Machine Learning Assisted Optimization Methods for Automated Antenna Design

This paper presents the basis of defining data-driven, machine learning assisted optimization models for automating the adjustment of physical parameters in EM software simulations controlled by AntennaCAT. Data from over 60,000 simulations have been collected for a rectangular patch antenna case study and used to design boundaries for rule-based models, and for increasing the reliability of machine learning based optimization models fed with unbalanced classes, or with sparse data. The data collection methodology, some early empirical results, and implications for automated tuning and design are discussed.