Abstract:

Process monitoring and fault diagnosis is an important problem in chemical processes.Aiming at the real chemical process with its complicated mechanism, nonlinear characteristics and the numerous predictor variables within serious multicollinearity, a novel fault diagnosis model was constructed by combination of isometric feature mapping (ISOMAP) with linear discriminant analysis (LDA).The resulting discriminate model based on this data-driven approach ISOMAP-LDA was divided into two steps. In the first step that performed the ISOMAP manifold learning algorithm was in the original high space for nonlinear dimensionality reduction and feature extraction, and in the second step that built the LDA discriminate model was built in making use of the extracted feature variables.Finally, the proposed ISOMAP-LDA approach was applied to the multiple faults diagnosis of Tennessee Eastman chemical process.The results showed that this method had great power in nonlinear dimensionality reduction and strong generalization ability.At the same time, the ISOMAP-LDA fault diagnosis model was more concise and could be used to observe the structure of the set of samples in the embedding space.

Key words:

流形学习, 集成等距特征映射, 线性判别分析, 故障诊断, Tennessee Eastman过程

摘要：

针对化工连续生产过程的时序性及非线性等特征，提出一种新的基于数据驱动的化工过程故障诊断方法：ISOMAP-LDA。首先实行流形学习算法ISOMAP，在保持量测数据几何结构特性下完成非线性降维，然后基于提取的嵌入变量张成的低维空间，选用线性判别分析（LDA）构造故障模式类的判别函数，负责各采样个体故障类型的判定。将该方法用于仿真化工Tennessee Eastman 过程的故障诊断，结果表明，ISOMAP-LDA方法不仅拥有较高的故障诊断能力，而且取得采样在低维空间的可视化表示。

关键词:

流形学习, 集成等距特征映射, 线性判别分析, 故障诊断, Tennessee Eastman过程