A deep reinforcement learning approach for maintenance planning of multi-component systems with complex structure

In recent years, the trend toward greater integration and complexity of mechanical systems has brought challenges to the formulation of preventive maintenance plans. It is very difficult to realize the traditional condition-based maintenance method that relies on calculating the optimal maintenance threshold to achieve optimal maintenance. However, in solving highly complex and challenging control and decision-making problems, the deep reinforcement learning (DRL) method shows its powerful ability and provides a new idea for the maintenance planning of complex systems. Numerical results show that DRL-based maintenance model can obtain optimization strategies through continuous exploration and realize the trade-off between component maintenance cost and the loss caused by system failure, whether in simple or complex multi-component systems. The policy minimizes the overall cost of the system by choosing actions that minimize the total long-term cost. The comparison with other maintenance strategies shows that the proposed model is superior to various baseline policies and reduces the system lifecycle cost.