Research On Fault Diagnosis Method Based On Local Fisher Discriminant Analysis

With the complication of the production process and equipment,the occurrence of faults is increasing,and the harm of the faults is becoming more and more serious.Therefore,the fault diagnosis technology is of great significance for ensuring the safety of the production process and equipment.Due to the development of computer distribution systems,a large amount of fault data has been recorded and saved,and data-driven fault diagnosis methods have gradually emerged.Among them,local Fisher discriminant analysis has received extensive attention in the field of fault diagnosis because it can better separate different types of faults.However,local Fisher discriminant analysis still exists limits,such as the inability to deal with nonlinear problem and small sample size problem.This paper puts forward a series of improvement measures around the local Fisher discriminant analysis to improve the performance of fault diagnosis.The main contents of the full text include:aiming at the diversity of the non-linear characteristics of the fault data,this paper proposes a multi-kernel local Fisher discriminant analysis model.The model integrates multiple kernel functions into a multi-kernel function through weight coefficients,and then projects the original fault data to a high-dimensional feature space through mapping of multi-kernel function,and finally reduces dimensionality in the high-dimensional feature space through local Fisher discriminant analysis.The model can comprehensively describe the nonlinear characteristics of different faults,thereby enhancing ability to extract nonlinear features of the model and improving the accuracy of fault diagnosis.Considering that the kernel mapping will cause the data dimension to become higher and lead to the small sample size problem,this paper adopts a matrix exponentialization strategy to improve the divergence matrix.This strategy makes the divergence matrix of the model always reversible,so that when solving the eigenvalues and eigenvectors,the discriminant information of the null space of the original divergence matrix is retained.In addition,the matrix indexation strategy enhances the edge effect between the samples,which strengthens the discriminative ability of the model.Aiming at the problem of parameter selection for multi-kernel local Fisher discriminant analysis,an improved particle swarm optimization algorithm is used to optimize the parameters.Improved particle swarm optimization algorithm enhances the convergence speed and obtains a better solution at the same time through a series of designs such as population initialization,control parameter adjustment and population diversification.The optimized multi-kernel local Fisher discriminant analysis not only has optimal parameter settings,but also adapts to different fault diagnosis scenarios,and has a certain versatility.The simulation experiments of TE process and the working process of 6S35ME-B9 marine diesel engine prove that the above method can greatly improve the accuracy of fault diagnosis and verify the effectiveness and practicability of the method proposed in this paper.