Anomaly Monitoring Methodology for DC/DC Converters Using Gaussian Process Regression and Gaussian Mixture Model

Healthy dc/dc converters in undersea observation networks are essential for ensuring that these systems operate reliably over decades. Any performance degradation or failures in the dc/dc converters can result in significant, potentially incalculable damage. To prevent such issues, continuous anomaly monitoring during operation is critical for synchronized and successful mission completion. This article introduces a method for dc/dc converter anomaly monitoring based on the complete output signal distribution. A specialized signal collection device is integrated into the monitoring system to capture and preprocess output signals, enabling reliable data acquisition in challenging environments like undersea operations. The method leverages Gaussian process regression (GPR) to determine monitoring thresholds and Gaussian mixture model (GMM) to detect anomalies. The robustness of the method is validated using high-order Superbuck converters, with comparative analysis against other monitoring methods, demonstrating its superior performance. Experimental results confirm that this method effectively identifies anomalous conditions across various types and orders of dc/dc converters, regardless of signal type, ensuring versatility and reliability in critical applications.