Research On Fault Diagnosis Technology Of Rolling Bearing Based On Wavelet Analysis And Convolutional Neural Network

In terms of mechanical fault diagnosis,due to the more sophisticated and complicated mechanical devices,the previous fault diagnosis technology has been difficult to obtain satisfactory results.Today,most of the fault diagnosis relies on complex and low-accuracy artificial feature extraction and Expert knowledge.In response to this problem,combined with the recent mechanical health monitoring into the "big data" era,the rolling bearing is taken as the research object,and the one-dimensional convolutional neural network algorithm model is built.The model automatically completes the extraction feature and fault diagnosis.Through the research on the occurrence and evolution of the failure of the rolling bearing during the operation,the basic aspects of the fault diagnosis of the rolling bearing are mastered,and the failure mode and the principle of vibration and noise signal generation are analyzed.Taking the vibration signal of the rolling bearing as the research object,the rolling bearing information acquisition experimental device is built.The faulty rolling bearing is used as the fault rolling bearing by differently using different parts of the deep groove ball bearing.The vibration acceleration sensor is used to collect the signals generated by the rolling bearing under various loads,and the acquired signal is preprocessed in the form of wavelet denoising.A wavelet adaptive noise reduction processing method is designed.Through the singular spectrum distribution of different signal and noise ratios of different layers and the same layer after wavelet decomposition of the vibration signal,the optimal noise reduction decomposition layer is determined to obtain the optimal noise reduction effect of the vibration signal.According to the method of image feature extraction and recognition by convolutional neural network,a one-dimensional convolutional neural network model for vibration signal feature extraction and fault classification is proposed.The model has two layers of large convolution kernels,which use the batch normalization algorithm to make the model have fast training speed and very high classification accuracy.The one-dimensional convolutional neural network(AWT-1DCNN)is improved and the adaptive batch normalization is added due to the small amount of data,the lack of signal data,and the like,which may cause the accuracy of the fault diagnosis and recognition of the algorithm to decrease.The improved model is the TI-1DCNN model.The algorithm model is aimed at the above questions,and can train small batch data,add Dropout to reduce the loss of information data,and increase the data processing capability of the algorithm itself.At the same time,the variable load adaptability is greatly improved.Using data visualization technology,the process of diagnosing bearing signals based on the training interference AWT-1DCNN model is demonstrated.