Machine Learning-Based Condition Monitoring for DC-Link Capacitors in AC/DC/AC Converters

Power electronic converters play a pivotal role in diverse industries. Yet, reliability of these converters has been a critical concern since its early applications. Thus, this article proposes a machine learning (ML)-based condition monitoring (CM) system for dc-link capacitors (DLCs) in ac/dc/ac converters. The capacitance of DLC is chosen as the health indicator, and the power spectral density (PSD) of dc-link voltage ripple is proposed as a diagnostic tool. This technique offers resilience against measurement noise and eliminates the necessity for additional filtering techniques. Moreover, the proposed method does not require a high sampling frequency unlike the similar studies. Another notable advantage of this approach is that no additional circuitry is required, as it solely relies on the available signals. Moreover, the effectiveness of the ML-based CM system is evaluated through simulation results and an experimental validation on a laboratory prototype. The experimental results pointed out the high accuracy of the proposed CM system.