Research On Key Techniques Of Rotating Machinery Fault Diagnosis Based On One-Dimensional Convolutional Neural Networks

With the development of information technology and the improvement of industrial intelligence level,intelligent data-driven fault diagnosis technology facing complex systems has gradually become a research hotspot.In this paper,the key components of rotating machinery(bearings and rotors)are taken as research objects.In the context of big data,the fault diagnosis methods have some key issues,such as complex artificial feature extraction,complex network caused by large amount of input data,and insufficient network training due to a small number of target task samples.To address these key issues,one-dimensional convolutional neural networks(1D CNNs)fault diagnosis method combining multi-sensor,compressed sensing and transfer learning is respectively proposed,based on the onedimensional characteristics of vibration signals and the advantages of convolutional neural networks.According to different applications and monitoring requirements,the effectiveness and superiority of the proposed method are proved by rolling bearing and rotor experiments.The specific research content of this paper is as follows:(1)The vibration signal has one-dimensional characteristics.For the problem that the traditional two-dimensional convolutional neural network destroys its spatial characteristics,we proposed 1D CNNs with overlapping pooling,dropout and adam optimization algorithms.It can automatically extract features from a single sensor and the vibration signal collected from the key components of the rotating machine is used for fault diagnosis,which proves that the method has high test accuracy.(2)For the problem that single sensor has limitations and multi-sensor processing technology relies too much on prior knowledge and feature extraction,the 1D CNNs based on multi-sensor fault diagnosis method is proposed.Firstly,data layer fusion is performed on the information from multiple sensors,and then the feature extraction and fault diagnosis are performed on the fused data by using 1D CNNs.The bearing experiment proves that the method can effectively identify and classify different faults.(3)In order to solve the problem of large data storage,transmission and computational pressure caused by large amount of data collected during the big data monitoring process,the 1D CNNs fault diagnosis method based on compressed sensing is proposed.Experiments show that the proposed method can ensure the accuracy and reduce the diagnosis time and the network complexity without reconstructing the compressed signal.This provides a reference for the practical application of compressed sensing theory in the intelligent fault diagnosis of rotating machinery.(4)In order to solve the insufficient network training caused by a small number of target task samples,1D CNNs fault diagnosis method based on transfer learning is proposed.The method takes the data collected under one working condition as the source task data,and the target data is diagnosed under small sample.The experimental results show that the proposed method has higher accuracy even under small sample.