Research On Rolling Bearing Fault Diagnosis Based On Machine Learning

As a key component in mechanical equipment,the state of rolling bearings affects the stable operation of the entire mechanical system.After long-term high load operation of rolling bearings,the probability of bearing failure increases.If remedial measures are not taken in a timely manner,the system may be paralyzed.Therefore,it is particularly important to carry out fault diagnosis research on rolling bearings,which can reduce equipment maintenance costs and avoid production accidents to a certain extent.The rolling bearing will produce a large number of vibration signals reflecting its own operating state during operation.However,due to the influence of other mechanical parts and working environment,the collected signals are often mixed with a large amount of noise and other redundant information,which makes the rolling bearing signals show high dimensionality and nonlinearity,resulting in "curse of dimensionality",and the efficiency of bearing fault detection directly will be very low.In response to the above issues,this paper takes rolling bearings as the research object,studies and improves feature extraction algorithms in machine learning and combines support vector machine(SVM)for fault identification.Finally,a complete rolling bearing fault diagnosis method is summarized.The main research works of this paper are as follows:(1)The collected rolling bearing fault signals have strong nonlinearity and non-stationarity,and using Euclidean distance to measure the similarity between samples will be limited,resulting in unsatisfactory results in extracting significant features from fault signals.Aiming at the above problems,a local linear embedding algorithm based on feature correlation is proposed.The algorithm first constructs a new feature set by processing the original data in segments to obtain the standard deviation within the segment.While retaining the periodic characteristics of the original rolling bearing fault signal,it achieves the purpose of preliminary dimension reduction and signal noise reduction.Then,it uses mutual information(MI)to evaluate the similarity between samples,so as to improve the accuracy of neighborhood selection.Finally,it reconstructs the high-dimensional local structure in the low dimensional space,Thereby obtaining significant features of high-dimensional samples.Validate the effectiveness of the algorithm in extracting features on the Case Western Reserve University(CWRU)bearing dataset.(2)Aiming at the problem of uneven spatial distribution of collected rolling bearing data,a feature extraction algorithm based on improved local metric learning was proposed.Firstly,the variational modal decomposition is used to denoise the rolling bearing fault signal to reduce the impact of noise on the subsequent feature extraction effect.Then,the robust local metric space is learned from each training sample and its relative neighborhood.The local metric matrix learned from each training sample is constructed into an integrated metric matrix,and the sample’s neighborhood is updated repeatedly to learn the optimal subspace,To keep samples in the same class as close as possible,and separate samples from different classes as much as possible.Finally,comparative experiments were conducted in the CWRU destructive dataset.(3)Combining the SVM algorithm with the improved method proposed in this paper,a complete rolling bearing fault diagnosis method is designed and tested on the self-collected dataset Data1 and Data2 in the laboratory.Realize the classification and identification of different rolling bearing fault data types,with a higher recognition accuracy compared to other methods.