Detection of Failure Precursors in Multilayer Ceramic Capacitors Based on Symbolic Time Series Analysis

Abstract

In electronic systems, capacitors are often used to perform such functions as noise reduction, signal filtering, and DC blocking. Multilayer ceramic capacitors (MLCCs) that consist of multiple ceramic layers and electrodes have relatively higher capacitance in comparison to regular capacitors. While increased need for high capacitance requires a thinner thickness of layers and electrodes in an MLCC, their reliability becomes a concern. This paper presents an approach to detect failure precursors of MLCCs using Symbolic Time Series Analysis (STSA). Capacitance and dissipation factor readings from 10 MLCCs were obtained in-situ under temperature-humidity-bias conditions. We converted each reading into a finite number of symbols and calculated transition matrices from every symbol transition of an MLCC. In this way, a training dataset was constructed from all of the symbol transitions of the normal MLCCs. The difference between the training dataset and the transition matrices was characterized by Euclidian norm, and later utilized as an anomaly index. The anomaly indices from the failed MLCCs provided failure precursors earlier than the actual times to failure, while those from the normal MLCCs did not. Thus, detection of failure precursors can enable users to diagnose a system whose performance variable shows gradual changes over time.