Remaining Life Prediction Method of Rolling Bearing Based on TD3 Optimization Model Updating Strategy

To address the problems of model degradation and low prediction accuracy in the of update procedure the rolling bearing life prediction model, the updating strategy of twin delayed deep deterministic policy gradient(TD3) is suggested to optimize the prediction model, and it applied to rolling bearing remaining useful life prediction. Continuous wavelet transform is used to build a dataset by converting the rolling bearing vibration data into two-dimensional time-frequency spectrum diagram. Deep reinforcement learning is introduced to design the elements of state, action, reward value and loss function, which are employed to training of the prediction model, this model is composed convolutional neural network (CNN) and bidirectional long short-term memory. The delayed update and soft update techniques of the TD3 algorithm are used to reduce the update frequency of the CNN feature network. A new loss function is proposed to establish a new model updating strategy for solving the prediction model degradation problem. The experimental study using the PHM2012 dataset and the finding indicate that, in comparison to the compared methodologies, the suggested prediction method with model optimization strategy can reduce the root mean square error and mean square error by more than 23.8% and 20.8% respectively, reducing the prediction error by at least 18.6% and improve the scores by more than 0.213.