Research And Software Implementation Of Bearing Life Prediction Method Based On LSTM

With the rapid development of large-scale mechanical equipment such as industrial CNC machine tools and aero-engines towards high precision and high intelligence,it is an extremely important task to manage the health of the equipment and ensure the safe and stable operation of the equipment.Bearings are key components of mechanical equipment,and their performance will directly determine the health of the equipment.Predicting the remaining life of bearings can provide effective decision support for equipment maintenance plans,so as to avoid safety accidents,which is of great significance to the health management of mechanical equipment.Currently,signal processing methods and data-driven methods are widely used in bearing life prediction.However,mechanical equipment has been in the background of strong noise for a long time,and its feature signal extraction is difficult,and the traditional regression prediction accuracy is low.And under the new operating conditions,the small amount of data and the poor generalization performance of the model bring difficulties to life prediction.In view of the above problems,this paper explores the prediction of the remaining service life of bearings.The main contents are as follows:（1）Aiming at the problem that the traditional prediction method is affected by the bearing signal noise in actual working conditions,which leads to the limited prediction accuracy,a bearing residual life method based on the IPVMD-LSTM model is proposed.The IPVMD-LSTM model proposed in this paper fully considers the characteristics of bearing cyclostationarity and impulsiveness,and constructs a synthetic index,which is used as the objective function to optimize the parameters of VMD through PSO,so as to achieve a better noise reduction effect.The overall performance is improved by 5.83%.At the same time,the time series characteristics of the actual working condition data are fully considered,and the time series characteristics are extracted through the long shortterm memory network LSTM for prediction.The experimental results show that the IPVMD-LSTM method in this paper has a significant improvement in the prediction accuracy,and its RMSE is reduced by 2.81% compared with the traditional method.（2）Aiming at the problems of the small amount of data and the inconsistent distribution of data characteristics in new operating conditions,a multi-source domain transfer（MC-MDA-LSTM）life prediction model combining MMD and CORAL is proposed.The MC-MDA-LSTM model proposed in this paper designs a domain-shared feature extraction LSTM network by taking existing multiple working conditions as multiple source domains,and sets a specific feature extraction network for each source domain,combining MMD and CORAL The advantage of measuring the difference of data distribution is to adapt the source domain and the target domain through adversarial learning,and to narrow the distribution between the two,so as to better extract the domain-invariant features.Compared with the traditional transfer method,MAPE and RMSE are reduced by 1.67% and 2.06%,respectively.（3）The bearing life prediction software is designed and implemented.The front and back ends of the software were built through Spring Boot and Vue.js,and demand analysis,architecture design,and software testing were carried out.The software provides an interactive visual interface,which is convenient for relevant personnel to perform algorithm testing and model training,greatly improving development work efficiency,and providing technical support for data and model management,which has great use value.