Off-Design Mission Performance Prediction for Unmanned Aerial Vehicles Based on Machine Learning

This paper presents an approach to estimate the off-design mission performance of unmanned aerial vehicles (UAVs) using supervised machine learning. The basis of this work is a procedure for the mission-based design of civil UAVs. The procedure optimally tailors UAVs to exemplary design missions by evaluating their mission performance (sensor data quality, fuel demand, and detection probability of targets to be searched) inside a conceptual design optimization loop. This is achieved by carrying out mission simulations using high-resolution terrain data for sensor data evaluation. However, in order to weigh possible design alternatives against each other, the off-design mission performance on terrains other than the one used for the design is of great interest. This study investigates how the mission performance of a UAV changes when operating the UAV on terrains which are different from the terrain it was designed for. For this purpose, the design optimization procedure is used to generate various different designs, each of them tailored to an individual design mission with an individual terrain. Scalar terrain attributes such as hilliness, size of the search area, and important elevation characteristics are captured for all terrains and stored in a terrain database. Following completion, the off-design mission performance of each design is evaluated on each terrain of the previously compiled terrain database. The obtained data is used to train supervised machine learning models in order to predict the off-design mission performance based on the associated changes of the terrain characteristics. All developed machine learning models are tested on a previously unseen data set. Very good performance is observed for all models. A use case is provided to demonstrate the application of the developed models.     «

This paper presents an approach to estimate the off-design mission performance of unmanned aerial vehicles (UAVs) using supervised machine learning. The basis of this work is a procedure for the mission-based design of civil UAVs. The procedure optimally tailors UAVs to exemplary design missions by evaluating their mission performance (sensor data quality, fuel demand, and detection probability of targets to be searched) inside a conceptual design optimization loop. This is achieved by carrying...     »