Research On Feature Extraction And Evaluation Method For Remaining Useful Life Prediction Of Rolling Bearings

As an important part of mechanical equipment,rolling bearings can predict their remaining useful life,which can provide guidance for the operation and maintenance of equipment,ensure the safety of equipment use,and avoid accidents.Therefore,it is of important practical significance to carry out research on the remaining life prediction.Using machine learning to accurately predict the remaining useful life of rolling bearings is currently the main research method.However,there is a huge difference in the life of rolling bearings.Even under the same model,batch,and operating conditions,the life of the bearing may still differ by several times.In addition,the life of the bearing will also be affected by materials,processing technology,installation technology,operating conditions,etc.,which poses a huge challenge for predicting the life of the bearing.In the current research on bearing remaining service life prediction,the focus of the research is to establish the relationship between the features extracted from the bearing vibration signal and its service life.However,the influence of the timing information implicit in the signal characteristics on the prediction of the remaining service life of the bearing,the influence of the volatility of the bearing signal characteristics in the time direction on the prediction of the remaining service life of the bearing,The influence of trend consistency between the same characteristics of different bearings on the prediction of the remaining service life of the bearing needs to be further studied.For this reason,the main work of this paper and the research results obtained are as follows:(1)Aiming at the problem of how to mine the time sequence information implicit in the signal that has an impact on the life prediction,a time sequence feature extraction method for the prediction of the remaining service life of rolling bearings is proposed.PMCCNN-LSTM model is constructed to effectively extract the timing features.First,through the structured PMCCNN unit,the signal is feature extracted,so that it can contain timing information and form timing features.Then use multi-layer LSTM unit to further explore the timing characteristics.Finally,make predictions.The example verifies that the method can effectively extract the timing information in the signal characteristics.The use of time series features can obtain better prediction results.(2)Aiming at the problem of how to effectively reduce the volatility of bearing signal characteristics in the time direction,while reducing the influence of human factors.An iterative generation feature extraction method for predicting the remaining useful life of rolling bearings is proposed.First,the convolutional self-encoding network is used as a feature extraction model to extract features from the signal,reducing the number of features and the network parameters required for subsequent calculations.Then build a multi-layer two-way long and short-term memory network to generate features,use it to generate smooth features,and better retain the advantages of the original trend,effectively reducing the volatility of features.Finally,an example proves that the method can effectively reduce volatility and at the same time effectively reduce forecast errors.(3)Aiming at the commonly used bearing life prediction feature evaluation indicators,which did not consider the trend consistency between the same features of different bearings,a feature evaluation indicator oriented to the prediction of the remaining useful life of rolling bearings is proposed.In order to quantitatively evaluate this trend consistency,referring to the index evaluation process,with the help of the correlation formula,a new characteristic evaluation index—the trend consistency index is designed.And apply it to the features extracted by the deep learning model.Finally,an example is used to verify that the indicator can evaluate trend consistency,and at the same time,to a certain extent,it improves the interpretability of the features extracted by the deep learning model.(4)Integrating the above research results,an online prediction system for the remaining service life of rolling bearings has been developed.The system receives the collected bearing state data,performs pre-processing such as downsampling,and predicts the remaining life of the rolling bearing online.At the same time,it provides users with a variety of visualization results such as life prediction curve,predicted value,characteristic trend,etc.,providing an intuitive reference for bearing maintenance.