This is partially responsible for the

coincides with the replacement interval length) and

the second is at 203 years. The second minimum is
a mathematical result that has little meaning in real
life. The first minimum, which occurs at 43 years, is
the one of interest. Considering that the component
has already survived for 24 years, the next inspection
should be within 19 years.
It is evident from Fig. 6 (crack growth) that preventive replacement time is estimated at 91 years

18000

16000

Fig. 5. Average inspection cost/year for

corrosion degradation mechanism.

Average cycle cost ($/year)

14000

12000

10000

Average cycle cost

Replacement time
Failure time

8000

6000

4000

2000

0
0

50

100

150

Inspection interval (years)

200

250

216

Khan, Haddara, and Bhattacharya

6000

Average cycle cost ($/year)

5000

4000

Average cycle cost

Replacement time
Failure time

3000

Fig. 6. Average cycle cost ($/year) for

cracking degradation mechanism.

2000

1000

0
0

100

200

300

400

500

600

Inspection interval (years)

rather long inspection interval. The second reason for

the long inspection interval is the fact that we have
used a relatively slow crack growth rate.
Considering both degradation mechanisms to be
acting at the same time, it is evident that the autoclave
should be inspected at a maximum of 19 years from
now. The analysis suggested here can be used repeatedly to modify the inspection interval based on newer
inspection data.
6. A PIPELINE
Pandey (1998) studied the probabilistic models
for the assessment of the condition of a pipeline using in-line inspection data. He used a segment of
an oil and gas pipeline of 914 mm diameter and
8.74 mm wall thickness to illustrate his approach. We
will consider the same case considered by Pandey
(1998) to illustrate the approach developed in the
present work. Input data used in this example are
given in Table II. Pandey (1998) showed that the
time of inspection should be between 5 and 20 years
depending upon the repair criteria (considering corrosion to be the only degradation mechanism). He also
concluded that a single inspection after 10 years with
a repair criterion of 1.4 would minimize the pipeline
failure probability to 0.01 per km over 25 years.
7. RESULTS AND DISCUSSIONS
This is a comparatively new segment of the oil
and gas pipeline and had only one regular inspection.

As may be seen from Fig. 7 (for thinning), the peaks

of the prior and the posterior functions are very much
separated from each other. The prior is more spread as
compared to the posterior (perfect and imperfect inspections). The posterior distribution for imperfect inspections shows the least spread. The results for crack
growth rate are shown in Fig. 8. In this case the peaks
of the posterior functions occur at degradation rates
of 0.2 and 0.195 mm/year for imperfect and prefect
inspections, respectively. These peaks are away from
the prior peak, which occured at a degradation rate
of 0.12 mm/year.
A detailed cost analysis is conducted for both
degradation mechanisms based on the posterior for
imperfect inspections. The results are shown in Figs. 9
and 10. When corrosion is the dominant degradation
mechanism, it was found that the preventive replacement time is 14 years and the failure time is 26 years
(see Fig. 9). The results considering SCC to be the
dominant mechanism are 23 years for the replacement
and 62 years for failure (see Fig. 10). It is evident that
the thinning degradation mechanism is fast and thus
must be used as the basis in inspection planning. The
cost curve has two minima, one at 10 years and the
other at 22 years. The first minimum occurring at an
interval of 10 years (though not the global minimum)
is a more conservative basis for inspection. Since the
pipeline has already been in operation for 5 years,
the next inspection is due in 5 years. The measurements obtained from the next inspection can then be
used to update the distribution and to predict a better

Risk-Based Integrity and Inspection Modeling

217

Background information

5 years of operation
Design temperature 120 C and pressure 7.0 MPa
Operating temperature 40 C and pressure 5.7 Barg
914 mm diameter; 1 km length, 8.74 mm wall thickness
Steel grade 60
Stress corrosion cracking
Internal thinning
One inspection, defect of depth (mean) 0.5 mm, CoV 0.20

Pipe dimensions
Material of construction
Active damage mechanism
Inspection history
Table II. Input Details for Segment of
Oil and Gas Pipeline

Other Parameters
Yield strength
Operating pressure
Corrosion rate
SCC rate
Corrosion allowance
n (power function for crack
growth rate)
Crack length-to-depth ratio
Inspection cost
Preventive replacement cost
Failure cost

estimate for the following inspection time. If data obtained from the next inspection support the current
posterior then the following inspection will coincide
with the preventive replacement of this pipeline. The
optimum inspection intervals are calculated, based

Values
Mean 461 MPa, coefficient of variance (CoV) 0.20
Mean 5.7 MPa, CoV 0.05
Mean 0.15 mm/year; CoV 0.25
Mean 0.1 mm/year; CoV 0.25
2.5 mm
0.50
6
$5,000
$10,000
$100,000

on SCC degradation mechanism, as 20 and 55 years.

Because corrosion effects are more dominant in the
case of a pipeline, the SSC-based inspection interval
is not of importance. However, in the next inspection
(5 years from now) inspection for crack growth should

50
45
40
prior
postprior
post_imp_prior
post_imp_prior_CDF

35

Fig. 7. Prior and posterior distribution

for corrosion degradation mechanism.

Density

30
25
20
15
10
5
0
0

0.05

0.1

0.15

0.2

0.25

Corrosion rate (mm/year)

0.3

0.35

0.4

0.45

218

Khan, Haddara, and Bhattacharya

14

1.2

12

0.8
8
Density

prior
postprior
post_imp_prior
post_imp_prior_CDF

0.6

0.4

Cumulative density function

10

Fig. 8. Prior and posterior density

distribution for cracking growth in
pipeline.

0.2

0
0

0.5

1.5

2.5

Crack growth rate (mm/year)

also be conducted to verify the posterior and subsequent analysis.

8. CONCLUSIONS
In this work, we have presented a method for the
determination of optimal maintenance intervals for
process components. The method takes into account
the random nature of material degradation of pro-

cess components. The material degradation rate can

be linear or nonlinear. The method allows the updating of the probability density function for the material degradation, using a Bayesian approach. Thus,
the failure rate can be adjusted based on actual measurements made during inspections. The updated failure rate obtained reflects the changes in operating
and environmental conditions. The inspection intervals obtained are more reliable because they are
based on actual measurements. The method takes into

7000

6000

Average cycle cost ($/year)

5000

4000

Average cycle cost

Replacement time
Failure time

Fig. 9. Average inspection (cost/year) for

corrosion degradation on pipeline.

3000

2000

1000

0
0

10

15
Inspection interval (years)

20

25

30

Risk-Based Integrity and Inspection Modeling

219

1800

1600

Fig. 10. Average inspection cost ($/year)

for crack degradation mechanism in
pipeline.

Average cost per cycle ($/year)

1400

1200
Average cycle cost
Replacement time
Repair time

1000

800

600

400

200

0
0

10

20

30

40

50

60

70

80

90

Inspection interval (years)

consideration the random error that may be associated with the measurements obtained during
inspection.
The method optimizes maintenance intervals
based on the risk associated with component failure.
The optimization criterion is based on a level of risk
that satisfies the acceptable risk criteria. Using risk
as an optimization criterion allows the maintenance
interval to be a function of both the probability of failure as well as the consequences of this failure. Proper
attention can then be devoted to the maintenance of
critical equipment.
The gamma distribution seems to describe the
material degradation process well. Results obtained
using the assumption that measurements obtained
during inspections are imperfect are more realistic
than those assuming perfect measurements.
The method has been applied to two case studies:
an autoclave and a pipeline. The results obtained for
these two case studies are in general agreement with
results discussed in the literature. Results of the case
studies presented above show that the method produces reliable estimates for the inspection intervals.
The optimal inspection intervals given by the model
for the case studies considered are reasonable and in
general agreement with results obtained using different approaches.
The main disadvantage of the method is that
it is computationally intensive. For the autoclave
case study, the simulation using 200 samples (coded
in MATLAB) took about 8 hours, whereas for the

pipeline case study, it took about 5 hours of computation time on a personal computer with an Intel Pentium 4 (clocks at 1.81GHz) and 256 MB RAM.
However, the method can be easily programmed
and the data required can be easily obtained from
operational records and original design documents.
ACKNOWLEDGMENTS
The authors gratefully acknowledge the financial support of Natural Science and Engineering Research Council (NSERC) under a discovery grant and
Canada Foundation of Innovation (CFI) new opportunity award.
NOMENCLATURE
A Crack depth, mm
D Diameter of the component (vessel,
pipe, etc.), mm
C Corrosion rate, mm/year
Ccr Crack growth rate, mm/year
cmax Corrosion allowance, mm
ci (, k) Costs incurred during unit time (i
1, i) as a function of replacement
percentage and inspectional interval, $
C(, k) Expected average costs per unit
time as function of replacement
percentage and inspection interval,
$/year

220

Khan, Haddara, and Bhattacharya

E(X) Expectation of the random variable
X
f X (X) Probability density function of the
random variable X
G State function, dimensionless
Ga(x | , ) Gamma density of X as a function
of x with the shape parameter and
the scale parameter
Ig(x | , ) Inverted gamma density of X as a
function of x with the shape parameter and the scale parameter
KIC Material
fracture
toughness,
MPa(mm)1/2
lcr Characteristic crack length, mm
L( | x) Likelihood of the variable given
the measurement x
m Safety margin, the amount of degradation at which the component is assumed to fail, dimensionless
N Crack growth exponent; for example, PWHT components n = 0.25
and for non-PWHT components n
= 0.50.
P Operating pressure, kPa
P( | y) Conditional probability of failure
due to wall loss y, given degradation
rate
Rl/a Crack length-to-depth (thickness)
ratio; it is considered as 6
R Material resistance, MPa
R(, ) Risk factor, dollar per unit time
E|R(, )| Expected average risk factor, dollar
per unit
S Material strength, MPa
St Applied stress, MPa
thickness Thickness of the material of construction, mm
X(t) Random variable in stochastic process
Y(t) Variable in stochastic process consisting of variable X(t) and an error,
X(t) +
Y e Geometric factor, dimensionless
GREEK SYMBOLS
Shape parameter for gamma and inverted gamma densities
Scale parameter for gamma and inverted gamma densities
E Normally distribution error with mean
0 and standard deviation

k Difference of error between two successive inspections, k and k1

Mean of the uncertain gamma density
(random variable X)
Standard deviation of the random variable X
Coefficient of variance
Replacement stage, percentage of
safety margin at which the component
should be replaced
{m, } Ratio of replacement to failure level
and calculated for each time interval
Time interval
 (a) Gamma function
(a, x) Incomplete gamma function
() Prior density of the variable
tk Time interval between two inspections
REFERENCES
American Petroleum Institute (API). (2002). Risk Based Inspection: API Recommended Practice 580. Washington, DC: American Petroleum Institute.
Brown, S. J., & May, I. L. (2000). Risk-based hazardous release
prevention by inspection & maintenance. Journal of Pressure
Vessel Technology, 122(8), 362367.
Cagno, E., Cron, F., Mancini, M., & Ruggeri, F. (2000). Using AHP
in determining the prior distributions on gas pipeline failures in
a robust Bayesian approach. Reliability Engineering & System
Safety, 67, 275284.
Concord Environmental Corporation. (1993). Risk Assessment
Techniques for Pipeline Systems. Report for Pipeline environmental Committee of Canadian Association of Petroleum
Producers. Calgary, Alberta: Concord Environmental
Corporation.
Desjardins, G. (2002). Improved data quality opens way for predicting corrosion growth and severity. Pipeline & Gas Journal,
December, 2832.
Dey, P. K. (2004). Decision support for inspection and maintenance:
A case study of oil pipelines. IEEE Transaction of Engineering
Management, 51(1), 4756.
Dey, P. K., & Gupta, S. S. (2001). Risk based model aids selection
of pipeline inspection, maintenance strategies. Oil and Gas
Journal, July 9, 5460.
Faber, M. H., Engelund, S., Sorensen, J. D., & Bloch, A. (2000).
Simplified and generic risk based inspection planning. In Proceeding OMAE 2000. New Orleans, LA: OMAE.
Faber, M. H., & Sorensen, J. D. (1999). Aspect of inspection
planningQuality and quantity. In Proceeding of ICASP 1998.
Sydney, Australia: ICASP.
Faber, M. H., Straub, D., & Goyet, J. (2003). Unified approach to
risk based inspection planning for offshore production facilities. Journal of OMAE, 125, 126131.
Geary, W. (2002). Risk Based Inspection: A Case Study Evaluation
of Offshore Process Plant. Report HSL/2002/20. Sheffield, UK:
Health and Safety Laboratory.
Goyet, J., Straub, D., & Faber, M. H. (2002). Risk based inspection
planning methodology and application to an offshore structure.
Revue Francasie de Genie Civil (English version), 6(3), 489
503.
Harr, M. E. (1987). Reliability Based Design in Civil Engineering.
New York: Dover Publications.

Risk-Based Integrity and Inspection Modeling

Hill, R. T. (1992). Pipeline risk analysis. Institute of Chemical
Engineers Symposium Series, 130, 637670.
Kallen, M. J. (2002). Risk based inspection in the process and
refining industry. Masters thesis, Faculty of Information
Technology and Science, Technical University of Delft, Delft,
The Netherlands.
Kallen, M. J., & van Noortwijk, J. M. (2003). Optimal maintenance
decisions under imperfect inspection. In Proceedings of the
European Safety and Reliability Conference. Maastricht, The
Netherlands.
Khan, F. I., & Haddara, M. (2003a). RBM: A new approach for
process plant inspection and maintenance. Presented at Proceedings of AIChEs Loss Prevention Conference, April 13,
New Orleans, LA.
Khan, F. I., & Haddara, M. (2003b). Risk-based maintenance
(RBM): A quantitative approach for maintenance/inspection
scheduling and planning. Journal of Loss Prevention in Process
Industries, 16, 561573.
Khan, F. I., & Haddara, M. (2004). Risk-based maintenance of
ethylene oxide production facilities. Journal of Hazardous
Materials, A108, 147159.
Khan, F. I., Sadiq, R., & Haddara, M. (2004). Risk-based inspection
and maintenance (RBIM): Multi-attribute decision-making
with aggregative risk analysis. Transaction of IChemE Process
Safety and Environmental Protection, 86 (B2), 398411.
Lotsberg, I., Sigurdsson, G., & Wold, P. T. (1998). Probabilistic inspection planning of the Asgaard A FPSO hull structure with
respect to fatigue. Presented at Proceeding of OMAE 1999,
Newfoundland, Canada.
Montgomery, R. L., & Serratella, C. (2002). Risk-based maintenance: A new vision for asst integrity management. Pressure
Vessel and Piping, 444, 151165.

221
Muhlbauer, W. K. (1992). Pipeline Risk Management Manual.
Houston, TX: Gulf Publishing Company.
Nessim, M. A., & Stephens, M. J. (1995). Risk based optimization
of pipeline integrity maintenance. OMAE, V, 303314.
Nessim, M. A., Stephens, M. J., & Zimmerman, T. J. E. (2000). Risk
based maintenance planning for offshore pipelines. Proceedings of the Annual Offshore Technology Conference, 2, 791
800.
Newby, M., & Dagg, R. (2002). Optimal inspection policies in the
presence of covariates. Presented at Proceedings of ESREL
2002, Lyon, France.
Pandey, M. D. (1998). Probabilistic models for condition assessment of oil and gas pipelines. NDT&E International, 31(5),
349358.
Straub, D., & Faber, M. H. (2000). Generic risk based inspection
planning for components subject to corrosion. Presented at
Proceeding of ESRA Workshop on Risk Based Inspection Planning, Zurich, Switzerland.
van Noortwijk, J. M. (2000). Optimal maintenance decisions on
the basis of uncertain failure probabilities. Journal of Quality
Maintenance Engineering, 6(2), 113122.
van Noortwijk, J. M. (2003). Explicit formulas for the variance of
discounted life-cycle cost. Reliability Engineering and System
Safety, 80(2), 185195.
van Noortwijk, J. M., & van Gelder, P. H. A. J. M. (1996). Optimal
maintenance decisions for berm breakwaters. Structural Safety,
18(4), 293309.
Wagner, H. M. (1975). Principle of Operations Research, 2nd ed.
Englewood Cliff, NJ: Prentice-Hall.
Willcocks, J., & Bai, Y. (2000). Risk based inspection and integrity
management of pipeline systems. International Society of
Offshore and PolarEngineers, II, 285294.