基于主元分析的多传感器故障检测

Fault Detection for Multi-Sensor Based on PCA

针对动态系统的压力、温度、流量等传感器数据,给出了一种基于主元分析法的传感器故障检测与诊断方法。该方法能够在对测量参数相关性分析的基础上,将传感器测量值所组成的测量空间分解为主元和残差两个子空间,通过传感器实际测量数据与正常数据矩阵在残差子空间投影的比较,对传感器的故障进行检测与诊断。通过双容水箱被控系统的传感器进行检测,结果表明主元分析法对传感器具有很好的故障检测和故障诊断能力。

A sensor fault detection and diagnosis method is presented using principal component analysis （PCA） for pressure, temperature and flux data of dynamic system. The method divides the measure space into two subspaces： principal component subspace （PCS） and residual subspace （RS） based on the correlation analysis of measure data. It can detect and diagnose sensor faults by comparing the projections in RS of actual sensors measurement and normal data matrix. Examples of using PCA for fault detection and diagnosis of the four types of sensor faults in double water tanks are given, which are based on fault simulation. The results show that the PCA approach has good performance in the fault detection and diagnosis of sensors.