Study On Multi-Fault Prediction Of Rolling Bearing Based On Fusion Feature

At present,rotary machinery and equipment are widely used in various fields of production.As an important component of mechanical equipment,rolling bearings are prone to failure.Once the failure occurs,it may cause a series of reactions to enterprises,such as production equipment shutdown,economic losses and casualties.Therefore,it is necessary to carry out scientific quality management of rolling bearings,the most critical step is to monitor and accurately predict the trend of failure,to achieve preventive failure control.In this thesis,an intelligent prediction method based on manifold learning and LBSA-BTSVM is proposed to predict various fault states of rolling bearings,and the accurate prediction of bearing faults is realized.By establishing an active "monitoring-predicting-controlling" mechanism,it is practical to control the maintenance measures before the failures spread,which is of practical significance for the reliable operation of equipment and the maximization of corporate and social benefits.First,VMD and wavelet packet thresholding are selected to denoise the initial bearing signal.The vibration signal of the bearing not only contains the effective information but also contains a lot of noise,which makes it difficult to extract the effective features,so it needs to be de-noised.In this process,the strong adaptive decomposition of VMD is combined with the good denoising effect of wavelet packet.The bearing signal is decomposed into IMF components by VMD,and then the processed signal is obtained by the threshold denoising of wavelet packet,which is used as the signal basis in the feature extraction stage.Then,bearing signal fusion features after noise reduction are extracted.The information represented by single feature is limited and can not reflect the complete information of bearing signal.Therefore,multi-feature extraction is carried out by combining statistical parameters,energy,permutation entropy,power spectrum entropy and sample entropy.A multidimensional feature set that comprehensively reflects the fault type information is constructed by using the complementarity between different indicators.However,it will result in large amount of calculation and low prediction efficiency.Therefore,dimensionality reduction is carried out by t-SNE in manifold learning to obtain the best fusion essential vector,which is used as input of bearing fault prediction model.Finally,the accurate identification of the fault state is accomplished by a prediction model based on LBSA-BTSVM.For the bearing fault,it is difficult to obtain a large number of samples and belongs to the multi-classification problem.The binary tree support vector machine is selected to predict the bearing fault state,but its prediction effect depends on the selection of relevant parameters.LBSA algorithm has great advantages in dealing with optimization problems,so it is used to optimize the parameters of BTSVM prediction model in order to improve the accuracy of fault prediction.Then the validity of the model is verified by experimental analysis.According to the prediction results of rolling bearing faults,effective fault control measures are proposed.