Automating PRA modeling through generative AI and open-source digital toolkits

Zheng, X. ; Shibamoto, Yasuteru 

Probabilistic Risk Assessment (PRA) modeling often requires extensive manual effort, specialized expertise, and the use of proprietary software, which can impose significant cost and accessibility barriers. This study investigates the feasibility of automating PRA model development by leveraging generative AI and open-source digital toolkits. A prototype workflow was developed that integrates the Google Gemini API with Python scripts, and employs the OpenPSA model exchange format alongside the SCRAM tool for model representation and analysis. A test implementation using Google AI Studio demonstrated the successful generation of PRA structures, such as event trees, from natural language descriptions. These results indicate that combining large language models with open-source tools can streamline PRA modeling, reduce development time, and lower the entry threshold for new practitioners. This approach holds promise not only for improving modeling efficiency and reducing costs in industry applications, but also for supporting education and training of the next generation of risk analysts by making PRA more accessible and interactive.