Abstract—In earlier papers, the authors identified critical parameters to be used in any effective identification of aircraft vortex encounters. Various techniques of pure fuzzy logic and hybrid soft-computing approaches were used to model and successfully classify vortex encounters. In this paper, the authors consider pure neural networks models having different architectures to identify aircraft encounters of wing-tip vortices. The automatic identification of airplane vortex encounters using neural networks gives excellent accuracy when compared with manual approaches. The highest accuracies are obtained by probabilistic neural networks. They are about 93%, 73% and 83% for the overall training, the overall testing and the overall average, respectively. The achieved results confirm the effectiveness of some neural network techniques and the choice of the critical parameters to automatically identify wing-tip vortices. Index Terms—Wing tip vortices, vortex encounters, neural networks (NN), flight data records. I. INTRODUCTION Recent studies on the automatic identification of aircraft vortex encounters have shown a great potential in using soft-computing approaches. In [1], the authors used 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 data from flight data recorders (FDRs) and pilot reports. The fuzzy rules were derived from a collection of 54 pilot reports [2] of vortex encounters and 210 records of flight events from FDRs. An average success rate of identification of 83.7% was obtained. In [3], 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 sets of parameters and factors, and they achieved an average identification accuracy of 84.2%. This paper builds on the previous results achieved with FL and ANFIS approaches, respectively, in the automatic identification of aircraft vortex encounters. The authors continue to investigate machine learning by using various Manuscript received October 18, 2018; revised December 26, 2018. The authors are with the Dubai Men’s College, Higher Colleges of Technology, Dubai, UAE (e-mail: aziz.almahadin@hct.ac.ae, faouzi.bouslama@hct.ac.ae). architectures of pure neural networks (NNs). These NNs are constructed based on a similar reduced set of parameters as in [1]-[3]. The paper is structured as follows. Section II introduces the airplane wing tip vortex problem including the selection of the critical parameters relevant to this investigation. Section III explains the identification classes and input vectors. Section IV provides details on the proposed NNs investigated in this research. Section V is the conclusion. II. WING TIP VORTICES AND CRITICAL PARAMETERS All airplanes generate wing tip vortices due to the pressure difference between wing upper and lower surfaces, Fig. 1. This vortex may cause danger to following aircraft [4]. Hence, it is important to identify actual vortex encounters in order to introduce mitigation measures. The potential hazard of a vortex on a following airplane can vary depending on a number of parameters such as the type of following and the leading airplanes, the flight phase, the airplane weight, the wing size, airplane configuration and the weather conditions. Encountering vortex can be hazardous during flight, in particular, at landing and takeoff flight phases, where the airplanes are required to fly within constrained flight paths, which makes vortex encounter avoidance and recovery more difficult [2] and [4]. Fig. 1. Airplane wing tip vortices. For the purpose of this research, 181 FDR records are collected which were reported to contain vortex encounters. In addition, another 29 records are utilized which were reported to contain other flight events such as wind shear and hard landing [2]. These later records are used to compare flight events and to test the appropriateness of the various NN techniques to discriminate vortex encounters from other flight events. FDRs contain over one thousand parameters, but only 8 are found to be relevant to this investigation, Table I [2]. Recognition of Airplane Wing-Tip Vortices Encounters Using Neural Networks Aziz Al-Mahadin and Faouzi Bouslama International Journal of Machine Learning and Computing, Vol. 9, No. 2, April 2019 115 doi: 10.18178/ijmlc.2019.9.2.774