Webinar: Application of Unsupervised Machine Learning Techniques in Prognostics of Power Electronics
Tuesday, July 10, 2018
This short-course-style talk will focus on the application of unsupervised machine learning techniques in the data-driven prognostics of power electronic devices. Anomaly/fault detection is an integral part of prognostics and health management. In many engineering problems that are not well understood, it is often difficult and costly to achieve fully-labeled training datasets. This makes unsupervised/semi-supervised machine learning a preferable choice under such circumstances. Unsupervised machine learning techniques commonly used in anomaly detection, such as principal component analysis (PCA) and k-means clustering will be briefly reviewed and discussed. Implementation of these unsupervised machine learning techniques, in combination with statistical assessment of extracted principal components on time-series of recorded degradation data of IGBT modules and GaN HEMTs will be discussed, to determine the probability of anomaly of test data. The implementation of particle filter techniques coupled with anomaly detection techniques described above, will be presented.
Yizhou Lu is currently working towards a Ph.D. degree in the Department of Mechanical Engineering under the direction of Prof. Aris Christou. He received the Bachelor of Science degree in mechanical engineering from Shanghai Jiao Tong University (SJTU), Shanghai, China, in 2013.