Generation of maintenance and troubleshooting strategies for complex equipment based on Bayesian network reinforcement learning

Abstract

To solve the problem of long decision-making time and high total cost of generated strategies in traditional heuristic maintenance and troubleshooting decision-making methods, a method for complex equipment maintenance and troubleshooting strategy generation based on Bayesian network combined with reinforcement learning is proposed . In order to better utilize the knowledge of complex equipment models, Bayesian network is used to express maintenance and troubleshooting knowledge, and in order to be closer to the actual situation of complex equipment, the failure probability analyzed according to the failure mode, impact and criticality is used as the prior probability of the Bayesian network after reasonable transformation; in order to generate maintenance and troubleshooting strategies using the decision-making process of reinforcement learning, a method for converting the maintenance and troubleshooting decision problem into a reinforcement learning problem is proposed ; in order to better solve the transformed reinforcement learning problem, the observation-repair action pair is introduced to reduce the problem scale, and the action mask is set to process the dynamic action space. The simulation verification results show that under the unified performance index , the proposed method obtains higher index values than the traditional method, which proves the effectiveness and superiority of the method.