An Effective Online Failure Prediction in DC-to-DC Converter Using XGBoost Algorithm and LabVIEW

In this paper, an online failure detection method for DC-to-DC converter using a machine learning algorithm is presented. The key feature of the proposed method is to detect multiple catastrophic failures by observing the abnormal conditions in the circuit parameters during its operation. The independent features are obtained online employing a data acquisition instrument using the LabVIEW tool, and a tree-based ensemble learning algorithm is used for failure prediction and classification. The conventional method of using a single prediction model often mislead to false prediction and belated failure detection. To enhance the effectiveness of the failure prediction algorithm, a group of different data-driven ensemble models is trained, and using Receiver Operating Characteristic (ROC) curve, the best prediction model is identified. Extreme gradient boost-based ensemble algorithm achieves the highest accuracy rate of 98.9% in multiclass classification among the ensemble models.