Research On Composite Fault Diagnosis Technology Based On Extracting Time-Frequency Waveform Features Of Rotor Equipment Vibration

With the widespread popularity of Industry 4.0,the field of intelligent manufacturing has developed rapidly,and there is a demand for intelligent diagnosis and operation and maintenance of mechanical equipment production,manufacturing,and maintenance.During the operation of mechanical equipment,unpredictable problems with components may lead to the collapse of the entire mechanical system,leading to safety issues and economic losses.Intelligent fault diagnosis for mechanical equipment has become increasingly important in this context.This article first introduces international and domestic policies and strategies related to intelligent manufacturing,as well as the importance of artificial intelligence and data-driven technology in intelligent manufacturing;Then the existing research methods in the field of fault diagnosis are investigated,and a simple summary is made;Subsequently,based on the advantages of existing research,optimization was carried out for the two most important links in the field of fault diagnosisfeature extraction and fault diagnosis models.Solutions were proposed for a series of problems in composite fault diagnosis in the field of fault diagnosis;Finally,this article successfully implemented deep feature mining on composite fault datasets using this solution and explored intelligent diagnostic methods based on data-driven and artificial intelligence.The integrated operation of data collection,feature extraction,feature selection,fault classification,and model deployment for composite fault diagnosis was achieved,making a certain contribution to improving the intelligent level of production and manufacturing.The main work of this article is as follows:(1)A fault feature extraction method based on filter banks,called F-Fbank feature extraction,is proposed to address the problem of poor performance and classification accuracy in feature extraction of fourteen types of composite faults in the composite fault dataset,which is commonly used for single faults.This method draws inspiration from the idea of Fbank feature extraction in speech recognition and can better extract the features of composite fault signals.Simulation results show that after F-Fbank feature extraction for fourteen types of composite faults,the generated feature images have obvious feature differentiation in the divided L,M,R,and A-I regions,which can provide good feature input for the composite fault diagnosis model.(2)In response to the problem that the current fault diagnosis models obtained using deep learning methods are generally large and difficult to deploy in resource-constrained front-end environments,various convolutional neural network models proposed in previous studies were combined to perform fault classification tests on the composite fault Fbank feature dataset obtained through F-Fbank feature extraction.LeNet was selected as the baseline model,and some network structures were optimized based on it to achieve high accuracy Small model size,and better front-end deployment support.Finally,the LeNet model(LeNet-F)designed and implemented for composite fault diagnosis was able to achieve 97.41%accuracy in fault classification testing with a model size of 2.4MB.The model was deployed on the K210 development board,proving the feasibility of implementing composite fault diagnosis on resource-limited front-end devices.