Volume 17, Issue 1 (2020)
ioh 2020, 17(1): 990-1001 |
Back to browse issues page
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Mirzaei Aliabadi M, Askaripoor T, Ghamari F, Aghaei H. An investigation of the relationship between human and organizational factors in occupational accidents using Bayesian network approach: A case study in mining accidents. ioh 2020; 17 (1) :990-1001
URL: http://ioh.iums.ac.ir/article-1-2760-en.html
URL: http://ioh.iums.ac.ir/article-1-2760-en.html
Arak University of Medical Sciences , Hamedaghaeih@gmail.com
Abstract: (2335 Views)
Background and aims: Human errors are major causes of the accident that occurring in the industries. However, attributing incidents to human error, regardless of the nature of human error, cannot be useful in preventing accidents. Identifying organizational and supervisory factors that affecting human errors, as well as determining the interactions between these factors, can be used in the management of appropriate control strategies to reduce the accidents. The Human Factors Analysis and Classification System framework (HFACS) is one of the most important and comprehensive qualitative tools to identify human and organizational contributing factors involved in an accident. Until now, several studies have tried to integrate the HFACS with a quantitative analysis tool in order to determine the interactions between human and organizational factors to reduce accidents. There are many types of quantitative tools that researchers usually used for this purpose. Fuzzy analytical hierarchy process, analytical network process, and artificial neural network are the most used analytical quantitative tools in this regard. Powerful graphical probability-based modeling approaches have been less well considered for quantitative analysis of the interaction and relationship between different variables. Bayesian network (BN) is one of the most important quantitative tools in this regard. BN is a probabilistic graphical model that uses for various types of inference such as diagnostic and predictive. Belief updating or sensitivity analysis is one of the exclusive feature of BN that researchers using this feature can examine the sensitivity of one “target variable” to changes in other variables. In the modeling, sensitivity analysis is used to rank the influence of input variables on the predicting of output variables. This study aimed to integrate the HFACS framework and BN to identify different factors that influence unsafe acts and determine the relationships and interactions among identified those factors to provide appropriate intervention strategies for preventing accidents in the future.
Methods: In this study, the accidents occurred in one of the largest mines in Iran that occurred during a period of 5 years (2011-2015) were collected, and then accidents with serious consequences such as fatalities, disabling injuries, or considerable property damage were screened. In the next step, all contributing factors in each accident were identified using an accident analysis team by root cause analysis (RCA) approach. RCA is a problem-solving approach that is applied to identify the root causes of problems. A total of 250 accidents analysis results were collected and classified in one of the 13 groups of the HFACS framework, and a database was created. According to the structure of the HFACS framework, the BN model was developed. HFACS is a 4 levels hierarchy of human and organizational errors, in which higher levels can influence directly lower levels and this pattern can help to the develop a BN graphical model. Causal factors at the 4 levels of the HFACS consist the nodes of the BN model. In the next step, for each node, states were defined that show different values of the variable. In this study, except for unsafe acts node that had three states (skill based, decision, and perceptual), other nodes had two states; yes (node involved in an accident) and no (node not involved in an accident). The main hypothesis of the HFACS framework is that deficiency at the higher level casual factors can lead to deficiency at the lower level casual factors. Hence, in the present study, all causal factors (parents nodes) at the higher level were connected to the lower level causal factors (child nodes) edge with arcs. For instance, causal factors of unsafe supervision (level 3) that include inadequate supervision, planned inappropriate operations, failure to correct a known problem, and supervisory violations are parents of environmental factors, personnel factors, and condition of operator nods which belong to preconditions for unsafe acts (level 2). After the graphical structure of the BN model was developed, using database that obtained in the previous section and the expectation–maximization (EM) algorithm model was trained. In a BN the conditional probability tables (CPTs) are used to determine quantitative relationships among a set of variables. The EM algorithm is one of the common methods to calculate. There are several approaches for conducting a sensitivity analysis but the mutual information (MI) approach is most common. In order to determine the factors with greatest impact on unsafe acts, the MI approach was used and the sensitivity analysis was performed. In probability theory, the MI of two random variables is a measure of the mutual dependence between the two variables. In the current study, Netica version 5.24 was used to perform calculations and analyses.
Results: The results of this study showed that at the level of unsafe acts, skill-based errors (%67.3) had the highest prior probability technique errors were the most skill based errors that were detected. Also at the level of unsafe conditions, environmental factors (%74.8) had the highest prior probability. Inadequate installation and improper housekeeping were the most frequently identified environmental factors that led to accidents. At the levels of unsafe supervision and organizational influences, inappropriate planned operation (%60.6) and organizational processes (%35.3) had the highest prior probability, respectively. Inadequate task/safety plan from unsafe supervision level and lack of standard operation procedures from organizational influences level were the most frequently identified deficiency in the selected accidents. The results of the sensitivity analysis demonstrated that the environmental factors from level 2, inappropriate planned operation from level 3, and organizational processes from level 4 had the greatest impact on unsafe acts. Based on the analysis results, several strategies were made to reduce the unsafe acts of employees.
Conclusion: In the current study, by integrating the HFACS framework as a qualitative tool and BN as a powerful quantitative tool, a human factors analysis model was developed. The results of this study indicated that the environmental factors and inappropriate planned operation had the most effect on the unsafe acts. Although organizational influences play a role as indirect factors on the unsafe acts, paying attention to eliminating defects at this level can be useful in reducing accidents. Different forms of unsafe acts require various interventions, therefore, the use of BN model can be helpful in determining strategies tailored to the specificities of the unsafe acts.
Methods: In this study, the accidents occurred in one of the largest mines in Iran that occurred during a period of 5 years (2011-2015) were collected, and then accidents with serious consequences such as fatalities, disabling injuries, or considerable property damage were screened. In the next step, all contributing factors in each accident were identified using an accident analysis team by root cause analysis (RCA) approach. RCA is a problem-solving approach that is applied to identify the root causes of problems. A total of 250 accidents analysis results were collected and classified in one of the 13 groups of the HFACS framework, and a database was created. According to the structure of the HFACS framework, the BN model was developed. HFACS is a 4 levels hierarchy of human and organizational errors, in which higher levels can influence directly lower levels and this pattern can help to the develop a BN graphical model. Causal factors at the 4 levels of the HFACS consist the nodes of the BN model. In the next step, for each node, states were defined that show different values of the variable. In this study, except for unsafe acts node that had three states (skill based, decision, and perceptual), other nodes had two states; yes (node involved in an accident) and no (node not involved in an accident). The main hypothesis of the HFACS framework is that deficiency at the higher level casual factors can lead to deficiency at the lower level casual factors. Hence, in the present study, all causal factors (parents nodes) at the higher level were connected to the lower level causal factors (child nodes) edge with arcs. For instance, causal factors of unsafe supervision (level 3) that include inadequate supervision, planned inappropriate operations, failure to correct a known problem, and supervisory violations are parents of environmental factors, personnel factors, and condition of operator nods which belong to preconditions for unsafe acts (level 2). After the graphical structure of the BN model was developed, using database that obtained in the previous section and the expectation–maximization (EM) algorithm model was trained. In a BN the conditional probability tables (CPTs) are used to determine quantitative relationships among a set of variables. The EM algorithm is one of the common methods to calculate. There are several approaches for conducting a sensitivity analysis but the mutual information (MI) approach is most common. In order to determine the factors with greatest impact on unsafe acts, the MI approach was used and the sensitivity analysis was performed. In probability theory, the MI of two random variables is a measure of the mutual dependence between the two variables. In the current study, Netica version 5.24 was used to perform calculations and analyses.
Results: The results of this study showed that at the level of unsafe acts, skill-based errors (%67.3) had the highest prior probability technique errors were the most skill based errors that were detected. Also at the level of unsafe conditions, environmental factors (%74.8) had the highest prior probability. Inadequate installation and improper housekeeping were the most frequently identified environmental factors that led to accidents. At the levels of unsafe supervision and organizational influences, inappropriate planned operation (%60.6) and organizational processes (%35.3) had the highest prior probability, respectively. Inadequate task/safety plan from unsafe supervision level and lack of standard operation procedures from organizational influences level were the most frequently identified deficiency in the selected accidents. The results of the sensitivity analysis demonstrated that the environmental factors from level 2, inappropriate planned operation from level 3, and organizational processes from level 4 had the greatest impact on unsafe acts. Based on the analysis results, several strategies were made to reduce the unsafe acts of employees.
Conclusion: In the current study, by integrating the HFACS framework as a qualitative tool and BN as a powerful quantitative tool, a human factors analysis model was developed. The results of this study indicated that the environmental factors and inappropriate planned operation had the most effect on the unsafe acts. Although organizational influences play a role as indirect factors on the unsafe acts, paying attention to eliminating defects at this level can be useful in reducing accidents. Different forms of unsafe acts require various interventions, therefore, the use of BN model can be helpful in determining strategies tailored to the specificities of the unsafe acts.
Type of Study: Research |
Subject:
Safety
Received: 2019/03/15 | Accepted: 2019/12/4 | Published: 2020/09/23
Received: 2019/03/15 | Accepted: 2019/12/4 | Published: 2020/09/23
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
