Deep neural network surrogates for next-generation twin-engine STOL airlifter design
Gabriele Morra, Vincenzo Cusati, Fabrizio Nicolosi, Giovanni Cerino · Aerospace Science and Technology · 2026
The operational demands of Agile Combat Employment and fleet recapitalization are driving interest in next-generation tactical transports. While twin-engine solutions promise reduced operational and maintenance costs compared to four-engine baselines, they present significant design challenges for achieving robust Short Take-Off and Landing ( stol ) performance from austere, semi-prepared runways. This work addresses the inherent antagonism between maximizing payload and minimizing critical field length through a certification-constrained multi-objective optimization framework. A high-fidelity dataset of approximately 16,500 converged designs was generated using the Pacelab APD & SysArc conceptual design suite to train a Deep Neural Network surrogate model with high validation fidelity against physics-based predictions. The surrogate was integrated with an NSGA-II genetic algorithm to efficiently map the multi-objective Pareto front, bounded by key constraints including One Engine Inoperative climb gradient, minimum control speed margin, static stability margin, and maximum takeoff weight. The resulting Pareto front reveals direct trade-offs between runway performance and payload capacity for twin-engine, T-tail C-130J-class configurations. This optimization workflow, combining certification constraints with surrogate modeling, accelerates conceptual design of stol transports, reduces early design risks, and extends to adjacent mission profiles and design space explorations.