Research On Industrial Process Fault Diagnosis Based On Bayesian Classifier

In the industrial process of actual production,people need to perform timely and accurate fault diagnosis on the entire industrial process to ensure the safe production of the entire industrial process.Fault diagnosis methods have developed vigorously in the past few decades.Among them,the data-driven fault diagnosis method performs real-time fault diagnosis based on on-site monitoring data.It does not require accurate and detailed mathematical models.It is very suitable for coupling,relevance,and correlation in this article.Industrial processes of great complexity.In the data-driven fault diagnosis methods,there are mainly signal processing,information fusion,neural networks in machine learning methods,support vector machines,and Bayesian classifiers in this article.The classification basis of Bayesian classifiers is It is the Bayesian formula.It is used to judge the posterior probability of each test data sample belonging to various fault types through the Bayesian formula.The calculation method is simple and the data sample is easy to handle.It has been widely used in the fault diagnosis of industrial processes.This paper takes TE process as the research object of industrial process fault diagnosis.TE process is an industrial simulation process.It is a simulation based on an actual industrial process.The primary purpose of the simulation is to provide application objects for various fault diagnosis methods.,Has been used by many scholars to verify the feasibility of fault diagnosis methods.Therefore,in this paper,the fault diagnosis of the TE process is studied based on the Bayesian classifier,and compared with other methods to verify the feasibility and accuracy of the method.In the actual on-site industrial process,most of the signal variables directly acquired by the sensor are continuous attribute variables.If the discrete attribute Bayesian classifier is used,the original data needs to be discretely processed.In this paper,considering the statistical attributes of the variable signal,it is assumed that each attribute variable obeys the Gaussian distribution,and the original data is discrete based on the three principles.化.After the actual collected raw data is subjected to the above-mentioned discretization processing,a part of the discretized data set is first selected as the training data set,and the naive Bayes classifier fault diagnosis model is trained for fault diagnosis.Then,the above-mentioned discretization processing is also performed on the data samples collected in real time,and input into the established fault diagnosis model.Finally,determine the fault category corresponding to the data sample and evaluate the fault diagnosis accuracy.The above method is applied to the fault diagnosis of partial faults in the TE process.The simulation results show that the diagnosis accuracy of the fault diagnosis is high,and the results show that the method is feasible.Although the continuous attribute data is discretized,the application of discrete attribute Bayes classifier in fault diagnosis can be realized.However,after the original data is discretized,the loss of data information may occur.And if the number of discrete intervals is too large,it will increase the complexity of subsequent calculations.To solve the above problems,the kernel density estimation method is used in the establishment of the continuous attribute naive Bayes classifier.Apply the trained fault diagnosis model to part of the fault diagnosis of TE process,respectively determine the fault category of each test data sample and evaluate its overall fault diagnosis performance,and compare it with other methods.The comparison result shows the fault of this method.The diagnosis accuracy is high.In order to verify the adaptability of the method,the method is applied to the actual process,and the simulation results show that the method is feasible.