Remaining Useful Life Prediction Method Of Mechanical Equipment Based On Deep Learning

Predicting the remaining life of mechanical equipment is an important task in condition-based maintenance.By predicting the degradation and remaining life of mechanical equipment in advance,it can effectively avoid failure accidents and solve the problem of maintenance waste through active maintenance.Although deep learning models are developing rapidly,there is still a large room for development in the field of remaining life prediction of mechanical equipment.Aiming at the limitations of existing prediction models and the complexity and diversity of data dimensions of different mechanical equipment,the following researches are carried out in this paper:(1)In view of the multi-dimensional monitoring data,this paper takes the turbofan engine as the research object.On the basis of analyzing its working principle and degradation reasons,the Transformer model is selected as the model for predicting the remaining life of the turbofan.The model can effectively solve the problem of memory degradation caused by the large distance between sequence vectors,and the self-attention mechanism can enhance the feature selection ability of the network.In order to make the Transformer model fully applied to the remaining life prediction problem of mechanical equipment,this paper designs a method based on RP-Trans for remaining life prediction.First,the Prob Sparse self-attention mechanism and rotational position encoding are used to enhance the sequence position correlation,and solve the problem of high complexity caused by the multi-head self-attention mechanism.Then,this paper designs an improved self-attention extraction operation,which uses Dilated-SKConv convolution instead of the fully connected layer to connect the self-attention module,so that the model can obtain an exponentially growing receptive field,and it is beneficial for the model to receive longer data input.The proposed method is validated in a case study of a turbofan engine.Experimental results show that the proposed method outperforms the state-of-the-art methods.(2)For the single-dimensional monitoring data,this paper takes the bearing as the research object,and analyzes the structure and degradation principle of the bearing.index,and then input the health index into the improved Transformer prediction model,so as to realize the prediction of the remaining life of the bearing.The experience shows that the CNN-Trans model proposed in this paper uses the powerful local feature acquisition ability of CNN to construct the stage degradation health factor,combined with the multi-level feature fusion of the improved Transformer,fully learns the characteristics of vibration data,and effectively improves the residual life prediction accuracy of rolling bearings.To sum up,this paper uses deep learning to explore and study the remaining life prediction problem for mechanical equipment with different dimensional data,and proposes two prediction models based on deep learning and validates them.Compared with other methods,the effect is good.It has certain reference significance.