Maintenance Optimization for Production Systems With Polytype Engineers Under Limited Maintenance Capability: A Reinforcement Learning Approach

Maintenance optimization is a perennially interesting subject within the field of reliability engineering, playing an essential role in enhancing the system reliability. In practice, maintenance engineers are heterogeneous with different ability levels, and maintenance failure may occur when tasks are handled by junior engineers. Inspired by the above engineering reality, this article investigates the maintenance optimization for a production system with multiple machines connected in parallel. A practical engineering scenario is studied that there is a limited number of maintenance engineers and they are categorized into different professional levels owing to diverse maintenance capabilities. In addition, the case that junior engineers cause maintenance failure with a certain probability is presented. The Markov decision process as well as Markov process are utilized to describe the operation and maintenance process of the system. When the system reaches a higher state after maintenance, it generates the higher revenue in next operation period, though at the cost of increased maintenance expenses. To balance the revenue and cost, the optimal maintenance engineer allocation and maintenance level are determined at each regular inspection epoch. With the objective of maximizing the total expected reward, a Q-learning-based reinforcement learning algorithm is employed to solve the optimal maintenance policies effectively. Finally, numerical examples are presented to validate the constructed model, and plentiful sensitivity analyses are conducted to provide scientific management proposals.