Fault diagnosis method of rotating machinery based on deep Q-learning and continuous wavelet transform

To solve the problems of strong neural network feature learning and weak decision-making ability in fault diagnosis of rotating machinery，a convolutional neural network（CNN）is used to fit the Q function in reinforcement learning，and the learning strategy is implemented by the Q-learning algorithm. For fault diagnosis，a fault diagnosis method for rotating machinery based on deep Q-learning and continuous wavelet transform is proposed. A continuous wavelet transform is performed on the vibration signal to obtain a time-scale matrix，and an environmental state space is constructed for the interaction between the agent and the environment. CNN is used to fit the Q function in Q-learning to obtain a deep Q network to convert the environment. The returned state is input to the deep Q network to learn the specific state feature representation of the fault data，and the learning strategy is characterized accordingly. The agent uses ε-greedy mode to decide the action and reward generator to evaluate the action. The agent continuously interacts with the environment to maximize the Q function value and obtain the optimal strategy for fault diagnosis. This method combines the perceptual ability of deep learning and the decision-making ability of reinforcement learning，so as to effectively improve the diagnostic ability. The effectiveness of the proposed method is proved by the fault diagnosis experiments of gearbox under different working conditions and different sample sizes.