Machine learning approach for predicting key design parameters in UAV conceptual design

The initial concept of an Unmanned Aerial Vehicle (UAV) design is complicated and unique due to performance parameters like payload capacity, engine power, endurance, service altitude, etc. required to perform a wide range of missions. Empirical correlations between key design parameters can approximate initial characteristics but to explore the entire design space while considering sensitivities of interacting parameters, comprehensive, time consuming and computationally expensive trade-off studies are required to converge the early concept appraisal. The current paper explores the potential of Machine Learning (ML) techniques for rapid and accurate estimation of UAV design parameters in the conceptual phase by extracting knowledge from UAVs already in service. An ML framework based on five different regression models is formulated to estimate the parameters significant to mission profile using database of fixed-wing UAVs key design attributes. The predictive performance of the presented ML approach shows excellent agreement with the actual values during validation and comparatively, turns out to be more accurate than the existing methodology based on empirical correlations. Overall, ML techniques have a great potential for being applied as a surrogate model for evaluating novel UAV design concepts using less computational time and resources.