Abstract:

Accurate numerical modeling of nonlinear flight dynamics is a fundamental requirement for the analysis, simulation, and control of unmanned aerial vehicle (UAV) systems. Classical numerical solvers such as Runge - Kutta methods provide reliable solutions but often suffer from high computational cost when applied to strongly nonlinear operating regimes. Recent advances in artificial intelligence have enabled learning-based numerical approaches that enhance efficiency while maintaining acceptable accuracy. In previous work, a machine learning–driven numerical framework was designed to improve nonlinear dynamic analysis in aerospace applications. This paper presents a comprehensive numerical validation of that designed framework when applied to nonlinear UAV flight dynamics. The proposed approach integrates learning-based correction mechanisms with physically grounded numerical structures, enabling improved convergence, reduced numerical error, and enhanced stability. Numerical validation is conducted using representative flight datasets obtained from industrial aviation operations. Extensive simulations benchmarked against a fourth-order Runge–Kutta (RK4) solver demonstrate that the proposed framework achieves comparable accuracy with significantly lower computational cost. The results establish the framework as a reliable and scalable numerical solution for advanced UAV modeling and future autonomous flight applications.