Fault Diagnosis Research Based On Improved Residual Shrinkage Network Under Unbalanced Data

With the increasing complexity of modern process industry,a large amount of process data needs to be monitored and recorded to ensure the normal operation of equipment.The production of large amounts of data has promoted the development of data-driven based fault diagnosis techniques.However,there are still some deficiencies in the current chemical process fault diagnosis methods.For example,the characteristics of some fault categories are small and difficult to distinguish.The fault diagnosis effect is poor under the condition of data imbalance.Therefore,in view of the above problems in chemical process fault diagnosis,this paper presents the following research work:(1)There are many kinds of faults in the process of chemical engineering,and the characteristic difference between some faults is small.The traditional fault diagnosis model can not distinguish these faults well.In order to solve this problem,this paper proposes a fault diagnosis model based on improved residual shrinkage network and bidirectional gated recurrent nerve unit.By constructing a multi-scale feature extraction structure with dynamic adaptive ability,the model can fully mine the static feature information in the data and extract the time-related information in the data combined with bidirectional gated recurrent nerve unit.The final model can fully extract the feature information and avoid the problem that the important features are covered when the number of signs increases,which improves the diagnosis ability of partial faults with similar features.(2)An improved residual shrinkage structure is proposed to replace the tedious wavelet denoising.By combining the soft threshold method with the attention mechanism,the network model can directly process noisy data,and adaptively complete noise reduction in the feature extraction stage,which realizes the integration of data denoising and feature extraction,and reduces the dependence on complex signal processing knowledge.(3)In the actual production process,the number of fault samples is far less than that of normal samples,and the commonly used fault diagnosis methods have poor diagnostic effect under the condition of data imbalance.Aiming at this problem,this paper proposes a fault diagnosis method under the condition of data imbalance.Multi-Scale Gradients Conditional Generative Adversarial Network is generated by using multi-scale gradients conditional generative adversarial network.MSG-CGAN is used to up-sample a few classes in the unbalanced data to construct the class-balanced data set,and then the MDRSN-Bi GRU model proposed in Chapter 3 is used for fault diagnosis.The experimental results show that the proposed method can effectively improve the fault diagnosis effect under the condition of data imbalance。...