Abstract | During a flight, when a change in the operational conditions arises (e.g., new updated weather forecast, delay at reaching a given waypoint), different alternative trajectories can be computed with dedicated optimisation or prediction systems. These systems usually produce trajectories with trade-offs between expected fuel usage and delay. The pilot, or the dispatcher, considers these expected values in order to decide how to tactically operate the aircraft. This approach has two main challenges. Firstly, it requires the translation of arrival delay into parameters which are relevant for the airlines, such as on-time performance and cost of delay. Secondly, uncertainties in the system need to be estimated, such as holding time at arrival, or taxi-in time. Both of these estimations (airlines performance indicators and uncertainty) rely on the airline staff expertise. Finally, the crew faces a multi-criteria decision process as different objectives (cost, on-time performance) and constraints need to be considered. The use of prior to the flight estimations, such as the cost index of the operational flight plan, might not be relevant at the moment of reassessing the flight, as the situation has evolved (for example, the number of passengers who can potentially miss their connections will depend on the status of the fleet of the airline). In other cases, this expected cost of delay could be estimated by the crew or the dispatchers, but generally it is difficult to internalise the dynamics of cost due to IROPS on passengers, or even to estimate the cost of a potential curfew at the end of the day. Uncertainties such as the expected holding delay, distance flown at the arrival TMA, or taxi-in time, might lead to sub-optimal decisions, such as recovering delay, using extra fuel, which does not translate into economic benefit, as larger holding than anticipated might lead to passengers still missing their connection; or shorter distances flown in the TMA means that speed-ups performed during the cruise were unnecessary. Pilot3, a Clean Sky 2 Research and Innovation action, sets out to overcome these issues by developing a multi-criteria support decision tool, which combines explicit estimation of key performance indicators and estimation of ATM operational parameters. These estimators will be developed incrementally, from simple heuristics to machine learning models. Pilot3 prototype comprises five sub-systems: * An Alternatives Generator, which will compute the different alternatives to be considered by the pilot; fed by two independent sub-systems: * Performance Indicators Estimator, which provides the Alternatives Generator with information on how to estimate the impact of each solution for the different performance indicators; * Operational ATM Estimator, which provides the Alternative Generator with information on how to estimate some operational aspects such as tactical route amendments, expected arrival procedure, holding time in terminal airspace, distance flown (or flight time spent) in terminal airspace due to arrival sequencing and merging operations, or taxi-in time; * Performance Assessment Module, which, considering the expected results for each alternative on the different KPIs, is able to filter and rank the alternatives considering airlines and pilots preferences; and * Human Machine Interface, which will present these alternatives to the pilot and allow them to interact with the system. Pilot3 is led by the University of Westminster with the Universitat Politecnica de Catalunya, Innaxis and PACE Aerospace Engineering and Information Technology as partners. The Topic Manager is Thales AVS France SAS. With support from the Advisory Board, Pilot3 has already identified the key operational performance indicators that crew should consider when tactically adjusting their trajectories (on-time performance and total cost, including fuel, IROPs and others); and a literature review and filtering process on multi-criteria decision making techniques has been conducted to select the most suitable method for the different phases of the optimisation process (trajectory generation, filtering and ranking of alternatives). |
---|