Objectives: As industrial systems become increasingly complex and technologically advanced, the human role in ensuring safety and efficiency remains indispensable. This study aimed to present a comprehensive review of Human Reliability Assessment research published between 2010 and 2023. It compares human reliability assessment (HRA) methodologies with emerging technologies, such as artificial intelligence and lights-out manufacturing, identifies existing research gaps, and analyzes the analytical techniques employed and the industrial sectors addressed.
Methods: A systematic search of major scientific databases using domain-specific keywords yielded over 230 publications. After removing duplicate studies, 180 articles were selected for detailed analysis. Each article was evaluated based on methodology, industrial application, country, institutional affiliation, and publishing outlet.
Results: The results indicated that the most frequently applied approaches in HRA research were the systematic human error reduction and prediction approach (SHERPA), cognitive reliability and error analysis method (CREAM), and fuzzy mathematics. The United States, China, and South Korea emerged as leading contributors to this field. The findings revealed that neither qualitative nor quantitative methods alone are sufficient to fulfill the three core objectives of HRA: error identification, probability estimation, and control design. Therefore, a hybrid approach was recommended by integrating SHERPA and Tecnica Empirica Stima Errori Operatori (TESEO). SHERPA provides comprehensive coverage of error identification and the design of effective control measures, while TESEO facilitates rapid, conservative probability estimation. These methods provided a practical and efficient framework for achieving HRA objectives within operational constraints. Additionally, ten key research gaps were identified.
Conclusion: The SHERPA–TESEO hybrid framework presents a viable strategy for achieving the core goals of HRA. Nonetheless, a shift from static to dynamic and data-driven models is necessary in the context of intelligent environments and operator-free production. Recommended developments include revising SHERPA’s cognitive task classifications, recalibrating TESEO's adjustment factors, and integrating real-time data with human–artificial intelligence (AI) interaction. These advancements are expected to significantly enhance real-time prediction of human error risks and support timely intervention strategies.
