Identification of the Early Degradation of Bond Wire via Multidimensional Data Fusion Based on RMT

Insulated-gate bipolar transistor (IGBT) modules are the core components of modular multilevel converters (MMCs), and their health state affects the reliability of the whole system. Bond wire degradation is one of the main failure modes of IGBT modules, and initial degradation monitoring and identification are highly important. Traditional detection methods are based on a single electrical parameter, which cannot be recognized when the monitored quantity is disturbed. In this article, we propose a feature identification method for identifying bond wire degradation using multidimensional data fusion, which does not need to analyze the complex physical coupling relationships inside the IGBT but only mines potential information from the electrical quantity data of the external port. A high-dimensional random matrix is constructed on the basis of random matrix theory (RMT) and the sliding window method, and the distribution law of the empirical spectrum is used to extract the linear eigenvalue statistics. Principal component analysis (PCA) is used to construct the fusion feature quantity to monitor the state of the bond wire. The results show that the sensitivity of the proposed method in recognizing initial degradation is greater than that of other common monitoring methods. Moreover, it can still recognize accurately when a single electrical quantity is disturbed. Compared with traditional methods, significantly lower sampling frequencies are needed.