神经网络主元分析的传感器故障诊断方法

Sensor fault diagnosis method based on neural network principal component analysis

针对多传感器故障诊断问题,将神经网络引入主元分析(principal component analysis,PCA)模型之中,提出一基于主元分析的多传感器故障诊断模型。首先,应用传感器正常工作时测量的历史数据,由PCA模型得到所有传感器的预测值。其次,计算传感器系统的平方预期误差值(squared prediction error,SPE),根据系统的SPE值是否跳变,判定有无故障发生。通过分别重构单个传感器信号的SPE值来确定发生故障的传感器。最后,应用一个多传感器故障诊断仿真实例证明了该方案的可行性。

For the problem of sensor fault diagnosis,a sensor fault diagnosis model based on principal component analysis（PCA） and artificial neural network is proposed.Firstly,the forecasting values of sensors are available from historical data measured from sensors in fault-free condition based on PCA model.Secondly,the squared prediction error of the system is calculated,the fault occurred when the squared prediction error（SPE） is suddenly increased.Sensor values are reconstructed respectively to newly calculate the SPE to locate the faulty sensor.Finally,the method proposed is proved feasible and effective by a simulation of multiple sensor fault diagnosis.