Online condition monitoring for DC-link capacitors of three-level NPC converters using noninvasive signal injection

This paper introduces an online method for monitoring the health condition of the DC-link in three-level NPC converters. This method does not require hardware changes or modifications to the control algorithm. It involves injecting a zero-sequence signal, added to the system reference voltages to generate a neutral-point current injection. A square waveform with an interharmonic cycle frequency is used for the signal injection to minimize estimation errors caused by existing signals. This combined signal allows for the simultaneous estimation of the capacitor equivalent series resistance (ESR) and capacitance (C) at different frequency components using wavelet decomposition. Additionally, a balance between the magnitudes of the system reference voltage and the signal injection is established to prevent power quality issues. Besides, the capacitor parameters are estimated in the time-frequency domain, enabling tracking of capacitor degradation and wear-out failure. A comparison with a Fourier-based approach shows that this method provides highly accurate capacitor parameter estimation with less than 2% errors, instilling confidence in its results. Experimental results from a three-level NPC converter's laboratory setup demonstrate the proposed method's effectiveness.