Performance Degradation Evaluation And RUL Prediction Of Rolling Bearings Under Feature Optimization

The manufacturing industry is moving towards the direction of intelligent and digital in order to achieve high quality development.Complex rotating machinery and equipment are widely used in production and life.Rolling bearings,as one of the key parts of rotating machinery,inevitably and irretrievably degrade or even fail due to fatigue,wear and other reasons with the accumulation of running time in the production process.The failure of rolling bearing will lead to the paralysis of the whole mechanical system.Intelligent operation and maintenance using data-driven method to obtain decision support is a research hotspot in the field of mechanical fault diagnosis.Therefore,this paper studies the performance degradation and life prediction of rolling bearings.In view of the problems that the determination of sensitive feature set in the assessment of rolling bearings degradation relies heavily on the prior knowledge of experts and that the degradation model has low tolerance for outliers and false fluctuations,insufficient time sequence information acquisition ability,and difficulty in establishing accurate mapping relationship between monitoring data and degradation trend,the following work is carried out:(1)Aiming at the problem of relying on the prior knowledge of experts when determining the sensitive feature set of rolling bearings,an adaptive determination method of the sensitive feature set was proposed.Multi-features in the time domain and frequency domain were integrated to construct the feature matrix,and then the feature quality was quantized by the evaluation indexes of monotonicity,correlation and robustness.Finally,the sensitive feature set was determined adaptively by combining the comprehensive evaluation index and clustering algorithm.(2)Aiming at the problem of low tolerance of the rolling bearing degradation evaluation model to outliers and false fluctuations,a multi-strategy optimization Support Vector Data Description(SVDD)method for rolling bearing performance degradation evaluation under optimal characteristics was proposed.The optimization strategies included: Auto-Associative Kernel Regression(AAKR)algorithm was introduced to reconstruct sensitive features to reduce the influence of outliers and improve the data quality of input model.The synthetic kernel method is used to combine different types of kernel functions to construct multi-kernel functions to improve the learning generalization ability of the model.The particle swarm optimization algorithm is used to obtain the important parameters of the model.For SVDD models with multi-strategy optimization,only early samples in the health stage are used to complete model training,and First prediction time(FPT)is determined adaptively and interference from outliers and false fluctuations to the model is overcome.(3)Aiming at the problems of the Remaining Useful Life(RUL)prediction method of the existing data-driven rolling bearing,such as insufficient time sequence information acquisition ability and difficulty in establishing accurate mapping relationship between monitoring data and degradation trend,the RUL prediction method of the rolling bearing under sensitive characteristics was proposed.It integrates the Convolutional Neural Network(CNN)and Long Short Term Memory(LSTM)to explore the local and overall associated nonlinear information contained in temporal signals.Combined with the linear regression module to mine the linear information of the time sequence signal,the prediction of the remaining service life of the rolling bearing is realized.