Research On Fault Diagnosis Method Of Rolling Bearing Based On Generative Adversarial Network

Rolling bearing is one of the important parts of rotating machinery,which is widely used in various transmission systems.In practice,the operation of rotating machinery will be affected by the environment and dynamic load,which may lead to the damage of rolling bearings and other parts,thus affecting the operation status of the whole system.Therefore,it is of great significance to carry out the research of rolling bearing fault diagnosis for rotating machinery.Generative Adversarial Network is a popular data generation model in recent years.It can learn the data distribution of target samples and generate false data that similar to the distribution of target samples,which can solve the problem of insufficient data.This paper mainly studies the fault diagnosis technology of mechanical rolling bearings based on Generative Adversarial Network,and aiming at the disadvantages of poor quality and single for the generates data to improve the Generative Adversarial Network.The main research work is as follows:(1)This paper introduces a fault diagnosis algorithm based on Generative Adversarial Network.There are more normal data and less fault data in the vibration signals collected by rotating machinery in actual operation,which makes the traditional deep learning model have poor classification of unbalanced samples.The Generative Adversarial Network model was introduced into fault diagnosis of rolling bearings,the adversarial learning mechanism was used to expand a small amount of fault data,build a network with higher fault diagnosis accuracy.Finally,model is used to process the vibration data of rolling bearing,good results are obtained.(2)Aiming at the problem of weak feature extraction ability of Generative Adversarial Network,this paper combined the Conditional Generative Adversarial Network with residual network to optimize the structure of generator and discriminator,which improved the ability of feature extraction and feature learning of the network,and improved the accuracy of rolling bearing fault diagnosis.(3)Aiming at the problem of gradient disappearance and poor diversity of generated results in training process of Generative Adversarial Network,this paper proposes a residual Condition Generative Adversarial Network based on evolutionary algorithm.Different loss functions are introduced to select the optimal generator,which can avoid the problem of unstable training caused by single loss function,improve the stability and fault classification ability of the network.(4)Aiming at the problem that it is difficult to obtain data labels in actual industrial production,a fault diagnosis algorithm based on semi-supervised evolutionary residual Condition Generative Adversarial Network is proposed.This model combines the data generation,feature classification ability of evolutionary residual Condition Generative Adversarial Network and semi-supervised learning ability of semi-supervised mechanism.In the case of a small number of labeled samples,it can use the data information contained in unlabeled samples to train the network and improve the diagnosis and classification ability of the network.At the same time,the experimental platform is used to compare with other semi-supervised and supervised methods.The results show that this method has high accuracy and good performance in the case of only a small number of labeled samples.