An adaptive neural network algorithm for assessment ...

Journal of Loss Prevention in the Process Industries 24 (2011) 361e370

Contents lists available at ScienceDirect

Journal of Loss Prevention in the Process Industries journal homepage: www.elsevier.com/locate/jlp

An adaptive neural network algorithm for assessment and improvement of job satisfaction with respect to HSE and ergonomics program: The case of a gas refinery A. Azadeh a, *, M. Rouzbahman b, M. Saberi c, d, I. Mohammad Fam e a

Department of Industrial Engineering and Center of Excellence for Intelligent-Based Experimental Mechanic, College of Engineering, University of Tehran, P.O. Box 11365-4563, Tehran, Iran Department of Mechanical and Industrial Engineering, University of Toronto, Canada c Department of Industrial Engineering, University of Tafresh, Iran d Institute for Digital Ecosystems & Business Intelligence, Curtin University of Technology, Perth, Australia e Department of Occupational Health and Safety, Faculty of Health, University of Hamadan Medical Science, Hamadan, Iran b

a r t i c l e i n f o

a b s t r a c t

Article history: Received 14 January 2010 Received in revised form 31 January 2011 Accepted 31 January 2011

Researchers have been continuously trying to improve human performance with respect to Health, Safety and Environment (HSE) and ergonomics (hence HSEE). This study proposes an adaptive neural network (ANN) algorithm for measuring and improving job satisfaction among operators with respect to HSEE in a gas refinery. To achieve the objectives of this study, standard questionnaires with respect to HSEE are completed by operators. The average results for each category of HSEE are used as inputs and job satisfaction is used as output for the ANN algorithm. Moreover, ANN is used to rank operators performance with respect to HSEE and job satisfaction. Finally, Normal probability technique is used to identify outlier operators. Moreover, operators with inadequate job satisfaction with respect to HSEE are identified. This would help managers to see if operators are satisfied with their jobs in the context of HSEE. This is the first study that introduces an integrated ANN algorithm for assessment and improvement of human job satisfaction with respect to HSEE program in complex systems. Ó 2011 Elsevier Ltd. All rights reserved.

Keywords: Job satisfaction Health, safety and environment (HSE) Ergonomics Artificial neural network Human operators Assessment

1. Introduction This section presents an introduction of HSE and job satisfaction with respect to HSEE. Section 1.1 presents the concepts of HSEE and important features of each concept are discussed. Section 1.2 presents the definition of job satisfaction and the previous works on this concept are discussed. Finally, Section 1.3 stresses the relationship between job satisfaction and HSEE factors. Also, the innovation of the proposed algorithm of this study is discussed. 1.1. HSE-ergonomics HSE at the operational level will strive to eliminate or decrease injuries, adverse health influences and hurt to the environment. Effective application of ergonomics in work system design can

* Corresponding author. E-mail address: [email protected] (A. Azadeh). 0950-4230/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.jlp.2011.01.015

cause a balance between worker characteristics and task requirements. This can increase worker productivity, create improved worker safety (physical and mental) and job satisfaction. The principal of HSE is now well recognized. Several companies use the acronym HSE to describe health, safety and environment as one entity (Deng, 1999). Various studies have shown positive influences of applying ergonomic rules to the workplace including machine, job and environmental design (Azadeh, Fam, Khoshnoud, and Nikafrouz, 2008a; Ayoub, 1990a,b; Blanchard and Fabrychy, 1998, pp. 112e123; Shikdar and Sawaqed, 2004). Studies in ergonomics have also produced data and instructions for industrial applications (Blanning, 1984; Bryden and Hudson, 2005; Burri and Helander, 1991). However, there is still a low level of acceptance and few applications in industry. The main concern of work system design in context of ergonomics is improvement of machines and tools. Lack of utilization of the ergonomic rules could bring inefficiency to the workplace. Moreover, an ergonomically deficient workplace can cause physical and emotional stress, low productivity and poor quality of work conditions (Azadeh et al., 2008a; Burri and Helander, 1991; Cabrero-Canosa et al., 2003; Caldwell, Breton, &

362

A. Azadeh et al. / Journal of Loss Prevention in the Process Industries 24 (2011) 361e370

Holburn, 1998). It is believed that ergonomic defects in industry are main cause of health hazards in workplaces, low levels of safety and decreased workers’ productivity (Champoux and Brun, 2003). Although ergonomics applications have achieved significant momentum in developed countries, knowledge remains low in developing countries (Azadeh et al., 2008a). HSEE encourages employees to adopt a healthy and safe lifestyle. It develops and operates its facilities with due concern for the health and safety of its neighbors and collaborates with authorities in the preparation of emergency response plans. It contributes to eco-efficiency by continuously improving energy consumption and decreasing waste. It designs and develops products to have the minimum contrary influence on the environment throughout their life-cycle. It optimizes the relation between operator and machine in a manner that operator faces the least weariness and has the most efficiency (Azadeh et al., 2008a; Changchit and Holsapple, 2001; Chen and Yang, 2004). There are close relationship between health, safety, environment and ergonomics factors. Basically, ergonomics is concerned with all those factors that can affect people and their behavior (Azadeh et al., 2008a). Inappropriate design between man and machine could lead to decreased safety. Management error and harmful factors related to work environment could cause human error. HSE has defined human factors and ergonomics as the environmental, organizational and job factors, and human and individual characteristics which influence work behavior. Exact consideration of human factors improves health and safety by reducing the number of injures and unsafe behaviors at work. It also provides considerable benefits by decreasing the costs associated with work injuries and enhancing efficiency. Saksvik and Nytr (1996) presented an implementation of internal control (IC) of health, environment and safety (HES) in Norwegian enterprises. IC involves systematic actions that reduce stress and occupational hazards which will, in turn, prevent injuries and workplace absenteeism (Saksvik et al., 1996). Eklund (1997) presented the relationships between ergonomics and several factors such as work conditions, product design, ISO 9000, continuous improvements and TQM (Eklund, 1997). Azadeh, Nouri, and Mohammad Fam (2005) evaluated the impact of total system design factors (TSD) on human performance in a power plant (Azadeh et al., 2005). Azadeh, Keramati, Mohammad Fam, and Jamshidnejad (2006) described an integrated macroergonomics model for operation and maintenance of power plants (Azadeh et al., 2006). Torp and Moen (2006) presented the effects of implementing and improving occupational health and safety management system in small- and medium-sized companies. Mohammad Fam, Azadeh, and Azam Azadeh (2007) used nonparametric statistical analysis to investigate the impacts of total ergonomics factors on local factors (Mohammad Fam et al., 2007). Azadeh, Mohammad Fam, Sadjadi, and Hamidi (2008b) presented an integrated framework for designing and development of the integrated health, safety and environment (HSE) model in a gas refinery. It was shown that the total ergonomics model is superior to the conventional ergonomics approach (Azadeh et al., 2008b). Azadeh, Mohammad Fam, and Nouri (2008c) presented a framework for development of integrated intelligent human engineering environment in complex systems. Moreover, health, safety, environment and ergonomics (HSEE) were developed and introduced. Duijm et al. (2008) showed that HSE management would benefit greatly from existing management systems and also from the further development of meaningful safety performance indicators that identify the conditions prior to accidents and incidents. Mohammad Fam, Azadeh, Faridan, and Mahjoub (2008) used behavior sampling technique to evaluate the workers safety behavior in a gas treatment company (Mohammad Fam et al.,

2008). Azadeh, Mohammad, and Azadeh (2009) implemented a study in a gas treatment company to show the superiority of HSEE over conventional HSE. HSEE integrated the structure of human and organizational systems with a conventional HSE system. It resulted in enhanced reliability, availability, maintainability and safety. Hivik, moen, Mearns, and Haukelid (2009) reported a qualitative interview study of 31 employees, with and without leadership responsibility, employed in a Norwegian petroleum company to gain insight into how the workers conceptualized the HSE concept and different aspects of HSE culture. Hassim and Hurme (2010) presented an Inherent occupational health index for assessing the health risks of process routes during process research and development stage. The method takes into account both the hazard from the chemicals and the potential for the exposure of workers to the chemicals. The certification and implementation of occupational health and safety management system has become a priority for many organizations. The status of implementing occupational health and safety management systems (OHSMS) and important performance indicators of OHSMS in the printed circuit board (PCB) industry in Taiwan have been investigated by Chen, Wu, Chuang, and Mac (2009). Chang and Liang (2009) developed a model to evaluate the performance of process safety management systems of paint manufacturing facilities based on a three level multi-attribute approach. Einarsson and Brynjarsson (2008) suggested a human factor program approach through case studies from incidents and accidents in Iceland and Netherlands. From their observations, a more holistic system view was proposed involving authorities and contractors. 1.2. Job satisfaction Job satisfaction is a topic of wide interest to both practitioners and managers. It is one of the most commonly studied variable in organizational behavior research. It is also a main variable in both research and theory of organizational phenomena ranging from job design to supervision (Spector, 1997). The traditional model of job satisfaction focuses on all the feelings that an individual has about his/her job. However, what makes a job pleasing does not depend not only on the nature of the job, but also on the demands of individuals versus job satisfaction. Investigating job satisfaction is not a simple exercise as several variables and their interaction such as individual characteristics, social environment, job characteristics of the enterprise are involved. Explaining job satisfaction has been an enduring problem in the study of organization. The major motivation of behavioral scientists for studying job satisfaction has been to create a link between workers’ job satisfaction and job performance. However, investigations have provided various results. Job satisfaction is believed to be related to several important organizational behaviors such as turnover, absenteeism, and union activity (Fisher & Locke, 1992). The effects of job satisfaction are also linked to variables on an individual level. Several studies suggest that job satisfaction has a positive relationship with employees’ knowledge of their quality of life, and physical/mental health (House, 1981). Edosomwan (1986) described a case study that examined certain ergonomic factors’ impact on productivity and job satisfaction in a specific computer-aided printed circuit board assembly task (Edosomwan, 1986). Byrd, Sankar, and Loh (1996) developed a causal model relating technical orientation with job satisfaction of technical and managerial personnel. Shikdar and Das (2003) showed that the participative standard with feedback condition emerges as the optimum strategy for improving worker satisfaction and job attitudes in a repetitive industrial production task. Calisir and Gumussoy (2007) used a survey to investigate the impact of IT professionals’ demographic characteristics, work characteristics

A. Azadeh et al. / Journal of Loss Prevention in the Process Industries 24 (2011) 361e370

363

Table 1 The HSE and ergonomics questions used as input indicators in the algorithm. Health

Safety

Do you drink enough water in hot seasons? Do you think ISO 18000 has been helpful to increase your efficiency?

Do you think your work area needs changes to improve safety issues? Does explosion and fire risks exist in your work area?

Environment

Ergonomics

Does machine vibration transfer to your body? Do you think your work area has appropriate light as far as the quality of its color is concerned? Do pollutants and hazardous chemicals Do you think it is necessary to use personal Is your work area temperature exist in your work environment? protection items at work? between 20e26  C? Would there be any circumstance under which Is there sufficient air flow in your you would have to ignore safety measures? work area? Is there any possibility for accidents which Does noise bother you in your have not been considered in the safety procedures? work area?

and work stress on job satisfaction. Five components of job satisfaction were revealed by principal component analysis, and variables affecting each job satisfaction component were determined by regression analysis. Likert-type scale that employs ordinal values to represent linguistics terms has been very popular in the studies related to job satisfaction evaluation. Rasmani and Shahari (2007) showed that the ordinal values in Likert scale does not offer the best way in representing the linguistic terms. They proposed the use of fuzzy sets to represent linguistic terms in Likert-type scale. They further employed fuzzy conjoint method in job satisfaction evaluation. Chen (2008) examined relationships between achievement motivation and job characteristics on job satisfaction among IS personnel. Baradaran, Ghadami, and Malihi (2008) used multi objective model to select effective solutions for improving job satisfaction. Five components of job satisfaction were revealed as the objective functions by principal component analysis, and the effective solutions were determined by multi objective analysis. Park, Baker, and Lee (2008) tested the relationship between need for cognition and task complexity in the Korean civil engineering management industry. Moreover, job satisfaction was predicted by individual factor and team factor (team need for cognition). Acuña, Gómez, and Juristo (2009) investigated the relationships between personality, team processes, task characteristics, product quality and satisfaction in software development teams. 1.3. Job satisfaction versus HSE and ergonomics Although several studies have investigated job satisfaction, however, there is a lack and need for modeling and assessment of job satisfaction with respect to integrated HSEE. Most of such studies have used simple approaches to investigate job satisfaction such as causal models, simple surveys and statistical methods. This study presents an effective algorithm to determine the impacts of HSEE factors on job satisfaction. This is the first study that considers

Do you feel any backache after your daily work activities? Are the heights of chairs adjustable?

Is there sufficient space for free movement in your work space? Do you feel any pain in your neck after your daily work activities? Do you feel any pain in your eyes in your daily work activities?

the interaction of HSE and ergonomics factors with respect to job satisfaction by an intelligent algorithm. Moreover, this study proposes an adaptive neural network (ANN) algorithm for measuring and improving job satisfaction among operators with respect to HSEE in a gas refinery. Standard questionnaires have been distributed to operators in four working shifts in a gas refinery. Four groups of inputs (Health, Safety, Environment and Ergonomics) and one main output (job satisfaction) were defined and utilized by the questionnaires. Next, operators’ efficiency was evaluated by the proposed algorithm. The paper is organized as follows. Section 2 provides an introduction to adaptive neural network. Section 3 introduces the proposed algorithm for assessing efficiency of HSEE factors with respect to job satisfaction. Section 4 presents the experimentation procedure and the case study. Section 5 offers results and analysis and finally Section 6 provides the concluding remarks. 2. Artificial neural network (ANN) An ANN is an information-processing paradigm that is inspired by the way biological nervous systems, such as the brain, process information. ANNs, like people, learn by example. An ANN is configured for a specific application, such as pattern recognition, function approximator or data classification through a learning process. They are made up simple processing units, which are linked by weighted connections to form structures that are able to learn relationships between sets of variables. This heuristic method can be useful for non-linear process that has an unknown functional form. ANNs consist of an inter-connection of a number of neurons. There are many varieties of connections under study, however here we will discuss only one type of network which is called Multi Layer Perceptron (MLP). In this network the data flows forward to the output continuously without any feedback. Hidden nodes with

Table 2 Analysis of variance for two groups of operators. Source

Degrees of freedom

Sum squares

Mean square

F

P-value

Factor Error Total

1 68 69

0.014 9.821 9.836

0.014 0.144

0.10

0.754

Individual 95% CIs For Mean Based on Pooled St Dev Level C3 C4

N 35 35

Pooled St Dev ¼ 0.3800

Mean 1.6000 1.6286

StDev 0.4505 0.2931

—————þ———————þ———————þ——————— (—————————————*—————————————) (—————————————*—————————————) —————þ———————þ———————þ——————— 1.520 1.600

1.680

364

A. Azadeh et al. / Journal of Loss Prevention in the Process Industries 24 (2011) 361e370

appropriate non-linear transfer functions are used to process the information received by the input nodes. The MLP’s most popular learning rule is the error back propagation algorithm. Back propagation learning is a kind of supervised learning introduced by Werbos (1974) and later developed by Rumelhart and McClelland (1986). At the beginning of the learning stage all weights in the network are initialized to small random values. The algorithm uses a learning set, which consists of input-desired output pattern pairs. Each inputeoutput pair is obtained by the offline processing of historical data. These pairs are used to adjust the weights in the network to minimize the Sum Squared Error (SSE) which measures the difference between the real and the desired values over, all output neurons and all learning patterns. After computing SSE, the back propagation step computes the corrections to be applied to the weights. The attraction of MLP has been explained by the ability of the network to learn complex relationships between input and output patterns, which would be difficult to model with conventional algorithmic methods. Several studies in the empirical field have shown the superiority of ANN over conventional methods (Azadeh, Ghaderi, Anvari, & Saberi, 2007a; Azadeh, Ghaderi, Anvari, Saberi, & Izadbakhsh, 2007b; Chiang, Urban, & Baldridge, 1996; Hwarng, 2001; Indro, Jiang, Patuwo, & Zhang, 1999; Jhee & Lee, 1993; Hill, O’Connor, & Remus, 1996; Kohzadi, Boyd, Kermanshahi, & Kaastra, 1996; Tang, Almeida, & Fishwick, 1991; Tang & Fishwick, 1993; Stern, 1996). 3. Methodology An integrated ANN algorithm is proposed to measure job satisfaction with respect to HSEE factors as follows: Step 1. Determine validity and reliability of the questionnaire. In this step, analysis of variance (ANOVA) experiment is used to test the equality of treatment means between two randomly selected groups of operators with respect to significant job satisfaction factor. Step 2. Determination of input (S) and output (P) variables of the model. Step 3. Collect data set S in all available previous periods which describes the inputeoutput relationship for operators. Assume that there are n operators to be evaluated. Note that the current period data (Sc) does not belong to S. Step 4. Divide S into two subsets: train (S1) and test (S2) data. Step 5. Use ANN to estimate relation between inputs and output (Murat & Ceylan, 2006):  Select architecture and training parameters.  Train the model by using the train data (S1).  Evaluate the model by using the test data (S2).  Repeat these steps by using different architectures and training parameters.  Determine the relative error (MAPE1) of the learned ANN. If MAPE  0.20, then go to the next step. Otherwise, distribute questionnaires to more operators. A max error of 0.20 is defined to allow the extensive variations between operators due to skills, education and experience. This is a reasonable and logical estimate.  Select the best network architecture (ANN*) with minimum MAPE on the test data set. Step 6. Run ANN* for Sc.

1 Mean Absolute Percentage Error MAPE ¼ (N: the number of rows)

1 N

PN

i¼1

i Set point valuei jActual value j Set point value i

Table 3 Accumulated data used in the integrated ANN algorithm. Operator code

Health

Safety

Environment

Ergonomic

Job satisfaction

1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00 13.00 14.00 15.00 16.00 17.00 18.00 19.00 20.00 21.00 22.00 23.00 24.00 25.00 26.00 27.00 28.00 29.00 30.00 31.00 32.00 33.00 34.00 35.00 36.00 37.00 38.00 39.00 40.00 41.00 42.00 43.00 44.00 45.00 46.00 47.00 48.00 49.00 50.00 51.00 52.00 53.00 54.00 55.00 56.00 57.00 58.00 59.00 60.00 61.00 62.00 63.00 64.00 65.00 66.00 67.00 68.00 69.00 70.00

1.33 1.50 1.50 1.67 1.00 1.83 1.00 1.33 1.50 1.17 1.50 1.33 1.33 1.67 1.67 1.50 1.67 1.67 1.67 1.17 1.33 1.50 1.50 1.33 1.33 1.33 1.17 1.67 1.17 1.67 1.33 1.67 1.67 1.67 1.83 1.67 1.67 1.50 2.00 1.50 1.58 1.67 1.92 2.00 1.75 1.75 1.17 1.58 1.50 1.58 1.58 1.58 1.67 1.58 1.58 1.50 1.58 1.25 1.25 1.42 1.42 1.58 1.67 1.42 1.42 1.50 1.42 1.25 1.42 1.50

1.10 1.40 1.00 1.40 1.20 1.60 1.20 1.50 1.20 1.40 1.30 1.50 1.50 1.30 1.20 1.30 1.40 1.50 1.40 1.20 1.00 1.20 1.40 1.40 1.00 1.40 1.10 1.20 1.40 1.10 1.60 1.40 1.50 1.80 1.70 1.40 1.40 1.50 1.60 1.20 1.45 1.40 1.50 1.20 1.40 1.55 1.20 1.40 1.25 1.25 1.35 1.35 1.30 1.20 1.25 1.45 1.35 1.10 1.30 1.40 1.40 1.40 1.25 1.10 1.30 1.45 1.40 1.25 1.25 1.50

1.83 1.80 1.27 1.27 1.40 1.93 1.27 1.73 1.43 1.20 1.83 1.80 1.40 1.43 1.87 1.70 1.63 1.73 1.53 1.53 1.20 1.43 1.40 1.40 1.40 1.47 1.73 1.90 1.87 1.57 2.00 1.50 1.40 1.60 1.57 1.93 1.07 1.77 2.00 1.63 1.77 1.38 1.75 1.60 1.53 1.28 1.75 1.37 1.60 1.37 1.38 1.38 1.43 1.23 1.75 1.50 1.28 1.35 1.50 1.67 1.62 1.72 1.65 1.32 1.53 1.72 1.40 1.60 1.72 1.75

1.67 1.33 1.53 1.13 1.46 1.60 1.23 1.60 1.26 1.00 1.73 1.46 1.60 1.60 1.53 1.37 1.60 1.60 1.53 1.33 1.13 1.53 1.26 1.60 1.26 1.40 1.53 1.60 1.60 1.60 1.60 1.46 1.80 1.93 1.20 1.46 1.26 1.63 1.53 1.53 14.87 1.53 1.73 1.77 1.77 1.22 1.62 1.33 1.57 1.50 1.33 1.33 1.40 1.46 1.50 1.37 1.30 1.36 1.35 1.53 1.67 1.48 1.57 1.33 1.43 1.53 1.43 1.47 1.60 1.53

1.50 1.50 2.00 2.00 2.00 2.00 1.00 1.50 2.00 1.00 1.00 2.00 1.00 1.50 2.00 1.00 1.50 2.00 2.00 1.00 1.50 2.00 1.00 2.00 1.00 1.00 2.00 2.00 2.00 2.00 1.00 1.00 2.00 2.00 2.00 1.50 2.00 2.00 2.00 1.50 2.00 1.25 2.00 2.00 1.75 1.75 1.25 1.75 1.50 1.75 1.25 1.25 1.50 2.00 1.50 1.75 1.50 1.50 1.50 2.00 1.00 1.50 1.75 1.75 1.50 1.75 1.50 1.50 2.00 1.00

A. Azadeh et al. / Journal of Loss Prevention in the Process Industries 24 (2011) 361e370

Output 1 (job satisfaction) ANN-MLP Model number

Learning method

Number of neurons in first hidden layer

First transfer function

Second transfer function

MAPE

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

LM B BFG BR CGB CGF CGP GD GDA GDM GDX OSS SCG RP GDM LM GDX B BFG BR CGB CGF CGP GD GDA

41 49 21 15 61 89 78 25 90 28 62 44 46 76 86 48 19 95 48 2 7 32 84 74 18

Logsig Logsig Logsig Logsig Logsig Logsig Logsig Logsig Logsig Logsig Logsig Logsig Logsig Logsig tansig tansig tansig tansig tansig tansig tansig tansig tansig tansig tansig

Logsig Logsig Logsig Logsig Logsig Logsig Logsig Logsig Logsig Logsig Logsig Logsig Logsig Logsig purelin purelin purelin purelin purelin purelin purelin purelin purelin purelin purelin

0.22 0.24 0.22 0.22 0.21 0.21 0.21 0.22 0.23 0.24 0.24 0.21 0.21 0.22 0.21 0.22 0.20 0.24 0.21 0.22 0.20 0.22 0.22 0.22 0.22

29

36 43

50

57 64

71

78

85 92

99

largest Ei0 which indicates the operator with the best performance is identified. Suppose kth operator has the largest Ei0 and hence

  Ek0 ¼ max Ei0

(3)

Thus, the value of the shift for each operator is different and is calculated by

Shi ¼ Ek * PðANN*Þi =PðANN*Þk =

i ¼ 1; .; n

(4)

The effect of the scale of operators on their efficiencies is considered. Moreover, the unit used for correction is selected by notice of its scale (CRS) (Azadeh et al., 2007a; Costa and Markellos, 1997; Delgado, 2005). Step 9. Calculate efficiency scores. The efficiency scores take values between 0 and 1. This maximum score is assigned to the unit used for the correction (Azadeh et al., 2007a).

  Fi ¼ Pi = PðANN*Þi þ Shi

i ¼ 1; .; n

(5)

Step 10. Plot Normal probability of efficiency scores by ANN from step 9 for all operators to identify outlier operators. Step 11. Perform corrective actions with respect to outlier operators.

(1)

i ¼ 1; .; n

(2)

4. Experiment: the integrated ANN algorithm The applicability of the proposed adaptive algorithm is experimented in an actual gas refinery. It is shown how each step is applied to assess the relationship between HSEE program and job satisfaction in the gas refinery. Moreover, the efficiency of the proposed algorithm is shown by MAPE. Also, the qualitative features of the adaptive algorithm are compared with recent studies in this area to show its superiority and applicability.

0.3

2

0.25

1.8

output

0.2 0.15 min MAPE

0.1

22

Fig. 2. The comparison of 100 independent runs for the best model (17th ANN-MLP).

The above formula (2) calculates largest error by noting the operator scale (Constant Returns to Scale (CRS)). To this end, the

MAPE

15

Series1

Step 8. Shift frontier function from neural network for obtaining the effect of the largest positive error.

Ei0 ¼ Ei =PðANN*Þi

8

Run number of best ANN model

Step 7. Calculate the error between the actual output (Preal(i)) and ANN model output (PANN*(i)) in the selected period to assess the efficiency of operators (Sc):

i ¼ 1; .; n

0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 1

B: Batch training with weight and bias learning rules; BFG: BFGS quasi-Newton back propagation; BR: Bayesian regularization; C: Cyclical order incremental update; CGB: PowelleBeale conjugate gradient back propagation; CGF: FletcherePowell conjugate gradient back propagation; CGP: PolakeRibiére conjugate gradient back propagation; GD: Gradient descent back propagation; GDA: Gradient descent with adaptive learning rule back propagation; GDM: Gradient descent with momentum back propagation; GDX: Gradient descent with momentum and adaptive learning rule back propagation; LM: LevenbergeMarquardt back propagation; OSS: One step secant back propagation; RP: Resilient back propagation (Rprop); SCG: Scaled conjugate gradient back propagation.

Ei ¼ PrealðiÞ  PANN*ðiÞ

MAPE error

Table 4 Architecture of the 25 ANN-MLP models and their associated relative error (MAPE).

365

1.6 1.4 1.2

0.05

1

0 1 Series1

3

5

7

9

11

13

15

17

19

21

ANN Models

Fig. 1. Relative error estimate for the 25 ANN-MLP models.

23

25

1

10 19 28 37 46 55 64 73 82 91 100 109 118 127 136 145

Real output 1 ANN output 1

operator number

Fig. 3. The results of the selected ANN-MLP and original data for job satisfaction.

366

A. Azadeh et al. / Journal of Loss Prevention in the Process Industries 24 (2011) 361e370 Health

Table 5 Estimation of efficiency scores for all operators by the proposed algorithm. Operator code

Preal

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70

2 1.5 2 1.5 2 2 2 2 1.5 1.5 2 2 1.5 1.5 2 1 2 2 2 2 1 1.5 2 1 2 1 2 1 1.5 2 1 1.5 2 2 1 1.5 2 1 2 1 1 2 2 2 2 1 1 2 2 2 1.5 2 2 2 1.5 2 1.25 2 2 1.75 1.75 1.25 1.75 1.5 1.75 1.25 1.25 1.5 2 1.5

(i)

1

PANN(i)

Ei ¼ Preal(i)  PANN(i)

Fi

Rank

1.52 1.29 1.56 1.32 1.56 1.72 1.21 1.45 1.38 1.60 1.70 1.49 1.38 1.45 1.38 0.95 1.53 0.97 1.89 1.44 0.93 1.47 1.38 1.02 1.65 1.36 1.51 1.31 1.57 1.38 1.37 1.60 1.67 1.61 1.18 1.19 1.37 1.45 1.26 1.27 1.28 1.28 1.37 1.38 1.43 1.70 1.61 1.57 1.76 1.68 1.58 1.42 1.57 1.81 1.34 1.33 1.58 1.74 1.68 1.66 1.50 1.26 1.52 1.35 1.47 1.51 1.51 1.57 1.43 1.38

0.48 0.21 0.44 0.18 0.44 0.28 0.79 0.55 0.12 0.10 0.30 0.51 0.12 0.05 0.62 0.05 0.47 1.03 0.11 0.56 0.07 0.03 0.62 0.02 0.35 0.36 0.49 0.31 0.07 0.62 0.37 0.10 0.33 0.39 0.18 0.31 0.63 0.45 0.74 0.27 0.28 0.72 0.63 0.62 0.57 0.70 0.61 0.43 0.24 0.32 0.08 0.58 0.43 0.19 0.16 0.67 0.33 0.26 0.32 0.09 0.25 0.01 0.23 0.15 0.28 0.26 0.26 0.07 0.57 0.12

1.00 0.87 1.00 0.81 1.00 0.96 1.00 0.94 0.75 0.70 0.91 1.00 0.81 0.79 1.00 0.65 0.96 1.00 0.70 0.85 0.55 0.65 0.92 0.56 0.84 0.48 0.91 0.51 0.68 1.00 0.51 0.69 0.89 0.93 0.58 0.88 1.00 0.49 1.00 0.50 0.50 1.00 0.97 0.97 0.95 0.42 0.44 0.91 0.84 0.88 0.69 1.00 0.93 0.84 0.79 1.00 0.56 0.84 0.86 0.77 0.83 0.67 0.83 0.77 0.86 0.60 0.60 0.71 1.00 0.77

1 17 1 27 1 5 1 7 34 37 13 1 28 30 1 42 4 1 36 22 49 43 12 48 23 55 11 50 40 1 51 39 14 9 46 15 1 54 1 53 52 1 3 2 6 57 56 10 20 16 38 1 8 21 29 1 47 24 18 33 26 41 25 31 19 45 44 35 1 32

Safety

2

Environment

3

Ergonomics

Job Satisfaction

. . .

19

Fig. 4. The architecture of the preferred ANN model for job satisfaction.

4.1. The gas refinery The gas refinery is located in southern province of Hormozgan, Iran. The questionnaires were distributed to operators of four working shifts. Moreover, the job satisfaction of operators with respect to HSEE has been ranked by the proposed algorithm. ANN is used to predict the efficiency of HSEE from feedback received from operators (questionnaires). The results of ANN were verified by actual data via MAPE. 4.2. Data collection and analysis Four detailed questionnaires containing valuable information related to human factors, safety, environmental and health are developed and presented to operators. The score (weight) to each question was assigned between 0 and 1. The first questionnaire includes questions about pain and anthropometric issues, specifications of workstations and chairs, personal protection items, lighting, noise and overall satisfaction from system. The second questionnaire contains questions about training, work pressures, organizational structure, job satisfaction, occupational safety, and relationship with management and coworkers. The third questionnaire covers questions about temperature of work area, ventilation, official layout, analog and digital systems, work instructions, quality management systems, environment,

Fig. 5. Normal probability plot for efficiency scores of 70 operators.

A. Azadeh et al. / Journal of Loss Prevention in the Process Industries 24 (2011) 361e370 Table 6 AndersoneDarling Normality Test for the efficiency scores of all operators. A-squared P-value

1.938 0

safety and health. The forth questionnaire (Hendric questionnaire) includes questions about ergonomic factors and machine interface. The input indicators are divided to 4 main categories which are health, safety, environment, and ergonomic and job satisfaction is defined as output. The input indicators are included in Table 1. Furthermore, the average score for each indicator is computed by referring to the related questions in Table 1. 5. Computational results and analysis Step 1: To prove the validity of the questionnaires, a significant question was chosen and tested. Then, one way ANOVA is used to compare the means of two random groups of operators with respect to this factor. By using ANOVA, the assumption of equality of treatment means between two groups is not rejected. From this analysis, it is claimed that the questionnaire has validity. The results of ANOVA are shown in Table 2. Step 2: The next step involves determination of inputs and output variables for the algorithm. There are 4 main categories (health, safety, environment, ergonomic) as input variables. Then, for each category, the average score is used in the proposed ANN algorithm. The questions that are used as inputs are shown in Table 1. In addition, 1 main question is selected from the questionnaire as the output variable, namely, job satisfaction. The output question is stated as follows: “Are you satisfied with your job? “ The answer is selected as yes, no or I don’t know. The accumulated data is shown in Table 3. The reader should note that in order to get better answer, the data is converted to the range of 1 and 2 to eliminate 0 from further calculations. Moreover, for each category (health, safety, environment, ergonomic), the average score is used as the final score in the integrated ANN algorithm. Step 3: 150 rows of data are collected from 70 operators. This is accomplished by 70 single operators, 34 mixtures of 2 operators, 22 mixtures of three operators, 13 mixtures of 4 operators and

367

Table 7 Results of linear regression. Operator number

Real output (Job satisfaction)

Conventional regression

64 65 66 67 68 69 70 MAPE ¼ 0.44

1.75 1.50 1.75 1.50 1.50 2.00 1.00

0.77 0.80 0.85 0.84 0.80 0.87 0.84

Table 8 Results of second order regression. Operator number

Real output (Job satisfaction)

Conventional regression

64 65 66 67 68 69 70 MAPE ¼ 0.47

1.75 1.50 1.75 1.50 1.50 2.00 1.00

3.32 2.16 2.48 2.16 2.64 2.90 2.48

finally 11 mixtures of 5 operators. This study has only shown the results of 70 operators. Step 4: S1 represents 135 rows of train data and S2 represents 15 rows of test data. Step 5: To locate the optimum structure for this study, 25 distinct ANN models are tested. Maximum number of neurons in the first hidden layer is set to 100. Each model is replicated 100 times to take care of possible bias or noise. The architecture of the ANN-MLP models and their MAPE values are shown in Table 4. It seems the 17th ANN-MLP model for job satisfaction has the lowest MAPE and consequently it is chosen as the optimum model. This is also shown graphically in Fig. 1. Fig. 2 shows the result of 100 independent runs for the preferred model (17th ANN-MLP). It can be seen that the fluctuations are within a narrow range. Therefore, it verifies previous finding that the 17th ANN-MLP is an ideal candidate for the purpose of this study. Fig. 3 compares the selected 17th ANN-MLP with actual data. This figure validates that the preferred model is relatively close to actual data. Step 6: Therefore, the 17th ANN-MLP is selected for estimating the performance assessment of HSSE versus job satisfaction. To

Table 9 Results of logarithmic regression. Operator number

Real output (Job satisfaction)

Conventional regression

64 65 66 67 68 69 70 MAPE ¼ 0.59

0.55 0.40 0.55 0.40 0.40 0.69 0.00

0.17 0.18 0.21 0.21 0.17 0.23 0.21

Table 10 Comparison of ANN and conventional regressions.

Fig. 6. Normal probability plot for efficiency scores non-outlier operators.

Models

Linear Regression

Second Order Regression

Logarithmic Regression

The Adaptive ANN Algorithm

MAPE

0.44

0.47

0.59

0.20

368

A. Azadeh et al. / Journal of Loss Prevention in the Process Industries 24 (2011) 361e370

Table 11 The features of the integrated algorithm versus other studies. Feature Study or method

The Adaptive ANN Algorithm of this study Delphi Methods Statistical methods Edosomwan (1986) Byrd et al. (1996) Saksvik and Nytr (1996) Torp and Moen (2006) Azadeh et al. (2006) Rasmani and Shahari (2007) Calisir and Gumussoy (2007) Mohammad Fam et al. (2007) Mohammad Fam et al., (2008) Baradaran et al. (2008) Azadeh et al. (2008b)

HSE Ergonomics Job satisfaction Integrated Data complexity Intelligent High precision Data pre-processing and modeling modeling evaluation HSE-ergonomics and non-linearity modeling and reliability post-processing versus job and forecasting satisfaction O

O

O O

O O

O

O

O O O O O O O O O

O

O

O

O

O

O

? ? O O

O

O O

O

O

identify an ANN output for each operator, the preferred model is tested for all operators. Fig. 3 presents the distance between the results of the preferred ANN and original data. This figure validates that the preferred model is relatively close to actual data. Appendix 1 shows the ANN train code. Appendix 2 shows the ANN test code and finally Appendix 3 presents the efficiency code. Steps 7e9: The results of these steps are shown in Table 5. This table shows the efficiency scores for all operators by the proposed algorithm. Moreover, all operators are ranked according to their efficiency scores. Fig. 4 presents the ANN architecture of the preferred network with respect to HSSE and job satisfaction (17thModel in Table 4). Steps 10 and 11: The Normal probability plot for all operators are plotted and shown in Fig. 5. This is to identify the outlier operators. As shown from this figure, several operators are identified as outliers. Management should take proper corrective actions with respect to these operators. This may be done through proper on-the-job training, simulator classes, etc. The result of Normality test is shown in Table 6. As shown in Table 6, the Normality assumption was rejected among 70 operators. The outlier operators should be found and management should take proper corrective actions with respect to these operators. Fig. 6 shows Normal probability plot for nonoutlier operators. The assumption of Normality for non-outlier operators is not rejected because of a p-value of 0.122. There are several reasons for existence of outlier operators. Some of these operators might not have good knowledge of HSE and need to have on-the-job training courses. A few might have answered the questions incorrectly due to job insecurity. Management should provide means to investigate assignable causes that have provided this inefficiency among several operators with respect to HSEE. Operators 2, 4e7, 11, 14, 16, 17, 24, 27, 29, 30, 33, 36e45, 48, 51, 52, 55e57, 60, 68, and 70 are considered as outlier operators. Therefore, 33 operators are not satisfied with their jobs with respect to HSE and ergonomics factors. As mentioned, proper corrective actions should be taken by the management to resolve this issue. 5.1. Comparison of ANN with conventional regression Conventional regression analysis is the most used statistical tool to explain the variation of a dependent variable Y in terms of the

? ? ? O ? ? ? ? ?

variation of explanatory variables X as: Y ¼ f(X) where f(X). Conventional regression is used for prediction of job satisfaction among operators of the gas refinery. 63 operators are used to find the best regression model and job satisfaction is estimated for the remaining 7 operators. The results of forecasting by linear, second order and logarithmic regressions are shown in Tables 7e9, respectively. The results of the ANN algorithm are compared with the above methods with respect to mean absolute percentage error (MAPE) (Table 10). Clearly the adaptive algorithm provides better solutions (lower relative error) than conventional regression approaches. 6. Conclusion HSE and ergonomics programs play an important role in enhancement of safety and human and organizational productivity. A highly unique flexible and adaptive ANN algorithm was proposed to measure and rank the job satisfaction scores with respect to HSE and ergonomics factors. The proposed algorithm is ideal because of its non-linearity, flexibility and universal approximations. Moreover, HSE and ergonomics factors were considered as input variables and job satisfaction was considered as output variable. The proposed algorithm is composed of eleven distinct steps. To show its applicability and superiority, it was applied to control room operations of a gas refinery. The efficiency score between 1 and 2 was devised to show job satisfaction with respect to the performance of HSE and ergonomics programs. The operators were then ranked by the algorithm. The Normal probability plot for all operators was then plotted to identify outlier operators. The existence of outlier operators shows that they are not satisfied with existing HSE and ergonomics programs. Management should provide means to investigate assignable causes that have provided this inefficiency among several operators. The results of the intelligent algorithm were compared with conventional regression methods with respect to MAPE (Table 10). Moreover, it was shown that the adaptive algorithm provided better solutions than conventional regression approaches. Table 11 presents the features of the adaptive algorithm. It is capable of integrated assessment of HSE and ergonomics factors with respect to job satisfaction. It is also capable of handling data non-linearity and complexity. The algorithm pre-processes and post-processes the data to eliminate non-normalized behavior. It also provides high reliability and accuracy because it identifies the best ANN configuration with lowest relative error.

A. Azadeh et al. / Journal of Loss Prevention in the Process Industries 24 (2011) 361e370

369

Acknowledgment

References

The authors are grateful for the valuable comments and suggestion from the respected reviewers. Their valuable comments and suggestions have enhanced the strength and significance of our paper.

Acuña, S. T., Gómez, M., & Juristo, N. (2009). How do personality, team processes and task characteristics relate to job satisfaction and software quality? Information and Software Technology, 51(3), 627e639. Ayoub, M. A. (1990a). Ergonomic deficiencies: I. Pain at work. Journal of Occupational Medicine, 32(1), 52e57. Ayoub, M. A. (1990b). Ergonomic deficiencies: II. Probable causes. Journal of Occupational Medicine, 32(2), 131e136. Azadeh, A., Nouri, J., & Mohammad Fam, I. (2005). The impacts of macroergonomics on environmental protection and human performance in power plants. Iranian j Env health Sci Eng, 2(1), 60e66. Azadeh, A., Keramati, A., Mohammad Fam, I., & Jamshidnejad, B. (2006). Enhancing the availability and reliability of power plants through macroergonomics approach. Journal of Scientific and Industrial Research, 65, 873e878. Azadeh, A., Ghaderi, S. F., Anvari, M., & Saberi, M. (2007a). Performance assessment of electric power generations using an adaptive neural network algorithm. Energy Policy, 35, 3155e3166. Azadeh, A., Ghaderi, S. F., Anvari, M., Saberi, M., & Izadbakhsh, H. (2007b). An integrated artificial neural network and fuzzy clustering algorithm for performance assessment of decision making units. Applied Mathematics and Computation, 187, 584e599. Azadeh, A., Mohammad Fam, I., Khoshnoud, M., & Nikafrouz, M. (2008a). Design and implementation of a fuzzy expert system for performance assessment of an integrated health, safety, environment (HSE) and ergonomics system: the case of a gas refinery. Information Sciences, 178, 4280e4300. Azadeh, A., Mohammad Fam, I., Sadjadi, M., & Hamidi, Y. (2008b). An integrated and multivariate model along with designing experiments approach for assessment of micro- and macro-ergonomic factors: the case of a gas refinery. J Res Health Sci, 8(2), 28e39. Azadeh, A., Mohammad Fam, I., & Nouri, M. A. (2008c). Integrated health, safety, environment and ergonomics management system (HSE-MS): an efficient substitution for conventional HSE-MS. Journal of Scientific and Industrial Research, 67, 403e411. Azadeh, A., Mohammad Fam, I., & Azadeh, M. A. (2009). Integrated HSEE management systems for industry: a case study in gas refinery. Journal of the Chinese Institute of Engineers, 32(2). Baradaran, V., Ghadami, S., & Malihi, S. E. (2008). A Multi objective approach for selecting solutions to improve job satisfaction an empirical case analysis. In IEEE international conference on industrial engineering and engineering management (pp. 1945e1948). IEEM. Blanchard, W., & Fabrychy, J. (1998). System engineering and analysis. USA: PrenticeHall International, Inc.. pp. 112e123. Blanning, R. W. (1984). Management applications of expert systems. Information and Management, 7, 311e316. Bryden, R., & Hudson, P. T. W. (2005). Because we want. Safety and Health Practitioner, 23, 51e54. Burri, G. J., & Helander, M. G. (1991). A field study of productivity improvements in the manufacturing of circuit boards. International Journal of Industrial Ergonomics, 7, 207e215. Byrd, T. A., Sankar, C. S., & Loh, L. (1996). Strategies to improve job satisfaction of U.S. technical personnel. EMJ e Engineering Management Journal, 8(4), 33e42. Cabrero-Canosa, M., Castro-Pereiro, M., Grãna-Ramos, M., Hernandez-Pereira, E., MoretBonillo, V., Martin-Egana, M., et al. (2003). An intelligent system for the detection and interpretation of sleep apneas. Expert Systems with Applications, 24, 335e349. Caldwell, N. H. M., Breton, B. C., & Holburn, D. M. (1998). Remote instrument diagnosis on the Internet. IEEE Intelligent Systems and Their Applications, 13, 70e76. Calisir, F., & Gumussoy, C. A. (2007). Determinants of job satisfaction in the IT, industry IEEM. In: 2007 IEEE International Conference on Industrial Engineering and Engineering Management (pp. 1960e1964). Champoux, D., & Brun, J. J. (2003). Occupational health and safety management in small size enterprises: an overview of the situation and avenues for intervention and research. Safety Science, 41, 301e318. Chang, J. I., & Liang, C. L. (2009). Performance evaluation of process safety management systems of paint manufacturing facilities. Journal of Loss Prevention in the Process Industries, 22, 398e402. Changchit, C., & Holsapple, C. W. (2001). Supporting managers’ internal control evaluations: an expert system and experimental results. Decision Support Systems, 30, 437e449. Chen, L. H. (2008). Job satisfaction among information system (IS) personnel. Computers in Human Behavior, 24(1), 105e118. Chen, J. R., & Yang, Y. T. (2004). A predictive risk index for safety performance in process industries. Journal of Loss Prevention in the Process Industries, 17, 233e242. Chen, C. Y., Wu, G. S., Chuang, K. J., & Mac, C. M. (2009). A comparative analysis of the factors affecting the implementation of occupational health and safety management systems in the printed circuit board industry in Taiwan. Journal of Loss Prevention in the Process Industries, 22, 210e215. Chiang, W. C., Urban, T. L., & Baldridge, G. W. (1996). A neural network approach to mutual fund net asset value forecasting, Omega. The International Journal of Management Science, 24, 205e215. Costa, A., & Markellos, R. N. (1997). Evaluating public transport efficiency with neural network models. Transportation Research, 5(5), 301e312. Delgado, F. J. (2005). Measuring efficiency with neural networks, an application to the public sector. Economics Bulletin, 3(15), 1e10.

Appendix 1. Train codes ptr ¼ xlsread(‘s.xls’,‘data’,‘a1:ee4’); ttr ¼ xlsread(‘s.xls’,‘data’,‘a5:ee5’); pte ¼ xlsread(‘s.xls’,‘data’,‘ef1:et4’); tte ¼ xlsread(‘s.xls’,‘data’,‘ef5:et5’); A ¼ zeros(100,15);W ¼ zeros(603,100); for i ¼ 1:100 net ¼ newff(ptr,ttr,[i 1],{},‘trainscg’);net.performFcn ¼ ‘sse’; net.trainParam.goal ¼ 0.1; net.trainParam.show ¼ 20; net.trainParam.epochs ¼ 5000; net.trainParam.mc ¼ 0.95;randn(‘seed’,192736547);net ¼ init (net); net ¼ train(net,ptr,ttr); A(i,:) ¼ sim(net,pte); for j ¼ 1:size(getx(net)) y ¼ getx(net); w(j,i) ¼ y(j); end end for i ¼ 1:100 e ¼ A(i,:)-tte; mape(i) ¼ mean(abs(e./tte)); end xlswrite(‘scg{}.xls’, A,‘output’); xlswrite(‘scg{}.xls’,w,‘w’); xlswrite(‘scg{}.xls’, transpose(mape),‘MAPE’);

Appendix 2. Test codes p ¼ xlsread(‘S.xls’,‘data’,‘a1:et4’); t ¼ xlsread(‘S.xls’,‘data’,‘a5:et5’); net ¼ newff(p,t,[150 1],{‘logsig’ ‘logsig’},‘trainlm’); net ¼ setx(net,X); output ¼ sim(net,p); xlswrite(‘output4.xls’, output,‘output’);

Appendix 3. Efficiency codes p ¼ xlsread(‘S.xls’,‘data’,‘a1:et4’); t ¼ xlsread(‘S.xls’,‘data’,‘a5:et5’); net ¼ newff(P,t,[i 1],{‘tansig’ ‘purelin’},‘trainlm’); net ¼ setx(net,X); ANN ¼ sim(net,P); for i ¼ 1:150 v(i) ¼ mean(t)  t(i)/150; s ¼ sum(v); w(i) ¼ v(i)/s; E(i) ¼ t(i)  ANN(i); E1(i) ¼ E(i)/w(i); [c,k] ¼ max(E1); sh(i) ¼ E(k)*(w(i)/w(k)); F(i) ¼ t(i)/(ANN(i)þsh(i)); end xlswrite(F);

370

A. Azadeh et al. / Journal of Loss Prevention in the Process Industries 24 (2011) 361e370

Deng, H. (1999). Multicriteria analysis with fuzzy pairwise comparison. International Journal of Approximate Reasoning, 21(3), 215e231. Duijm, N. J., Fiévez, C., Gerbec, M., Hauptmanns, U., & Konstandinidou, M. (2008). Management of health, safety and environment in process industry. Safety Science, 46(6), 908e920. Edosomwan, J. A. (1986). Role of ergonomics in improving and job satisfaction on computer-aided assembly tasks. In: Proceedings, American Institute of Industrial Engineers, annual conference and convention (pp. 378e381). Einarsson, S., & Brynjarsson, B. (2008). Improving human factors, incident and accident reporting and safety management systems in the Seveso industry. Journal of Loss Prevention in the Process Industries, 21, 550e554. Eklund, J. (1997). Eergonomics, quality and continuous improvement e conceptual and empirical relationships in an industrial context. Ergonomics, 40(10), 982e1001. Fisher, C. D., & Locke, E. A. (1992). The new look in job satisfaction research and theory. In C. J. Cranny, P. C. Smith, & E. F. Stone (Eds.), Job satisfaction: How people feel about their jobs and how it affects their performance (pp. 165e194). NY: Lexington Books. Hassim, M. H., & Hurme, M. (2010). Inherent occupational health assessment during process research and development stage. Journal of Loss Prevention in the Process Industries, 23, 127e138. Hill, T., O’Connor, M., & Remus, W. (1996). Neural network models for time series forecasts. Management Science, 42(7), 1082e1092. Hivik, D., Moen, B. E., Mearns, K., & Haukelid, K. (2009). An explorative study of health, safety and environment culture in a Norwegian petroleum company. Safety Science, 47(7), 992e1001. House, J. S. (1981). Work stress and social support. Reading, MA: Addison Wesley Publishing Company. Hwarng, B. H. (2001). Insights into neural-network forecasting of time series corresponding to ARMA (p;q) structures. Omega, 29, 273e289. Indro, D. C., Jiang, C. X., Patuwo, B. E., & Zhang, G. P. (1999). Predicting mutual fund performance using artificial neural networks, Omega. The International Journal of Management Science, 27, 373e380. Jhee, W. C., & Lee, J. K. (1993). Performance of neural networks in managerial forecasting. Intelligent Systems in Accounting Finance and Management, 2, 55e71. Kohzadi, N., Boyd, M. S., Kermanshahi, B., & Kaastra, I. (1996). A comparison of artificial neural network and time series models for forecasting commodity prices. Neurocomputing, 10(3), 169e181.

Mohammad Fam, I., Azadeh, A., & Azam Azadeh, M. (2007). Modeling an integrated health, safety and ergonomics, management system: application to power plants. J Res Health Sci, 7(2), 1e10. Mohammad Fam, I., Azadeh, A., Faridan, A. M., & Mahjoub, H. (2008). Safety behaviours assessment in process industry: a case in gas refinery. Journal of the Chinese Institute of Industrial Engineers, 25(4), 298e305. Murat, Y. S., & Ceylan, H. (2006). Use of artificial neural networks for transport energy demand modeling. Energy Policy, 34(12), 3165e3172. Park, H. S., Baker, C., & Lee, D. W. (2008). Need for cognition, task complexity and job satisfaction. Journal of Management in Engineering, 24(2), 111e117. Rasmani, K. A., & Shahari, N. A. (2007). Job satisfaction evaluation using fuzzy approach. In: Proceedings e third international conference on natural computation, ICNC (pp. 544e548), Vol. 4. Rumelhart, D. E., & McClelland, J. L. (1986). Parallel distributed processing: Explorations in the microstructure of cognition. In Foundations, . Cambridge, MA: MIT Press. Saksvik, P., & Nytr, K. (1996). Implementation of internal control (IC) of health, environment and safety (HES) in Norwegian enterprises. Safety Science, 23(1), 53e61. Shikdar, A. A., & Das, B. (2003). A strategy for improving worker satisfaction and job attitudes in a repetitive industrial task: application of production standards and performance feedback. Ergonomics, 46(5), 466e481. Shikdar, A. A., & Sawaqed, M. N. (2004). Ergonomics and occupational health and safety in the oil industry: a managers’ response. Computers & Industrial Engineering, 47, 223e232. Spector, P. E. (1997). Job satisfaction: Application, assessment, causes, and consequences. London: SAGE Publications. Stern, H. S. (1996). Neural networks in applied statistic. Technometrics, 38(3), 205e214. Tang, Z., & Fishwick, P. A. (1993). Back-propagation neural nets as models for time series forecasting. ORSA Journal on Computing, 5(4), 374e385. Tang, Z., Almeida, C., & Fishwick, P. (1991). A time series forecasting using neural networks vs. BoxeJenkins methodology. Simulation, 57(5), 303e310. Torp, S., & Moen, B. E. (2006). The effects of occupational health and safety management on work environment and health: a prospective study. Applied Ergonomics, 37(6), 775e783. Werbos, P. I. (1974). Beyond regression: New tools for prediction and analysis in the behavior sciences. Ph.D. thesis. Cambridge, MA: Harvard University.