Enhancing Process Safety Management: A Novel Application of GPT-Based Statistical AI Models for Driving Improvement in Process Safety Barrier Health

Abstract Objectives/Scope This study is aimed at enhancing process safety management practices through the innovative use of artificial intelligence to analyze and monitor preventative and mitigative risk control scheme barrier condition with the objective of optimizing functional assurance tasks, critical spare requirements, and offer a forward prediction of condition. This work focused on integrating these insights into a comprehensive process safety management framework that allows for the proactive delivery of major accident hazard prevention principles. Methods, Procedures, Process The methodology employed a two-step process. The initial phase involved the use of GPT-based models (GPT: Generative Pre-trained Transformers). These models were tasked with the categorization and interrogation of process safety events, safety critical assurance performance testing results, equipment performance and process safety barrier failures/impairments to pre-determined risk control barriers. Following this AI-driven categorization, the study carried out a rigorous trend analysis inclusive of sentiment analysis to drive insight to improve safety culture and site morale. From there, statistical modelling techniques were applied to these events including linear regression, log-log models, and non-linear equations to build reliability models to assess the condition of the respective process safety barrier. Reduction in likelihood of major incident is assured by the robust predictive capabilities which determine forward asset resource needs, prioritize remedial actions and inventory/spares requirements to protect and enhance the process safety barrier condition, reducing likelihood of major incident. Results, Observations, Conclusions Results indicate that AI models effectively categorize events and that the statistical models provide a robust projection of barrier health with key areas of focus identified to maintain and improve condition. The connectivity between reliability modelling, task optimization, safety observations and inventory requirements provided an end-to-end model that allows for simulation of impact of change in input variables and can be calibrated to meet asset owner's outcome requirements. The combined approach not only enhances the accuracy of our understanding of barrier conditions but also facilitates a deeper understanding of equipment trends, leading to more informed decision-making in major accident hazard prevention. Novel/Additive Information This study introduces a new approach to process safety and incident management by integrating GPT-based AI models with traditional statistical analysis to enhance the categorization and interpretation of process safety performance data. This dual-layered method offers a novel contribution to the energy industry by providing a more nuanced understanding of process safety data and predictive insights that were previously unattainable. The integration of AI and statistical methods sets a new benchmark for predictive process safety management, offering significant potential benefits including enhanced barrier control.