Research On Rolling Bearing Fault Diagnosis Method Based On One-Dimensional Convolutional Neural Network And Transfer Learning

Nowadays,rotating machinery is increasingly showing the development trend of high precision,high efficiency and high automation.Rolling bearings as the key support component in rotating machinery,fault monitoring and diagnosis are essential to ensuring the efficient and safe operation of industrial production systems.In the context of industrial big data intelligent manufacturing era,how to efficiently and accurately dig out the information containing the health status of rolling bearings from the massive high-dimensional and rich variety of industrial big data is an urgent problem to be solved.Last few years,data-driven intelligent fault diagnosis methods for bearings have received extensive attention and research.Therefore,based on Onedimensional Convolutional Neural Networks(1DCNN)and Transfer Learning(TL),this paper conducts research on rolling bearing fault identification methods in complex environments.The main research work is as follows:(1)Aiming at the problems of noise in the original signal of rolling bearings and low diagnostic accuracy and efficiency of traditional convolutional neural network,a rolling bearing fault diagnosis method combining the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)and the improved multi-scale one-dimensional convolutional neural network is proposed.This method uses CEEMDAN and Pearson correlation coefficient method to denoise and reconstruct the original vibration signal At the same time,a multi-scale 1DCNN is designed and improved,which effectively preserves the characteristic information of the data and improves the accuracy of fault identification.The experimental results on the two datasets show that the method can preserve the sensitive features in the original data and complete the fault identification task,and has good generalization and noise resistance.(2)To address the problem of low accuracy of 1DCNN fault discrimination due to feature domain shifts occurring under variable load conditions,a domain-adaptive adversarial migration 1DCNN based on an attention mechanism is proposed,which uses a channel attention mechanism to reduce redundant feature information and makes the model focus more on features relevant to the classification task.In addition,an improved loss function is constructed to measure the distribution of features between the source and target domains,which improves the accuracy of cross-domain fault identification.The results show that the method shows better accuracy and generalization than traditional methods.(3)An intelligent fault condition monitoring and diagnosis system is developed in the Lab VIEW virtual environment based on a double span bearing fault simulation experiment bench.The system adopts eddy current sensors to collect vibration signals,and the ADLINK DAQ2214 data acquisition card completes the A/D conversion.At the same time,the algorithm designed in the paper is added to the mode recognition module,and a combination of Python and Lab VIEW is used to complete the data processing and analysis,thus realizing the status monitoring and fault identification of the experimental bench signals.