Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

ARTICLE IN PRESS

Journal of Loss Prevention in the Process Industries 19 (2006) 542–552


www.elsevier.com/locate/jlp

An integrated off-on line approach for increasing stability and


effectiveness of automated controlled systems based on pump
dependability—case study: Offshore industry
V. Ebrahimipoura,, K. Suzukia, A. Azadehb
a
System Analysis Laboratory, Department of Systems Engineering, Okayama University Address: 700-8530, 3-1-1 Tsushima Naka, Okayama, Japan
b
Department of Industrial Engineering and Institute of Energy Management and Planning Faculty of Engineering, University of Tehran,
P.O. Box 11365-4563, Iran

Abstract

Automated controlled systems are vulnerable to faults. Faults can be amplified by the closed loop control systems and they can
develop into malfunction of the loop. A control loop failure will easily cause production stop or malfunction at a petrochemical plant. A
way to achieve a stable and effective automated system is to enhance equipment dependability. This paper presents a standard
methodology for the analysis and improvement of pump performance to enhance total operational effectiveness and stability in offshore
industry based on dependability. Furthermore, it is shown how a reliability–safety analysis can be conducted through equipment
dependability indicators to facilitate the mitigation of hazard frequency in a plant. The main idea is to employ principle component
analysis (PCA) and importance analysis (IA) to provide insight on the pumps performance. The pumps of offshore industries are
considered according to OREDA classification. The approach identifies the critical pump and their fault through which the major
hazards could initiate in the process. At first PCA is used for assessing the performance of the pumps and ranking them. IA is then
performed for the worst pump which could have most impact on the overall system effectiveness to classify their components based on
the component criticality measures (CCM). The analysis of the classified components can ferret out the leading causes and common-
cause events to pave a way toward improving pump performance through design optimization and online fault detection which
ultimately enhance overall operational effectiveness.
r 2006 Published by Elsevier Ltd.

Keywords: Dependability; Reliability; Maintainability; Principle component analysis; Importance analysis; Online fault detection

1. Introduction 1998; Azadeh &Ebrahimipour, 2002; Ebrahimipour &


Suzuki, 2004). Equipment plays an important role in the
With the growing complexity of automated industrial overall performance of manufacturing systems. In fact,
systems, such as offshore and chemical plants, and the equipment performance is correlated with the overall
increasingly stringent safety and reliability requirements, effectiveness of a manufacturing system. According to US
attention has been focused on dependability assessment National Research Council Report (2000), one of the
and effectiveness improvement. Major factors influencing research priorities of US manufacturing is equipment
the overall effectiveness and efficiency of an industrial reliability and maintenance. It classifies the equipment
organization are identified as technology, equipment, reliability and maintenance into people-related needs and
management, personnel, rules and procedures (Tompkins, equipment-related needs. Furthermore, the quality perfor-
Bozer, Frazelle, Tanchoco, Trevino, 1996; Blanchard, mance of an industrial organization is often assessed by
reliability and safety of its equipment. Fordham, Baxter,
Corresponding author. Tel.: +81 86 251 8058; fax: +81 86 251 8059. Hunter, Malkow (2003) concluded that efficiency, relia-
E-mail addresses: vahid@syslab.sys.okayama-u.ac.jp bility and availability play the most important role in
(V. Ebrahimipour), aazadeh@ut.ac.ir, ali@azadeh.com (A. Azadeh). effectiveness improvement of power and nuclear plants.

0950-4230/$ - see front matter r 2006 Published by Elsevier Ltd.


doi:10.1016/j.jlp.2005.12.007
ARTICLE IN PRESS
V. Ebrahimipour et al. / Journal of Loss Prevention in the Process Industries 19 (2006) 542–552 543

Ingruber (1999) considered efficiency and reliability aspects maintenance is performed under stated conditions and
for improving of harmonic-control amplifiers. Hinkel using stated procedures and resources; and (4) safety—
(2004) introduced a procedure to reach a high-performance a measure of the continuous delivery of proper service free
of pump through enhancing efficiency and reliability from occurrences of catastrophic failures.
indicators. Moustafa, Abdel Maksoud and Sadek (2004) The integrated approach takes four steps which are:
regarded maintainability as a key factor in their analysis (1) assessing the performance of pumps based on avail-
and proposed multi-variate models for making mainte- ability, maintainability and reliability; (2) ranking pumps
nance policies for proper and efficient functions of the to recognize the critical pump based on PCA scores;
manufacturing system. Generic Markov models for avail- (3) importance analysis for the components of the critical
ability and failure analysis in petroleum refineries were pump by using the component criticality measures;
presented by Chochran, Arvindh and Krishnamurthy (4) improving pump performance through design optimi-
(2001). A Petri net was introduced for failure detection as zation and online fault detection.
an effective tool for preventive maintenance in reliability The paper is organized as follows. In Section 2, the
enhancement (Yang, 2003). The impact of equipment models are presented. In Section 3, we introduce the
failures on total effectiveness is an important concern for integrated approach discussed in this paper. Section 4
oil & gas industry. According to CCPS Report (2001), the contains our conclusions and directions.
most frequent accidents in petrochemical plants are
dedicated to fire and explosion. This report shows that 2. Models
the most contributing factor is equipment failure. Barton
and Nolan concluded that in the most of the 263 typical 2.1. PCA
case histories, main problems arose from a basic lack of
understanding of thermo chemistry of the reactions Principal component analysis (PCA) is widely used in
coupled with inadequate engineering design and inade- multivariate statistics such as factor analysis. It is used to
quate operational and maintenance procedures (Banerjee, reduce the number of variables under study and conse-
2003). The significant equipment failures in a petrochem- quently ranking and analysis of decision-making units
ical plant are related to pumps, compressors and piping (DMUs), such as industries, universities, hospitals, cities,
(Lees, 1996). Today, many petrochemical factories utilize etc. (Zhu, 1998). Wang, Xu and Lu (2003) proposed several
advanced methods to enhance their knowledge and under- capability indices and quality measures to summarize
standing about the pump performance and its impact on process performance using PCA. Chen, Kao, Kuo, Wang
process behavior. Therefore, the need for an integrated and Jang (1996) proposed a fuzzy clustering and classifica-
approach for continuous assessment and the improvement tion model for productivity analysis of machinery industry.
of process availability based on pump dependability A multivariate approach was used among 128 manufactur-
becomes essential. The objective of this paper is to present ing organization to indicate that man/machine interfaces
a framework for assessing and improving pumps based on are significant contributors to reducing the negative effect
dependability. Furthermore, it is shown how a reliability– of system complexity (Guimaraes, Matensson, Stahre and
safety analysis can be conducted through equipment Igbaria, 1999).
dependability indicators to facilitate the mitigation of The first objective of PCA is to identify a new set of
hazard frequency in a plant. The structure and methodol- variables such that each new variable, called a principal
ogy of this paper could be particularly useful for component, is a linear combination of original variables.
petrochemical and chemical industries. The results of such Second, the first new variable Y1 accounts for the
studies would help managers to gain a better understanding maximum variance in the sample data and so on. Third,
and improve existing systems with respect to equipment the new variables (principal components) are uncorrelated.
conditions. PCA is performed by identifying Eigen structure of the
This study has identified dependability indicators as covariance or singular value decomposition of the original
major technical factors which have the most contribution data. Here, the former approach will be discussed. It is
on overall process effectiveness. Dependability is defined as assumed there are p variables (indicators) and n DMUs and
the trustworthiness of a system such that reliance can suppose X ¼ ðx1 :::xp Þnp is a n  p matrix composed by x0ij s
justifiably be placed on the service it provides. Depend- and defined as the value of jth indicator for ith DMU and
ability is an overall ability which has other measures such therefore xm ¼ ðx1m . . . xnm ÞT ðm ¼ 1; . . . ; pÞ. Furthermore,
as reliability, availability, maintainability and safety. These suppose X^ ¼ ðx^ 1 :::x^ p Þnp is the standardized matrix of X ¼
terms are defined respectively: (1) reliability—the prob- ðx1 . . . xp Þnp with x^ 0ij s defined as the value of jth
ability that a equipment will perform in a satisfactory standardized indicator for ith DMU and therefore
manner; (2) availability—the probability that an equip- x^ m ¼ ðx^ 1m :::x^ nm ÞT . PCA is performed to identify new
ment will be operational at a given time; (3) maintain- independent variables or principal components (defined
ability—the probability that a given active maintenance as yj for j ¼ 1yp), which are, respectively, different linear
action, for an item under given conditions of use, can be combination of x^ 1 . . . x^ p . As mentioned, this is achieved by
carried out within a stated time interval when the identifying Eigen structure of the covariance of the original
ARTICLE IN PRESS
544 V. Ebrahimipour et al. / Journal of Loss Prevention in the Process Industries 19 (2006) 542–552

data. The principal components are defined by a n  p At the same time, the z vector (z1yzn) where zj shows the
matrix Y ¼ ðy1 . . . yp Þnp composed of y0ij s which are shown score of jth DMUs is given by
by (Minhas and Jacobs, (1996): ,
Xp

y1 ¼ l 11 x^ 1 þ l 12 x^ 2 þ . . . þ l 1p x^ P ; wj ¼ lj lj ¼ lj =p; j ¼ 1 . . . p, (7)
j¼1
y2 ¼ l 21 x^ 1 þ l 22 x^ 2 þ . . . þ l 2p x^ P X
p
  zi ¼ w j yj ; i ¼ 1 . . . n. (8)
j¼1
 
  2.2. Importance Analysis
yP ¼ l p1 x^ 1 þ l p2 x^ 2 þ ::: þ l pp x^ P ;
Component importance analysis is a key part of the
where l jm is the coefficient of mth variable for the jth system reliability quantification process. It enables the
principal component. The l jm ‘s are estimated such that the weakest areas of a system to be identified and indicates
following conditions (1, 2 and 3) are met: modifications which will improve the system reliability and
safety. Efforts to improve reliability can then be connected
1. y1 accounts for the maximum variance in the data, on those areas whose contribution indicates that by
y2accounts for the maximum variance that has not upgrading them, the maximum improvement in system
been accounted by y1, and so on. reliability can be achieved (Birolini, 2004; Henley &
12
2 2
p 2 Kumamoto, 1981). Importance measures assign a numerical
2. l m þ l m þ . . . þ l m ¼ 1; m ¼ 1 . . . p. (1) value between 0 and 1 to each system component or minimal
CS; the 1 signifies the highest level of importance. This
3. l m1 :l n1 þ l m2 :l n2 þ . . . þ l mp :l np ¼ 0 for all man signifies system susceptibility to component or minimal cut-
n ¼ 1 . . . p. ð2Þ set failure. Measures of importance can be categorized as
either deterministic or probabilistic. Deterministic measures
assess the importance of a component or minimal cut-set
For obtaining the lij0 s and consequently p vectors (yj)
without considering component reliability. Although these
(j ¼ 1yp) and PCA scores the following steps are
measures can be useful early in the design phase when
performed:
information concerning component–failure probability is
Step 1: Calculate the sample mean vector x and
limited, generally probabilistic measures of importance are
covariance matrix S:
preferred because they provide more valuable information.
x ¼ ðx̄1 . . . x̄p Þ1p . (3) Probabilistic measures of importance can be further
categorized as either ‘measures concerned with system
1X n
unreliability (contributing to failure frequency)’ or ‘mea-
In which; x ¼ E ðX Þ and x̄j ¼ xij for j ¼ 1 . . . p,
n i¼1 sures concerned with system unavailability (contributing to
failure probability)’. System availability is only of interest
(4)
when SF can be tolerated. Numerous measures of impor-
S ¼ ðsjq Þpp ¼ EfðX  xÞT ðX xÞg for q ¼ 1 . . . p. (5) tance have been developed to enable an analyst to assess the
roles a component failure can play in the deterioration of the
Step 2: Calculate the sample correlation system state. For coherent system, the component criticality
pffiffiffiffiffi pffiffiffiffiffi pffiffiffiffiffi
matrix.R ¼ C 1 = sjj :S:C 1 = sjj where C 1 = sjj is a p  p measure is defined as: the probability that component i is
pffiffiffiffiffi
diagonal matrix whose jth diagonal element is 1= sjj for critical to system and i has failed, weighted by the system
j ¼ 1yp.   unavailability at time t (Beeson & Andrews, 2003):
Step 3: Solve the following equation:R  lI p  ¼ 0 where
G i ðqÞ:qi
Ip is a p  p identity matrix. We obtain the ordered p I Ci ¼ , (9)
characteristic roots (eigen values) l1  l2  . . .  lp with QSys ðtÞ
Spj¼1 lj ¼ p and the related p characteristic vectors (eigen-
  dQSys ðtÞ
vectors) l m m m Gi ðqÞ ¼ , (10)
1 ; l 2 ; :::; l p (m ¼ 1yp). dqi
Those characteristic vectors compose the principal where QSys ðtÞ is system unavailability function; probability
components yj. The components in Eigenvectors are, that a system is in a failed state at time t and qi is defined as
respectively, the coefficients in each corresponding yj: failure probability of component i.
X
p
ym ¼ lm ^ ij
j x for m ¼ 1 . . . p and i ¼ 1 . . . n. (6) 3. Integrated Model
j¼1

Step 4: Calculate the weights (wj) of the principal To achieve the objectives of this paper, a comprehensive
components and PCA scores (zi of each DMU (i ¼ 1yn). study was conducted to locate five technical indicators,
ARTICLE IN PRESS
V. Ebrahimipour et al. / Journal of Loss Prevention in the Process Industries 19 (2006) 542–552 545

which influence equipment performance. These indicators available time minus down time. The standard factors such
are related to equipment dependability. The five indicators as maximum time, down time, time to repair, mean time
were identified as major indicators impacting equip- between failures, and available time, considered as shaping
ment conditions in manufacturing systems (Azadeh parameters, cover the three equipment conditions which
&Ebrahimipour, 2004; Corrie, 1986; Blanchard, 1998; are: (1) critical failure—a failure which causes immediate
Birolini, 2004; Nakajima, 1991). The five indicators are and complete loss of a system’s capability of providing its
categorized into three classes. The first class reflects output; (2) degraded failure—a failure which does not
maintainability of equipment and is measured by indica- critical, but which prevents the system from providing its
tors 1, 2 and 3. Maintainability can be considered by 3 output within specifications; (3) incipient failure—a failure
types of measures which are: (1) maintenance time which does not immediately cause loss of a system’s
including corrective time and preventive time; (2) main- capability of providing its out put, but which, if not
tenance labor hours expended in the process; (3) main- attended to, could result in a critical or degraded failure in
tenance frequency. In this study, we regard maintenance the near future.
labor hours. The second class deals with equipment The pump units, in this study, are selected according to
reliability and is measured by indicator 4 that pertains to the format of Offshore Reliability Data Handbook
the type of equipment and the environment in which the (OREDA, 2002). At first stage, a comparative study is
equipment is operating. The last class represents inherent conducted through PCA by considering the selected five
availability which is defined as the probability that a system indicators. This shows the weak and strong points of
is operating satisfactorily at any point in time and offshore industry in regard to pump units’ performance.
considers only operating time and corrective time. Max- Furthermore, PCA identified which pump units had the
imum time is defined as maximum allowable time dedicated major impact on the overall operational effectiveness of the
to equipment per period (106 h). Active maintenance time is industry.
a portion of downtime when corrective or preventive At second stage, IA analyzed the worst pump unit
maintenance activities are being accomplished. Logistics selected by PCA to classify their components to identify
delay time (LDT) is a portion of downtime when the common-cause events and errors which lead to improve
system is not operational because of delays associated with pump reliability and safety. Fig. 1 depicts the stages that
the support capability; waiting for a spare part, waiting for should be taken.
the availability of test equipment, waiting for use of a The selected pump unit indicators are listed as follows:
special facility. Administrative time is a portion of down-
time due to delay for reasons of an administrative nature x1: maintenance labor hours in critical condition,
for example, the unavailability of personnel, organizational x2 : maintenance labor hours in degraded condition,
constraints, etc. Available time is defined as maximum time x3 : maintenance labor hours in incipient condition,
minus down time. Down time is obtained by integration of x4 : reliability,
inefficient time and administrative time. Operating time is x5 : inherent availability,

Develop Identify the


Identify Define PCA model pump unit
pump units dependability and rank which is the
pump units lowest rank

Identify the most


important indicator or
improving factor by
Eigenvalue analysis

Identify the some


directions for the areas Identify the areas in the Develop IA for
based on the improving pump unit to be improved components of the worst
factors pump unit

Fig. 1. The integrated approach for assessing and improving stages of pump units based on dependability.
ARTICLE IN PRESS
546 V. Ebrahimipour et al. / Journal of Loss Prevention in the Process Industries 19 (2006) 542–552

3.1. Assessment and Ranking presented in the last three column of Table 3. It should be
noted that the coefficients are retrieved from the eigenvec-
As mentioned, the format of OREDA is used to tors for the respective principal components.
categorize the pump units. Therefore, the structure and Step 4: Evaluate the principal components and aggre-
modeling approach may be used for the petrochemical gated weights. The values of principal components and
industry in particular or other industries in general. The consequently their aggregated weights are presented in
assessment process examines the pump units (level 4), that Table 4.
are numbered from 1.3.1.1, 1.3.1.2 to 1.3.3.1 according to
Taxonomy structure. The process is continued up to
3.2. Improving Model
specify the status of the worst pump unit. Furthermore,
the information about the five indicators in regard to pump
The PCA results of the 4-digit pump units are presented
units (level 4) 1.3.1.1, 1.3.1.2 to 1.3.3.1 are collected, and
in Table 4. It shows the worst and best ranks respectively,
structured.
with regard to the five selected indicators. The sea water lift
PCA is used to rank and analyze the data. Consequently,
centrifugal pump got the worst rank 19. By referring to the
it identifies the weak and strong points and introduces
coefficient and weights of the principal components, the
improving indicators for pump units (DMUs). In the next
pump is further analyzed in order to know how we can
section, PCA approach for the four-digit code is described.
improve the sea water lift centrifugal pump performance.
PCA is achieved through a set of well-defined steps as
The most influential indicators for the 4-digit pump units
follows:
may be calculated from eigenvalues and eigenvectors. It
Step 1: Standardize the indicators x1–x5. The indicators
can be observed that the first three principal components
are standardized and are shown in Table 1. They are
conform to between 82 percent of the sample variance
standardized through predefined mean and standard
(Tables 3). To identify the most influential or important
deviation for each indicator.
indicators, aggregated weights must be computed. It
Step 2: Evaluate the correlation matrix. This matrix
should be noted that the p aggregated weights ðw~ m Þ for
shows the values of linear correlation between indicators
m ¼ 1yp show the importance of each indicator and are
x^ 1 –x^ 5 (Table 2).
computed as follows:
Step 3: Calculate eigenvalues, eigenvectors and propor-
tion of the sample variance for all the five principal X
p

components (new variables). The Eigenvalues and propor- w~ m ¼ wj l jm . (11)


j¼1
tion of the sample variance for all the five indicators
(principal components) are presented in Table 3. It is noted Applying the above formulation to the 4-digit pumps,
that first three principal components y1, y2 and y3 account the aggregated weights w~ m (where m ¼ 1y 5) are evaluated
for 82 percent of the sample variance. Therefore, the and shown in Table 5. It can be observed that w~ 4 , reliability
coefficients of the first three principal components are is the most important indicator respectively in the 4-digit

Table 1
Standardized matrix for the five indicators of four -digit DMUs

Pump unit Four-digit code x^ 1 x^ 2 x^ 3 x^ 4 x^ 5

Chemical injection centrifugal 1.3.1.1 0.928 0.365 1.290 0.533 0.634


Combined function centrifugal 1.3.1.2 0.363 0.973 1.638 0.512 0.496
Condensate processing centrifugal 1.3.1.4 0.247 1.197 0.561 0.399 0.196
Cooling system centrifugal 1.3.1.5 0.456 1.148 1.290 0.527 0.634
Crude oil handling centrifugal 1.3.1.6 0.458 1.430 0.481 0.332 0.020
Flare, vent & blow-down centrifugal 1.3.1.8 0.379 1.148 1.290 0.533 0.634
Gas processing centrifugal 1.3.1.9 0.443 0.618 0.521 0.252 0.482
Gas treatment centrifugal 1.3.1.11 0.928 1.148 2.735 0.533 0.634
Heating medium centrifugal 1.3.1.12 0.416 0.426 0.002 0.478 0.561
Oil export centrifugal 1.3.1.14 0.504 1.837 0.481 0.570 2.161
Oil processing centrifugal 1.3.1.15 0.601 1.680 0.425 0.499 0.505
Oily water treatment centrifugal 1.3.1.16 0.403 0.531 1.009 0.346 0.184
Sea water lift centrifugal 1.3.1.17 3.458 0.019 0.008 0.123 2.031
Water fire fighting centrifugal 1.3.1.18 1.369 1.220 0.370 0.397 0.137
Water injection centrifugal 1.3.1.19 0.183 0.668 0.290 3.674 2.537
Chemical injection reciprocating 1.3.2.1 0.669 0.531 0.083 0.757 0.419
Gas processing reciprocating 1.3.2.2 0.662 0.583 0.093 1.253 0.446
Gas treatment reciprocating 1.3.2.3 0.495 0.432 0.616 0.509 0.551
Oily water treatment rotary 1.3.3.1 0.618 0.839 0.988 0.527 0.588
ARTICLE IN PRESS
V. Ebrahimipour et al. / Journal of Loss Prevention in the Process Industries 19 (2006) 542–552 547

Table 2 Table 5
Correlation matrix for four -digit DMUs Aggregated weights for each of the five indicators

x^ 1 x^ 2 x^ 3 x^ 4 x^ 5 SIC w~ 1 w~ 2 w~ 3 w~ 4 w~ 5

x^ 1 1.000 0.233 0.081 0.081 0.595 4-digit Code 0.038 0.236 0.242 0.312 0.284
x^ 2 0.233 1.000 0.070 0.180 0.479
x^ 3 0.081 0.070 1.000 0.087 0.152
x^ 4 0.081 0.180 0.087 1.000 0.674
x^ 5 0.595 0.479 0.152 0.674 1.000
G ffi ð1  bÞ2 l2 þ bl. The beta-factor method assumes
each component has two types of failures: (1) independent
primary failure rate lI; (2) dependent failures of the
Table 3 common-cause type lc. The total failure rate for a
Eigen values for the four -digit DMUs component is then l ¼ li+lc. A parameter b is thus
Variable Eigen values (lj) Weight (wj) Eigenvectors
defined b ¼ lc =ðlc þ li Þ. Hence, the parameter b is the
fraction of the total failure rate attributable to dependent
l1p l2p l3p failures. It is clear that one major way to keep a sea water
lift centrifugal pump online, in its safest and productive
1 2.222 0.444 0.432 0.150 0.620
2 0.968 0.194 0.410 0.152 0.227 mode of operation is to improve b (decreasing common-
3 0.932 0.186 0.164 0.974 0.143 cause failures) so that critical components can be tested,
4 0.768 0.154 0.455 0.024 0.733 maintained and repaired without shutting the plant down
5 0.110 0.022 0.642 0.068 0.079 or requiring hardship maintenance operations which
compromise the safety of the plant. In order to improve
b which leads to decrease risk, we should identify the
Table 4 overriding risk of the sea water lift centrifugal pump and
The values and scores of principal components for the four -digit DMUs then the components have most contributions in the risks.
Four -digit DMU y1 Y2 y3 Zi (Scores) Rank
Table 6 shows the failure rate and repair data for
maintainable components of the pump unit to identify
1.3.1.1 1.112 1.143 0.236 0.188 12 the criticality measures. Table 7 reports the three major
1.3.1.2 0.525 1.711 0.653 0.562 4 failure modes having most occurrence rates and the
1.3.1.4 0.421 0.379 0.476 0.381 14
component contributions for each failure mode based on
1.3.1.5 1.132 1.120 0.273 0.403 10
1.3.1.6 0.699 0.176 0.922 0.618 15 criticality measures. Considering the total relative con-
1.3.1.8 1.495 0.995 0.239 0.628 2 tribution in Table 7 specifies the critical components in the
1.3.1.9 0.784 0.695 0.117 0.534 5 sea water lift centrifugal pump unit that result in directly or
1.3.1.11 1.072 3.008 0.004 1.101 1 indirectly undesired accidents such as fire and explosion in
1.3.1.12 0.933 0.152 0.039 0.441 8
1.3.1.14 2.696 0.020 0.210 1.375 18
the plant. The four critical components (seals, control
1.3.1.15 0.192 0.557 0.354 0.375 13 instruments, valves and pipes) are recognized as substantial
1.3.1.16 0.833 0.837 0.246 0.297 11 initiators in the sea water lift centrifugal pump unit
1.3.1.17 2.847 0.643 1.890 1.433 19 unavailability. The major causes which have the compo-
1.3.1.18 0.884 0.030 1.481 0.759 16 nents failed are presented in Table 8. The causes allow us to
1.3.1.19 3.701 0.069 2.587 1.102 17
identify the nature of common-cause failures. Table 8
1.3.2.1 0.445 0.147 1.069 0.445 6
1.3.2.2 0.257 0.158 1.440 0.444 7 shows the five causes (mechanical failure, leakage, instru-
1.3.2.3 1.077 0.435 0.077 0.414 9 ment failure, out of adjustment and wear) have the most
1.3.3.1 1.390 0.715 0.282 0.585 3 impact on the sea water lift centrifugal pump failure. In
order to improve b, the nature of the five causes is further
analyzed. According to the EDF classification to analyze
dependent (mainly common-cause) failures, the systematic
pumps. In fact, it is proven that for the pump units failures generally result from either: (1) environmental
selected, reliability factor plays an important role in system hazards—events related to the environment outside the
effectiveness. In other words, it is intuitively expected that facility or inside the facility but, in any case, outside the
pump unreliable will increase overall process hazard. studied elementary system; (2) from design errors—made
Therefore, it can be expected that improving sea water lift during the component and elementary system design
centrifugal pump reliability enhances both the pump studies (operating principles, components used, operating
performance and overall system safety. and test procedure definition); (3) manufacture errors—
Reliability is impacted by the failure rate of device made in manufacturing the elementary system components;
itself. This can be shown by the beta-factor method. For (4) assembly errors—committed in the course of the
example, in a redundant system composed two pumps component assembly (at the factory, on the site) and
in standby, the pump failure rate upon demand is during preoperational tests performed on the components
ARTICLE IN PRESS
548 V. Ebrahimipour et al. / Journal of Loss Prevention in the Process Industries 19 (2006) 542–552

Table 6
Failure and repair rates for the sea water lift centrifugal pump unit components

No. Maintainable item Subdivision in centrifugal pump unit Failure rate (per 106) MTTR (h)

1 Actuating device Control & monitoring 19.027 0.317


2 Casing Pump 24.902 0.415
3 Control unit Control & monitoring 38.949 0.650
4 Filter Lubrication system 20.304 0.339
5 Instruments, general Lubrication system & pump 31.414 0.524
6 Instruments, flow Control & monitoring 68.447 1.142
7 Instrument, pressure Control & monitoring 89.773 1.497
8 Instrument, vibration Control & monitoring 57.976 0.967
9 Instruments, temperature Control & monitoring 83.516 1.393
10 Monitoring Control & monitoring 22.220 0.371
11 Oil Lubrication system 22.603 0.377
12 Piping Lubrication system 81.856 1.365
13 Radial bearing Pump 25.157 0.420
14 Seals Pump 201.638 3.363
15 Subunit Pump 61.296 1.022
16 Valves Pump & control 79.302 1.323

Table 7
Critical components and criticality measures for the three overriding failure modes

No. Maintainable item External leakage Abnormal instrument Service problems Total relative
(%) reading (%) (%) contribution (%)

1 Actuating device 0.000 0.510 0.620 1.130


2 Casing 0.900 0.000 0.050 0.950
3 Control unit 0.310 1.180 0.260 1.750
4 Filter 0.310 0.000 0.460 0.770
5 Instruments, general 0.000 1.130 0.510 1.640
6 Instruments, flow 0.050 3.950 0.210 4.210
7 Instrument, pressure 0.240 4.900 0.560 5.700
8 Instrument, vibration 0.050 2.410 0.560 3.020
9 Instruments, temperature 0.100 4.540 0.690 5.330
10 Monitoring 0.000 1.140 0.180 1.320
11 Oil 0.260 0.000 0.260 0.520
12 Piping 3.830 0.000 0.410 4.240
13 Radial bearing 0.080 0.000 0.100 0.180
14 Seals 13.270 0.000 0.360 13.630
15 Subunit 0.340 0.150 0.460 0.950
16 Valves 2.660 0.000 1.280 3.940
Total 22.40 19.91 6.97 49.28

and elementary systems, or (5) operation error; committed the design and operating errors, which are: (1) fault
when operating the components and elementary system avoidance/prevention; (2) fault tolerance; (3) fault removal.
previously found fit for service (Villemeur, 1992). There- For design optimization of the sea water lift centrifugal
fore, the five failure categories can be utilized to the nature pump, fault avoidance should lie with as follows:
of the failure causes. The knowledge on common-cause
failures derived from the operating experience of industrial  Control instrument failures and abnormal instrument
systems can help us to estimate their natures. The statistical reading emphasize wrong diagnosis during incidents or
data obtained by analyzing event reports of 933 incidents accidents. A control system equipped by proactive
corresponding to 84 common-cause failures over a period adaptive fault detection technique is able to identify
of 218 reactor-years specifies how common-cause failures type and location of faults (sensors and mechanical
are distributed among the previously defined categories components) and then make a proper decision for fault
(Villemeur, 1992). The results derived from incident recovery.
statistics emphasize to design and operating errors. There-  The configuration of the mechanical components (seals,
fore, the five causes could belong to the design and valves and pipes) and their connections in the sea water
operating errors. There are three techniques for decreasing lift centrifugal pump unit is such that the failure of one
ARTICLE IN PRESS
V. Ebrahimipour et al. / Journal of Loss Prevention in the Process Industries 19 (2006) 542–552 549

Table 8
Causes and their contributions in the three failure modes

No. Causes External leakage Abnormal instrument Service problems Total relative
(%) reading (%) contribution (%)

1 Blockage/plugged 0.670 1.440 0.720 2.830


2 Breakage 0.510 0.770 0.310 1.590
3 Burst 0.000 0.000 0.000 0.000
4 Clearance/alignment failure 0.300 0.260 0.050 0.610
5 Contamination 0.050 0.000 0.560 0.610
6 Corrosion 0.820 0.510 0.770 2.100
7 Electrical failure-general 0.000 0.620 0.260 0.880
8 Erosion 0.150 0.000 0.000 0.150
9 Fatigue 0.150 0.050 0.000 0.200
10 Faulty signal/indication/ 0.000 3.340 0.150 3.490
alarm
11 Instrument failure-general 0.310 7.740 2.000 10.050
12 Leakage 9.600 0.210 0.360 10.170
13 Material failure-general 0.930 0.000 0.360 1.290
14 Mechanical failure-general 8.980 0.670 1.490 11.140
15 Misc.external influence 0.920 0.460 1.130 2.510
16 Out of adjustment 0.000 3.390 0.360 3.750
17 Wear 1.440 0.410 0.560 2.410
Total 24.83 19.87 9.08 53.78

of them induces the failure of the whole unit or several where P, v, h and E(P) are pressure, velocity, elevation and
parts of unit. This risk may be concealed by an pressure energy drop, respectively (Cheremisinoff, 2000).
apparently satisfactory redundancy level for instance, E(P) shows the amount of energy generated by a
the use of double mechanical seals, piping and centrifugal pump in the form of a pump head discharge.
emergency isolation valves to prevent a serious leak at In order to identify the amount of energy needed, the
the pump. system energy is regarded. Based on the pump head-
 The design of the four components is such that access to discharge and system head curves (hloss), the point at which
them is difficult and tagging, disassembly and reassem- the centrifugal pump unit should operate to produce the
bly operations are problematic. Therefore, the prob- amount of flow and provide the energy, is selected.
ability of operational errors will obviously increase. According to the energy equation for a system fluid, total
energy in terms of head and head loss is defined as follows:
In order to decrease operating errors through fault P1 =Y þ v21 =2g þ h1 ¼ P2 =Y þ v22 =2g þ h2 þ hloss . (13)
tolerance and removal techniques, we should identify the
impact of the abnormality of the sea water lift centrifugal For steady state, flow v1 ¼ v2 therefore, pressure energy
pump on the process. Therefore, the sea water lift drop includes elevation energy and momentum pressure
centrifugal pump is further analyzed to know about the energy. The pressure energy drop is proportional to flow
pertinent fault propagating during operating period and rate (q) defined as hloss ¼ kq2 (Fluid Mechanics, 2004). The
how we can recognize it. In general, a pump is a device that abnormality/failure of the pump increases risk in the plant
converts electrical energy to mechanical energy. The in two ways: (1) directly, by the immediate consequences of
centrifugal pump is one the most widely used pumps for the pump failure, such as leaks or spills of hazards material
transferring liquids. This is for a number of reasons. or physical damage from failed parts; (2) indirectly, by
Centrifugal pumps are very quiet in comparison to other fluctuating flow in system through pressure energy drop
pumps and have a relatively low operating and main- resulted from efficiency loss (Lees, 1996). Pump efficiency is
tenance costs. The principal operation of a sea water lift a function of the head discharge (E(P)). The efficiency of
centrifugal is to convert fluid velocity into pressure energy the centrifugal pump can be computed by finding the ratio
to lift water. The total pressure energy needed to lift the of the power provided to the pump and the power
water across a system consists of two key components produced by the unit. There are two types of power
which are: (1) a static pressure difference due to the transformations that occur in the centrifugal pump:
elevation of the fluid; (2) a pressure differential due to the (1) electrical power, which is transferred into mechanical
charge of momentum. This can be shown by Eq. (12): power via the pump motor; and (2) mechanical power, that
rotates the shaft, turns the impeller and transfer power to
P2 v2 P1 v2 the fluid. The two power transformations are affected by
þ h2 þ 2 ¼ þ h1 þ 1 þ EðPÞ, (12)
g 2g g 2g the hydraulic losses which relate to the construction of the
ARTICLE IN PRESS
550 V. Ebrahimipour et al. / Journal of Loss Prevention in the Process Industries 19 (2006) 542–552

pump and are caused by the friction between the fluid and
∆P4 the walls, mechanical losses in the bearings and seals which
reduce the power transmission, and volumetric losses
due to leakage of fluid between the back surface of the
∆P3 impeller hub plate and the casing or through other pump
components. These losses and pump efficiencies can be
Flow rate

shown by the following equation:


∆P2
q w p  p1
Zv ¼ ; Zh ¼ and Zm ¼ , (14)
q þ q1 w þ w1 p
∆P1
where q, ql, w, wl, p and pl are volume flow out of the pump,
leakage volume flow, specific work from the pump, specific
work lost due to hydraulic friction, power transferred from
the motor to the shaft and power lost in transmission.
Table 8 shows that leakage (10.17%) and mechanical
Y1 Y2 Y3 Y4 Travel
failure (11.14%) are the most major causes in the sea water
Fig. 2. Flow vs. travel for equal-percentage valve. lift centrifugal pump which directly make mechanical and
volumetric losses that are highly correlated to the pump
performance and pump head discharge. The direct result of

Fig. 3. The layout of refinery line including centrifugal pump, valve actuator and heat exchanger.
ARTICLE IN PRESS
V. Ebrahimipour et al. / Journal of Loss Prevention in the Process Industries 19 (2006) 542–552 551

Flow rate in Valve Actuator (CV303) one of the six leading causes1 of runway reaction,
3500
mentioned on typical case histories taken courtesy of
3000 Barton and Nolan in 1984, which is a major subgroup of
2500 fires and explosions in petrochemical plant (Banerjee,
Kmol/Hour

2000 2003). Figs. 4 and 5 show the fault propagation starting


1500 from pump (GA201) up to heat exchanger (E301).
1000
4. Conclusion
500
0 The integrated model proposed in this paper identified
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58
the sea water lift centrifugal pump unit as a weak point and
Time
reliability as the improving technical character of the pump
Fig. 4. Jump in flow rate caused by pressure drop. unit. Furthermore, the model concluded design and
operating errors (systematic error) to be the nature of the
five causes mentioned in Table 8. Design optimization can
Temperature of outflow of Heat exchanger (TI301) be performed in regard to the four critical components
150 mentioned in Table 7. We identified that the centrifugal
Temperature degree (C)

145 pump abnormality during operating period substantially


140
makes flow fluctuation. This fault is amplified by valve and
pipeline and then initiates hazards, such as runway reaction
135
which ultimately could make fire ball or explosion, in a
130 petrochemical process. The model showed that the cost-
125 effective way to eliminate the fault while system is
120 operating online fault detection through valve actuator
115
flow analysis. The fault can be detected when an
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 unexpected jump in valve flow is discovered by operator.
Time In summary, this paper presents a standard methodology
for the assessment and improving of offshore pump units’
Fig. 5. Influence of the jump of flow input on thermal output of heat
exchanger. performance based on dependability. The results of such
studies would help designers and engineers in order to
know about the hazards which are initiated by the worst
equipment. This knowledge can facilitate the engineering
this pump abnormality in the process is flow fluctuation. strategies for inherently safer process design. The modeling
Flow fluctuation has a high impact on system pressure approach may be extended to include external units
energy drop (hloss ¼ kq2) so that it can be considered as a (competitors) to identify standings and weak and strong
fault which propagates in system very fast and easily, machine factors in the big picture. As the systems, such as
especially through valve and pipe lines. Automated control oil & gas industry, grow and the complexity explodes this
process is so vulnerable to the fault. The fault can topic elevates in importance.
be amplified by the closed loop control process and
it can develop into malfunction of the loop. The References
closed loop may alternatively hide the fault from being
observed until eventually developed to a state where Azadeh, A., Ebrahimipour, V. (2002). An integrated approach for
assessment of manufacturing sectors based on machine performance:
process failure is inevitable. The impact of the system
The cases of automotive and food and beverages industries. Proceeding
pressure drop caused by the centrifugal pump abnormality of the second international conference of the manufacturing complexity
can be detected online by valve actuator flow analysis. network, University of Cambridge, Cambridge UK, (pp. 468).
Fig. 2 shows the inherent valve characteristic ‘‘flow vs. Azadeh, A., & Ebrahimipour, V. (2004). An integrated approach for
travel’’ for equal-percentage valve plotted for various assessment and ranking of manufacturing systems based on machine
pressure drops. performance. International Journal of Industrial Engineering, 11(4),
349–363.
It can be observed that how system pressure drop makes Banerjee, S. (2003). Industrial hazards and plant safety. New York: Taylor
a jump in the flow passing through valve cross section. We & Francis Books.
traced this kind of fault (pressure drop) through valve Beeson, S., & Andrews, J. D. (2003). Importance measures for non-
actuator flow analysis in a simulated refinery line by visual coherent-system. IEEE Transaction on Reliability, 52(3), 301–309.
omega simulation 2. Fig. 3 depicts the simulated layout of 1
the centrifugal pump, valve actuator and heat exchanger Typical case histories taken courtesy of Barton and Nolan introduce
the leading causes of runways as follows: (1) inadequate cooling, (2)
connected to a reactor in a refinery line. The fault decreases incorrect charging, (3) incorrect agitation, (4) impurity exotherm, (5)
inflow rate inputting to heat exchanger and then increases unknown exotherm/decomposition, (6) inadequate temperature control
its temperature outflow. The high temperature outflow is (Banerjee, 2003).
ARTICLE IN PRESS
552 V. Ebrahimipour et al. / Journal of Loss Prevention in the Process Industries 19 (2006) 542–552

Birolini, A. (2004). Reliability engineering; theory and practice. New York, Ingruber, B. (1999). Reliability and efficiency aspects of harmonic-control
Berlin, Heidelberg: Springer. amplifiers. IEEE Microwave and Guide Wave Letters, 9(11), 464–466.
Blanchard, B. S. (1998). System engineering management (2nd ed.). NJ: Lees, F. P. (1996). Loss prevention in the process industries. Loughborough,
Wiley & Sons. Oxford, UK: Reed Educational and Professional Publishing Ltd.
Chen, L. H., Kao, C., Kuo, S., Wang, T. Y., & Jang, Y. C. (1996). Moustafa, M. S., Abdel Maksoud, E. Y., & Sadek, S. (2004). Optimal
Productivity diagnosis via fuzzy clustering and classification: An major and minimal maintenance policies for deteriorating systems.
application to machinery industry Omega, 24(3), 309–319. Reliability Engineering and System Safety, 83, 363–368.
Cheremisinoff, N. P. (2000). Hand book of chemical processing equipment. Nakajima, S. (1991). Maintenance management and control. (2nd ed.).
Woburn: Butterworth-Heinemann. Handbook of industrial engineering.
Chochran, J. K., Arvindh, M., & Krishnamurthy, V. (2001). Generic OREDA Participants. (2002). OREDA Handbook (4th ed.). Trondhim:
Markov models for availability estimation and failure characterization OREDA Participants.
in petroleum refineries. Computers and Operations Research, 28, 1–12. Tompkins, J. A., Bozer, W., Frazelle, E. H., Tanchoco, J. M. A., &
Corrie, E. V. (1986). Cost control begins with budgeting in modern cost Trevino, J. (1996). Facilities planning (2nd ed.). New York: Wiley.
engineering. New York: McGraw Hill. Villemeur, A. (1992). Reliability, availability, maintainability and safety
Ebrahimipour, V., Suzuki, K. (2004). An artificial intelligence approach to assessment. Chichester: Wiley Ltd.
assess and anticipate the impact of safety management on the chemical Wang, W., Xu, Z., & Lu, J. W. (2003). Three improved neural network for
system’s performance from the view point of efficiency and maintain- air quality forecasting. Engineering Computations, 20(2), 192–210.
ability. Conference proceeding of the 10th APPCChE congress, Japan. Yang, S. K. (2003). A conditional-based failure- prediction and
Fluid Mechanics (2004). The engineering tool box. http://www.engineer- processing—Scheme for preventive maintenance. IEEE Transactions
ingtoolbox.com/21_614.html on Reliability, 52(3), 373–383.
Fordham, R. J., Baxter, D., Hunter, C., & Malkow, T. (2003). The impact Zhu, J. (1998). Data envelopment analysis vs. principal component
of increasing demand for efficiency and reliability on the performance analysis: An illustrative study of economic performance of chinese
of waste-to-energy plants. Materials at High Temperatures, Institute for cities. European Journal of Operation Research, 111, 50–61.
Energy, 20, 19–25.
Guimaraes, T., Matensson, N., Stahre, J., & Igbaria, M. (1999).
Empirically testing the impact of manufacturing complexity on
performance. International Journal of Operations and Production Further reading
Management, 19(12), 1254–1269.
Henley, E. J., & Kumamoto, H. (1981). Reliability engineering and risk Charnes, A., Cooper, W. W., & Rhodes, E. (1995). Measuring efficiency of
assessment. New Jersey: Prentice-Hall. decision making units. European Journal of Operation Research, 3,
Hinkel, T. (2004). A Pump reliability and efficiency program; how one 429–444.
company used high-performance thermoplastic to upgrade the hydraulic US Department of Labor Multifactor Productivity Measures for Three-
performance of 1200 pumps. Wear Products Group, EGC crop: digit SIC Manufacturing Industries. (2000). Report 948, December.
Houston, TX. US Department of Labor, Bureau of Labor Statistics.

You might also like