Prediction of Capacitor’s Accelerated Aging Based on Advanced Measurements and Deep Neural Network Techniques

摘要

Capacitors are widely used in electronic systems and have a key function in electromagnetic compatibility (EMC) compliance. However, the aging of capacitors results in an alteration of their parameters, which could pose a threat on the normal operation of systems as well as their EMC compliance. Normally, accelerated aging is employed to shorten the experiment time. After the aging, the capacitance and equivalent series resistance (ESR) are measured to evaluate the aging process. In this article, a new continuous characterization measurement setup is implemented in which the accelerated aging of the capacitors under test (CUTs) is continuously monitored during the overall accelerated aging process. It significantly improves the continuity of the measurement and eliminates the errors attributed to the interrupting of the aging process. This method is validated by comparing measurement results from the new measurement method with the results of the conventional method. This was done by subjecting two types of film capacitors to thermal and electrical stress in order to evaluate the accelerated aging effects. Furthermore, a conditional deep neural network with a dropout technique is proposed to predict the accelerated aging conditions of the capacitors. Instead of only forecasting the failure threshold, the proposed network is able to dynamically predict the accelerated aging conditions at different elevated temperatures and voltages. This leads to a serious reduction in the total measurement time from 1000 to 200 h.