ABSTRACT Human reliability analysis (HRA) plays a pivotal role in modeling and mitigating human errors within reliability engineering. The success likelihood index method (SLIM) is a widely utilized HRA tool for quantifying human error probabilities (HEPs) of operational tasks. However, the traditional SLIM predominantly relies on subjective judgments from a small group of experts, which often leads to unreliable HEP estimates due to inherent individual biases and prejudices. Besides, conflicts and divergences are prevalent among experts, so making the integration of a consensus model into the SLIM is indispensable for achieving the recognized HEP results. Nevertheless, the existing SLIM variants rarely account for trust relationships among experts and individual independence preservation during the consensus reaching process (CRP), which impairs their applicability in large group HRA scenarios. To address these critical limitations, this paper proposes a novel large group SLIM model to quantify HEPs by integrating flexible linguistic expressions (FLEs) and an extended two‐stage consensus model. First, the FLEs are employed to represent experts’ complex and diversified assessments of tasks states, which can effectively accommodate the ambiguity and vagueness of subjective judgments and allow interdisciplinary experts to express nuanced opinions without forced conformity. Second, a network partition algorithm is utilized to cluster large group experts based on their trust relationships, yielding stable subgroups rooted in professional credibility that are impervious to dynamic changes of expert opinions. Third, an improved two‐stage CRP is constructed, which considers both expert trust relationships and individual independence, to yield consistent and reliable HEP estimates. Finally, a practical case of cargo tank cleaning operations in chemical tanker ships is implemented to validate the effectiveness and feasibility of the proposed SLIM model, supplemented by a comparison analysis and some simulation experiments to verify its robustness and superiority.
Huang et al. (Wed,) studied this question.