基于KPI子空间的工业过程自适应故障检测

Adaptive Fault Detection in Industrial Process Based on KPI Subspace

针对具有时变特性的工业过程发生的故障是否影响关键性能指标的问题,提出一种基于关键性能指标(Key Performance Indicators,KPI)子空间的自适应故障检测方法(AFD-KPIS)。将正常过程数据空间分解为主元和残差空间。根据过程变量和关键性能指标(KPI)的相关性将主元空间分为KPI相关子空间和KPI无关子空间,并将残差空间Q统计量并入KPI无关子空间,在KPI两个子空间分别构建T2和k近邻故障检测统计量,在线检测判断为正常的数据通过时间窗更新到建模数据中进行用于模型的更新。通过数值案例和TE过程的仿真验证所提方法的有效性。

An adaptive fault detection method(AFD-KPIS)based on key performance indicators(KPI)subspace is proposed to solve the problem of whether the faults occurring in time-varying industrial processes affect the key performance indicators.The normal process data space is decomposed into a component and a residual space.According to the correlation between process variables and key performance indicators(KPI),the principal component space is divided into KPI-correlated subspace and KPI-uncorrelated subspace,and the residual space Q statistics are incorporated into the KPI-uncorrelated subspace,and T2 and K-nearest neighbor fault detection statistics are constructed in the two KPI subspaces,respectively.The data judged as normal by online detection is updated to the modeling data through the time window for updating the model.The effectiveness of the proposed method is verified by numerical examples and TE process simulation.