Relative Discriminant Analysis Based Fault Diagnosis Analysis In Industrial Processes

With modern industrial processes becoming more and more complicated,once in the event of fault,the whole system will be in abnormal situation,leading to systen paralysis,huge economic losses and even casualties.Therefore,it is necessary to monitor the industry processes,take effective actions to diagnose the faults and take immediate operations to avoid remedying the system,which are all of great importance to the safety and reliablity of the industry processes.Bias of data location and increase in data variations are two typical disturbances,which in general,simultaneously exist in the fault process.This thesis focuses on the analysis of these two fault disturbances and further studies how to improve the performance of fault diagnosis.In this regard,my research work mainly focuses on the following aspects:1.Considering that single method cannot well extract information of different disturbances,a nested-loop fisher discriminant analysis(NeLFDA)algorithm and relative changes(RC)algorithm are effectively combined for analyzing the fault characteristics,which is named as NeLFDA-RC two-step strategy.The proposed fault diagnosis method based on this two-step strategy can not only effectively extract the variations of data location but also analyze the variance variations,which can help improve the fault diagnosis performance.2.Considering that not all the variables are affected by disturbance and those unaffected normal variables may joepardize the extraction of fault features,a faulty variable selection strategy is proposed based on sparse FDFDA algorithm to select the significantly influenced faulty variables,which can help deeply study the fault process characteristics.3.Considering that the faulty variables and normal variables contain different process characteristics,a fault diagnosis method based on concurrent analysis of fault information and normal information is proposed,which establishs different fault diagnosis models for faulty variables and normal variables respectively and then intergrate the dianogsis results of these two kinds of models by a probabilistic fault diagnosis strategy.