You are on page 1of 14

Computers and Chemical Engineering 152 (2021) 107362

Contents lists available at ScienceDirect

Computers and Chemical Engineering


journal homepage: www.elsevier.com/locate/compchemeng

Optimum Maintenance Interval Determination for Field Instrument


Devices in Oil and Gas Industries Based on Expected Utility Theory
Reza Abbasinejad a, Farzad Hourfar b,∗, Ali Elkamel b,∗
a
Technical Inspection Department, Sarkhoon and Qeshm Gas Treating Company, Bandarabbas, Iran
b
Department of Chemical Engineering, University of Waterloo, Waterloo, Ontario, Canada

a r t i c l e i n f o a b s t r a c t

Article history: With the increasing demand for complicated control and monitoring systems in chemical process plants
Received 14 December 2020 based on tremendous improvements in the digital world, the number of devices for monitoring and con-
Revised 10 April 2021
trolling the plants has risen dramatically. In such huge networks of a myriad of instrument devices, de-
Accepted 4 May 2021
termination of an optimum proactive maintenance interval (PMI) has become vital for enhancing plant
Available online 27 May 2021
safety and reducing maintenance costs. PMI optimization is a matter of considerable benefit variation,
Keywords: process reliability, personnel safety, and environmental concerns. In this paper, a comprehensive formula
Expected utility theory and a concise formula are developed and applied for obtaining the optimum PMI based on the expected
Game theory utility theory. The proposed formulas are applicable for determining the most affordable and effective
Proactive maintenance maintenance plan providing the lowest cost and acceptable reliability, availability, and safety in oil and
Optimum proactive maintenance interval gas industries in terms of an optimal decision-making problem. By considering variable failure rates and
Instrument devices
components’ wear and tear, a comprehensive formula has been developed. Moreover, for constant failure
Oil and gas industries
rate and returning to the “as-new” condition after each proactive maintenance, a concise and “easy-to-
use” formula has been obtained. Finally, it has been shown through practical implementation of the pre-
sented methodology in a typical gas refinery that about 41% benefit is achieved in one year, compared
with the previous heuristic maintenance strategy in the same duration.
© 2021 Elsevier Ltd. All rights reserved.

1. Introduction file to maximize the net present value obtained from a horizon-
tal well before gas break-through. Moreover, (Gu and Hoo 2015)
In recent years, due to the increased demand for the safe and introduced a control strategy to closed-loop model-based control
efficient operation of oil and gas plants, while complying with en- of the hydraulic fracturing process in oil and gas deep under-
vironmental regulations, the concept of process control has gained ground resources. Furthermore, (Siddhamshetty et al., 2018) devel-
momentum. Moreover, improvements in and facilitating the use of oped a framework for closed-loop controlling of a hydraulic frac-
process monitoring and control systems due to the recent emer- turing process in conventional oil reservoirs. In process control sys-
gence of new technologies in the digital world of information tech- tems, control and monitoring activities are performed based on the
nology (IT) have made great contributions to this demand. The process variables. These variables include the pressure and flow
concepts of “process control” in oil and gas industries includes rate of fluids, the temperature of flame or materials, the liquid
many applications in both upstream and downstream domains. level in tanks, and other quantitative items which are measured
For instance, (Narasingam and Kwon 2017) used dynamic mode by sensors, transferred on the industrial networks by transmitters,
decomposition for model predictive controlling of highly nonlin- and actuate the final elements (Abbasinejad et al. 2020). The men-
ear hydraulic fracturing processes in oil reservoirs. (Carlsen, Ny- tioned devices used for measuring, transmitting, and manipulating
gaard, and Nikolaou 2013) presented an assessment of various con- the variables related to process control and monitoring purposes
trol strategies in unexpected gas influx occurrence during drilling. are known as field instrument devices. A substantial part of instru-
(Siddhamshetty and Kwon 2018) designed a model-based feed- ment device costs corresponds to their maintenance management.
back control system for computing the optimal oil production pro- Huge amounts of instrument devices require a large number of ac-
tivities and logistics and, as a result, a substantial budget is neces-
sary for their maintenance, repair, and loss. Improper maintenance

Corresponding Authors. activities will damage the device, which may even lead to device
E-mail addresses: f.hourfar@ut.ac.ir (F. Hourfar), aelkamel@uwaterloo.ca (A. Elka- replacement. Instrument devices are generally expensive and in-
mel).

https://doi.org/10.1016/j.compchemeng.2021.107362
0098-1354/© 2021 Elsevier Ltd. All rights reserved.
R. Abbasinejad, F. Hourfar and A. Elkamel Computers and Chemical Engineering 152 (2021) 107362

stalling new devices sooner than the expected lifespan would be a vanced maintenance programs such as predictive maintenance and
noticeable loss for the owner. In addition, a failure in instrument condition-based maintenance (Salahshoor et al. 2014). In the fol-
devices can lead to production interruption, plant shutdown, safety lowing paragraphs, some relevant achievements in the field of
hazards, or environmental contaminations (Lundgren, Skoogh, and optimum PMI determination are presented. In (Basker and Hus-
Bokrantz 2018). However, it is obvious that unnecessary mainte- band 1978), a method has been introduced for optimal inspec-
nance imposes immense and unjustified expenses on plant own- tion frequency determination, which calculates the optimal num-
ers. In general, maintenance strategies are divided into two main ber of inspection in a known duration based on the Jardin for-
groups: 1) reactive approaches and, 2) proactive approaches. Reac- mula for profit per time unit expression (Jardine 1973) and queue
tive maintenance as an unplanned activity is done after failure oc- theory. The inspection methodology, introduced in that paper con-
currence. But, proactive maintenance is planned such that proper sists of minor repairs as a subset of preventive maintenance strat-
activities are done before probable failure. Proactive maintenance egy. In (Basker and Husband 1978) failure consequence loss was
can be applied in the form of preventive maintenance or predictive introduced properly, but human and environmental consequences
maintenance. Preventive maintenance is planned either in fixed were not considered. The other shortage of that research is that
time intervals, or based on the in-service duration of the equip- the authors just used constant failure rate and did not consider a
ment (Exner et al. 2017). Maintenance interval is the time dura- more appropriate failure probability function such as Weibull prob-
tion between successive maintenance actions (Percy and Kobbacy ability density function. Moreover, (Silver and Fiechter 1992) pro-
20 0 0). On the other hand, predictive maintenance is dependent on posed a method for preventive maintenance interval determina-
the actual condition of the equipment. (Exner et al. 2017). By op- tion, based on two possible lifetimes and a dynamic programming
timizing the proactive maintenance interval (PMI), minimum ex- formulation. Also, (Hariga 1996) developed a model for the single-
penses for maximum exploitation of instrument devices in the best machine inspection problem based on preventive maintenance ap-
condition will be obtained. proach, and maximized the expected profit per unit of time by op-
In recent decades, many improvements have occurred in timal and heuristic solutions. Although in (Hariga 1996) Weibull
maintenance planning. Early methods were in the form of re- model has been used, just a numerical solution -not an analyti-
active maintenance. Interruption in the production is probable cal one- for finding the optimum maintenance interval has been
when applying this maintenance policy. Nowadays, by switch- introduced. (Duarte, Craveiro, and Trigo 2006) have utilized Pois-
ing to upgraded maintenance programs such as applying proac- son failure distribution and queue theory to find optimum preven-
tive policy (preventive or predictive maintenance), the risk of tive maintenance intervals in the series components system. In (D.
undesired shutdown has been reduced, especially in strategic Q. Nguyen, Brammer, and Bagajewicz 2008), a new methodology
processes such as oil, gas, petrochemical, and power plants for assessing the economic value of the planning rules for pre-
(Moore 2004). ventive maintenance based on the use of a Monte Carlo simula-
Generally, refineries and petrochemical industries deal with a tion was presented. Moreover, (D. Nguyen and Bagajewicz 2010)
large number of flammable, explosive, and poisonous materials, used genetic algorithm (GA) to obtain an economically optimal
which are in the vicinity of high temperature and pressure. It is preventive maintenance frequency for different equipment. Both of
obvious that, in such processes, a small failure in a device can have the mentioned papers did not provide closed-form solutions. In
unimaginable and catastrophic consequences. Therefore, providing (Rezaei 2017), based on the Poisson process model, a preventive
a safe environment with high reliability in these industries is an maintenance technique has been developed to find minimal and
urgent obligation. Evidently, achieving a high level of safety and perfect repair types. In (Pashnina 2018), a method for finding cali-
reliability depends on the establishment of suitable PMIs for field bration intervals for custody transfer measuring devices, based on
devices. measuring errors, has been developed. In addition, (Ye et al. 2019)
Activities during the past two decades in maintenance fields used mixed-integer nonlinear programming framework for the op-
provide evidence for the effectiveness of applying new mainte- timal selection of inspection and maintenance frequency. This re-
nance strategies. Improvements in maintenance planning are a re- search has introduced a selection procedure, not a closed-form so-
sponse to different industrial projects’ complexity, size, and di- lution. In (Sharifi and Taghipour 2020), an optimal inspection in-
versity. Meanwhile, experts’ growing knowledge about the sub- terval for K out of N systems with non-identical components was
stantial effects of optimum maintenance intervals on the environ- specified, based on considering possible system states and cal-
ment, personnel safety, and product quantity and quality can also culating expenses for each interval. The proposed approach con-
help the application of modern maintenance policies (Fitouhi and sists of an algorithm for minimizing total inspection cost. Recently,
Nourelfath 2012), (Van Horenbeek et al. 2014), (Mo, Bil, and Sinha (Gordon et al. 2020) combined process safety, machine learning,
2015). and mathematical optimization for integrated maintenance plan-
Unexpected failures usually have many detrimental impacts on ning and process optimization. Based on the above explanations,
the benefit, environment, and even personnel health and safety. in some of the studied papers Weibull failure probability function,
Studies by (Khan and Abbasi 1998) and (Kumar 1998) illustrate the which enhances the quality of the obtained results, has not been
close relationship between the irregular maintenance and the oc- applied. Some others, are not able to result in closed-form solu-
currence of sudden failures. In addition, plant profitability has a di- tions or explicit formulas for calculating the optimum PMI.
rect and close relationship with devices’ reliability and availability. In this paper, a comprehensive solution for computing the op-
Moreover, product quality highly depends on the health of the de- timum PMI is developed. Using EUT along with the Weibull fail-
vice. The main goal in an effective PMI determination is to design ure probability distribution function, an exact closed-form solu-
and implement a comprehensive program which improves devices’ tion for determining the optimal PMI has been proposed. Under
availability and performance and controls the failure mechanism certain conditions, it has been demonstrated that the exponen-
rate, while meets safety and environmental standards and reduces tial distribution (a simplified form of Weibull function) can pro-
production expenses. Obviously, the mentioned objectives are ac- vide the optimal PMI more quickly. One of the main advantages of
cessible by employing a proper approach which copes with de- the proposed formulas besides being straightforward, is their need
vices’ failures and determines optimum maintenance and inspec- for few resources such as time, money, and manpower to yield
tion intervals. applicable results. PMI calculation, either by using recorded plant
PMI management approaches differ according to plant and data, data recorded in other similar plants, or manufacturer data,
process conditions and can vary from periodic visits to ad- is performed through the introduced procedure. A case study on

2
R. Abbasinejad, F. Hourfar and A. Elkamel Computers and Chemical Engineering 152 (2021) 107362

four types of instrument devices in a typical gas refinery located From mean time to failure (MTTF), λ can be calculated:
in the south of Iran has been considered. By studying each de-
1 1
vice’s Weibull parameters, maintenance expenses, and failure con- λ= = . (5)
sequences, the optimized PMIs (time interval in which the device
MT T F η
should be serviced and inspected) have been achieved.
2.1.2. Proactive maintenance cost
The proactive maintenance cost consists of two parts: produc-
2. Optimum proactive maintenance interval (PMI) tion loss due to maintenance time production interruption, and de-
determination vice maintenance activity cost. The proactive maintenance cost is
as Eq. 6:
2.1. Definition of concepts
k = MT T M · P P T · P P PU + MAC, (6)
2.1.1. Failure probability function where MTTM is the mean time to maintenance, PPT is production
The failure probability distribution for instrument devices can per time, PPPU is production price per unit, and MAC is mainte-
be calculated using statistical methods over enough data collected nance activity cost. The maintenance cost for any device in any
during plant operation. If a proper distribution is not achieved, the country and region can be easily found on the price list published
results will not be reliable. The reliability of the generated distribu- by plan and budget authorities in that region.
tions depends on factors such as data size, which should be large
enough for making correct judgments based on it. In other words,
2.1.3. Failure consequence loss
the determination of each device probability function parameters
By failure scenario creation and failure consequence assess-
should be based on a relatively large statistical population.
ment, failure consequence loss will be obtained. The failure con-
One of the most frequently applied distributions for device fail-
sequence loss has four parts: production loss due to failure, repair
ure probability modeling is the Weibull distribution. A major char-
activity cost, human dangers’ penalty, and environment dangers’
acteristic of the Weibull distribution is its relatively great flexibil-
penalty. The failure consequence loss is as Eq. 7:
ity and degree of freedom. This leads to the capability of model-
ing devices’ failure probability during their lifecycle and covering a L = MT T R · P P T · P P PU + RC + HDP + EDP, (7)
wide range of statistical distributions from Gaussian to exponential
where MTTR is the mean time to repair, PPT is production per time,
(Galar and Kumar 2017; Rogova, Lodewijks, and Lundteigen 2017).
PPPU is production price per unit, RC is repair cost, HDP is human
Generally, the data recorded during failures are used for the recog-
danger penalty, and EDP is the environmental danger penalty.
nition of Weibull distribution function parameters. Eq. 1 presents
the Weibull probability distribution function (Galar and Kumar
2017). 2.2. Developing the interval determination formula

β  t β −1 −( ηt )β In this step, the optimum interval calculation formula for in-


f (t ) = e . (1)
η η strument device proactive maintenance will be developed. Proac-
tive maintenance consists of inspection, cleaning, adjustment, over-
In Eq. 1, β > 0 is known as the shaping parameter, and η > 0 haul, and other required maintenance activities to prevent unnec-
is the scaling parameter in the distribution. β is a very important essary shutdown commands which originate from device failures
parameter which shows the class of failures as: 1) infant mortality and also aims to decrease the operational risk. Although the fre-
phase for (β < 1), 2) useful life phase for (β = 1), or 3) wear-out quency of planned (preventive) maintenance is usually given by
phase for (β > 1). the manufacturers, the values in the operational manuals are just
The Weibull cumulative distribution function which can be de- an average and should be modified according to actual plant con-
rived from Eq. 1 is referred to as (Galar and Kumar 2017): ditions (Legg 2017; Wong 2010).
β PMI determination and calculation is a matter of cost and ben-
F (t ) = 1 − e−( η ) .
t
(2) efit. Over-maintenance of a device imposes unnecessary costs of
Another familiar form of Eq. 2 is: maintenance. On the other hand, by leaving the device for a long
time without checking and inspection, that instrument may en-
β
F (t ) = 1 − e−(λt ) , (3) counter more unexpected failures. Therefore, PMI determination is
a kind of decision-making related to risk, uncertainties, and bene-
where λ = η1 is the failure rate. fit. As a result, the owner is involved in rational decision-making
In case of ambiguity for the extraction of statistical distribu- according to future events (Jahanian 2017).
tion due to any reason such as the lack of enough data, the PMI determination as a rational decision-making problem un-
Weibull distribution makes it possible to gradually (and by increas- der uncertainty can be achieved using the expected utility theory
ing the statistical data) reach an acceptable answer. In case of the (EUT). The idea of EUT was proposed by Daniel Bernoulli in 1738
absence of manufacturer-provided parameters, Weibull curve fit- in his well-known St. Petersburg paradox. The original idea was
ting by having the failure data of a device or identical devices is then improved by (Ramsey 1926), (Von Neumann and Morgenstern
possible (Galar and Kumar 2017; Johnson 1950). Weibull param- 2007), and (Savage 1972) in the 20th century. EUT describes the be-
eters’ extraction from failure data is available in (Campbell 2015; havior of rational people in decision-making. The general meaning
Galar and Kumar 2017). Another way for Weibull curve fitting is of “Utility” is happiness or satisfaction (Common and Stagl 2005).
the use of available functions such as Wblfit() function in MAT- Although more realistic descriptive models, such as the prospect
LAB. By setting β = 1 and η = λ1 , a widely used failure proba- theory, are available nowadays, the EUT has been utilized as a
bility model for equipment and devices as exponential density normative model only to define the rational decision made by
function ((CCPS) 2010; Hauge, Håbrekke, and Lundteigen 2010; people (Jahanian 2017). For example, the EUT model is used by
Jahanian 2015; Wang, Tomovic, and Liu 2015) will be obtained. The (Jahanian 2017) to find the safety integrated level (SIL) for safety
exponential failure probability distribution is: instrumented functions. In this paper, the mentioned concept with
appropriate changes in the methodology is adopted to find the op-
Fconstant f ailure rate (t ) = 1 − e−λt . (4)
timum PMI in the process plants.

3
R. Abbasinejad, F. Hourfar and A. Elkamel Computers and Chemical Engineering 152 (2021) 107362

EUT has been used in the field of maintenance in several stud- where C(Ok ) and Pf (Ok ) are the cost and the probability of device
ies. (Kapliński 2013) has applied EUT for maximization of the failure in case of option Ok selection, respectively. In addition, I0
expected value of “utility” by using a decision tree for sequen- reflects the value of initial properties (initial capital investment), B
tial decisions to find the best strategy. (Allah Bukhsh et al. 2019) is the benefit in case of no failure, L is the failure loss, L is a com-
have used EUT to rank and prioritize bridges for maintenance. bination of production loss, repair cost, and failure cost related to
(Cappello, Zonta, and Glišić 2016) have also utilized structural human and environmental damages, and u(.) illustrates decision-
health monitoring data and decision trees for the life management maker personality which is concave for risk-avoiders, convex for
of bridges. risk-takers, and linear for risk-neutrals. Eq. 10 is a conditional ex-
In general, EUT is based on the game theory which is, it- pectation of utility which should be maximized in case of selecting
self, used for maintenance planning in some studies. For instance, option Ok . Oil and gas plants are subject to tough regulatory code
(Chang et al. 2019) introduced a service-oriented multi-player requirements. Therefore, decision-makers in those plants should
maintenance grouping strategy for a complex component system comply with obligations, and their decision is not based on their
based on the Stackelberg-Nash game model. (Min et al. 2013) de- preferences. Then, their decision-making behavior can be taken as
veloped an approach for power generation group company main- risk-neutral (Jahanian 2017).
tenance planning by using game theory and finding the optimal Without loss of generality in Eq. 10 (because we need to set
strategy profile by Nash equilibrium of the game. Also, based on the derivative of Eq. 10 equal to zero), u(.) is taken as the identity
game theory and Nash equilibrium and comparing the construc- function u(x)=x; then, Eq. 10 would be (Jahanian 2017):
tion and improvement game and rational and non-rational player,  
E [U Ok ] = I0 − C (Ok ) − Pf (Ok ) · L(Ok ) + 1 − Pf (Ok ) · B(Ok ) . (11)
(Tayeb, Benatchba, and Messiaid 2018) found the best maintenance
strategy. (Chattopadhyay 2004) used the Cournot-Nash equilibrium In Eq. 11, Pf 1 (Chalupa 2017; Hauge, Håbrekke, and
to maximize a power generation company’s profit in a group of Lundteigen 2010; John and Grebe 2004) which leads to
companies by obtaining the best maintenance strategy. Further- ( 1 − P f ( Ok ) ) ∼
= 1. Therefore, (1 − Pf (Ok ) ) · B(Ok ) will be B(Ok ).
more, (Hernandez, Seepersad, and Mistree 2002) adopted the game All right-side terms in Eq. 11 are average amounts because the
theory in the designing phase to manage maintenance costs. expected utility should be valid for every PMI, and each monetary
If a rational person has some options to choose from, based on variable should include the effect of all moments inside the PMI
EUT, the problem will be maximizing the expected value of utility duration. Then Eq. 11 will be:
by selecting the most appropriate option (Eq. 8).
  E [U Ok ] = I0 − C (Ok ) − L(Ok ) · Pf (Ok ) + B(Ok )
E U OOptimum = max{E [U |Ok ]k = 1, . . . , n} where E [U Ok ]
 = I0 − C (Ok ) − L(Ok ) · Pf (Ok ) + B(Ok ) . (12)
= u ( O k , x ) · p( x | O k ) , (8)
x

where the expected value of utility according to option Ok is k = 1, . . . , m


shown as E[U|Ok ]. In addition, u(Ok , x ) is the utility according to Ok
In Eq. 12, one should calculate the differential for obtaining the
option and outcome x, and p(x|Ok ) is the conditional probability
optimal value of E[U Ok ]. Therefore, the first term (I0 ) of the equa-
according to outcome x while selecting option Ok (Jahanian 2017).
tion can be removed as constant terms with zero differential. Thus,
For the expected value in the continuous space, options Ok will
Eq. 12 will be reduced to:
change to continuous variable O, the outcome will be represented
by probability density function f(x), and the expected value will be 
GBI (Ok ) = −C (Ok ) − L(Ok ) · Pf (Ok ) + B(Ok ) . (13)
obtained by Eq. (9):
E [UO] = ∫ u(O, x ) · f (x ) · dx. (9) GBI is the gain related to an option, in which “BI” stands for the
concept of “Benefit Included” . It has been shown in the following para-
In continuous space, the main goal is to find OOptimum . The opti- graphs that term B may be ignored, without any adverse conse-
mum option maximizes E[U|O]. For determining a periodic inspec- quence in computation of the optimal solution (Eq. 19). Since C,
tion plan in a preventive maintenance program, utility is repre- L and B are positive numbers, the right side of Eq. 13 is negative
sented by benefit and failure (Hariga 1996). The failure rate can be and it implies the reverse relation of gain and cost and loss. In
obtained from either the manufacturers’ datasheets or the in-site case of failure occurrence, loss is the summation of the costs cor-
recorded data. responding to the production reduction, stop and repair costs, and
In this study, failure costs are minimized, while the cost of the assets-human-environmental damages. Eq. 13 in the continu-
proactive maintenance is minimized by determining the optimum ous space would be:
PMI. For a PMI planner in the maintenance planning department,

as a decision-maker, there are n options to choose from. Each op- GBI (O ) = −C (O ) − L(O ) · Pf (O ) + B(O ) . (14)
tion Ok (PMI) is based on some costs and benefits. Costs include:
1) initial general costs and expenses spent for maintenance plan- The goal of Eq. 14 is to find the optimum PMI. Therefore, in
ning, such as the required manpower for maintenance planning, 2) this case, the option is finding and selecting the best PMI τ . Then,
option cost which is the cost related to the selected PMI for instru- Eq. 14 will be:
ment devices, 3) the risk or probability of failure multiplied by the O = τ, (15)
resulted loss related to the chosen PMI. Benefits include the ben-
efit in case of no failure. When the expected utility is maximized,
the best PMI (option) has been selected. GBI (τ ) = −C (τ ) − L(τ ) · Pf (τ ) + B(τ ). (16)
To define utility expectation for PMI determination, Eq. 10 is
defined as: Now, by considering an arbitrary duration D (as illustrated in
  Fig. 1), the duration length would be the number of PMIs in the
E [U Ok ] = u(I0 ) − u(C (Ok ) ) − Pf (Ok ) · u(L(Ok ) ) + 1 − Pf (Ok ) · u(B ), duration D, multiplied by the PMI duration. The goal is to max-
(10) imize the average gain over duration D, which is chosen much
larger than proactive maintenance interval.

k = 1, . . . , n D=n·τ . (17)

4
R. Abbasinejad, F. Hourfar and A. Elkamel Computers and Chemical Engineering 152 (2021) 107362

Fig. 1. An arbitrary duration D consisting of multiple PMIs.

If the average benefit in time unit is b, and under the assump- η. Failure probability in an interval is the average probability of
tion Pf 1, then the average benefit in duration D will be: all interval moments’ failure probability. Then, from Eq. 21, Eq. 22,
and Eq. 23, we will have:
B(τ ) = b · D. (18)

D·L 1 τ (β −1)
It can be observed that the average benefit in duration D is in- L (τ ) · P f (τ ) = · β β ·τ
τ β (1 + β ) η
dependent of τ . Since the differential value of G(τ ) is needed to

calculate the optimum PMI, B(τ ) as a constant value can be re- 1 τ ( β −1 )
· β β ·τ . (24)
moved from Eq. 16. So the modified Eq. 16 is rewritten as: β (1 + β ) η
G(τ ) = −C (τ ) − L(τ ) · Pf (τ ) (19) From Eq. 24, one can obtain:
By decreasing the PMI, more proactive maintenance activities
2
1 τ (β −1)
are done and by increasing the interval, fewer proactive mainte- L (τ ) · P f (τ ) = D · L · τ · β β . (25)
nance activities are performed in a certain duration. Therefore, as
β (1 + β ) η
mentioned earlier about the average essence of monetary terms in From Eq. 19, Eq. 20, and Eq. 25, the variable part of the PMI
Eq. 19, and by using Eq. 17 and considering k as the device aver- gain would be:
age proactive maintenance expense which should be paid for the
2
maintenance, the average cost of maintenance C (τ ) in a certain D·k 1 τ (β −1)
G (τ ) = − −D·L·τ · β . (26)
duration D would be: τ β (1 + β ) ηβ
D·k
C (τ ) = n · k = . (20) Eq. 26 can be rearranged and simplified as:
τ
D·k D·L·τ
Loss occurrence in duration D should also be stated in the av- G (τ ) = − − · τ (2β −2) . (27)
τ (1 + β )2 η2β
erage. Thus, by assuming that L is the average loss in each de-
vice failure occurrence, τ is much smaller than MTTF, and by using To find the optimum gain according to the PMI length, the
Weibull failure probability distribution, the average loss in duration derivation of gain over τ will be set to zero.
D is:  
d − Dτ·k − D·L·2τ 2β · τ (2β −2)
L (τ ) = (n · L ) · P f (τ ) . (21) dG(τ ) (1+β ) η
= =0 . (28)
dτ dτ
From (Rogova, Lodewijks, and Lundteigen 2017), Pf (τ ) can be
Then:
calculated by Eq. 22:
dG(τ ) D·k D · L · ( 2β − 1 )
1 = 2 − · τ (2β −2) = 0 . (29)
Pf (τ ) = (Z (τ ) · τ ) . (22) dτ τ (1 + β )2 η2β
β (1 + β )
In other words:
Since Pf (τ )  1, for gain estimation in Eq. 19, the right side
k · η 2β · ( 1 + β )
2
of Eq. 22 should be much smaller than 1. The infant mortality τ 2β = , (30)
phase in many applications is not the case since the main lifecy- L · ( 2β − 1 )
cle of the device happens in the useful life and wear-out phase or,
(Moubray 2001). Therefore, we can assume that β ≥ 1. Also, Z (τ )
21β
21β
k · η 2β · ( 1 + β )
2
is equal to (Rogova, Lodewijks, and Lundteigen 2017): 1 k
τ = = η · (1 + β ) · β
L · ( 2β − 1 ) L · ( 2β − 1 )
τ (β −1)
Z (τ ) = β . (23) . W hen β ≥ 1and τ < η .
ηβ (31)

From Eq. 23 and β ≥ 1, the condition β (1+ Eq. 31 is a closed-form comprehensive formula for calculating
β ) (Z (τ ) · τ )  1
1

(β −1 ) β the optimum PMI based on Weibull failure probability distribution.


would be β (1+β ) · β ·
1 τ · τ = τ  1 or τ < 1. Then, two
β η β
(1+β )η η To better understand the concept and application of the ob-
conditions for the developed formula’s validity are β ≥ 1 and τ < tained formula, the cost curve (−G(τ ) ) for different parameters

5
R. Abbasinejad, F. Hourfar and A. Elkamel Computers and Chemical Engineering 152 (2021) 107362

(β , k, L, η ) and the optimum PMI curve are illustrated in presented in Table 1. Optimum τ curves obviously follow the min-
Figs. 2(a), 2(b), 2(c), and 2(d). For each figure, just one parameter imum cost trajectory in Figs. 2(a), 2(b), 2(c), and 2(d).
changes while the others are fixed. The fixed parameters and the In many PMI and failure probability-related applications such as
range of variable parameters are routine examples which can be (Duarte, Craveiro, and Trigo 2006), (Rezaei 2017), (Jahanian 2017),
seen in industrial plants. In Figs. 2(a), 2(b), 2(c), and 2(d), we have: (Sharifi and Taghipour 2020), and (Basker and Husband 1978), a
1 1
Optimum τ = η · (1 + β ) β · ( L·(2βk −1) ) 2β , and the curve groups of constant failure rate with an exponential probability density func-
D·L·τ (2β −2) are illustrated for one
tion, and “as new” device condition after periodic maintenance
−G(τ ) = cost (τ ) = Dτ·k + 2 2β · τ are assumed. Therefore, it would be useful and practical to obtain
(1+β ) η
changing parameter, while others are fixed. The range of the the optimum PMI calculation formula based on the mentioned as-
changing parameters and the value of the fixed parameters are sumptions.

Fig. 2. Illustrating the effect of Weibull parameters’ change on the calculated PMI by showing the cost curve groups (-G(τ )) according to different values of each parameter
while the other parameters are constant. (a) For parameterβ , (b) For parameter k, (c) For parameter L, (d) For parameter η.

6
R. Abbasinejad, F. Hourfar and A. Elkamel Computers and Chemical Engineering 152 (2021) 107362

Fig. 2. Continued

Table 1
Changing parameters and fixed parameters in the curve groups of −G(τ ) related to Eq. 31.

Varying parameters Fixed parameters


Fig.
Range of variations Value Value

2(a) 1 ≤ β ≤ 10 β =1, 1.25, 1.5, 1.75, 2, 2.5, 3, and 10 η=2190 k=1 L=60
2(b) 1 ≤ k ≤ 100 k=1, 2, 5, 10, 20, 25, 40, and 50 β =1.5 η=2190 L=1000
2(c) 10 ≤ L ≤ 10 0 0 L=10, 20, 30, 40, 50, 100, 200, 500, and 1000 β =1.5 η=2190 k=1
2(d) 300 ≤ η ≤ 8400 η=300, 600, 1200, 2400, 3600, 4800, 6000, 7200 and 8400 β =1.5 k=1 L=60

7
R. Abbasinejad, F. Hourfar and A. Elkamel Computers and Chemical Engineering 152 (2021) 107362

By setting β = 1 and η = 1/λ in the Weibull cumulative distri- considering the mentioned assumptions, the problem will be sim-
bution function, it is possible to achieve the constant failure rate plified and, as a result, the formula will be simpler. Eq. 41 is a
and device “as new” condition after the periodic maintenance for- user-friendly formula which, under the above conditions, can be
mula; however, due to Eq. 31 verification, the new formula will simply used to obtain the optimum PMI.
be obtained by EUT, and then, by setting Eq. 31 parameters to the Cost curves (−G(τ ) ) for different parameters ( k, L, λ ) and
mentioned values, and comparing the results, the obtained formu- optimum PMI curves are illustrated
 in Figs. 3(a), 3(b), and 3(c). In
las will be double-checked.
each figure, Optimum τ = λ2 k
L and the curve group of −G(τ ) =
By taking the failure rate constant and assuming the device “as
D·L·λ2 ·τ
new” after periodic maintenance, all steps of the procedure are cost (τ ) = τ +
D·k
4 are depicted for one variable parameter,
identical to the comprehensive formula from Eq. 1 to Eq. 21. In while the other parameters are fixed. The range of the changing
this case, in Eq. 21 Pf (τ ) (failure probability average over the en- variables and the value of the fixed variables are illustrated in
tire interval moments) will be calculated by: Table 2. The optimum τ curve obviously follows the minimum cost

trajectory in Figs. 3 (a), 3(b), and 3(c).
1 τ  −λt

Pf (τ ) = 1−e dt . (32) Now, to verify the calculated PMI from Eq. 41, the constant fail-
τ 0 ure rate (β = 1 ) and η = 1/λ are applied to Eq. 31:
Loss occurrence in duration D should also be stated as average.
21β
1 k β =1 & η=1/λ
Consequently, by considering that L is the loss for each device fail- τ = η · (1 + β ) β · ⇒
ure occurrence, and τ is much smaller than MTTF, the average loss L · ( 2β − 1 )
in duration D will be:
12 

1 1 k 2 k
D·L 1 τ   τ = · (1 × 2 ) ·1 = . (42)
L (τ ) = (n · L ) · P f (τ ) = · 1−e −λt
dt . (33) λ L · (2 − 1 ) λ L
τ τ 0
It can be seen that Eq. 42 is completely identical to Eq. 41.
Failure probability over an interval is equal to the average of
failure probability over time in the interval; therefore: 2.3. Methodology



D·L 1 τ   1 τ  
L (τ ) · P f (τ ) = · 1 − e−λt dt · 1 − e−λt dt . The use of the industrial plants’ data and formulas developed
τ τ 0 τ 0 in the paper as a methodology is depicted as a flowchart (Fig. 4).
(34) It should be noted that due to inherent uncertainties and
also equipment aging during the plant lifecycle, variations in fail-
As mentioned earlier, τ is much smaller than MTTF or λ . There- 1
ure probability function parameters may occur. As a result, in
fore, λτ will be much smaller than one. Thus, the exponential fail-
Eq. 31 parameters β and η, and in Eq. 41 parameter λ may change.
ure probability distribution function can be estimated as:
In other words, PMI can be updated frequently based on the new

plant conditions or available information.
(λτ )2 (λτ )3
1 − e−λτ = 1 − 1 − λτ + − + ... ∼
= λτ , when λτ  1.
2! 3!
3. Case study
(35)
By considering Eq. 32 and Eq. 33, we will have: The fixed interval preventive maintenance (PM) strategy is used

τ

τ
for most devices by the maintenance department in the selected
D·L 1 1 gas refinery studied in this research. By updating the failure rate in
L (τ ) · P f (τ ) = λt dt ·· · λt dt
·
τ τ 0 τ 0 a device lifecycle using the data collected from device maintenance
2 2 records and applying it to the proposed formulas, the presented
D·L λt τ λt τ D · L · λ2 · τ
= · = . (36) approach can be used as a condition-based maintenance interval
τ 3 2 0 2 0 4 determination strategy.
From Eq. 19, Eq. 20, and Eq. 36, the variable part of the PMI
gain would be: 3.1. Case study roadmap

D·k D·L·λ ·τ 2
G (τ ) = − − . (37) In this case study, firstly, 20 instrument devices were selected
τ 4 from four types of field instruments based on the available failure
To find the optimum gain according to the PMI length selection, records in maintenance department database (In total, 80 instru-
the derivation of gain over τ will be set to zero. ments were considered in this study). The selected instrument de-
  vices used in this study are as follows:
d − Dτ·k − D·L·4λ ·τ
2
dG(τ )
= =0 . (38) - Flow transmitter
dτ dτ
- Temperature transmitter
Then:
- Level switch
dG(τ ) D·k D · L · λ2 - Control valve positioner
= 2 − =0 . (39)
dτ τ 4
The instruments were selected from different process units of
In other words:
the gas refinery (separation unit, dehydration unit, stabilizing unit,
4k and output unit). Secondly, failure rates of instrument devices were
τ2 = , (40)
L · λ2 obtained from refinery maintenance records database, and based
or: on maintenance costs, and failure loss in one year (from April 2018
  to April 2019), optimum PMIs were calculated for each device type
4k 2 k
τ= = , when λτ  1. (41) by Eq. 31. Thirdly, for the next year (starting from April 2019 until
Lλ2 λ L April 2020), the maintenance costs and failure loss related to each
Again, by assuming a constant failure rate, the optimum PMI instrument device type were calculated while applying the PMI ob-
determination formula has been calculated. It is obvious that by tained from Eq. 31. Finally, the annual benefit for each instrument

8
R. Abbasinejad, F. Hourfar and A. Elkamel Computers and Chemical Engineering 152 (2021) 107362

Table 2
The range of the changing variables and the value of the fixed variables in the curve groups of −G(τ ) in Eq. 41.

Varying parameters Fixed parameters


Fig.
Range of variations Value Value

3(a) 300 ≤ 1/λ ≤ 8400 1/λ =300, 600, 1200, 2400, 3600, 4800, 6000, 7200 and 8400 k=1 L=60
3(b) 1 ≤ k ≤ 100 k=1, 2, 5, 10, 20, 25, 40, 50, and 100 1/λ=2190 L=1000
3(c) 10 ≤ L ≤ 10 0 0 L=10, 20, 30, 40, 50, 100, 200, 500, and 1000 1/λ=2190 k=1

Table 3 Table 4
Average accumulative distribution function parameters for each instrument de- Mean time to maintenance and average maintenance activity cost.
vice type.
Device type MTTM per device Average MAC per
Device type Scale parameter η (hour) Shape parameter β (hours) device ($)

Flow transmitter 9620 1.96 Flow transmitter 0.2 20


Temperature transmitter 9340 1.90 Temperature transmitter 0.2 20
Level switch 11136 1.83 Level switch 0.3 30
Control valve actuator 6892 1.85 Control valve actuator 0.8 60

device type was computed by comparing the new results with the
Table 5
ones calculated based on the former heuristic maintenance policy.
The production of the production per time and the pro-
duction price per unit.
3.2. Determination of failure probability function parameters
PPPU( Unit )
$
· PPT( )
Unit of production
Unit of production Hour

According to the information recorded in the maintenance de- Separation 630 ($/Hour)
partment database, Weibull accumulative distribution function pa- Dehydration 200 ($/Hour)
rameters were identified using failures data related to each device Stabilizing 2350 ($/Hour)
Output 800 ($/Hour)
type. Table 3 illustrates the average Weibull failure probability den-
sity function parameters related to each device type.
lated. Using Eq. 43, by dividing the total PM cost for each device
3.3. Periodical maintenance cost determination
type by the number of devices, the average total PM cost for each
device can be obtained.
For the typical gas refinery studied here, the selected instru-
ment devices’ PM costs are calculated by Eq. 6. The average of cal- 
kDevice T ype(Average) = MT T MDevice T ype(Average) ·
culated PM costs and repair time, for each instrument device type,  
are presented in Table 4. Production per time and production price P P T(Re f inery Unit,Device T ype) · P P PU(Re f inery Unit,Device T ype) ·
per unit are shown in Table 5. Table 6 lists the number of devices
Re f inery Unit  
per each refinery process unit.
Number o f device typeRe f inery Unit + MACDevice T ype(Average ) /
As mentioned earlier, the total PM cost per each device, by us- Number o f device type .
ing Eq. 6, and refinery data for the selected devices, can be calcu- (43)

Fig. 3. Illustrating the effect of Weibull parameters’ change on the calculated PMI by showing the cost curve group (-G(τ )) according to different values of each parameter,
while the other parameters are constant. (a) For parameter λ1 , (b) For parameter k, (c) For parameter L.

9
R. Abbasinejad, F. Hourfar and A. Elkamel Computers and Chemical Engineering 152 (2021) 107362

Fig. 3. Continued

Table 6
Number of devices per each gas refinery process unit.

Unit Flow Transmitter Temperature Transmitter Level Switch Control Valve Actuator

Separation 8 0 8 10
Dehydration 8 8 10 8
Stabilizing 2 12 2 2
Output 2 0 0 0

10
R. Abbasinejad, F. Hourfar and A. Elkamel Computers and Chemical Engineering 152 (2021) 107362

Fig. 4. Illustration of the presented methodology for determining the optimum PMI.

Table 7
Table 8
Average of each instrument device type PM cost for
Mean time to repair and average repair cost.
each interval.
Device type MTTR per device Average RC per device
Device type Average PM Cost ($)
(hours) ($)
Flow transmitter 149
Flow transmitter 2 30
Temperature transmitter 318
Temperature 3 30
Level switch 206
transmitter
Control valve actuator 564
Level switch 3 40
Control valve actuator 4 70

Table 7 lists the summarized results for the average PM cost of


each instrument device type for each interval. Table 9
Each device type’s financial failure losses and consequences.
3.4. Failure consequence loss determination Device type Average financial losses and consequences ($)

Flow transmitter 1324


The failure consequence cost can be calculated by Eq. 7, MTTR Temperature transmitter 4500
and RC are illustrated in Table 8, PPT and PPPU multiplication can Level switch 1801
be found in Table 5, and the average failure consequence loss for Control valve actuator 2590
each type of instrument device can be seen in Table 9. For the

11
R. Abbasinejad, F. Hourfar and A. Elkamel Computers and Chemical Engineering 152 (2021) 107362

Table 10
The PMI obtained by the proposed approach for the selected instrument devices.

Device type Average PMI (Hours) Validity check β ≥ 1 and τ < η

Flow transmitter 7298 1.96>1 and 7298<9620


Temperature transmitter 6212 1.90>1 and 6212<9340
Level switch 8323 1.83>1 and 8323<11136
Control valve actuator 6147 1.85>1 and 6147<6892

Table 11
Comparison of the heuristic maintenance interval and the average calculated maintenance interval.

Device type Previous average PMI (Days) Average of calculated PMI (Days)

Flow transmitter 90 304


Temperature transmitter 60 259
Level switch 180 347
Control valve actuator 90 256

Table 12 ically smaller to reduce the risk of device/equipment failure and to


Comparison of the devices’ failures during a year before and a year after applying
guarantee the plant safetey.
the proposed technique.
Table 9 presents a summary of the results for the average fail-
Device type Failures occurring in a Failures occurring in a ure loss of each instrument device type for each interval.
year before applying year after applying the
the proposed technique proposed technique

Flow transmitter 2 4 3.5. Calculating the optimum PMI


Temperature 2 3
transmitter By Eq. 31 and based on Tables 3, 7, 8, and 9, the average op-
Level switch 3 4
Control valve actuator 2 6
timum PMIs for each instrument device type were calculated. The
conditions for validation of Eq. 31, used for obtaining the PMI, as
stated in Section 2-2 are β ≥ 1 and τ < η. The results and validity
checking are presented in Table 10.
studied process units and devices, the human danger penalty and Table 11 presents the previous heuristic PMIs and the average of
environment danger penalty are not the case and can be ignored. calculated PMI by the proposed formula for each instrument device
Here, the total failure loss per each device, by using Eq. 7, and type.
refinery data for the selected devices, can be calculated. In Eq. 44, In addition, the number of failures for each field instrument
by dividing the total failure loss for each device type by the num- type in two consecutive years is presented in Table 12.
ber of devices, the average total failure loss for each device can be
obtained.
3.6. Calculating the obtained annual benefit by applying the
 presented technique to compute the optimal PMI
LF low T ransmitter (Average) = MT T RDevice T ype(Average) ·
 
P P T(Re f inery Unit,Device T ype) · P P PU(Re f inery Unit,Device T ype) · By calculating the total maintenance cost added to the failure
Re f inery Unit  loss for each device type in two consecutive years, and obtain-
Number o f device typeRe f inery Unit
ing their difference, the annual benefits achieved from the calcu-
+RC + HDP + EDP )/Number o f device type lated PMI implementation are obtained. Related information can
(44) be found in Table 13.
From Table 13, it is concluded that almost 737.56$, equivalent
It should be noted that the instrumnets which we had permis- to 40.75% of maintenance and failure expenses per year per instru-
sion to work with in this research were those which did not have ment device has been saved for the gas plant owner by using the
any negative effect on personnel and environment factors, accord- PMI obtained via the proposed technique, in comparison with the
ing to the studies of gas plant HSE department. So in this case previous heuristic PMI applied by the plant experts. Furthermore,
study, HDP and EDP are assumed to be zero. However, for the cases it can be observed that although an increase in PMI results in a
in which the mentioned penalties are active, PMI becomes dramat- higher value of total failure loss (from 22,611$ to 42,330$), the

Table 13
Annual benefit obtained while applying the calculated PMI and the related detailed information.

Device type PM cost for a year Failure loss for a PM cost for a year Failure loss for a Total benefit while Total benefit while
before applying the year before after applying the year after applying applying the applying the
proposed technique applying the proposed technique the proposed calculated PMI per calculated PMI per
($) proposed technique ($) technique ($) year ($) year (%)
($)

Flow transmitter 11,952 2,708 2,988 5,416 6,256 42.67


Temperature 38,160 9,060 6,360 13,590 27,270 57.75
transmitter
Level switch 8,244 5,523 4,122 7,364 2,281 16.57
Control valve 45,120 5,320 11,280 15,960 23,200 46.00
actuator
Total 103,476 22,611 24,750 42,330 59,007 40.75

12
R. Abbasinejad, F. Hourfar and A. Elkamel Computers and Chemical Engineering 152 (2021) 107362

significant reduction in total PM cost (from 103.476$ to 24.750$) Common, Michael, Stagl, Sigrid, 2005. Ecological Economics: An Introduction. Cam-
leads to a substantial benefit in the second year. bridge University Press.
Duarte, José A Caldeira, Craveiro, João C Taborda A, Trigo, Tomás Pedro, 2006. Op-
timization of the Preventive Maintenance Plan of a Series Components System.
4. Conclusion International Journal of Pressure Vessels and Piping 83 (4), 244–248.
Exner, Konrad, Schnürmacher, Christian, Adolphy, Sebastian, Stark, Rainer, 2017.
Proactive Maintenance as Success Factor for Use-Oriented Product-Service Sys-
The tradeoff between the cost and benefit of a large number of
tems. Procedia CIRP 64, 330–335.
instrument devices’ PM in oil and gas plants is a challenging issue. Fitouhi, Mohamed-Chahir, Nourelfath, Mustapha, 2012. Integrating Noncyclical Pre-
In this research, two techniques were developed for determining ventive Maintenance Scheduling and Production Planning for a Single Machine.
International Journal of Production Economics 136 (2), 344–351. http://www.
the optimum PMI based on EUT. The comprehensive formula in-
sciencedirect.com/science/article/pii/S0925527311005378.
cluded the degradation effect on devices and the concise one could Galar, Diego, Kumar, Uday, 2017. EMaintenance: Essential Electronic Tools for Effi-
be used for devices with a relatively constant failure rate. These ciency. Academic Press.
formulas could be used by both fixed and condition-based main- Gordon, Christopher Ampofo Kwadwo, Baris Burnak, Melis Onel, and Efstratios N
Pistikopoulos. 2020. “Data-Driven Prescriptive Maintenance: Failure Prediction
tenance policies since in condition-based methods, the PMI can be Using Ensemble Support Vector Classification for Optimal Process and Mainte-
refreshed by the proposed formulas at any time according to the nance Scheduling.” Industrial & Engineering Chemistry Research.
updated parameters. To verify the introduced technique, compre- Gu, Qiuying, Hoo, Karlene A, 2015. Model-Based Closed-Loop Control of the Hy-
draulic Fracturing Process. Industrial & Engineering Chemistry Research 54 (5),
hensive and concise formulas were developed separately. Then, by 1585–1594.
setting the comprehensive formula parameters based on the con- Hariga, M.A., 1996. A Maintenance Inspection Model for a Single Machine with Gen-
stant failure rate condition, the results were verified. Finally, in the eral Failure Distribution. Microelectronics Reliability 36 (3), 353–358.
Hauge, Stein, Håbrekke, Solfrid, Lundteigen, Mary Ann, 2010. Reliability Prediction
practical stage, the comprehensive formula was applied to a typi- Method for Safety Instrumented Systems–PDS Example Collection, 2010 Edition.
cal gas refinery for the selected instrument devices. Using the data SINTEF Report A 17956, 42–50.
collected by refinery maintenance department and also financial Hernandez, Gabriel, Conner Seepersad, Carolyn, Mistree, Farrokh, 2002. Designing
for Maintenance: A Game Theoretic Approach. Engineering Optimization 34 (6),
data during one year for the selected instrument devices, the opti-
561–577.
mum PMI for each type of instrument devices were obtained. For Van Horenbeek, Adriaan, Kellens, Karel, Pintelon, Liliane, Duflou, Joost R, 2014. Eco-
the subsequent year, the obtained PMIs were used for the selected nomic and Environmental Aware Maintenance Optimization. Procedia CIRP 15,
343–348. http://www.sciencedirect.com/science/article/pii/S2212827114004715.
instrument devices. By comparing the costs and losses in two con-
Jahanian, Hamid., 2015. Generalizing PFD Formulas of IEC 61508 for KooN Configu-
secutive years, it was observed that 737.56 $ (equivalent to 40.75% rations. ISA transactions 55, 168–174.
of maintenance and failure expenses) benefit per instrument de- Jahanian, Hamid, 2017. Optimization, a Rational Approach to SIL Determination. Pro-
vice per year was achieved. cess Safety and Environmental Protection 109, 452–464.
Jardine, A.K.S., 1973. Determination of Optimal Maintenance Times. The Plant Engi-
neer 13 (6), 109–114.
Declaration of Competing Interest John, C., and R. A. Grebe. 2004. Failure Modes, Effects and Diagnostic Analysis(Ros
04/02-07 R001.
Johnson, Leonard G., 1950. The Median Ranks of Sample Values in Their Popula-
The authors declare that they have no known competing finan-
tion with an Application to Certain Fatigue Studies. In: Bulletin of the Ameri-
cial interests or personal relationships that could have appeared to can Mathematical Society. AMER MATHEMATICAL SOC 201 CHARLES ST, PROVI-
influence the work reported in this paper. DENCE, p. 247 RI 02940-2213.
Kapliński, Oleg., 2013. The Utility Theory in Maintenance and Repair Strategy. Pro-
cedia Engineering 54, 604–614.
CRediT authorship contribution statement Khan, Faisal Irshad, Abbasi, Shahid Abbas, 1998. Risk Assessment in Chemical Pro-
cess Industries. Discovery Publishing House.
Reza Abbasinejad: Conceptualization, Methodology, Software, Kumar, Uday., 1998. Maintenance Strategies for Mechanized and Automated Min-
ing Systems: A Reliability and Risk Analysis Based Approach. Journal of mines,
Formal analysis, Visualization, Investigation, Writing – original metals and fuels 46 (11), 343–347.
draft. Farzad Hourfar: Conceptualization, Methodology, Validation, Legg, Roger., 2017. Air Conditioning System Design. Butterworth-Heinemann.
Visualization, Investigation, Writing – review & editing. Ali Elka- Lundgren, Camilla, Skoogh, Anders, Bokrantz, Jon, 2018. Quantifying the Effects of
Maintenance–a Literature Review of Maintenance Models. Procedia CIRP 72,
mel: Writing – review & editing, Supervision, Project administra- 1305–1310.
tion. Min, C G, Kim, M K, Park, J K, Yoon, Y T, 2013. Game-Theory-Based Generation Main-
tenance Scheduling in Electricity Markets. Energy 55, 310–318.
References Mo, John P T, Bil, Cees, Sinha, Arvind, 2015. Engineering Systems Acquisition and
Support" - Systems Design. Woodhead Publishing, Oxford, pp. 51–73.
(CCPS) Center for Chemical Process Safety, 2010. Guidelines for Safe and Reliable Moore, Ron. 2004. “Making Common Sense Common Practice.”
Instrumented Protective Systems. John Wiley & Sons. Moubray, John., 2001. Reliability-Centered Maintenance. Industrial Press Inc.
Abbasinejad, Reza, Hourfar, Farzad, Madhuranthakam, Chandra Mouli R, Elka- Narasingam, Abhinav, Kwon, Joseph Sang-Il, 2017. Development of Local Dynamic
mel, Ali, 2020. A Data Size Reduction Approach Applicable in Process Control Mode Decomposition with Control: Application to Model Predictive Control of
System of Oil and Gas Plants. Sustainability 12 (2), 639. Hydraulic Fracturing. Computers & Chemical Engineering 106, 501–511.
Allah Bukhsh, Zaharah, et al., 2019. Network Level Bridges Maintenance Planning Von Neumann, John, Morgenstern, Oskar, 2007. Theory of Games and Economic Be-
Using Multi-Attribute Utility Theory. Structure and infrastructure engineering havior (Commemorative Edition). Princeton university press.
15 (7), 872–885. Nguyen, Duy Quang, Brammer, Christopher, Bagajewicz, Miguel, 2008. New Tool for
Basker, B A, Husband, T M, 1978. Determination of Optimal Overhaul Intervals the Evaluation of the Scheduling of Preventive Maintenance for Chemical Pro-
and Inspection Frequencies—A Case Study. Microelectronics Reliability 17 (2), cess Plants. Industrial & engineering chemistry research 47 (6), 1910–1924.
313–315. Nguyen, DuyQuang, Bagajewicz, Miguel, 2010. Optimization of Preventive Mainte-
Campbell, John., 2015. Complete Casting Handbook: Metal Casting Processes, Metal- nance in Chemical Process Plants. Industrial & engineering chemistry research
lurgy, Techniques and Design. Butterworth-Heinemann. 49 (9), 4329–4339.
Cappello, Carlo, Zonta, Daniele, Glišić, Branko, 2016. Expected Utility Theory for Pashnina, Nadezhda., 2018. Determination of Optimal Calibration Intervals by Bal-
Monitoring-Based Decision-Making. Proceedings of the IEEE 104 (8), 1647–1661. ancing Financial Exposure against Measurement Costs. Flow Measurement and
Carlsen, Liv A, Nygaard, Gerhard, Nikolaou, Michael, 2013. Evaluation of Control Instrumentation 60, 115–123.
Methods for Drilling Operations with Unexpected Gas Influx. Journal of Process Percy, David F, Kobbacy, Khairy A H, 20 0 0. Determining Economical Maintenance
Control 23 (3), 306–316. Intervals. International Journal of Production Economics 67 (1), 87–94.
Chalupa, R., 2017. Failure Modes, Effects and Diagnostic Analysis (ROS Ramsey, F.P., 1926. Truth and Probability. Reprinted In. Studies in subjective proba-
11/02-058 R001) https://www.emerson.com/documents/automation/ bility 61–92.
product- certificate- 644- hart- temperature- transmitter- option- code- qt- device- Rezaei, Esmaeil., 2017. A New Model for the Optimization of Periodic Inspection In-
label- sw- rev- 1- 1- x- rosemount- en- 89376.pdf. tervals with Failure Interaction: A Case Study for a Turbine Rotor. Case studies
Chang, Fengtian, et al., 2019. A Service-Oriented Multi-Player Maintenance Group- in engineering failure analysis 9, 148–156.
ing Strategy for Complex Multi-Component System Based on Game Theory. Ad- Rogova, Elena, Lodewijks, Gabriel, Lundteigen, Mary Ann, 2017. Analytical Formulas
vanced Engineering Informatics 42, 100970. of PFD and PFH Calculation for Systems with Nonconstant Failure Rates. Pro-
Chattopadhyay, Deb., 2004. A Game Theoretic Model for Strategic Maintenance and ceedings of the Institution of Mechanical Engineers, Part O: Journal of Risk and
Dispatch Decisions. IEEE Transactions on Power Systems 19 (4), 2014–2021. Reliability 231 (4), 373–382.

13
R. Abbasinejad, F. Hourfar and A. Elkamel Computers and Chemical Engineering 152 (2021) 107362

Salahshoor, Karim, Hoorfar, Farzad, Abbasinejad, Reza, Ataei., Seyed Saeed, 2014. De- Silver, Edward A, Fiechter, Claude-Nicolas, 1992. A Simple Case of Preventive Main-
sign and Implementation of a Risk Based Maintenance Program for Determining tenance Decision-Making with Limited Historical Data. International journal of
Optimum Maintenance Intervals for Field Instruments in Sarkhun Gas Refinery. production economics 27 (3), 241–250.
Journal of Control 7 (4), 19–29. Tayeb, Fatima Benbouzid-Si, Benatchba, Karima, Messiaid, Abd-Essalam, 2018. Game
Savage, Leonard J., 1972. The Foundations of Statistics. Courier Corporation. Theory-Based Integration of Scheduling with Flexible and Periodic Maintenance
Sharifi, Mani, Taghipour, Sharareh, 2020. Optimal Inspection Interval for a Planning in the Permutation Flowshop Sequencing Problem. Operational Re-
K-out-of-n System with Non-Identical Components. Journal of Manufacturing search 18 (1), 221–255.
Systems 55, 233–247. Wang, Shaoping, Tomovic, Mileta, Liu, Hong, 2015. Commercial Aircraft Hydraulic
Siddhamshetty, Prashanth, Sang-Il Kwon, Joseph, 2018. Model-Based Feedback Con- Systems: Shanghai Jiao Tong University Press Aerospace Series. Academic Press.
trol of Oil Production in Oil-Rim Reservoirs under Gas Coning Conditions. Com- Wong, William., 2010. The Risk Management of Safety and Dependability: A Guide
puters & Chemical Engineering 112, 112–120. for Directors, Managers and Engineers. Elsevier.
Siddhamshetty, Prashanth, Yang, Seeyub, Sang-Il Kwon, Joseph, 2018. Modeling of Ye, Yixin, Grossmann, Ignacio E, Pinto, Jose M, Ramaswamy, Sivaraman, 2019. Mod-
Hydraulic Fracturing and Designing of Online Pumping Schedules to Achieve eling for Reliability Optimization of System Design and Maintenance Based on
Uniform Proppant Concentration in Conventional Oil Reservoirs. Computers & Markov Chain Theory. Computers & Chemical Engineering 124, 381–404.
Chemical Engineering 114, 306–317.

14

You might also like