Research On Prediction Method Of Rolling Bearing Performance Degradation Trend Based On Deep Learning

Rolling bearings are one of the key components of mechanical equipment,and their working status and performance directly affect the normal and safe operation of mechanical equipment.If the health or damage of rolling bearings cannot be predicted in time,it will not only affect the formulation of maintenance and replacement plans for mechanical equipment,but also cause serious economic losses and even safety accidents.Therefore,accurate prediction of the performance degradation trend of rolling bearings helps engineers to reasonably formulate bearing maintenance strategies and replacement plans to ensure the safe service of mechanical equipment and improve economic benefits.Based on the original vibration data,this thesis first uses the Multi-resolution singular value decomposition(MRSVD)method to process the original vibration data of the rolling bearing to obtain a mixed domain sensitive feature set with good monotonicity and high correlation.Then standardize the features and reduce the dimensionality to obtain the degradation index to locate the rapid degradation stage,and the mixed domain sensitive feature index is used as the input of the prediction model.Finally,a hybrid deep network is used to predict the degradation trend of rolling bearings.The research method in this thesis is a data-driven method.The main research content is: from data processing to the establishment of degradation indicators,and then build an effective deep learning model to predict the degradation trend of rolling bearings.The main research content is as follows:(1)In view of the noise interference in the original vibration signal of the rolling bearing,it affects the construction of the index.Use the self-optimizing multi-resolution singular value decomposition(MRSVD)method to process the raw vibration data of rolling bearings,and decompose them to obtain approximate signals containing degradation information and detailed signals containing a large amount of noise information,which are used in the index evaluation criteria The monotonicity of is used as the evaluation standard of the root mean square characteristic of each decomposition layer,and the optimal decomposition layer is determined to establish an appropriate degradation index.Simulation experiments are used to verify the comparison and prove the advantages of the proposed method.(2)There is a lot of information loss in the way of establishing indicators for a single feature,which results in insufficient reflection of the running state of the bearing,and the different sensitivity of each feature to the degradation process of the rolling bearing.First,use the MRSVD optimal decomposition layer to process the rolling bearing vibration data,and obtain approximate signals to extract features from the data samples,including: time-domain statistical features and frequency-domain statistical features,energy,spectral entropy,and 56 feature quantities,and These features are screened according to the degradation index evaluation criteria,and a mixed domain sensitive feature set with good monotonicity and high correlation is selected.Then standardize the selected feature set,use the principal component analysis method to reduce the dimension of the feature set,establish the performance degradation index of rolling bearing,and verify the feasibility and effectiveness of the method through experimental data.(3)Aiming at the problem of accurate and effective prediction of rolling bearing degradation trend,a hybrid deep network-based rolling bearing degradation trend prediction method is proposed.First of all,the rapid degradation moment of rolling bearing is located by the performance degradation index and the triple standard deviation method.Then,the sensitive feature set extracted and screened in the mixed domain after processing by the multi-resolution singular value decomposition method is divided into training samples and test samples;finally,the hybrid depth of the integration of the attention convolutional neural network and the bidirectional long-short-term memory network is constructed The network uses the attention convolutional neural network to learn the feature correlation of the multi-dimensional feature set,fully self-extracts and expresses the degradation features,and combines the two-way long and short-term memory network to make full use of the advantages of context information to obtain more accurate and effective rolling bearing degradation trend prediction results.The article concludes with a summary of the work done in this thesis,and looks forward to the next research.