Bearing Fault Diagnosis Based On Deep Learning

With the intelligent development of modern machinery and equipment in the field of transportation,its fault diagnosis technology is facing new challenges.As an important component of high-speed train running part,bearing is the most important factor in fault diagnosis due to its complexity and uncertainty.Bearing operation under complex conditions may occur compound fault,that is,two or more fault occur simultaneously.Traditional bearing fault diagnosis methods have limited extraction of fault features and may lose useful information.Deep learning technology has the ability to adaptively extract features,which provides new research ideas for bearing fault diagnosis.This thesis based on the vibration signal data samples of different bearing faults,a bearing fault diagnosis model based on deep learning is built for training learning,so as to achieve the effect of diagnosing bearing faults.It is difficult to realize the collection of each type of compound fault data in actual operation scenarios,and most of the compound fault diagnosis methods are an extension of the single fault diagnosis method.Therefore,the capsule network structure is improved,and a bearing fault diagnosis model based on multi-convolution capsule network is proposed to solve the problems of compound fault data acquisition,vibration signal instability,and low bearing fault diagnosis rate.Firstly,the vibration signal is processed by wavelet transform to get the spectrum,and then CNN is used to extract features of the spectrum.Finally,the features are sent to the capsule layer.The encoder in the capsule network converts the scalar input into a vector output.The decoder performs feature reconstruction,outputs a reconstructed image,and calculates the loss.Experiments show that the model has better effect on the diagnosis of bearing faults than other methods.In order to study the status of bearings of trains,the vibration signals collected from them need to be studied and analyzed.How to effectively extract useful information from vibration signal data for bearing fault diagnosis is of great significance.Therefore,aiming at the problems of signal nonlinearity and low fault diagnosis rate,a beaeing fault diagnosis model based on deep convolutional neural network is designed.The model uses the ability of a Convolution Neural Network to adaptively learn features to extract features,which avoids the loss of useful information when manually extracting features.First,wavelet transform is performed on the collected vibration signals to generate a wavelet time-frequency map,and then features are extracted from the wavelet time-frequency map by using a convolutional neural network and input to the deep convolutional network to learn features.Finally,classification is performed by a trained classifier.Experimental results show that the recognition rate of bearing fault diagnosis models based on deep convolutional neural networks is much higher than the existing methods.