Research On Rolling Bearing Fault Detection Based On Wavelet Packet And Clustering Algorithm

Rolling bearings are essential parts of rotating machinery systems,and their proper operation is critical to the performance and life of the mechanical system.Once a failure occurs,it may lead to the mechanical system not working properly and affect life safety.Through the fault diagnosis of machinery and equipment,its normal operation can be effectively ensured so as to maximize its value.This topic studies the theory and methods related to rolling bearing fault diagnosis,and explores the characteristics of vibration signals exhibited by its different fault states.In this paper,several groups of comparison experiments are conducted in the rolling bearing dataset of Western Reserve University,and the results show that the I-Canopy-Kmeans algorithm proposed in this paper performs well in each group of experiments,and its evaluation indexes all achieve better results with the highest accuracy rate of 97.97%.This indicates that the algorithm in this paper can be effectively applied to the field of machine fault detection with high feasibility and practicality.This paper focuses on the following aspects:For the clustering algorithm in fault diagnosis needs to set the number of clusters according to specific problems,requires a lot of computational resources and time,and is sensitive to noisy data,etc.,the use of wavelet packet transform is proposed along with the introduction of Canopy algorithm.Firstly,this paper introduces wavelet packet transform to reduce the dimensionality of the original signal,which can effectively reduce the dimensionality of the data and the influence of noise.Secondly,this paper uses the Canopy algorithm to coarse cluster the data,which avoids the problem that the number of clusters needs to be set in advance in the traditional clustering algorithm,and at the same time reduces the computational effort and improves the efficiency of the algorithm.Finally,this paper uses Kmeans algorithm to further cluster these clustering centroids to get the final clustering results.Compared with the traditional clustering algorithm,the method using wavelet packet transform and Canopy algorithm has better noise immunity and operation efficiency,and can diagnose the fault of rolling bearings more accurately.An improved Canopy-Kmeans algorithm is proposed for the problems that the selection of the initial centroid,the size of the specified k-value and the noise largely affect the final results of clustering in the traditional Kmeans algorithm.First,the initial centers are selected using the "furthest nearest" principle in the Canopy algorithm,thus avoiding the problem of poor clustering results due to the random selection of the initial centers in the traditional method.The basic idea of this principle is that the distance between any two Canopy centroids should be as far as possible when obtaining n Canopy;secondly,the selection of the threshold T1 and T2 in the optimized Canopy algorithm is done by using the Euclidean distance to find the mean of all data points and calculating the distance from the mean to all data points.The most suitable T1 and T2 are calculated by the crossvalidation method,and finally the number of generated Canopy is assigned to k,which is used as the initial centroid of clustering and Kmeans clustering is performed to achieve the optimal clustering effect.The experimental results show that the method proposed in this study can effectively avoid the problem that the traditional clustering algorithm tends to fall into the local optimal solution and make the prediction results more reliable and accurate.This study fully explores the key technology of rolling bearing fault detection,which can not only greatly improve the operational efficiency of mechanical equipment,but also ensure safe operation and bring higher economic benefits.In summary,the Canopy-Kmeans algorithm based on wavelet packet transform and optimization can provide an efficient and accurate rolling bearing fault diagnosis method,whose accuracy and applicability are significantly improved.