Dynamic Human Reliability Assessment enhanced with Bayesian Networks; a Comparison with Classical Approaches

The reliability study of a production system allows to obtain important information to determine its performance and understand how to prevent failures and dangerous situations. Similarly, the assessment of the Human Reliability Assessment (HRA) is crucial in assessing production performance, but its estimation is complex, often more complex than the equipment reliability. Complexity lies in the very nature of the human being, who reacts in an articulated way to different environmental stress, to changes in company policy and to psychological personal situations. In this study, a new methodology to assess human reliability is developed, then a comparison with another one method is drawn to observe how they behave in an ever-changing situation. This is innovative, especially considering how input data for the method have been achieved. In the present case, the comparison was carried out by means of a case study, under uncertain conditions, in a company producing machines for the recovery, recycling and recharging of refrigerant gas in automotive air conditioning equipment. The new methodology is based on Bayesian Network (BN) while the compared one is SLIM Method. In both cases, the PIFs (Performance Influence Factors) were evaluated as starting points, identified and evaluated by means of expert opinions and theory of belief functions (Dempster-Shafer Theory - DST). By taking Human Error Probability (HEP) values for each task of the process, it was possible to have an overall picture of the impact of the human factor at the process stages and a demonstration of which method is best suited to changing information.