From November 2018 to November 2020, MODAL has been contributing to the PERF-AI (Enhance Aircraft Performance and Optimisation through utilisation of Artificial Intelligence) project led by the company Safety Line. The project involved Benjamin Guedj and Vincent Vandewalle and hired Florent Dewez (as a postdoctoral researcher) and Arthur Talpaert (as a research engineer). The outputs are two research papers
- From industry-wide parameters to aircraft-centric on-flight inference: improving aeronautics performance prediction with machine learning. Dewez, F.; Guedj, B.; and Vandewalle, V. Data-Centric Engineering, 2020.
- An end-to-end data-driven optimisation framework for constrained trajectories. Dewez, F.; Guedj, B.; Talpaert, A.; and Vandewalle, V. 2020. Preprint
Here is the official description of the project.
Commercial aviation is already responsible of 3% of the total CO2 emissions, and with a constant growth rate of 5% per year, traffic will double within the next decade. Major improvements have been made in aircraft design and materials, engines performance and efficiency; still aircraftoperations remain unchanged for many years. With the digital transformation of many industries, airline operations have been just at the beginning of a major change. With the support of new technologies related to machine learning and artificial intelligence, in-flight connectivity, major improvements can be introduced to optimize flight trajectories.
The main objective of PERF-AI is to bring those new technologies to the field of aviation based on statistical analysis of flight data that are generated by aircraft throughout their lifecycle. Currently, aircraft manufacturers, flight management systems and flight preparation software providers are using a single manufacturer’s performance model that is the same for every aircraft of the same type, and also on a weather forecast that is computed long before the flight. The performance is based on manufacturer’s model that is derived from flight tests conducted on brand new aircraft during certification phase. The only corrections applied to those performance models are made through the fuel or performance factor, that is a single percentage applied to the whole flight, though it is only a measurement made during cruise phase and corresponds to a steady flight. PERF-AI will focus on the challenge of minimizing fuel consumption throughout the flight. The aim will be to provide a flight trajectory optimization prototype that implements new machine learning performance models. Minimizing the fuel consumption can be mathematically modelled as an optimal control problem, whose solution is expected to be as close as possible to the best trajectory in reality. This can only be achieved if the mathematical modelling of the problem is performed as accurately as possible, a requirement for this being the precise estimation of the aircraft’s behaviour. This motivates the search for narrow system identification techniques, which are the main topic of this call. Several machine learning methods will be identified and tested for this purpose. New techniques will be proposed in order to have the most accurate tool as possible. Moreover, high-level artificial intelligence techniques will use the machine learning models for the objective of the fuel use minimization.
This project has received funding from the Clean Sky 2 Joint Undertaking under the European Union’s Horizon 2020 research and innovation programme under grant agreement No 815914 (PERF-AI).