Machine Learning for EMC/SI/PI - Blackbox, Physics Recovery, and Decision Making

• Published in: IEEE electromagnetic compatibility magazine Vol. 12; no. 4; pp. 65 - 75
• Main Author: Jiang, Lijun
• Format: Journal Article Magazine Article
• Language: English
• Published: New York IEEE 2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Black boxes; Closed box; Data Driven; Data models; Decision making; Dimension Reduction; Dimensionality reduction; EMC; Machine learning; Machine learning algorithms; Mathematical analysis; Neural Network; Neural networks; Optimization; Recovery; Reinforcement learning; Reviews; Supervised learning; Task analysis; Unsupervised Learning
• ISSN: 2162-2264; 2162-2272
• DOI: 10.1109/MEMC.2023.10466473

Machine learning (ML) is one of today's most studied subjects in almost every research area. It provides interesting mathematical tools that could inspire us to rethink about the EMC/SI/PI engineering. This paper gives a preliminary review on machine learning methods for EMC/SI/PI technology developments. Sample examples from publications on EMC/SI/PI methodologies powered by machine learning methods are discussed. There are three major types of machine learning methods. From an EMC/SI/PI engineering point of view, supervised learning provides heterogeneous high dimensional surrogate blackbox model, unsupervised learning enables dimension reduction for physics recovery, and reinforcement learning uses rule-based decision making for optimizations. It is important to select proper machine learning tools and algorithms for various EMC/SI/PI tasks.