A method for fault detection on synchronous generators using modified principal component analysis

An online fault detection method for synchronous generators is presented. This method provides efficient and reliable fault detection using Principal Component Analysis (PCA). The traditional PCA technique is modified to enable multivariate modelling using the machine's phase voltages, excitation current and shaft voltage. Hotelling's statistic and the model residuals are used to detect incipient faults on the machine and avoid false alarms. The presented method is tested and validated using an experimental system.