Research On Automatic Classification Algorithm Of Rolling Bearing Fault Signals Based On Deep Learning

Rolling bearings are one of the most common components in rotating machinery and are the core components of much heavy machinery.Rolling bearings can reduce friction between rotating parts and make the machine work effectively.As a precision mechanical support element,whether the rolling bearing is running normally affects the performance of the entire equipment directly.Rolling bearing faults can lead to mechanical failures even serious accidents.Therefore,it is necessary to monitor the running state of the rolling bearings to detect any abnormalities and possible failures in time to ensure the normal operation of the rolling bearings.Although many important advances have been made in bearing fault diagnosis,we still need to explore better fault identification methods in the background of big data.This topic uses deep learning to study the intelligent classification of bearing fault signals.Firstly,the manifestations of common faults of rolling bearings are analyzed.The bearing fault signals are processed by spectrum analysis,envelope analysis and wavelet packet analysis respectively,for extracting the fault characteristics.The effects of the number of network layers,iterations and nodes of hidden layers on the performance of deep neural network(DNN)are studied.A deep neural network structure suitable for bearing fault signal identification is built.Based on the above foundations,an intelligent automatic classification algorithm for bearing faults based on deep learning is proposed.The algorithm extracts signal features from the frequency domain and sets a random number of labels for each sample to build a sample dataset.This sample dataset is used to train the constructed DNN,and the labels of the samples are adjusted by using the sub-signals to test the trained DNNs,then trains the DNN by modified sample library again.After iterations of DNN training and testing,fault samples with similar characteristics are grouped together in the same class.The algorithm proposed in this paper is tested with experimental data provided by Case Western Reserve University(CWRU)Bearing Data Center.The results show that the bearing data of the drive end 12kHz and 48kHz and the fan end 12kHz are automatically divided into 7,6 and 4 categories.This indicates that the algorithm proposed in this paper has strong automatic classification ability.The algorithm proposed in this paper can automatically cluster multiple fault signals without intervention of human knowledge,which provides certain technical support for intelligent identification and diagnosis of bearing faults.