Research On Fault Diagnosis Of Rotating Machinery Based On Vibration Signal Analysis

Rotating machinery has been widely used in modern industry.The stable operation of machinery directly affects the efficiency and safety of industrial production.Therefore,it is an effective means to diagnose the rotating machinery and find the hidden trouble in advance to ensure the safety of production personnel and avoid economic losses.The traditional fault diagnosis of rotating machinery requires manual preprocessing of signals and selection of appropriate fault features.Although a lot of achievements have been made in this kind of diagnosis scheme,it relies on a certain amount of expert experience and knowledge,which increases the difficulty of fault diagnosis.With the development of deep learning and the intellectualization of industrial equipment,a large number of equipment state data are constantly acquired.A large amount of industrial data and the excellent performance of deep learning in feature extraction and pattern recognition make deep learning provide a new research method for rotating machinery fault diagnosis.Therefore,this thesis takes rotating machinery as the research object.Aiming at motor eccentric fault and rolling bearing fault in rotating machinery,the end-to-end fault diagnosis based on traditional diagnosis scheme and deep learning was studied respectively by analyzing their vibration signals.The main research contents are as follows:1.In order to realize the fault diagnosis of motor eccentricity,the fault mechanism of motor eccentricity is analyzed.Considering the extra vibration component in the motor vibration signal caused by motor eccentricity,a time-spectrum energy feature based on Hilbert transform was designed,and the support vector machine with particle swarm optimization was used as the fault diagnosis classifier.The experimental results show that the designed time-frequency spectrum energy feature can effectively extract the fault information of motor eccentricity,and the diagnostic effect is better than the two compared fault features under the same classifier.2.Aiming at the diagnosis of various faults of rolling bearing in rotating machinery,a convolutional neural network based on multi-scale features is proposed to automatically extract fault characteristics for fault diagnosis.The proposed model directly combines the traditional manual design fault features with the classification model and can automatically extract fault features from the time-domain signals to realize fault diagnosis.Compared with the traditional convolution neural network,the proposed model uses convolution kernels of various sizes in the first layer to obtain features of various scales.Through experiments,it is found that the proposed diagnosis model can well realize fault diagnosis,and through the visualization of model output,it is verified that the model can classify different bearing faults in feature space.3.Aiming at the problems that rolling bearings in rotating machinery often encounter changes in workload and noise in vibration signals in actual scenes,a deep learning model optimization scheme based on adaptive batch standard layer and model parameter transfer is proposed.Based on the proposed multi-scale feature convolutional neural network model,this scheme uses the mean and variance of the signal in the target domain to replace the mean and variance of the source domain in the batch normalization layer of the model,and the feature extraction layer parameters trained in the source domain are transferred to the model under the target domain,which improves the diagnostic accuracy of the model under variable load and noisy signals.