改进PCA及其在过程监测与故障诊断中的应用

IMPROVED PCA WITH APPLICATION TO PROCESS MONITORING AND FAULT DIAGNOSIS

提出一种改进的主元分析 (PCA)方法 ,采用主元相关变量残差 (PVR)统计量代替通常的平方预测误差Q统计量 ,用于工业过程的监测与故障诊断。改进PCA避免了Q统计量的保守性 ,能够提供更详细的过程变化信息 ,从而有效识别正常工况改变与过程故障引起的T2 图变化。通过对双效蒸发过程的仿真监测 ,与普通PCA方法进行了比较 。

Principal component analysis (PCA) is an effective method to extract relationships between variables and thus has been widely applied to multivariate statistical process monitoring and fault diagnosis. An improved PCA is presented, which uses principal-component-related variable residual (PVR) statistic to replace Q-statistic in the conventional PCA. The improved PCA can avoid the conservation of Q statistical test and provide more explicit information about the process conditions. As a result, the root cause that violates the Hotelling T2 test but still satisfies Q test can be unambiguously identified, which is impossible in the conventional PCA. Then a simulated double-effect evaporator is monitored and diagnosed by using this proposed method and comparisons with the conventional PCA are made. The simulation result shows that the improved PCA is more sensitive to weak process changes and has an enhanced fault diagnosing performance.