Data-driven thrust prediction for UAV powertrains using artificial neural networks

This study develops an Artificial Neural Network (ANN)-based model to predict thrust in electric powertrain systems of Unmanned Aerial Vehicles (UAVs), aiming to reduce dependence on extensive experimental testing. Traditional thrust estimation approaches often require resource-intensive setups or simplified theoretical models with limited accuracy. To address this, an ANN model was trained on a comprehensive dataset covering 32 propeller configurations with varying diameters, pitches, and materials. The model, optimized using the Scaled Conjugate Gradient (SCG) algorithm, achieved high predictive accuracy with an R² value of 0.998 and low errors across key metrics. Additionally, new propeller configurations were generated through data interpolation, enabling thrust prediction without additional physical tests. The results demonstrate that ANN-based modeling provides a reliable, cost-effective, and scalable alternative to conventional methods, supporting faster evaluation and design of UAV powertrain systems.