Bearing Fault Diagnosis Based On Cyclostationay Analysis And Convolutional Neural Network

Bearings are important components in the transmission system of rotating machinery and play an important role in the stable operation of machinery.The traditional bearing fault diagnosis methods rely on expert experience to design features manually,which is difficult to meet the demand for accurate,fast and intelligent diagnosis of modern equipment.Therefore,in the context of the big data-driven Industry 4.0 era,it is important to deeply study the intelligent fault diagnosis of electromechanical equipment under complex working conditions.This paper takes rolling bearing as the research object,based on the theory of cyclostationary analysis and convolutional neural network,conducts an in-depth research on the fault diagnosis technology of rolling bearing under complex working conditions,and verifies the effectiveness and reliability of rolling bearing fault diagnosis method based on cyclostationary analysis and convolutional neural network through the combination of theoretical analysis,simulation,experimental verification and comparison.The main studies are as follows.(1)Aiming at the characteristics of bearing fault vibration signals,a rolling bearing fault diagnosis method based on spectral correlation density and convolutional neural network(SCD-CNN)is proposed.Firstly,the demodulation performance of the second-order cyclostationary analysis method is verified using simulated signals.Secondly,the experimental validation is performed using the standard bearing dataset of Case Western Reserve University and compared with the signal preprocessing methods based on short-time Fourier transform(STFT)and continuous wavelet transform(CWT),and the results show that the SCD-CNN method has a high fault identification accuracy.Finally,the output feature maps of each layer of the convolutional neural network and the classification process are visualized to further explain the advanced nature of the SCD-CNN method and verify that SCD-CNN is an effective method for intelligent fault diagnosis of bearings,and its performance is better than that of STFT-CNN and CWT-CNN methods.(2)The generalization capability of the SCD-CNN model is studied.First,the adaptation ability of SCD-CNN model under variable load conditions is investigated.Secondly,the generalization ability of SCD-CNN model in the case of few samples is researched.Finally,the robustness of the SCD-CNN model under the category imbalance condition is verified.The results show that the SCD-CNN model has good generalization performance and can maintain high fault detection rate under all three conditions.(3)Aiming at the problems of performance degradation and low fault recognition rate of SCD-CNN model in non-Gaussian noise environment,a rolling bearing fault diagnosis method based on cyclic correntropy spectral density and convolutional neural network(CCSD-CNN)is proposed.The model utilizes the noise reduction performance of correntropy on Gaussian noise and non-Gaussian noise,improves the noise suppression ability of convolutional neural network in non-Gaussian noise environment,and effectively improves the rolling bearing fault feature extraction and pattern recognition performance.To address the problem that the neural network classification process is difficult to explain,the CCSD-CNN model is used to reveal the process of fault classification under noise in detail by using data visualization techniques.