A machine learned go-around prediction model using pilot-in-the-loop simulations. (July 2022)

• Record Type: Journal Article
• Authors: Dhief, Imen; Alam, Sameer; Lilith, Nimrod; Mean, Chan Chea
• Is Part Of: Transportation research. Volume 140(2022)
• Publication Date: 2022-07
• Subjects: Flight approach phase -- Machine learning -- Safety metric -- Go-around prediction -- Pilot-in-the-loop simulations; Transportation -- Periodicals; Transportation -- Technological innovations -- Periodicals; 388.011
• DOI: 10.1016/j.trc.2022.103704
• Languages: English
• ISSNs: 0968-090X

Abstract

Abstract: Go-around manoeuvers are challenging for the management of arriving air traffic due to their complexity and unpredictability. In the present work, we propose to augment the air traffic control (ATC) environment with a predictive tool that assists Air traffic controllers (ATCOs) in the detection and prediction of the likelihood of go-around events. First, a safe and cost-effective flight approach data collection mechanism is introduced via a pilot-operated flight simulator under varying visibility conditions. Then, data are processed and a go-around probability assignment mechanism is introduced. Finally, a go-around probability prediction model is proposed. The proposed model updates the go-around probability for each new flight record, namely every second, and converges by detecting either a go-around or a successful landing. Experiments are conducted for two airports, namely Philadelphia International Airport (PHL) and Van Nuys Airport (KVNY), which feature different go-around procedures and are operated by different types of aircraft. Findings demonstrate that the trajectories of predicted go-around probabilities follow closely the computed probabilities regardless of the dissimilarities in the trajectory patterns. Furthermore, results suggest that the likelihood of go-around exceeds 93% when the go-around is initiated, for almost 50% of the flights in the test set. Moreover, the proposed model is able to detect the successful landing of aircraft. In fact, the system indicates low chances of go-around when the aircraft landed. For instance, the likelihood of a go-around at the touchdown is below 35% and 10% for KPHL and KVNY, respectively. The findings underscore the ability of the proposed model in providing accurate and timely go-around probabilities. This will help the ATCO to be prepared for re-sequencing arrival flights and clearing the go-around path, prior to the actual initiation of a go-around. Highlights: Air traffic control improved awareness with safety metric based on machine learning techniques. Assisting Air traffic controllers in detecting and predicting the likelihood of go-around events. Safe and cost-effective flight approach/landing data collection mechanism. A pilot-operated flight simulator under severe weather conditions. Provide air traffic controllers with a progressive sequence of go-around probabilities.