Quadcopter Mission Energy Prediction Using Machine Learning

Abstract:
Battery energy awareness is an important aspect of tasking Unmanned Aerial Systems (UAS) safely and efficiently. By considering energy expenditure during mission planning, flight plans are assigned to the UAS only if there is sufficient energy onboard to complete the mission. In this work, several methods are developed for predicting the energy consumed during a flight, and their accuracy is assessed. Three simulation-based models derived from momentum-theory, blade-element theory, and computational fluid dynamics (CFD) are considered in addition to two data-driven models derived from flight test data (linear regression and Kriging), and four multi-fidelity models (Optimized Kirchstein, Hover-corrected, Additive Bridge, and Predictor-Corrector). Each model is used to predict the energy consumption of a representative mission and their predictions are compared to the measured energy consumption. From this analysis, it is found that a linear regression model trained on flight test data is able to deliver predictions within 3% of the measured value and outperforms other simulation-based and multi-fidelity models despite a simple architecture. The high level of accuracy and low computational requirements make this linear regression model desirable for energy-aware mission planning and energy-leash computations.

Did you attend Forum 82? Click the preview below to access the full paper.