Rotorcraft low-noise trajectories design: black-box optimization using surrogates

Abstract

This paper addresses the noise-minimal trajectory optimization problem for a specific type of aircraft: rotorcraft. It relies on a realistic noise footprint computation software provided by industry that is black-box. Locally optimal trajectories are computed through a tailored solution approach based on the Mesh-Adaptive Direct Search algorithm. We propose multiple surrogates defined according to our knowledge of the problem, including a surrogate relying on the physics of the problem (approximating the rotorcraft noise model), and another based on a machine learning (neural network) method. The proposed solution approach is further enhanced by the computation of an appropriate starting guess through a path planning algorithm tailored to the problem, and by the reduction of the variable space domain. The performance of the proposed methodology both in terms of quality of the solutions (trajectories exhibiting significant noise reduction compared to those currently flown in practice) and computing time is illustrated through numerical experiments on realworld case studies.

Trajectory Optimization of Unmanned Aerial Vehicles in theElectromagnetic Environment

“…In this section we consider four sensors at positions (-4,10), (2,18), (6,8), and (1, 5) and investigate the effect of intermittent communications on the optimal paths, i. e. where the UAV reduces their emission by a certain amount for a certain amount of time; see Section 3.2 for details.…”

“…One key insight for modeling various objectives F is the fact that the act of trying to locate an agent or just determining the existence of an agent through receiving a signal is a process with uncertainty, as any signal is affected by some random background noise [6]. Thus, whenever a sensor measures e. g. the distance of a UAV to the sensor, the sensor realises a random variable.…”

“…This is precisely what is observed for this example in Figure 9a. The probability of detection of the agent peaks twice throughout the journey, at 1000 and 4000 seconds, occurring when the agent is close to sensors at positions (2,4) and (6,4), respectively.…”

“…Hui et al [11] provide a framework of joint optimization of target tracking and threat avoidance. Avoiding adversary sensors can also be seen as a case of an obstacle avoidance problem, see e. g. [4,6,19]. Joint optimization of trajectories and other decision has been considered only very recently; trajectories and signal strengths are considered by Zhou et al [29], while joint strategies for trajectory optimization and jamming resp.…”