Optimizing preventive maintenance with PPO, periodic, and quantile inspections for industrial reliability

Abstract

Preventive maintenance is a key factor in controlling maintenance costs and failure risk in degrading industrial systems. This study presents a comparative analysis of two condition-based maintenance strategies: Periodic Inspection and Replacement ( PIR ) and Quantile-Based Inspection and Replacement ( QIR ). Asset degradation is represented using a homogeneous Gamma process, capturing cumulative and stochastic deterioration behavior. Both strategies are formulated within a cost-minimization framework and optimized using Proximal Policy Optimization ( PPO ), a policy-gradient deep reinforcement learning algorithm that enables adaptive maintenance decision-making under uncertainty. Performance is evaluated using the maintenance cost per renewal cycle, long-run expected cost rate, cost variability, and a composite robustness–performance metric. Monte Carlo simulations are conducted to evaluate strategy behavior under diverse degradation trajectories and cost configurations. The numerical results show that PPO-based policies achieve lower expected costs and improved robustness compared with static maintenance strategies. QIR demonstrates enhanced robustness under highly stochastic degradation conditions, while PIR remains competitive for more predictable degradation processes. Sensitivity analyses confirm the stability of these results across varying model parameters.