基于改进的动态核独立成分分析的故障监测

Fault Detection Based on Improved Dynamic Kernel Independent Component Analysis

随着现代工业自动化设备中计算机技术与传感器技术的快速发展和广泛应用,基于数据驱动的故障诊断方法也获得了进一步的发展。由于实际工业过程数据的非高斯性、非线性和时序性,为了解决工业过程出现的这一问题,提出一种基于改进的动态核独立成分分析方法。首先,通过对自回归模型中测量矩阵引入时滞参数得到一个适用于动态系统的增广矩阵;然后,选择核函数,计算核矩阵,将增广矩阵映射到高维空间进行白化;最后,利用改进的快速ICA方法提取出独立成分实现对新的测试数据进行在线监测。实验结果表明,该方法具有不错的效果。

This paper presents an improved dynamic kernel independent component analysis method because of the non-Gaussian,nonlinear and timing of actual industrial process data. Firstly,this paper introduces a delay matrix for the dy-namic system by introducing the delay parameter into the measurement matrix in the autoregressive model.Then it selects the kernel function to calculate the kernel matrix and map the augmented matrix to the high-dimensional space for whiten- ing. Finally,The improved fast ICA method extracts independent components to enable on-line monitoring of new test data.