Investigation of Human Errors Using Fuzzy-Bayesian Belief Networks

It is an on-going process for the nuclear industry to implement newer safety measures based on lessons learnt from operating experience in both engineering and operation (such as establishing new regulations and providing additional measures for backup safety actions during external hazards). Even minor safety incidents in NPP are viewed critically and appropriate safety measures are undertaken to prevent such occurrences. Nevertheless, human errors are found to be one of the key contributors to accidents in NPPs. This is due to various reasons such as complex interfaces, workloads, and dynamic situations. Moreover, human error data is hard to be collected and classified; the factors which influence human behaviour are difficult to be identified. When attempting to reduce human error in such complex systems, cognitive dimensions (attention, memory, etc.), behavioural variables, environmental factors, organizational and technical factors associated with incidents/accidents must be clearly addressed. Though a number of methods are available to analyse and investigate the causes of human error, it seems that most of the existing methods are insufficient for comprehensive analysis of human activity-related contextual aspects of accidents. For this reason, the present study proposes a proactive modelling approach that combines fuzzy analytical hierarchy process (FAHP) and Bayesian belief networks (BBNs) with human factors analysis and classification system (HFACS), in order to identify the role of human errors. HFACS is a validated method to classify errors from current investigative records and to document the full range of potential human factor data in complex industries, such as avionics, rail and chemicals. FAHP implementation strengthens the HFACS system by offering an empirical context to ensure the quantitative evaluation of nuclear accidents. The proposed model is comprised of three stages. The first step is the empirical study of accident-related human factors, which utilizes HFACS to distinguish active or latent failures. In the next stage, which is a statistical analysis using BBN, the hierarchy of human factors defined in the first phase offers input for analysis. BBN strengthens HFACS capability by assessing the degree of human factors relationships. The fuzzy analytical hierarchy process methodology is introduced in the final phase to compute the conditional probabilities. BBN enhances the ability of HFACS by measuring the degree of relationships among the human factors. The application of fuzzy networks adopted in this study is to effectively model (1) uncertainty in human error (2) variations attributed to assessment by experts, (3) performance influencing variables and (4) cognitive human behaviour. A case study illustrates the model and indicates that the proposed model is capable of looking for critical contextual factors related to human activity that is not easily obtainable by using existing methods. Some of the advantages of implementing these accident analysis approaches are discussed. Finally, some considerations, including further work, associated with the FAHP-HFACS-BBN are discussed and concluded in this paper. The study of human reliability in the nuclear field focuses on understanding the causes and propagation mechanisms of operator errors and its implications for effective modelling in risk assessment studies.