Research On Fault Diagnosis Method Of Rolling Bearing Under The Condition Of Early Weak Sample Unbalance

Fault diagnosis is a technology to ensure the safe operation of mechanical devices,reduce the maintenance cost caused by faults in the life cycle,and improve the stability and reliability of mechanical devices.The rolling bearing is a core component of the rotating machinery device,which is widely used in gear box,crusher,gas turbine,generator and other devices.The performance of the rolling bearing will directly affect the operation state of the entire device,and then affect the safety and economic level of industrial production.Therefore,it is of great significance to study the fault diagnosis technology of rolling bearing.However,how to detect the early failure of rolling bearings during operation,accurately extract weak fault features,timely detect and diagnose faults in rolling bearings during operation;How to fault diagnose the data samples with only a few labels under the condition of difficulty in fault sample collection and incomplete data has become the focus and difficulty of current research.Based on the above background,taking the rolling bearing as the research object,combining with machine learning and migration learning,the research on the fault diagnosis technology of the rolling bearing is carried out in view of the weak fault features in the early stage and the small sample of label fault.The main contents are as follows:1.A dual-dimensional EKNN based early fault diagnosis method for rolling bearings was studied.Firstly,AE signals of early faults are collected to extract the characteristic information that can best represent the running state of the bearing in real time,and a mixed feature set is constructed,which is composed of the wave characteristics,time-domain and frequency-domain statistical characteristics of AE signals.Secondly,the membership matrix is used to calculate the parameters of in-class compactness and inter-class overlap,and the ratio of the two is used as the objective function for feature selection of SFS algorithm.Finally,a new density calculation method is applied to the traditional KNN classifier,which uses the EKNN classifier calculated in two dimensions of density and distance to reduce the sensitivity to K value and identify the optimal feature group more stably.The experimental results show that the proposed method is effective and can accurately diagnose the early faults.2.This paper studied a fault diagnosis method of TD-DCSAE based on deep transfer learning in the rare case of fault data of rolling bearing with label samples.First of all,using a new constraint to enhance sparse automatic encoder(CSAE)sparse,then stack multiple CSAE and form the depth of the constrained sparse automatic encoder(DCSAE)as generated against in the network to identify the network and to generate fight against network(GAN)mechanism,the iterative optimization,gradually improve the identification of the fault identification capability of the network.Then the parameter migration learning strategy is used to solve the problem of fault diagnosis with only a few tag samples.Finally,two data sets in different domains are used to verify the validity of the proposed method.Experimental results show that this method has good fault diagnosis performance,and is more effective than the existing intelligent method when there is only a small amount of label data.