Machine Learning-Assisted Tolerance Analysis and Its Application to Antennas

An efficient machine learning-assisted tolerance analysis (MLATA) method is proposed by applying machine learning (ML) methods into multiple layers of the antenna tolerance analysis. The computational time for operations including worst case performance searching, maximum input tolerance hypervolume searching and robust optimization is greatly reduced while maintaining high reliability due to the introduction of the ML methods. The surrogate models which are built using ML methods have been introduced to predict both antenna performance and tolerance of parameters at given design points. The Pareto front combining antenna performance, robustness and size has been obtained to guide trade-offs for antenna robust design. A planar inverted-L antenna for mobile terminals is simulated to validate the proposed MLATA method.