Research On Auditory Geometry Group Network And Its Application In Bearing Fault Transfer Diagnosis

Rolling bearing is the main part of rolling equipment which carries the rotating work.It is significant to make a precise diagnosis during machine working,which can guarantee industrial production,stabilize product performance,reduce or avoid major production accidents or even disasters.New large-scale equipment is becoming more and more complex,refined,and high-speed.At the same time,the large-scale the structure is becoming more and more complex.In traditional diagnosis methods,prior knowledge is often introduced as the basis of diagnosis.When it comes to new equipment,new scene and the requirements of industrial modernization,intelligent diagnosis has become a key problem that needs to be solved urgently.This paper focuses on intelligent diagnosis and focuses on fault diagnosis and transfer learning method based on convolution neural network.The contents and conclusions of this paper are as follows:1.Discuss the cause of rolling bearing fault,point out the features commonly used in classification task.Starting with the features,compare and discuss the treatment method of neural network and the principal component analysis.It is found that the neural network with auto-encoder structure improves the non-linear part lacking in principal component analysis.2.Comparing and studying the application of two kinds of neural networks in signal processing,pointing out their advantages and disadvantages.For the explanability,convolution network is selected as the basis of experiment.Using existing data to produce data sets,train the network to successfully classify faults.3.Three transfer learning methods are compared and studied,and their respective application scope is pointed out.Three methods are used for the experiment and the three methods are compared.This technique can be used for the diagnosis of a particular type of fault when the required number of samples is small but there are many other similar but different samples.4.In view of the lack of theoretical interpretation of neural network,transfer learning is guided by generating characteristic diagrams as theoretical explanations.Analysis methods such as energy calculation and characteristic frequency calculation are made for vibration signals.Finally,experimental verification of the characteristic diagram can explain part of the operation principle of convolution neural network in vibration signal diagnosis.Based on the theory and logical reasoning to make assumptions,through experimental verification,combined with the theoretical knowledge of neural network and rolling bearing vibration signal,it can effectively carry out the fault identification and migration diagnosis of vibration signal.The combination of deep learning and artificial intelligence technology is the focus of bearing vibration signal diagnosis,and also one of the most important topics in the mechanical industry.