Application of artificial neural networks in estimating the thrust generated by a drone propulsion system

Abstract

Artificial intelligence and machine learning have increasingly expanded into numerous fields, particularly for estimating reliable physical parameters. Dynamic systems modeling is a fundamental component of control design, as controller performance is directly linked to model fidelity. Although mathematical tools are crucial, the inherent complexity of non-linear systems presents significant challenges to traditional modeling. Consequently, models based on Artificial Neural Networks (ANNs) and deep learning algorithms have emerged as effective solutions to improve this design stage. Unmanned Aerial Vehicles (UAVs) have become indispensable in agriculture. however, their growing adoption, especially in logistics and cargo transport, introduces the challenge of relating propulsion systems of brushless motors and propeller propulsion systems to the vehicle's payload requirements. This relationship directly influences the technological evolution of these platforms. This relationship directly affects the technological evolution of these platforms. In this context, estimating the thrust generated by propulsion systems is a focus of study for the evolution of drones use in applications that require a higher payload. Therefore, paper proposes a feed-forward backpropagation ANN model to estimate the thrust of a motor-propeller assembly, utilizing experimental data to evaluate model performance and accuracy. After training, the created network achieved a Mean Absolute Percentage Error (MAPE) of 3.94% and a Mean Absolute Error (MAE) of 2.18 grams. These results demonstrate the potential of this approach for modeling complex, non-linear systems, offering a robust alternative to classical analytical modeling, which is often mathematically complex and constrained.