You are on page 1of 8

Efficiency Prediction of Control Room Operators

Based on Human Reliability Analysis and Dynamic


Decision-Making Style in the Process Industry
Esmaeil Zarei,a,b Iraj Mohammadfam,a Mostafa Mirzaei Aliabadi,a Ali Jamshidi,c and
Fakhradin Ghasemia
a
Department of Occupational Health and Safety Engineering, School of Public Health and Research Center for Health Sciences,
Hamadan University of Medical Sciences, Hamadan, Iran; mohammadfam@umsha.ac.ir (for correspondence)
b
Student Research Center, Hamadan University of Medical Sciences, Hamadan, Iran
c
Railway Engineering Section, Delft University of Technology, Stevinweg, The Netherlands
Published online 14 September 2015 in Wiley Online Library (wileyonlinelibrary.com). DOI 10.1002/prs.11782

Process industries have a potential for the occurrence of Statistical analysis of 489 major accidents from 1985 to
major accidents. These accidents can have severe adverse 2001, in the European Union revealed that the petrochemical
effects on human health and the environment and they can industries ranked second (17%), after the general chemicals
cause extensive damage to equipment and buildings. During (32%) in the number of accidents, and 70% of the major acci-
major process upsets, central control rooms are among the dents took place when the plants were in normal operation
most stressful workplaces in the world. Therefore, Human status [4]. Analysis of 1,129 reported accidents in nine gas
Reliability Analysis and Dynamic Decision-Making Styles refineries in Iran from 2001 to 2007, revealed that the more
(DDMSs) play an important role in safety management in than 94% of accidents had human error as one of the key
these industries. This study employs the intelligent Adaptive causal factors rather than systemic failure [5]; moreover,
Neuro Fuzzy Inference System model associated with two results show that the human error is involved; in over 90%
questionnaires along with Cognitive Reliability Error Analysis of failures in the nuclear industry [4,6,7], over 80% of the fail-
Method to analyze the Human Reliability Influencing Factors ures in the chemical and petrochemical industries, over 75%
(HRIFs) and DDMSs of the control room operators and to marine casualties [8,9], and over 70% of aviation accidents
determine the efficiency of operators as well as their domi- [10]. Statistics mentioned shows that the causes of many acci-
nant and efficient decision-making styles. Nine influencing dents in process industries, directly or indirectly were related
factors on human reliability and five DDMSs are evaluated to human error. In addition, studies revealed that a signifi-
and the correlation between the HRIFs is investigated. Effi- cant number of worker’s behaviors were unsafe, which is
ciency of the operators, according to the HRIFs, is determined one of the main causes of accidents in the process industry
and they are ranked. Next, the most dominant and efficient [11]. Therefore, it can be concluded that the human reliabil-
of the DDMSs among the operators was identified. Finally, ity, plays a major role in the safe performance of individual
an intelligent algorithm for determining the efficiency of a and subsequently in the safe operation of the process indus-
control room’s operators is developed. VC 2015 American Institute tries [12]. Human Reliability Analysis (HRA) is an analysis
of Chemical Engineers Process Saf Prog 35: 192–199, 2016 technique based on human performance studies and consid-
Keywords: human reliability analysis; decision making; ers many Human Reliability Influencing Factors (HRIFs) in
efficiency prediction; process industry order to predict human performance in contributing to acci-
dents, error, and risk of tasks [13,14].
Dynamic Decision-Making Style (DDMS) is interdepend-
INTRODUCTION ent decision-making process that takes place in an environ-
The petrochemical industries must safely manage the ment that changes over time either due to the previous
processing of hazardous materials or disastrous used during actions of the decision maker or due to events that are out-
process, potentially disastrous accidents can occur [1]. The side the control of the decision maker [15,16]. There are a lot
increasing scale of process plants handling ever-increasing of tasks in the control rooms of petrochemical industries
quantities of highly explosive and flammable materials cre- with such conditions [1] in which operators must gather
ates the potential for large scale accidents if the facilities are more information and are forced to take action at the same
not properly designed and safely operated [2,3]. time. Information obtained and decisions made are con-
stantly changing during their shiftwork. The accuracy of deci-
This work was supported by Student Research Center of Hamadan
sions of the operators are crucially important in terms of
University of Medical Sciences. safety, because failure in their decisions may cause deviation
and flaws in the normal chemical operations and eventually
C 2015 American Institute of Chemical Engineers
V may lead to catastrophic events. The petrochemical

192 June 2016 Process Safety Progress (Vol.35, No.2)


Table 1. Some example of questions of HRIFs and DDMSs questionnaires.

Response
HRIFs Questionnaire
1. Does the noise in the control Never Little Somewhat Very
room makes you uncomfortable?
2. Do you have any pain or fatigue Never Never Somewhat Very
in your eyes during the weekday?
3. How would you evaluate the No Low Moderate High
current workload level?
4. Is it possible to adjust height of No Rarely Often Yes
your work station for doing different
tasks and operations?
5. Are there situations where you have No Rarely Often Yes
to violate safety rules due to working pressure?
DDMSs Questionnaire
1. Applying of creativity and innovation
2. Applying rules and regulations
3. Time pressure
4. Work pace
5. Need for planning

industries are developing with remarkable growth in Iran Instruments


and failures are expected consequences of the increasing
complexity of these industries. Parallel to the development HRIFs
of the industries and the technologies and the use of reliable Data were collected via a self-administered questionnaire
systems and engineering designed, the reliance of such sys- about the HRIFs. The initial design of the questionnaire was
tems can cause biases and failures in decision making [17]. based on taxonomy of performance influencing factors for
There are many studies that discuss the importance of HRA of emergency tasks [20]. A questionnaire of 40 items
human factors and their impact on human reliability. How- was used in order to assess HRIFs in the present study. The
ever, to the best knowledge of the authors, there has been questionnaire uses a 1–4 scale, with most response points
no study conducted on human reliability and its relationship labeled: 1 5 Yes, 2 5 Often, 3 5 Seldom, 4 5 Not at all. For
with decision-making style in the petrochemical plants. This items related to satisfaction during shift work, 1 5 com-
study employs the intelligent Adaptive Neuro Fuzzy Infer- pletely satisfied and 4 5 completely dissatisfied; for items
ence System (ANFIS) model associated with two question- related to work load rate and psychological pressure (stress),
naires. ANFIS is a method based on a set of fuzzy IF–THEN 1 5 High and 4 5 Not at all.
rules for membership functions that uses a hybrid learning
algorithm based on a collection of training data. Hence, Validity and Reliability Assessment of the Questionnaire
ANFIS is considered to be a universal estimator [18]. The Validity. After design of the questionnaire, the face and
main objectives of this study were: content validity was tested. Its validity was determined by
submitting the questionnaire to 15 operators and academics
 Determining, measuring, and assessing of the HRIFs using
and long-experienced industrial experts. The questionnaire
Cognitive Reliability Error Analysis Method along with a
was then corrected and revised.
self-administered questionnaire that including nine factors
as follow; (1) Risk Degree, (2) Physical and Mental Health Reliability. In the present study, reliability analysis was
Status, (3) Communication, (4) Stress Factors, (5) Skill carried out by calculating the Cronbach’s Alpha coefficient.
Level, (6) Anthropometry, (7) Human Error, (8) Job Satis- The Cronbach’s Alpha coefficient evaluates the internal con-
faction, and (9) Inherent Psychological Features (IPF). sistency of the questionnaire. The alphas can range from 0 to
 Survey and analysis the DDMSs of operators by Driver’s 1.0, where 1.0 indicates perfect internal consistency. Internal
questionnaire [19]. consistency measures how well all the questions on a scale
 Applying of the ANFIS to predict the efficiency of the
are correlated with each other, and high inter-item correla-
operators and their DDMSs according to the HRIFs.
tions suggest that all the questions measure the same factor
of interest. Table 2 shows results of calculation of the
METHOD AND MATERIALS Cronbach’s Alpha values.

Setting and Participants DDMSs


This cross-sectional study was performed on control To assess DDMSs of the operators, the standard question-
rooms in one of the largest petrochemical complex in Iran. naire of Driver was used [19]. This questionnaire measures
This complex consists of six large companies and each com- five DDMSs: (1) Decisive style, (2) Hierarchic style, (3) Sys-
pany has different units and several control rooms. In this temic style, (4) Flexible style, and (5) Integrative style. It con-
complex, all operators were male and the sample size was tains 17 factors: (1) Time pressure, (2) Excessive work load,
62 operators who answered both the HRIFs and DDMSs (3) Work pace, (4) Effectiveness on the system, (5) Prefer-
questionnaires. Table 1 shows some example of questions ence to use logical reasoning, (6) Applying creativity and
about HRIFs and DDMSs questionnaires. innovation, (7) Applying rules and regulations, (8) Lack of

Process Safety Progress (Vol.35, No.2) Published on behalf of the AIChE DOI 10.1002/prs June 2016 193
Table 2. Cronbach’s alpha value of the questionnaire of the HRIFs.

0.71 0.79 0.79 0.63 0.64 0.71 0.62 0.79


Job satisfaction Communication Physical and Stress Skill Anthropometry Risk Total
mental health factor level degree Cronbach’s
alphas

Figure 1. Overview of the proposed intelligent algorithm. [Color figure can be viewed in the online issue, which is available
at wileyonlinelibrary.com.]

familiarity with working rules and regulations, (9) The con- tionnaire analysis guidelines. All statistical analysis was car-
trol limits, (10) Power rate, (11) Need for planning, (12) ried out using SPSS 16 (SPSS Inc., Chicago, IL) and MINITAB
Complexity of the inputs data, (13) Uncertainty, (14) Involve- at 5% significance level.
ment degree with others, (15) Need for reason and logic in
doing assignments, (16) Preference to use several methods, PROPOSED INTELLIGENT ANFIS ALGORITHM
and (17) Involvement of others. The flow chart in Figure 1 shows the steps of the pro-
The participants are asked to judge the degree of impor- posed methodology.
tance of each factor, relying on their experience in dealing The following sections describe the details of every step.
with the emergency situations. Degree of importance of each
factor is from 1 to 7, where 1 indicates the lowest and 7 indi- Input-Output Selection and Data Collection
cates the highest importance. ANFIS is used to estimate the HRIFs according to the deci-
Finally, after the participants complete the questionnaires, sion-making styles. The inputs for ANFIS are the data col-
DDMS of the each person is determined according to ques- lected with the questionnaire survey and to ensure robust

194 June 2016 Published on behalf of the AIChE DOI 10.1002/prs Process Safety Progress (Vol.35, No.2)
Table 3. The frequency of demographic characteristics of control room operators.

Job Experience Education


(year) Degree Age (year) Height (m) Wight (kg)
Number (%) Groups Number (%) Groups Number (%) Groups Number (%) Groups Number (%) Groups
9 (15) 1–5 30 (48) Diploma 20 (32) 25–35 25 (40) 155–170 9 (15) 55–65
15 (24) 5–10 9 (15) Associate Degree 31 (50) 35–45 32 (52) 170–185 15 (24) 65–75
25 (40) 10–15 21 (34) Bachelor 11 (18) 45–55 5 (8) >185 23 (37) 75–85
10 (16) 15–20 2 (3) Master 10 (16) 85–100
3 (5) >20 5 (8) >100

Table 4. Frequency of dynamic decision-making style (DDMSs) of operators.

DDMSs H I H-I D I-F F-D F S


Frequency 30 14 10 2 3 2 3 0

H, Hierarchic; I, Integrative; D, Decisive; F, Flexible; S, Systematic.

and fair testing of the ANFIS, data were randomly divided


into two sets: train and test data, where 35% of the collected Fi 5Pi =ðPðANFISÞi 1Shi Þ i51; :::; n (5)
data were chosen as the test data. To obtain the network
with the best quality of training and assessment, different
parameters including inference functions and radius influ- Participants Ranking and Determine the Best DDMS
ence of clustering are tuned. HRIFs are the outputs of the After calculation of ESs of the participants, they are
ANFIS analysis. ranked based on their scores (Fi) and the best DDMS (Best is
the smallest difference between expected level of perform-
ance and actual performance) as determined among the effi-
cient participants. This style is compared with the dominant
Calculating the Efficiency Score style determined from DDMSs questionnaire and thus the
Adopting a comparative assessment strategy, the effi- efficiency or inefficiency of dominant style can be
ciency of an operator can be interpreted as the difference determined.
between each individual and the average expected perform-
ance of the group. The expected level of the performance is RESULTS AND DISCUSSION
determined by the trained ANFIS. Therefore, the difference
between actual performance and ANFIS output can be DDMS and HRIFs Results
served as an efficiency score (ES). To obtain the ESs of the The characteristics of study population are shown in
participants, error (Ei) between the actual outputs and ANFIS Table 3. The mean age of study population was 38.52 years
outputs is calculated for all participants (range: 27–54 years) and the most number of operators had
diploma degree 30 (48%) and their average job experience
Ei 5PrealðiÞ 2PANFISðiÞ i51; :::; n (1) was 11 years (range: 3–23 years; Table 3).
DDMS of the operators was obtained by the questionnaire
of Driver and frequency of decision-making styles extracted
where PrealðiÞ is the actual HRIF for participant i and PANFISðiÞ by means of descriptive statistics (Table 4). According to
is the estimate of HRIF for the participant i obtained from Table 4, the hierarchical style is the dominant DDMS among
ANFIS. the operators. In case if the frequency of two styles is the
Then, normalizing by the reference value leads to
same, a combination of two styles may be considered as
dominant style. Systematic and Decisive styles had the lowest
Ei0 5Ei =PðANFISÞ i i51; :::; n (2) frequency (F 5 0, F 5 2, respectively) among all styles.
Among the combined styles, H-I (F 5 10) and F-D (F 5 2)
where Ei0 is the relative error of participant i. Assume that styles had highest and lowest frequency, respectively
the participant k has the largest value of the Ei0 , thus we (Table 4).
have: In the next step, average of HRIFs for all DDMS calcu-
lated and the dominant style based on influencing factors is
Ek0 5maxðEi0 Þ (3) determined. Table 5 shows the dominant DDMS relying on
HRIFs. In the hierarchical style, operator makes maximum
Let us define transition size for each of the participant use of the data for adopting a single best response. As a
(Shi) as follows: result, they had high skill level. Low focus and lack of plan-
ning are disadvantages of the flexible style [17]. As an illus-
Shi 5Ek  PðANFISÞi =PðANFISÞk i51; :::; n (4) tration, in the present study for skill level and human error,
dominant styles were specified hierarchical and flexible
Finally, calculating the ESs (Fi) is as follows: styles, that the results of present study confirm this (Table 5).

Process Safety Progress (Vol.35, No.2) Published on behalf of the AIChE DOI 10.1002/prs June 2016 195
Table 5. Dominant dynamic decision-making styles (DDMSs) according to HRIFs.

DDMSs I-F F F-D H F F-D I-F D I-F


HRIFs* Job Human Anthropometry Skill Stress IPF Communication Physical & Risk
satisfaction Error Level Factors Mental Health Degree
Status

*Human Reliability Influencing Factors.

Table 6. Significant relationships between HRIFs and


demographic characteristics.

Human Job Risk Skill


Error Satisfaction Degree level
Education 0.011 0.028
degree
Age 0.029 0.041
Job 0.034 0.026
experience

Table 7. Relationship and correlations between HRIFs. Figure 2. .Minimum MAPE obtained from ANFIS training for
the operators. [Color figure can be viewed in the online
Risk Human Human issue, which is available at wileyonlinelibrary.com.]
Risk Degree and Error Error
Degree and Human and Stress and Skill
HRIFs Skill Level Error Factors Level
and long experienced industrial experts. These categories
PCC* 0.85 0.89 0.79 0.88 were considered as input variables (IV). In addition, nine
P-Value 0.022 0.023 0.021 0.031 factors influencing the HRA were considered as output varia-
bles (OV). Factor analysis results reveal the following inputs:
*Pearson Correlation Coefficients.
 Input variable 1: Effectiveness on the system, Preference
to use logical reasoning, Applying of creativity and inno-
vation, Applying rules and regulations.
Table 6 shows significant relationships between HRIFs  Input variable 2: Time pressure, Excessive workload,
and demographic characteristics. The statistical analyses were Work pace.
conducted with the Kruskal–Wallis test and results indicate  Input variable 3: Lack of familiarity with working and the
that education degree variable have a significant relationship rules and regulations, Control limits, Power rate, Need for
with human error (P 5 0.011) and job satisfaction planning.
(P 5 0.028), age variable have a significant relationship with  Input variable 4: Complexity of the inputs (data), Uncer-
risk degree (P 5 0.029) and skill level (P 5 0.041), job experi- tainty, Involvement degree with others
ence have a significant relationship with job satisfaction  Input variable 5: Need for reason and logic in doing
(P 5 0.034) and skill level (P 5 0.026). The relationship assignments
between HRIFs and other demographic characteristics of  Input variable 6: Preference to use several methods,
operators were not significant (P > 0.05). Involvement of others
The relationship and correlation between the HRIFs by
means of Kruskal–Wallis test and Pearson Correlation Coeffi- In the training phase, in general, 24 overall structures of
cients (PCC) were investigated (Table 7). The results indicate ANFIS are considered for each of OVs. Since the parameter
that there is significant relationship between factors of risk of effectiveness clustering possessing a value between (1,
degree and skill level, risk degree and human error, human 0.1), to get the lowest relative error, this numerical range
error and stress factors, skill level and human error (P < searched at steps of 0.1. Generally for each of the 24 overall
0.05). The PCC and P-values between these factors are structures of the proposed ANFIS, 819 runs were done and
shown in Table 7. The strongest relationship is related to risk at the end, the best structure of the ANFIS, which had the
degree and human error. lowest Mean Absolute Percentage Error (MAPE), was chosen.
The results of ANFIS training error is shows in Figure 2.
ANFIS Result
ES of Operators
Input and Output Variables for the ANFIS Tool in Algorithm By implementation of the ANFIS for each nine OVs, a
In the present study, based on closeness and similarity numerical value is obtained for each operator. According to
rate of the factors and also to avoid of the scattering and bet- the mentioned method for calculation of operators’ ESs in
ter understand the results, the 17 factors influencing the the previous section, by having an output and the output
DDMS were classified into six categories using Factor Analy- obtained from ANFIS and follow-up the mentioned steps,
sis Method and expert opinions that consists of 13 academics numerical value of the ESs are calculated (Table 8).

196 June 2016 Published on behalf of the AIChE DOI 10.1002/prs Process Safety Progress (Vol.35, No.2)
Table 8. Efficiency values of operators obtained from ANFIS according to output variables (OV).

Operator N DDMS OV1 OV2 OV3 OV4 OV5 OV6 OV7 OV8 OV9
1 H 1.00 0.71 0.46 0.49 0.42 0.54 0.58 0.71 0.59
2 H 1.00 0.66 0.56 0.46 0.43 0.68 0.58 0.82 0.37
3 H 1.00 0.69 0.45 0.49 0.64 0.65 0.41 0.83 0.51
4 H 1.00 0.70 0.59 0.50 0.52 0.70 0.74 0.71 0.56
5 H 1.00 0.66 0.59 0.55 0.51 0.64 0.74 0.71 0.56
6 H 1.00 0.72 0.52 0.53 0.71 0.64 0.59 0.71 0.65
7 H 1.00 0.76 0.54 0.42 0.65 0.74 0.58 0.71 0.69
8 H 1.00 0.66 0.64 0.51 0.74 0.89 0.73 0.72 0.56
9 H 1.00 0.69 0.50 0.51 0.51 0.75 0.41 0.83 0.63
10 H 1.00 0.72 0.50 0.54 0.51 0.65 0.58 0.55 0.70
11 H 1.00 0.67 0.50 0.46 0.68 0.64 0.74 0.55 0.63
12 H 1.00 0.68 0.68 0.52 0.58 0.70 0.57 0.55 0.71
13 H 1.00 0.54 0.58 0.59 0.50 0.75 0.58 0.83 0.56
14 H 1.00 0.69 0.50 0.53 0.40 0.68 0.58 0.55 0.56
15 H 1.00 0.63 0.54 0.53 0.70 0.73 0.59 0.83 0.61
16 H 1.00 0.67 0.54 0.43 0.63 0.73 0.58 0.79 0.56
17 H 1.00 0.63 0.37 0.52 0.50 0.63 0.40 0.83 0.62
18 H 1.00 0.71 0.54 0.44 0.57 0.68 0.58 0.55 0.63
19 H 1.00 0.65 0.55 0.46 0.51 0.68 0.58 0.55 0.47
20 H 1.00 0.55 0.49 0.69 0.51 0.56 0.73 0.79 0.61
21 H 1.00 0.48 0.44 0.47 0.57 0.58 0.58 0.71 0.56
22 H 1.00 0.47 0.44 0.48 0.57 0.61 0.58 0.72 0.56
23 H 1.00 0.57 0.54 0.53 0.39 0.63 0.41 0.71 0.63
24 H 1.00 0.55 0.61 0.45 0.57 0.46 0.58 0.55 0.51
25 H 1.00 0.67 0.57 0.45 0.67 0.69 0.57 0.73 0.62
26 H 1.00 0.60 0.49 0.53 0.56 0.59 0.74 0.56 0.56
27 H 1.00 0.75 0.49 0.53 0.57 0.62 0.58 0.72 0.68
28 H 1.00 0.59 0.58 0.57 0.56 0.70 0.58 0.56 0.62
29 H 1.00 0.58 0.53 0.51 0.39 0.65 0.59 0.72 0.46
30 H 1.00 0.62 0.53 0.45 0.56 0.73 0.59 0.84 0.69
31 I 1.00 0.72 0.60 0.53 0.56 0.71 0.74 0.56 0.70
32 I 1.00 0.75 0.57 0.52 0.56 0.91 0.41 0.84 0.72
33 I 1.00 0.72 0.48 0.57 0.38 0.71 0.42 0.72 0.38
34 I 1.00 0.68 0.52 0.46 0.65 0.74 0.59 0.56 0.64
35 I 1.00 0.73 0.61 0.62 0.55 0.72 0.59 0.57 0.68
36 I 1.00 0.63 0.43 0.47 0.61 0.56 0.59 0.72 0.66
37 I 1.00 0.76 0.53 0.54 0.38 0.75 0.42 0.84 0.63
38 I 1.00 0.78 0.62 0.50 0.48 0.80 0.74 0.56 0.68
39 I 1.00 0.61 0.42 0.47 0.47 0.64 0.74 0.84 0.56
40 I 1.00 0.64 0.48 0.48 0.38 0.62 0.59 0.56 0.47
41 I 1.00 0.73 0.59 0.63 0.55 0.76 0.68 0.72 0.60
42 I 1.00 0.49 0.52 0.53 0.47 0.64 0.59 0.80 0.63
43 I 1.00 0.66 0.48 0.54 0.72 0.70 0.74 0.80 0.63
44 I 1.00 0.69 0.51 0.61 0.54 0.79 0.59 0.56 0.72
45 H-I 1.00 0.75 0.51 0.33 0.46 0.68 0.59 0.72 0.63
46 H-I 1.00 0.54 0.47 0.54 0.48 0.73 0.59 0.72 0.64
47 H-I 1.00 0.69 0.52 0.49 0.48 0.67 0.59 0.72 0.67
48 H-I 1.00 0.59 0.34 0.53 0.37 0.81 0.59 0.84 0.66
49 H-I 1.00 0.56 0.55 0.61 0.67 0.76 0.75 0.80 0.57
50 H-I 1.00 0.64 0.51 0.64 0.54 0.76 0.74 0.56 0.56
51 H-I 1.00 0.71 0.46 0.57 0.46 0.74 0.68 0.72 0.57
52 H-I 1.00 0.56 0.46 0.32 0.36 0.46 0.42 0.57 0.57
53 H-I 1.00 0.74 0.55 0.53 0.54 0.79 0.59 0.72 0.60
54 H-I 1.00 0.70 0.46 0.48 0.36 0.55 0.59 0.84 0.61
55 F 1.00 0.69 0.55 0.57 0.37 0.69 0.74 0.57 0.64
56 F 1.00 0.65 0.50 0.53 0.45 0.71 0.58 0.73 0.57
57 F 0.50 0.53 0.41 0.86 0.83 0.55 0.52 0.71 0.85
58 I-F 1.00 1.00 0.26 0.59 0.69 1.00 1.00 0.35 0.68
59 I-F 1.00 0.44 0.63 0.45 0.59 0.49 0.26 0.43 0.80
60 I-F 0.59 1.00 0.78 0.58 1.00 0.86 0.57 0.69 0.69
61 F-D 0.67 0.38 0.53 0.43 0.44 0.66 0.51 1.00 0.70
62 D 0.69 0.86 0.86 0.47 0.76 0.39 0.51 0.40 1.00

DDMS, dynamic decision-making style;


OV1, risk degree; OV2, physical and mental health; OV3, communication; OV4, inherent psychological features (IPF); OV5,
stress factor; OV6, skill level; OV7, anthropometry; OV8, human error; OV9, job satisfaction.
Figure 3 shows total efficiencies obtained from ANFIS for using ANFIS is found. Then, with the sum of scores of peo-
each operator, accordingly, ESs of the first 30 operators, ple in the OVs, total rank of each operator was determined.
those who had hierarchical style were more than the other Total operator rankings obtained from ANFIS are shown in
operators. Thus, they had the best performance among all Figure 4.
studied operators. According to the results, dominant DDMS based on maxi-
Operator 26 (ES 5 0.74, Style 5 H) and operator 52 (ES mum efficiency was the hierarchical style (OVs). Prevailing
5 0.51, Style 5 H-I) had the highest and lowest ESs, respec- style that was specified in the previous step (Standard ques-
tively. These findings will help to other operators who had tionnaire of Driver) was hierarchical style that is consistent
lower degree of efficiency, to push their decision-making with ANFIS results.
style toward the first 30 operators, those who had hierarchi-
cal style and subsequently the best performance. CONCLUSION
In the following, the influence of the DDMSs on each of Petrochemical plants are considered as one of the
the OVs is investigated. Average efficiency of operators in mother industries because they can play a major role in the
lieu of each of OVs is shown in Table 9. According to the economic progress of any country. In these plants, human
results, the DDMSs had the highest influence on the OV1 error is recognized as the main cause of accidents. There-
(Risk degree, AE 5 0.98) and lowest influence on the OV5 fore, achieving a higher efficiency and reducing risk of such
(Stress factor, AE 5 0.51). hazardous systems, to a considerable extent, depends on
the level of human reliability. Hence, understanding the fac-
Operators Ranking tors that affect human reliability is a major concern.
In this step, after calculation of ESs of the operator in The present study proposes an intelligent algorithm to
each OV, the rank of each operator among all the OVs analyze the HRIFs and DDMSs. In the first step, through the
standard questionnaire of Driver the hierarchical style was
determined as the prevailing style of operator decision mak-
ing. In the next step, efficiency operator was calculated by
ANFIS method and they were ranked based on ES. The
results of both techniques showed that the dominant style of
the operators of control rooms is hierarchical. In addition,
DDMS had the highest and lowest influence on risk degree
and stress factor, respectively.
The first 30 operators, those who had hierarchical style,
had the best performance among all studied operators.
Therefore, it is better that other operators be encouraged to
adopt this decision-making style during doing their tasks and
operations.
Figure 3. Total efficiencies obtained from ANFIS for each Since often operations that have been carried out in con-
operator. [Color figure can be viewed in the online issue, trol rooms of the petrochemical industry are similar to each
which is available at wileyonlinelibrary.com.] other, so the findings of the present study can be general-
ized and used in other petrochemical industries in
everywhere.

Table 9. Average efficiency of the operators for each of the output variables (HRIFs).

Output Variables OV1 OV2 OV3 OV4 OV5 OV6 OV7 OV8 OV9
Average Efficiency 0.98 0.65 0.52 0.69 0.51 0.54 0.68 0.61 0.62

Figure 4. Total ranking of operators obtained from ANFIS. [Color figure can be viewed in the online issue, which is available
at wileyonlinelibrary.com.]

198 June 2016 Published on behalf of the AIChE DOI 10.1002/prs Process Safety Progress (Vol.35, No.2)
ACKNOWLEDGMENTS 10. R.L. Helmreich, On error management: Lessons from avi-
The authors announce no conflicts of interest. The authors are ation, BMJ: Br Med J 320 (2000), 781–785.
grateful for the valuable comments and suggestions from the 11. I. Mohammadfam and E. Zarei, Safety risk modeling and
respected reviewers. Their valuable comments and suggestions major accidents analysis of hydrogen and natural gas
have enhanced the strength and significance of our article. releases: A comprehensive risk analysis framework, Int J
Hydrogen Energy (2015) (in press), http://dx.doi.org/
10.1016/j.ijhydene.2015.07.117.
LITERATURE CITED 12. A. Noroozi, N. Khakzad, F. Khan, S. MacKinnon, and
1. T. Deacon, P. Amyotte, and F. Khan, Human error risk anal- R. Abbassi, The role of human error in risk analysis:
ysis in offshore emergencies, Saf Sci 48 (2010), 803–818. Application to pre- and post-maintenance procedures
2. A. Dormohammadi, E. Zarei, M.B. Delkhosh, and A. of process facilities, Reliab Eng Syst Saf 199 (2013),
Gholami, Risk analysis by means of a QRA approach on 251–258.
a LPG cylinder filling installation, Process Saf Prog 33 13. R. Abbassi, F. Khan, V. Garaniya, S. Chai, C. Chin, and
(2014), 77–84. K.A. Hossain, An integrated method for human error
3. E. Zarei, M.J. Jafari, and N. Badri, Risk assessment of vapor probability assessment during the maintenance of off-
cloud explosions in a hydrogen production facility with con- shore facilities, Process Saf Environ 94 (2015), 172–
sequence modeling, J Res Health Sci 13 (2013), 181–187. 179.
4. Z. Nivolianitou, M. Konstandinidou, and C. Michalis, Sta- 14. H.S. Blackman, Human Reliability and Safety Analysis
tistical analysis of major accidents in petrochemical indus- Data Handbook, Wiley, New York, 1994.
try notified to the major accident reporting system 15. B. Brehmer, Dynamic decision making: Human control of
(MARS), J Hazard Mater 137 (2006), 1–7. complex systems, Acta Psychol 81 (1992), 211–241.
5. R. Mehrdad, A. Bolouri, and A. Shakibmanesh, Analysis 16. W. Edwards, Dynamic decision theory and probabilistic
of accidents in nine Iranian gas refineries: 2007–2011, Int information processings, Hum Fact: J Hum Factors Ergon
J Occup Environ Med 4 (2013), 205–210. Soc 4 (1962), 59–74.
6. E. Hollnagel, Human reliability Analysis: Context and 17. S. French, T. Bedford, S.J. Pollard, and E. Soane, Human
Control, Academic Press, London, 1993. reliability analysis: A critique and review for managers,
7. D. Gertman, B. Halbert, M. Parrish, M. Sattison, Saf Sci 49 (2011), 753–763.
D.Brownson, and J. Tortorelli, Review of Findings for 18. J.S.R. Jang and C.T. Sun, Neuro-Fuzzy and Soft Comput-
Human Performance Contribution to Risk in Operating ing: A Computational Approach to Learning and Machine
Events, United States Nuclear Regulatory Commission, Intelligence, Prentice-Hall, Inc., Upper Saddle River, New
Washington, DC, 2002. Jersey, 1996.
8. S.G. Kariuki and K. L€ owe, Integrating human factors into pro- 19. M.J. Driver, K.R. Brousseau, and P.L. Hunsaker, The
cess hazard analysis, Reliab Eng Syst Saf 92 (2007), 1764– Dynamic Decision Maker: Five Decision Styles for Execu-
1773. tive and Business Success, iUniverse, Bloomington, Indi-
9. J. Ren, I. Jenkinson, J. Wang, D.L. Xu, and J.B. Yang, A ana, 1998.
methodology to model causal relationships on offshore 20. J.W. Kim and W. Jung, A taxonomy of performance influ-
safety assessment focusing on human and organizational encing factors for human reliability analysis of emergency
factors, J Saf Res 39 (2008), 87–100. tasks, J Loss Prev Process Ind 16 (2003), 479–495.

Process Safety Progress (Vol.35, No.2) Published on behalf of the AIChE DOI 10.1002/prs June 2016 199

You might also like