Abstract—The encounter of vortices generated by a leading aircraft during takeoff or landing can be a source of hazard to a following aircraft. In spite of airport efforts to keep safe separation distances between aircrafts, a number of them encounter severe vortices each year. It has been challenging to accurately identify those encounters by manual approaches. To mitigate the impact of vortex encounters on an aircraft, it is important that more reliable identification techniques be developed. This research is a contribution towards the automatic identification of vortex encounters using artificial neural networks. Multilayer feedforward neural networks are trained using the back-propagation learning algorithm to classify flight events into either vortex encounters or other events. Using salient inputs such as aircraft roll angle, normal acceleration and lateral acceleration, the neural networks are able to achieve an overall average identification rate of about 88%. These results confirm the authors’ earlier assumption on using a reduced set of critical inputs to properly classify aircraft vortex encounters. Index Terms—Vortex encounter, flight data recorder (FDR), neural networks (NN), multilayer feed-forward (MLFF) network. I. INTRODUCTION Aircraft encounter various types of turbulences during a flight. One of the most hazardous turbulence is caused by the wing tip vortices. This type of turbulence is critical to flight safety as its decay is slow and can produce a significant rotational airflow that severely influence a following aircraft. In fact, aircraft safety is greatly affected by wake vortices generated by a leading aircraft. An aircraft wake vortex [1] is naturally produced by all types of aircraft. The severity of vortex encounter can vary depending on parameters such as the type of leading and following aircraft, the flight phase, the aircraft weight, the wing size, the configuration, and the weather conditions. Encountering a vortex can be hazardous during flight, in particular, at landing and takeoff flight phases, where the aircraft are required to fly within confined flight paths, which makes vortex encounter avoidance and recovery more difficult. In fact, the wake vortex hazard is one of the main factors Manuscript received June 20, 2018; revised October 23, 2018. This work is an interdisciplinary research work between the Computer Information Science Department (CIS) and the Aviation Engineering Department at Dubai Men’s College, the Higher Colleges of Technology, UAE. Faouzi Bouslama is with the CIS Department, Dubai Men’s College, the Higher Colleges of Technology, PO Box 15825, Dubai, UAE (e-mail: faouzi.bouslama@hct.ac.ae). Aziz Al-Mahadin is with the Aviation Engineering Department, Dubai Men’s College, the Higher Colleges of Technology, PO Box 15825, Dubai, UAE (e-mail: aziz.almahadin@hct.ac.ae). defining safe separation minima between two aircrafts. The international wake vortex separation rules are based upon the aircraft weight categorization whether it is Heavy, Medium or Light. However, such categorization has become inappropriate which led some countries to introduce their local separation standards [2]. Moreover, the vortex separations sometime unnecessarily reduce airports capacity. In fact, for any vortex separation modifications, there is a need for comprehensive relevant investigations to ensure safety and appropriateness. For this reason, it is necessary to examine and accurately identify actual vortex encounters. The identification of vortex encounters has been conducted manually in most cases. Very often, pilots are requested to report any vortex encounters, hence providing vital information to vortex analysis. The complete analysis report is supplemented by radar and meteorological information. Consequently, a flight data recording (FDR) analyst carries out a manual analysis of flight data to confirm the vortex encounters. Nevertheless, the manual analysis agreement with pilot reporting of vortex encounters is appoximately in the range of 55 to 70% [3]. In some studies, however, the focus has been on the automatic identification. Various modeling and classification approches were used to better capture uncertainties and complexities in data and also to reduce subjective human judgment errors. In [4], the authors reconstructed FDRs time histories using neural networks and established the concept of virtual flight data recorder. In [5], the authors evaluated the performance of neural networks and fuzzy logic re-constructors for the development of a virtual flight data recorder. They stated that the main drawback of their method was that specific flight data at each flight phase were needed for effective training of the neural network. More recent studies have shown more potential in using soft-computing approaches in the identification of vortex encounters. In [6], the authors use fuzzy logic (FL) to model and identify vortex encounters. FL tolerates data imprecision and cope well with complexities in modeling the vortex encounters. Fuzzy linguistic variables were used to model FDR data. The fuzzy rules were derived from a collection of 54 pilot reports of vortex encounters and 210 records of flight events from FDRs. An average success rate of identification of 83.7% was obtained. In [7], a neuro-fuzzy identification system was used to classify vortex encounters. Artificial neural networks integrated with fuzzy systems have been used as a solution in the automatic tuning of the membership functions of fuzzy linguistic variables and applied to various problems. The authors used a hybrid Adaptive Neuro-Fuzzy Inference System (ANFIS) to automatically tune the parameters of the fuzzy membership functions. They investigated various neuro-fuzzy models having different Airplane Vortex Encounters Identification Using Multilayer Feed-Forward Neural Networks Faouzi Bouslama and Aziz Al-Mahadin International Journal of Machine Learning and Computing, Vol. 9, No. 1, February 2019 1 doi: 10.18178/ijmlc.2019.9.1.757