<![CDATA[Training gliders are the entry platform of pilot education, yet domestic design capability for this class in Indonesia remains limited. This study develops the preliminary design of a single-seat glider for the early solo phase of training and a machine-learning (ML) surrogate of its steady glide performance. The design follows Raymer's method with a NACA 6409 section analysed with NeuralFoil and a take-off mass of 420.3 kg based on the CS-22 occupant allowance. The maximum aerodynamic efficiency is 21.67 and the stall speed 20.56 m/s. From a 600 m release in still air, the range and endurance are 13.0 km and 10.4 min at the theoretical best-glide speed, or 12.8 km and 9.5 min with a 1.1 stall-speed margin; the 20 min requirement is not met and needs a release height of about 1268 m or thermal lift. Five regressors were trained on 2000 Latin Hypercube samples generated by deterministic analytical equations, so their accuracy measures approximation fidelity rather than physical validation. Gaussian Process Regression was the most accurate (10-fold cross-validated MAPE 0.005%, 95th-percentile error below 0.08%) and its prediction intervals were conservative. Sobol' indices show that parasite drag and aspect ratio govern aerodynamic efficiency, while wing loading and maximum lift coefficient govern stall speed.Keywords—glider, preliminary design, flight performance, surrogate modelling, sensitivity analysis]]>
Copyrights © 2026
