AI-based Exploration of Socio-technical Safety Barrier Dynamics

Safety barriers are deployed in a system as a safeguard against undesired events. The prevalent safety barrier concepts focus on component level failures. These concepts may be insufficient for complex systems characterized by limited data & emergent behavior. Recent research developments advocate reframing safety barriers as sociotechnical systems, integrating human, organizational, & technical aspects to address emerging risk within the broader context of safety-II & resilience engineering. This study builds on this perspective & investigates the operational safety challenges in a complex hydrogen system. However, quantitative assessment of socio-technical safety barriers is challenging & has not been extensively studied. This study uses synthetic datasets generated from an assumed system dynamics model of a hydrogen refueling station maintenance procedure to demonstrate the feasibility of data-driven identification of socio-technical safety barrier dynamics. These datasets are used for system identification through SINDy algorithm. The two different models are identified with one focusing on interpretability to enable dynamic evaluation of the socio-technical safety barriers & other with focus on predictive performance of the system model. The first model establishes mathematical equations that enable quantitative representation and comparative analysis of socio-technical safety barriers under assumed system dynamics, while second model can be used for predicting & simulating the system dynamics. This approach illustrates the potential to strengthen barrier performance evaluation beyond conventional models. The findings contribute to advancing socio-technical safety barrier analysis methodologies for hydrogen systems, promoting adaptive & resilient strategies.

Keywords: socio-technical safety barriers, emerging risk, hydrogen safety, artificial intelligence, system identification.