Research On Fast Classification Method Of Rolling Bearing State Based On Improved Broad Model Transfer

As one of the key components of rotary mechanical equipment,rolling bearing is widely used in aviation,transportation,machinery production and other fields,it can prevent the accident occurrence of the equipment to diagnose the rolling bearing faults effectively.In the tasks of fault diagnosis for rolling bearing,deep learning network is mostly the basis for intelligent fault diagnosis technology,whose training is time-consuming.In addition,the vibration signal is widely used to infer the health status of the machine.When rolling bearing works,the load often changes according to the needs.The change of the load will affect the probability distribution of vibration signal in the feature space,then the effect of diagnosis will be affected.Therefore,it is very important to quickly realize the fault diagnosis for rolling bearing under different loads.Taking both broad learning network and model transfer learning method as the core technology,this thesis establishes the fault diagnosis model to realize the quick classification for rolling bearing states under different loads.For the time-consuming problem that the training in deep learning network,a fast classification method of rolling bearing states is proposed based on improved broad learning system(BLS).The original vibration signal of rolling bearing is pre-processed by fast Fourier transform to obtain the frequency-domain amplitude sequence,which is used as the input of BLS.It is proposed to build the enhanced nodes windows of BLS in a cyclic and extended way,and introduce the Maxout activation function into enhancement layer to construct the improved BLS.Combined with genetic algorithm(GA),the network parameters of improved BLS are optimized to realize the quick classification for rolling bearing states under different loads.The above method enhances the ability of feature extraction and generalization in BLS,but do not take it into account that the large difference of rolling bearing data probability distribution under different loads.Therefore,aiming at this problem,a fast classification method of rolling bearing states is further proposed based on improved broad model transfer network.The frequency-domain amplitude sequences of rolling bearing under a certain(some)load are selected as the source domain,and the frequency-domain amplitude sequences of other loads are selected as the target domain.The source domain data are put into improved BLS for extracting features and training,with the help of GA,the source domain network parameters are optimized,and a pre-training model is constructed.It is proposed to transfer the network parameters of the pre-training model,the weight parameters in the feature layer and the enhancement layer to the target domain network,the output layer of the target domain network is fine-tuned by ridge regression method,and establish the fault diagnosis model of rolling bearing under different loads.