Research On Intelligent Fault Diagnosis Of Rotating Machinery Under Strong Noise Less Sample Variable Working Conditions

At present,all kinds of rotating machines used in aerospace,shipping and transportation,wind power generation and other fields are becoming more and more complex,and the level of automation and intelligence is getting higher and higher.Once the the key components such as gears and bearings are damaged,it will lead to abnormal operation,shutdown of the whole machine,or cause casualties.Therefore,it is of great significance to conduct intelligent fault diagnosis for rotating machines.As the increasing maturity of mechanical fault diagnosis technology incorporating artificial intelligence,the higher performance intelligent fault diagnosis models could be constructed to further ensure the reliable and safe operation of rotating machinery.However,most of the existing models are driven by the monitoring data from known ideal conditions,and do not consider the characteristics of noisy and variable working conditions of rotating machines in real industrial scenarios,which makes the existing models generally have some defects such as low generality and poor self-adaptability.To overcome the shortcomings of the existing fault diagnosis methods,this paper takes deep learning as the technical core and selfsupervised learning and deep reinforcement learning as the theoretical basis to carry out the research on intelligent fault diagnosis technology for rotating machinery under strong noise interference,few labeled fault samples and variable working conditions,respectively.The main contents of this dissertation are summarized as follows:(1)To address the problem that the fault features of rotating machinery condition signal are easily disturbed under strong noise interference,which makes it difficult to achieve high-precision fault diagnosis,a fault diagnosis method based on mel-frequency cepstrum coefficient(MFCC)feature matrix and paralle multi fusion convolutional neural network(PMFCNN)is proposed.Compared with the traditional data-driven fault diagnosis method,this method can achieve a high-precision fault diagnosis for rotating machinery under strong noise interference.The designed MFCC feature matrix can reduce the effect of noise interference on fault characteristics.PMFCNN has a variety of network fusion layers and structural integration layers,which can further explore the fault features from input data,and improve the stability and robustness of fault diagnosis under strong noise interference.Finally,the effectiveness of the proposed method is validated using the bearing fault data set and gearbox fault data set.(2)To address the problem that it is difficult to obtain many labeled fault samples for building a high-performance fault diagnosis model in practice due to the tedious process of labeling rotating machinery fault data,a fault diagnosis method based on self-supervised convolutional neural network(SS-CNN)is proposed.Compared with the traditional supervised convolutional neural networks,SS-CNN can be trained by using unlabeled fault samples.After the initial network training is completed,only a small number of labeled fault samples need to be applied to optimize the network parameters to build a high-performance fault diagnosis model.Finally,the effectiveness of the proposed method is verified using the gearbox fault dataset.(3)To address the problem of poor adaptivity of fault diagnosis models due to variable operating conditions affecting the distribution of condition signal features,a fault diagnosis method based on adaptive capsule network(Acapnet)is proposed.Compared with the traditional fault diagnosis method,this method can realize the adaptive fault diagnosis under changing working conditions.By replacing the traditional fully connected neurons with capsule neurons,the feature extraction capability of Acapnet is improved.By introducing deep reinforcement learning algorithm and designing online feature dictionary,action execution method and reward function,the model can be updated online using the monitoring data of unknown labels under variable working conditions.Finally,the effectiveness of the proposed method is verified using two bearing fault data sets which contains multiple working conditions.(4)A fault diagnosis system for rotating machinery is designed and developed based on the above research findings.The overall system architecture and functional modules are introduced one by one.Moreover,paddle shaft drive system test bench,turbine generator set platform and aviation hydraulic pump platform are used to verify the effectiveness of the theoretical research and the feasibility of the system.