Analyze Potential of Dynamic Aircraft Wake Separation via Data-driven Aircraft Wake Region Detection

The recategorization of aircraft weight has demonstrated promising benefits in improving runway efficiency through safe reduction of aircraft wake separation. However, present implemented separation reduction schemes often remain conservative and time-independent; further research is warranted to develop dynamic wake separation in relation to meteorological conditions and aircraft pairs. This paper investigates the potential of a deep learning approach for near real-time aircraft wake region detection and consequently studies the exploratory dynamic temporal wake separation at Hong Kong International Airport (HKIA). The YOLO v5 model is applied as the benchmark for this wake region detection task. An improvement of the loss function is proposed to improve the detection performance and decision safety. Next, we examine the dynamic wake separation related to crosswinds by assessing the wake duration in the final approach path. The computational results indicate the superior regressive performance and the confidence of our proposed enhancement to the benchmarking model. Moreover, the further wake separation analysis based on wake region assessment reveals the effect of strong crosswinds in reducing wake separation time. This provides strong potential for online and near real-time wake vortex monitoring and the development of dynamic wake separation suggestion system.