Fault Diagnosis Of Non-Gaussian Stochastic Systems Based On Generalized Correntropy Criterion

In recent years,control systems have been moving toward large-scale and intelligent.Complex system structure,harsh working environment and long-term operation may lead to system failure,which threatens life safety and causes economic loss.Therefore,fault diagnosis is an essential part of ensuring safe and reliable operation of the system.In the actual industrial process,the complexity of system modeling and the inevitable non-Gaussian noises are two major problems in fault diagnosis technology.Based on the generalized correntropy criterion,model-based and data-based fault diagnosis methods are proposed for non-Gaussian stochastic systems in this paper.For non-Gaussian stochastic systems,where sensor failure may occur and the model is known,the generalized correntropy criterion,which can fully characterize the statistical properties of non-Gaussian random variables,is used to establish a filter.Secondly,a group of generalized correntropy filters are used to detect residual signals to determine whether the system has fault.Subsequently,a fault location function is established based on the state estimation error to determine the fault sensor.Finally,the signal reconstruction is realized through a set of filters,and a fault tolerant control algorithm is established to make the system run stably.In order to solve the problem of difficult system modeling,dynamic principal component analysis(DPCA)method is widely used in fault diagnosis.However,the traditional DPCA fault diagnosis method is based on the Gaussian assumption of the system.For non-Gaussian systems,the diagnostic results are not ideal.To solve this problem,this paper proposes a DPCA fault detection method based on generalized correntropy criterion.The method is mainly divided into two steps: offline modeling and online detection.The offline model is mainly to establish a generalized mutual entropy DPCA model and use the kernel density estimation method to obtain the control limit.During the online inspection process,firstly,collecting non-Gaussian stochastic system process variable data,and then an online generalized correntropy DPCA model was establish based on these data.Secondly,calculate the statistics,and finally compare the statistics of the online model with the control limits in the offline model.When the statistics exceed the control limit,the system is in a fault state at this time.