Research On Rolling Bearing Remaining Useful Life Prediction Method Based On Dense Connection Network

Rolling bearings are the main components of rotating machinery and equipment,and play a role in carrying loads and reducing friction in mechanical transmission systems.Remaining Useful Life(RUL)prediction of rolling bearings is a key aspect of bearing maintenance and directly affects the productivity and safety of the equipment.For this reason,it is necessary to study effective methods to accurately estimate the RUL of rolling bearings in order to develop reasonable maintenance plans and take relevant preventive measures.Therefore,this paper takes rolling bearings as the research object and carries out research on the method of RUL prediction for rolling bearings based on deep learning theory as follows.(1)Aiming at the problems such as the disappearance of information and gradient in the process of information transmission in deep network and the determination of the first time of fault,a prediction method of RUL combining a densely connected network with dilated causal convolution and an exponential model is proposed.The method uses the continuous wavelet transform to transform the time-domain signal into a time-frequency image as the input to the network model,and uses the constructed dilated causal convolutional densely connected network model to automatically extract features from the time-frequency image to construct health indicators.The proposed model not only inherits the advantages of the densely connected network model,but also captures the time-series characteristics of the time series.Finally,a synthetical index was constructed by fusing four time-domain statistical indicators to determine the First Predicting Time(FPT)of failure,and the obtained health indicators were fitted with an exponential model to predict the RUL of rolling bearings.The two rolling bearing datasets are compared and the results show the good performance of the proposed method in rolling bearing RUL prediction.(2)Aiming at the problems of large parameters in deep network and over-fitting in the training process,a prediction method of RUL based on Channel Attention-based Residual Dilated Causal Convolution Densely Connected Network is proposed.The method uses the time-frequency signal of vibration data as the input of the proposed model,and automatically obtains the weights of each feature channel using the introduced residual connection and channel attention mechanisms to automatically raise the weights of useful features and lower the weights of irrelevant features,and the residual connection helps train the deep network to prevent overfitting.Finally,the mapped RUL was obtained directly from the proposed model and the degradation trend of the rolling bearing.The two rolling bearing datasets are compared and the results show the good performance of the proposed method in rolling bearing RUL prediction.In summary,two methods for predicting the remaining life of rolling bearings based on densely connected networks are proposed in this paper,and the effectiveness of the proposed methods is verified using publicly available data sets.