Application Of DICA And Improved NBC In Distributed Process Monitoring And Fault Diagnosis

When the chemical production process is abnormal,the effective process monitoring and fault diagnosis model can accurately and quickly detect the fault and determine its type,providing a timely and reliable reference for the operator.In the modern process industry system,the production links are closely connected,the observation variables have high dimensionality,mutual coupling and self-correlation,which adds a lot of difficulty to the fault diagnosis research.This paper proposes a distributed process monitoring model based on dynamic independent component analysis(DICA)and a class-specific weighted naive Bayes classifier fault diagnosis model for the complex characteristics of the process industry.The system variables are divided into several sub-modules based on process knowledge and data-based methods,and the DICA process monitoring method is used to extract fault information from each sub-module to avoid the interference of irrelevant information when mapping global variables.Due to the difference in process monitoring results of different sub-modules,it is binarized into the conditional attributes of Bayesian classifier and the class conditional probability is calculated,and a class-specific weighted naive Bayes classification model is constructed for fault diagnosis.The class-specific weight matrix is learned by optimizing the objective function.Emphasis is placed on the decision-making role of key condition attributes to improve the accuracy of fault classification.This paper uses TE process data sets to first compare the fault diagnosis performance of knowledge-based and data-based sub-module division methods;then,through a single-fault distributed process monitoring simulation experiment,it is verified that the distributed process monitoring method based on DICA can effectively extract fault information from key sub-modules;finally,through multi-fault classification simulation experiments,it is verified that the class-specific weighted naive Bayes classification can accurately classify the faults and has certain explanatory power.The experiment results prove that the distributed process monitoring and fault diagnosis model proposed in this paper is an effective method.