Low-cost fabrication and comparative evaluation of machine learning algorithms for flexible PDMS-based hexagonal patch antenna.

Abstract

This article focuses on the design and development of a flexible hexagonal microstrip patch antenna fabricated on a Polydimethylsiloxane (PDMS) substrate for potential use in 5G and wearable applications. The antenna geometry was selected to operate in a higher-order resonant mode to enhance performance under mechanical bending. To accelerate the design process and determine the most effective model for predicting optimal geometrical parameters that yield improved impedance matching at the target frequency, four supervised machine learning algorithms including Random Forest, XGBoost, CatBoost and LightGBM were evaluated and compared. These models were trained using a dataset generated from full-wave electromagnetic (EM) simulations, and the Random Forest model exhibited the best predictive accuracy, with an R² value of 0.99. Additionally, a low-cost and scalable fabrication approach was demonstrated to realize conductive traces on PDMS, and a functional prototype was manufactured and tested. The prototype demonstrated consistent performance, with measured gain of 3.2 dB. An electrical equivalent lumped-element circuit model was also formulated to analytically validate the antenna's EM simulation results. This integrated approach of data-driven optimization, low-cost fabrication, and circuit-level validation contributes to the practical realization of flexible RF antennas.