A methodology for enhancing the reliability of expert system applications in probabilistic risk assessment

In highly complex industries, capturing and employing expert systems is significantly important to an organization's success considering the advantages of knowledge-based systems. The two most important issues within the expert system applications in risk and reliability analysis are the acquisition of domain experts' professional knowledge and the reasoning and representation of the knowledge that might be expressed. The first issue can be correctly handled by employing a heterogeneous group of experts during the expert knowledge acquisition processes. The members of an expert panel regularly represent different experiences and knowledge. Subsequently, this diversity produces various sorts of information which may be known or unknown, accurate or inaccurate, and complete or incomplete based on its cross-functional and multidisciplinary nature. The second issue, as a promising tool for knowledge reasoning, still suffers from lack of deficiencies such as weight and certainty factor, and are insufficient to accurately represent complex rule-based expert systems. The outputs in current expert system applications in probabilistic risk assessment could not accurately represent the increasingly complex knowledge-based systems. The reason is the lack of certainty and self-assurance of experts when they are expressing their opinions. In this paper, a novel methodology is presented based on the concept of Z-numbers to overcome this issue. A case study in a high-tech process industry is provided in detail to demonstrate the application and feasibility of the proposed methodology.