Research On Fault Diagnosis Method Based On Probability Frame

Industry is the lifeblood of national economic development,and the safe,efficient,and sustainable operation of industrial production is the fundamental guarantee for maximum benefit.In this paper,we carry out research on fault diagnosis methods in the field of data driving,extract key feature information from high-dimensional industrial data,find faults that threaten industrial production in time,and ensure the safety of industrial production.Specific research contents are as follows:First,according to the characteristics of generating high-dimensional data in the modern process industry,a fault diagnosis method based on the probability frame for bhattacharyya bound and the minimum error minimax probability machine is proposed.This method takes the minimization of the upper bound of Bayesian classification error probability directly as the core idea,overcomes the shortcomings of traditional linear discriminant analysis method that is easily affected by the dominant category,and enhances the accuracy and interpretability of the model.In order to obtain sufficient information,Gram-Schmidt orthogonalization is used in this paper to obtain multidimensional projection vectors,so that the dimension of the data after dimension reduction is not limited to one dimension.Second,the objective function of the minimum error minimax probability machine was modified to adapt it to multi-category fault diagnosis.A sparse minimum error minimax probability machine with L-1 regularization term is proposed to overcome the phenomenon of overfitting the data due to high model complexity.Third,in view of the extremely strong dynamic correlation of industrial data,the dynamic expansion of the original data is applied in the fault diagnosis method based on probability framework,which can make full use of the dynamic information of the data in the process of fault diagnosis,and then significantly reduce the error rate of fault identification.At the same time,the information discrimination criteria are proposed to determine the dynamic expansion order and the dimension after dimensionality reduction of fault data,which greatly reduces unnecessary calculations.Finally,the performance and effectiveness of the proposed method are verified by experiments on multiple faults sets in the Tennessee-Eastman simulation process.