Machine Learning-Based Performance Prediction and Analysis of UAV Propellers under Step-Throttle Conditions

Abstract

Accurate performance prediction of unmanned aerial vehicle (UAV) propellers under dynamic operating conditions is crucial for propulsion system matching and flight efficiency optimization. Traditional static models often struggle to capture complex nonlinear aerodynamic behaviors. This study aims to significantly improve the prediction accuracy of propeller performance under various working conditions using a data-driven approach. To this end, a physical experimental test bench was established to conduct comprehensive step-throttle tests on two different models of propellers, systematically collecting real-world dynamic response data. Based on the experimental dataset, a machine learning-based prediction model was constructed. The model directly utilizes battery voltage and throttle command as input features to simultaneously and accurately predict key output metrics, including mechanical power, rotational speed, and overall efficiency. Experimental validation demonstrates that the proposed machine learning model exhibits excellent prediction accuracy and highly robust curve-fitting capabilities across different throttle steps. The results indicate that this data-driven framework effectively overcomes the limitations of conventional analytical methods, providing a highly reliable and efficient tool for evaluating and optimizing UAV propulsion systems.