基于非线性主元子空间的故障模式识别方法

Fault Pattern Recognition Based on Nonlinear Principal Component Subspace

针对多元统计过程监控中的故障源识别问题,提出一种非线性主元子空间方法识别故障模式。该方法对不同类型的故障数据进行核主元分析,获得描述数据主要变化的非线性主元子空间,以此为基础构造故障模式分类器。考虑到核主元分析的计算复杂性,提出一种基于特征样本的非线性主元子空间算法,使用基于克隆选择原理的免疫算法提取特征样本用于故障模式识别。在Tennessee Eastman过程上的仿真结果说明,非线性子空间方法能够比线性子空间方法更有效的识别故障模式。

To identify fault root cause in multivariate statistical process monitoring, nonlinear principal component subspace method was proposed to recognize fault pattern. Kernel principal component analysis was performed on different fault pattern datasets so that the nonlinear principal component subspace was available to describe data variance. The subspace classifier was constructed to identify fault pattern. In order to reduce the computation complexity, feature samples based nonlinear principal component subspace method was studied. Immune algorithm based on clonal selection principle was applied to compute feature samples, which were used for fault pattern recognition. The simulation results on Tennessee Eastman process show that nonlinear subspace method can identify fault pattern more effectively than linear subspace method.