A Deep Reinforcement Learning Method for Finding the Risk-Based Optimal Prescriptive Maintenance Policy of Degrading Safety Barriers

Safety barriers have to be properly maintained to limit the reduction of their performance due to age. In this work, we formulate the problem of finding the optimal maintenance actions to be performed on the safety barriers as a Sequential Decision Problem (SDP). The SDP is, then, solved by Deep Reinforcement Learning (DRL), with Imitation Learning (IL) and Proximal Policy Optimization (PPO). The resulting prescriptive maintenance policy accounts for the effects that the different maintenance actions have on the limitation of the degradation of the safety barriers and, ultimately, on the risk due to the escalation of the accidental scenarios. The novelty of the work lies in the formulation and efficient solution of the problem of risk-based prescriptive maintenance optimization for safety barriers, accounting for their age-dependent degradation. An application is shown with regards to a chemical facility exposed to Natural hazard-induced Technological (NaTech) risk, such as that due to earthquakes, and equipped with safety barriers whose performance degrades in time. The optimal prescriptive maintenance policy found by DRL is proven to outperform the typical maintenance strategy both in terms of cost and risk reduction.