Research On Hierarchical Intelligent Diagnosis Method Of Rolling Bearing

Rolling bearing is the key part of the gas turbine.Its operating environment is complex and it is also the fault source of frequent faults.Once the rolling bearing fails,it will affect the normal operation of other components,and even paralyze the whole system of the gas turbine.Therefore,it is of great significance to carry out the research on intelligent fault diagnosis of rolling bearing and master the health status information of rolling bearing in real time,with the intention of ensure its safe and efficient service.Taking the rolling bearing as the research object,based on the fault feature analysis and taking the intelligent algorithm as the carrier,this paper constructs three hierarchical and progressive rolling bearing diagnosis models,answers the three questions of whether the rolling bearing is faulty,where the fault location is and how the damage degree is and verifies the effectiveness of the proposed method on the rolling bearing test bench,The main research work is as follows:(1)Physical simulation and time-frequency characteristic analysis of rolling bearing fault.In this paper,the laser is used to artificially process fault bearings with different failure sizes on the inner ring,outer ring,and rolling body of rolling bearing.On the rolling bearing test bench,144 groups of rolling bearing vibration data of different fault positions and three fault degrees were collected at four speeds of 800r/min,1600r/min,2400r/min and 3200r/min and four loads of 0 N·m,1 N·m,2 N·m,and 3 N·m.The vibration signal of rolling bearing is analyzed in the time and frequency domain,and the time and frequency characteristics that can characterize the state of the rolling bearing are obtained.(2)Research on rolling bearing fault detection method based on random forest.The data set is expanded by a random sliding window,multiple decision trees are constructed by using time-domain characteristics,and the decision results are fused by integration method to establish a random forest model.The network search method is used to optimize the super parameters in the random forest,and the optimal model is obtained.The recognition rate of the fault state of the rolling bearing is more than 99.9%.Under the conditions of different sample sizes,different loads,different speeds and different data sets,the performance tests of the proposed rolling bearing fault detection method are carried out.The results show that without losing the detection accuracy,the model has low demand for the sample size of the training set,strong cross domain diagnosis ability,and can still maintain the minimum recognition rate of 99% under the variable data set.(3)Research on fault location method of rolling bearing based on FFG-ECNN.Focusing on the characteristics of rolling bearings,a weighted undirected frequency domain characteristic graph is proposed.The feature extraction method of weighted undirected graph based on frequency domain features(FFG)expands the frequency domain features of rolling bearings.An efficient convolutional neural network ECNN is proposed by introducing the Mish activation function and deep separable module with expansion factor.Under the condition of considering both model accuracy and parameter quantity,the expansion factor is optimized by grid search method,and the final rolling bearing fault location model is obtained.In the rolling bearing fault location identification task,the accuracy of the model can reach more than 99.2%.Finally,under the influence of sample imbalance and noise,the FFG-ECNN rolling bearing fault location model is tested.The results show that the rolling bearing fault location method proposed in this paper has high accuracy,low requirements for sample size,and can still maintain the minimum recognition rate of 99% in high noise environment.(4)Research on quantitative diagnosis method of rolling bearing based on the DFCNN.Aiming at the original vibration signal of rolling bearing,based on ECNN,the backbone network based on residual network,pyramid pool structure,path aggregation structure,and other feature fusion units are designed,and Xavier initialization is used instead of random initialization,a convolution neural network for depth feature fusion(DFCNN)is established.The effectiveness of the stochastic gradient descent method for network parameter optimization is verified,and the relationship between data set transmission times and model accuracy is discussed.Finally,fault tests under different sample sizes and different noise environments are carried out.It is found that the DFCNN model proposed in this paper can effectively identify the damage location and damage degree of rolling bearing with low requirements for sample size and excellent noise resistance.