Adaptive deviation surrogate modelling for mesh antenna surface accuracy prediction

Mesh antennas have become the preferred solution for large-scale deployable antenna due to their lightweight, large aperture, and high folding ratio. The cable network is a critical component, with its surface accuracy directly determines the wavelength and frequency of reflected electromagnetic waves. Currently, model-based surface accuracy monitoring and adjustment faces challenges in engineering consistency. To address this, we develop a hybrid mechanism-based and data-driven model that combines the high accuracy characteristic of data-driven approaches with the strong interpretability inherent in mechanism-based models by integrating deviation surrogate models with mechanism models. This integration ensures high consistency with physical prototypes while enabling stable and precise evaluation of cable network surface accuracy. We further propose an adaptive sampling deviation surrogate model that efficiently captures deviations between measured data and mechanism model results with minimal data input. The proposed adaptive sampling strategy, HDAS, explicitly accounts for sample space uniformity, local response complexity near sampling points, and high-dimensional mapping challenges, thereby achieving superior surrogate model accuracy with fewer samples. An engineering-oriented metric, DRMS_LOOCV, is proposed to quantitatively assess prediction accuracy and convergence behavior during adaptive sampling and serves as a termination criterion for training. A physical prototype was fabricated for data collection and model training, followed by surface accuracy evaluations under multiple loading conditions. Model errors were 0.45%, 1.38%, and 0.22% across three cases, demonstrating the precision and effectiveness of the proposed method.