Abstract:

To investigate the feasibility of applying artificial intelligence (AI)to inherent safety decision-making for chemical process, a decision-making framework comprising data collection, risk identification, and strategy development was developed to address delayed risk identification, the challenges of dynamically determining control parameters, and the lack of decision support under complex operating conditions. Deep learning, knowledge graph, and reinforcement learning were integrated into the safety design process. The framework was evaluated using representative process scenarios, including exothermic reaction systems, storage and transport of easily combustible light gases, high-pressure gas compression, acid-base neutralization reactions, and oxygen-containing organic reactions. The model is validated in terms of risk identification, path optimization, and strategy development using results from simulations and case studies. Results indicate that the model can dynamically identify the high-risk process pathways and conduct correlation analysis of the parameters. Decision outputs can be produced in the following scenarios: temperature control of the exothermic reactions, identification of the leakage risk zone, abnormal warning for high-pressure systems, pH drift control, and determination of explosion limits. Nevertheless, the model's application is restricted by the following: interpretability, data quality, interface compatibility, and regulatory integration. AI can assist with the integration of multiple data sources, identification of complex risks, and generation of design leads to integrate the multi-source data systems and complex risks. Its use is however determined by the particular process scenario. Engineering implementation will require improved model transparency, data harmonization, and system integration.

Key words: artificial intelligence (AI), inherent safety in chemical processes, inherent safety decision-making, typical process scenarios, applicability