Fault Diagnosis Of Complex Industrial Process Based On Bayesian Network

The safe and reliable operation of industrial processes is critical for the sustainable development and long-term profitability of modern industry.Therefore,fault diagnosis techniques that ensure the safe and reliable operation of industrial processes play a pivotal role in modern industrial processes.However,complex process characteristics,such as large-scale,complex variable characteristics,nonstationary characteristics,pose a serious challenge to fault diagnosis research.This paper studies the typical fault diagnosis problems such as key fault information extraction,fault identification and root cause variable traceability with complex process characteristics.The specific research content of this paper is as follows:(1)Considering the large number of fault-:independent variables and the redundant information between fault variables,a naive Bayesian classification based key fault feature extraction for fault diagnosis is proposed.Firstly,based on the mutual information,the key fault variables highly correlated with the fault are selected.Then,the key fault features independent of each other are extracted to further refine the key fault features.The diagnostic model constructed by the two-step progressive fault feature extraction strategy not only fully describes the fault information but also eliminates redundant information between features,which improves the performance of the fault diagnosis model and enhances the interpretability of the model.(2)Considering the mix:ing characteristics by dynamic and static fault information of industrial processes,the single static or dynamic fault information is not su:fficient to describe all fault types.Therefore,a distributed Bayesian network online fault diagnosis method based on collaborative analysis of static and dynamic information is proposed.By effectively combining dynamic and static fault information,the method deeply mines fault feature information,establishes a distributed diagnostic subnet based on fault features,and effectively decouples the mixed fault information for further fine-scale diagnosis in multi-fault classification and recognition.(3)Considering the adverse effects such as the causal relationship distortion caused by nonstationary characteristics in industrial process,the root cause fault variable will hardly be correctly located.Therefore,a hierarchy Bayesian network fault diagnosis method based on causal analysis is proposed.The diagnostic model includes a two-layer diagnostic network:the underlying network consists of multiple diagnostic subnets,which respectively describe the local fault transfer information of nonstationary variables;the upper layer network integrates the causal relationship of the underlying diagnostic subnets,and the overall cause relationship structure is constructed.This method can accurately locate the root cause fault variable and effectively identify the fault propagation path.