Research On Intelligent Fault Diagnosis Algorithm Of Bearing Under Variable Operating Conditions

In the process of gradually improving the level of industrial mechanization,in order to avoid the loss caused by mechanical equipment damage,it is more and more important to ensure the normal operation of large-scale mechanical equipment with increasingly precise and complex internal structure.As one of the key parts of mechanical equipment,rolling bearing belongs to vulnerable parts.Once damaged,it will seriously affect the normal operation of mechanical equipment.meanwhile,in practical industrial applications,mechanical equipment is usually in a more complex external environment,which makes the system fault signal present the characteristics of randomness,uncertainty and fuzziness.Therefore,it is of great significance to carry out condition monitoring and fault diagnosis of rolling bearing under complex working conditions.Aiming at the problem of fault diagnosis of rolling bearing under the condition of load,speed change and noise interference,the key technical problems of deep learning and transfer learning in bearing fault diagnosis are deeply studied in this paper,which effectively improves the bearing fault diagnosis effect under complex working conditions.The main contents of this paper are as follows:(1)Because of its strong feature extraction ability,convolutional neural network was first used in the field of two-dimensional image processing.In view of its ability to characterize the complex mapping relationship between signal features and mechanical operation state,scholars in the field began to apply it to mechanical fault diagnosis.In this paper,rolling bearing fault was simulated and data were collected.Experimental results verified the one-dimensional convolution neural network The feasibility of network in fault diagnosis of bearing.(2)Aiming at the problem of load variation and external noise interference,a bearing fault diagnosis algorithm based on training double interference is proposed.In the training stage,two kinds of external interference are introduced to solve the load and noise problems respectively: the kernel corrupt algorithm is used to introduce interference to prevent the mutual adaptation between neurons in the same layer,and the random noise layer RNL is used to introduce interference to enhance the antiinterference performance in the fault identification stage of the model.The experimental results show that the model has a good effect on bearing fault diagnosis when the load changes and strong noise interference.(3)Aiming at the problem of different frequency sensitive regions under different rotating speeds,the concept of multi granularity layer is proposed.The input data of the first layer are reconstructed with multi granularity.The sensitive features of different frequency regions reconstructed by convolution check are effectively extracted,and the high-dimensional features of time-domain signals are effectively extracted.The experimental results show that the model can get a better effect of bearing fault diagnosis when the speed changes.(4)Aiming at the problem of label free fault diagnosis for rolling bearing,a new method based on deep convolution transfer learning is proposed.In this method,the source domain samples are used to pre train the deep convolution migration model,and the domain invariant fault features of the data are extracted to obtain the general knowledge of the data.Then,the pre trained model is applied to the problem of fault diagnosis without labels,so as to improve the diagnosis effect of the model for unlabeled faults.Three different types of bearing data sets are used for experiments.The results show that the model proposed in this paper can realize the bearing fault diagnosis without label in the same health state under different equipment.