Thermal-Fluid Modeling and Kriging Surrogate-Assisted Lightweight Simulation of a Piston Engine for Large Unmanned Aerial Vehicles

Piston engines are the core power units of large unmanned aerial vehicles (UAVs), and the characteristics of their thermal-fluid working processes directly affect the power performance and operational reliability of UAVs. To address the difficulty of simultaneously ensuring simulation accuracy and computational efficiency in conventional piston engine models, a lightweight simulation method for the thermal-fluid modeling of piston engines is proposed. A Rotax 915 piston engine for large UAVs was selected as the target engine. A one-dimensional thermal-fluid working process model was first established in GT-Power, and a Kriging surrogate model was then constructed to develop a lightweight simulation framework, enabling the fast and accurate prediction of key engine response parameters. The proposed framework enables the rapid prediction of engine performance, propeller load, and lubrication responses under UAV mission-related operating conditions. The surrogate model delivered favorable prediction accuracy against the GT-Power model (R2 > 0.95, MAPE < 5%); flight data verification (excluding cold start) presented a MAPE of 11% and 27% for exhaust gas temperature and lubricating oil pressure, respectively. The proposed lightweight model was validated against actual UAV flight data, and the results show that it can effectively capture the variation trends of exhaust gas temperature and lubricating oil pressure. Further simulation results indicate that with increasing altitude, engine power decreases whereas brake specific fuel consumption (BSFC) increases. With increasing engine speed, engine power increases, BSFC decreases, and lubricating oil pressure declines. Under different flight phases, the takeoff phase shows the highest engine power and BSFC, the cruise phase corresponds to the maximum propeller torque and thrust, and the descent phase exhibits the lowest overall load level. The results provide a useful basis for performance evaluation, condition monitoring, and digital-twin-oriented modeling of piston engines for large UAVs.