The aviation industry needs led to an increase in the number of aircraft in the sky. When the number of flights within an airspace increases, the chance of a mid-air collision increases. Systems such as the Traffic Alert and Collision Avoidance System (TCAS) and Airborne Collision Avoidance System (ACAS) are currently used to alert pilots for potential mid-air collisions. The TCAS and the ACAS use algorithms to perform Aircraft Trajectory Predictions (ATPs) to detect potential conflicts between aircrafts. In this paper, three different aircraft trajectory prediction algorithms named Deep Neural Network (DNN), Random Forest (RF) and Extreme Gradient Boosting were implemented and evaluated in terms of their accuracy and robustness to predict the future aircraft heading. These algorithms were as well evaluated in the case of adversarial samples. Adversarial training is applied as defense method in order to increase the robustness of ATPs algorithms against the adversarial samples. Results showed that, comparing the three algorithm’s performance, the extreme gradient boosting algorithm was the most robust against adversarial samples and adversarial training may benefit the robustness of the algorithms against lower intense adversarial samples. The contributions of this paper concern the evaluation of different aircraft trajectory prediction algorithms, the exploration of the effects of adversarial attacks, and the effect of the defense against adversarial samples with low perturbation compared to no defense mechanism.
DOCUMENT
For long flights, the cruise is the longest phase and where the largest amount of fuel is consumed. An in-cruise optimization method has been implemented to calculate the optimal trajectory that reduces the flight cost. A three-dimensional grid has been created, coupling lateral navigation and vertical navigation profiles. With a dynamic analysis of the wind, the aircraft can perform a horizontal deviation or change altitudes via step climbs to reduce fuel consumption. As the number of waypoints and possible step climbs is increased, the number of flight trajectories increases exponentially; thus, a genetic algorithm has been implemented to reduce the total number of calculated trajectories compared to an exhaustive search. The aircraft’s model has been obtained from a performance database, which is currently used in the commercial flight management system studied in this paper. A 5% average flight cost reduction has been obtained.
MULTIFILE
Estimating the remaining useful life (RUL) of an asset lies at the heart of prognostics and health management (PHM) of many operations-critical industries such as aviation. Mod- ern methods of RUL estimation adopt techniques from deep learning (DL). However, most of these contemporary tech- niques deliver only single-point estimates for the RUL without reporting on the confidence of the prediction. This practice usually provides overly confident predictions that can have severe consequences in operational disruptions or even safety. To address this issue, we propose a technique for uncertainty quantification (UQ) based on Bayesian deep learning (BDL). The hyperparameters of the framework are tuned using a novel bi-objective Bayesian optimization method with objectives the predictive performance and predictive uncertainty. The method also integrates the data pre-processing steps into the hyperparameter optimization (HPO) stage, models the RUL as a Weibull distribution, and returns the survival curves of the monitored assets to allow informed decision-making. We vali- date this method on the widely used C-MAPSS dataset against a single-objective HPO baseline that aggregates the two ob- jectives through the harmonic mean (HM). We demonstrate the existence of trade-offs between the predictive performance and the predictive uncertainty and observe that the bi-objective HPO returns a larger number of hyperparameter configurations compared to the single-objective baseline. Furthermore, we see that with the proposed approach, it is possible to configure models for RUL estimation that exhibit better or comparable performance to the single-objective baseline when validated on the test sets.
DOCUMENT
