PERF-AI optimises flight trajectories and reduces emissions with artificial intelligence
The PERF-AI project has developed a fully functional prototype that analyses historical flight data to generate accurate performance models of individual aircraft.
Taking into account specific differences that arise in an aircraft’s lifetime and impact on their individual performance (such as ageing engines, modifications to aircraft structure etc), the software generates a more tailored, accurate performance model that can then be used to calculate flight trajectories that are optimised for fuel efficiency.
The data that’s collected can be used to inform flight crews with more accurate estimates of how much fuel they’ll need. Carrying less fuel on board means less weight, so less fuel burn. The technology also suggests more efficient flight trajectories, supporting pilots to make more eco-conscious decisions. In some cases, PERF-AI reported a reduction of 500 kg of carbon dioxide emissions for a medium-haul flight of two hours, representing up to 2-5% of cruise consumption.
The project started by collecting and setting up a large enough flight data set, then several machine learning algorithms were developed and tested to select the most relevant. Finally, a standalone application was built that can use any type of aircraft data and provide corresponding performance tables. In parallel, new types of optimisation techniques based on machine learning were also explored.
An impressive number of flights were analysed to create the dataset – 17000 flights for B737-800 type aircraft and 25000 flights on A320.
Despite the huge amount of data processed, the prototype doesn’t require particularly powerful computers, meaning it can be used not only to calculate optimum trajectories before flight but could also be used in the future to re-calculate during flight. It was tested on two different aircraft types, and demonstrated very good initial results.
It wasn’t all smooth sailing for the project, however. Some of the information that needed to be extracted from the data was not present among the available parameters. A workaround was needed and the use of flight dynamics was used to compute the missing elements.
One research paper has already been published in Data Centric Engineering Review from Cambridge University, and a second one has been submitted.
Discussions are ongoing regarding how to exploit the results of the project in a commercial form. The results of the optimisation are currently being tested with Transavia France.
The partners of the project were Safety Line, a Paris based SME, and INRIA Lille a research centre specialised in Machine Learning. The Topic manager was Thales Avionics and two airlines were represented at the advisory board: Transavia France and Lufthansa. Altogether, €705 000 was invested in PERF-AI, with €568550 coming from the European Commission’s Horizon 2020 funding under Clean Sky.