主元空间中的故障重构方法研究

主元分析 (PCA)作为一种数据驱动的统计建模方法，在化工产品质量控制与故障诊断方面获得了广泛研究和应用.利用故障子空间的概念，研究了基于T²统计量的故障重构问题，获得了主元空间中的完全重构、部分重构，以及可重构性的条件.为进一步在主元空间中进行故障分离和识别提供了可能.通过对双效蒸发过程的仿真监测，对不同传感器的故障类型、幅值等重要信息进行重构和波形估计，证实了所获结果的有效性.

FAULT RECONSTRUCTION APPROACH IN PRINCIPAL COMPONENT SUBSPACE

Principal component analysis (PCA) finds wide application in chemical process monitoring and product quality control as a data-driven modeling method.Based on the concept on fault subspace, the fault reconstruction issue was explored by using T~2 index, while the geometric method recently developed by Dunia et al focuses on the SPE index.However, some faults involving process fault and sensor fault that do not violate the PCA statistical model can only be detected by T~2 index.Thus the proposed reconstruction approach has superior performance in the general sense.The acquired results were then illustrated and verified by monitoring a simulated double-effect evaporator (DEE) process, where different sensor faults were reconstructed and fault wave/magnitude was estimated to judge the sensor fault type.