Class mean kernel principal component fault diagnosis method based on combination kernel function

Abstract

The invention discloses a method for diagnosing faults of class mean value kernel principal components based on combined kernel functions, including S1: selecting appropriate N types of initial data samples, mapping the original sample data to a high-dimensional feature space H, and obtaining a new set of N types of mapping data S2: Calculate the class mean vector of each type of mapping data; S3: Select an appropriate kernel function K(x, y), and adjust the flexibility factor in the function; S4: Perform principal component analysis on the class mean vector; S5: Calculate the class-mean flexible kernel matrix D=(drs)N×N, and then center the class-mean flexible kernel matrix D into S6: Solve the eigenvectors α1, α2, ..., αm obtained by solving the characteristic equation, and normalize . The established fault diagnosis method improves the detection rate of faults, and the introduction of the flexibility factor meets the dynamic balance requirements of the fault monitoring model for sensitivity and robustness, and has better process monitoring performance.

Description

technical field [0001] The invention belongs to the field of fault diagnosis, in particular to a fault diagnosis method based on combined kernel functions of class mean value kernel and pivotal component. Background technique [0002] Slow drift refers to the phenomenon that system parameters deviate from the rated value, and the normal slow drift of the system affects the results of fault diagnosis. As the operating conditions of equipment become more and more complex, the kernel principal component analysis method is commonly used to solve the problem of fault diagnosis in the nonlinear field, but the traditional kernel principal component analysis method is suitable for steady systems, and there are defects that cannot be changed after the model training is completed. When applied to dynamic time-varying systems, there will be a large number of false positives and false positives. To this end, researchers have successively proposed some KPCA methods to update the monitor...

AI Technical Summary

Problems solved by technology

As the operating conditions of equipment become more and more complex, the kernel principal component analysis method is commonly used to solve the problem of fault diagnosis in the nonlinear field, but the traditional kernel principal component analysis method is suitable for steady systems, and there are defects that cannot be changed after the model training is completed. When applied to dynamic time-varying systems, there will be a large number of false positives and false positives.

To this end, researchers have successively proposed some KPCA methods to update the monitoring model in real time to adapt to the dynamic and time-varying system, which improves the model update efficiency to a certain extent, realizes the rapid diagnosis of dynamic data, and reduces the occurrence of false positives. However, the model update process is relatively complicated, the problem of parameter selection has not been solved, and there is a lack of adaptive control for systems prone to parameter drift

In addition, the use of a single kernel function for fault diagnosis of slow drift systems also has the disadvantage of high fault misdiagnosis rate

Images