A Predictive Maintenance Approach for Power Converters Using Artificial Neural Networks

In this paper, a predictive maintenance approach using Artificial Neural Networks (ANN) to analyze the ripple waveforms is proposed to estimate the degradation status of electrolytic capacitors in power converters. An off-the-shelf full-brick DC-DC power converter was used to build a 380 V to 24 V power conversion system for investigation. Powered by advanced Digital Signal Processor (DSP), a universal ripple sensing module was developed to acquire and process the ripple waveforms of voltage and current on the electrolytic capacitors from the primary and secondary sides of the power conversion system. Accelerated aging test of the electrolytic capacitors was performed and ripple waveforms of the power conversion system with the electrolytic capacitors at different degradation status were obtained by the ripple sensing module. An ANN model was established by analyzing the evolution of ripple waveforms with the aging time. To verify the capability and accuracy of the ANN based predictive maintenance approach, independent aging tests were performed in parallel and the obtained ripple waveforms were used to determine the degradation status of electrolytic capacitors. The results show that predicted aging time agree well with experimental data.