Research On Bearing Fault Diagnosis Method Based On Improved LLE Algorithm

Bearing is an important component of various large-scale mechanical equipment,which has a significant impact on the performance and operating state of mechanical equipment.Therefore,monitoring the health of bearings is very important.The collected bearing vibration signal contains a large amount of information that can reflect the operating status of bearing equipment.But a great deal of data can also lead to information redundancy,increase computational complexity,and reduce the accuracy of fault identification.Therefore,before fault identification,it is necessary to extract features from the collected high-dimensional data to obtain significant features which can characterize bearing data and achieve efficient diagnosis of bearing faults.This paper mainly conducts in-depth research on the Local Linear Embedding(LLE)algorithm in manifold learning,improves on some of its existing problems,and uses Support Vector Machine(SVM)algorithm for classification and recognition,and proposes a complete set of bearing fault diagnosis methods.The main research work of this paper is as follows:(1)The LLE algorithm is sensitive to neighborhood parameters and only extracts features by mining a single geometric structure.To solve the above problems,the dual weight local linear embedding algorithm based on adaptive neighborhood(AN-DWLLE)algorithm is proposed.Firstly,the candidate neighborhoods of the sample are estimated using cosine similarity and the center point is calculated.According to the distances between the sample and the centers of the candidate neighborhoods,an appropriate neighborhood parameter is estimated.Then,the neighbor sequence structure is mined using the neighbor distance sequence information,and integrated with the local linear structure to obtain more robust weight.Finally,the weight is kept unchanged in the low dimensional space to achieve feature extraction of the data.A large number of experiments have been conducted on the Case Western Reserve University(CWRU)dataset,and the results show that the AN-DWLLE algorithm can extract more significant bearing signal features compared to other relevant methods.(2)Aiming at the problem that the LLE algorithm does not fully consider the sample neighborhood linearity when constructing low dimensional objective function,resulting in inaccurate low dimensional embedding results,the dual weight Local Linear Augmentation Embedding algorithm based on adaptive neighborhood(AN-DWLLAE)algorithm is proposed.Based on the AN-DWLLE algorithm,the neighborhood linear enhancement strategy is introduced into the construction of low dimensional objective functions.By constructing the mean and variance models of sample neighborhood linearity,significant features of the samples are obtained.The validity and robustness of the AN-DWLLAE algorithm for feature extraction are verified using CWRU data set.(3)Combining the algorithm proposed in the above study with the SVM algorithm,a complete set of bearing fault diagnosis method is proposed.A large number of experiments have been conducted on the dataset collected by our laboratory,and the results show that the proposed methods have significant advantages over other related methods,which can effectively extract significant features of bearing data and achieve higher fault identification accuracy.