부산대학교 도서관

Pump failures are severe accidents for vanadium redox flow batteries (VRFBs) since they will lead to permanent stack damage. Fault detection of VRFBs can help to detect faults immediately and minimize damage. This study reports a pump fault detection method without using flow rate sensors. A novel method based on the support vector machine (SVM) is proposed. First, the characteristic parameter is extracted from the voltage curve. Second, the magnitude of this characteristic parameter is affected by the state of charge (SOC) of the battery, so SOC is also selected as one of the fault detection variables. Finally, the parameters of the SVM are optimized, and the fault prediction results are obtained by SVM training. The obtained results show that this method has high accuracy in detecting the pump fault of the battery, and the classification accuracies were 100%, 99.1935%, and 98.3871% in the case of bilateral pump failure, positive pump failure, and negative pump failure, respectively.