Method for predicting the remaining useful life of mechanical equipment based on deep reinforcement learning

The reliability and safety of mechanical equipment are increasingly being considered as equipment gets bigger and more complex. The remaining useful life (RUL) prediction technology examines the operating data of the equipment. This technology can effectively enhance the safety and reliability of the operation of the equipment and provide key facts for the subsequent maintenance decisions of the equipment. This paper proposes a deep reinforcement learning-based RUL prediction algorithm. First, the autoencoder (AE) is used to extract key features from the original signals obtained from mechanical equipment. Using the extracted features to form the state variable of reinforcement learning, the reinforcement learning model is trained after setting an appropriate action space and reward function. The sequential interaction decision logic of the reinforcement learning method can naturally maintain the sequential dependence relationship between samples and decrease the volatility of prediction results. Then, the turbine engine data set CMPASS was employed to validate the proposed method. The proposed method is superior to the most current RUL prediction methods in root mean square error and Score, particularly for equipment near the end of degradation. © 2023 Chinese Academy of Sciences. All rights reserved.