Deep neural network-based lifetime diagnosis algorithm with electrical capacitor accelerated life test

Abstract

Power storage and conversion technologies are increasingly in demand for their energy efficiency and eco-friendliness, with capacitors being key in stabilizing and filtering voltage in these devices. However, during this process, thermal degradation phenomena often occur due to over-voltage and over-current. This degradation can lead to decreased performance, capacitor failure, and in severe cases, explosions. Thus, the precise monitoring and prediction of capacitor lifetime is paramount. In this study, we use accelerated life test data to create images using reference plots and compare the accuracy of deep neural network training through image fusion. This introduces a new methodology for monitoring the lifetime of capacitors. This approach involves collecting aging data through accelerated life tests and then generating images from time-series data composed of capacitor voltage, current, and resistance. These images are used to train the deep learning algorithm, extracting relevant features and predicting the remaining life of the capacitors. Our method demonstrates remarkable effectiveness, showing an impressive accuracy rate of 80% in the real-time monitoring of capacitors under various operating conditions. Ultimately, this deep neural network-based lifetime monitoring algorithm holds potential to be scaled and applied to diverse electronic systems, enhancing their reliability and safety.