Research On Fault Diagnosis Algorithm Of Auxiliary Classifier Generative Adversarial Network Based On Multi-generators

With the increasing integration and intelligence of industrial production,the operating conditions and working environment of modern mechanical equipment have become more changeable.While greatly improving industrial production efficiency,it also increases the probability of failure of mechanical equipment and its parts.In the production process,once mechanical equipment and its parts fail,it might cause huge economic losses and even casualties.Therefore,the research of effective and accurate fault diagnosis methods has become one of the current research hotspots.The development of sensor technology has made it possible to obtain massive amounts of data.At the same time,the development of computer technology has promoted the rapid development and wide application of data-driven fault diagnosis methods.As a newly emerging branch of data-driven methods,deep learning can handle complex nonlinear data and has a stronger ability to extract features.Therefore,deep learning is widely used in image processing,speech recognition,and fault diagnosis.Due to the complex working environment of the equipment,less fault data can be collected,resulting in a lack of labeled fault data.The existing fault diagnosis methods are difficult to achieve high-precision fault diagnosis.This thesis applies generative adversarial network to fault diagnosis,and expands the fault data set by generating new samples through its generator.This thesis starts from the background of fault diagnosis based on deep learning,which is based on the generative adversarial network models.Aiming at the problem of mode collapse and single input noise in the process of applying generative adversarial network to fault diagnosis,the fault diagnosis method based on generative adversarial network is improved and researched.1)Aiming at the problem of mode collapse when the generative adversarial network is applied to fault diagnosis,the fault diagnosis method of auxiliary classifier generative adversarial network(ACGAN)based on multi-generators is proposed.First,select the number of generators according to the number of failure types under actual working conditions to ensure that each generator can generate one type of fault data.Then replace the original single generator structure with a multi-generator structure,and use different generators to capture different high-probability modalities Finally,when the generator and the discriminator of the new model are alternately trained,different generators are encouraged to generate corresponding types of generated samples by assigning corresponding class labels.This method improves the feature extraction ability and generalization ability of the model,and solves the problem of mode collapse.Experiments show that the proposed new model can capture more high-probability modes and generate high-quality samples,and has good generalization ability and high diagnostic accuracy for fault data under different working conditions.2)Aiming at the problem of the single input noise of the generative adversarial network in the fault diagnosis process,a method of noise preprocessing is studied.According to the difference between the mean and variance,the random noise is classified and processed first,and the noise with different mean and variance is obtained after processing.Then,select random noise with different mean and variance from the processed noise and input it into the generator,and replace the original single noise input of the generator with a multi-Gaussian noise input.Multi-Gaussian noise can contain more effective features,improve the diversity of samples,and solve the problem of single input noise.Finally,each generator is guided to generate different types of fault data through class tags.Experimental results show that the proposed noise preprocessing method can carry more information and expand the search space.It also increases the randomness and diversity of the sample,thereby improving the accuracy of model fault diagnosis.