#CIFRE PhD position on Geometric Deep Learning - H/F

Job Description:

A PhD position on Geometric Deep Learning has opened within the Aerodynamics department of Airbus Civil Aircraft in Toulouse.

Mission of the team

The Aerodynamics Department is responsible for the aerodynamic characterization of the overall aircraft for all Airbus programmes. We generate these models using various techniques from the early design stage, and then update and validate them through wind-tunnel testing and flight-test. The models are delivered to various customers such as:

• Handling Qualities and Flight Control Laws

• Performance Team

• Aircraft Loads

• Airworthiness and International Authorities

Development of  these models relies on inputs obtained from various sources, such as CFD, wind tunnel tests and flight tests. Each of these sources has its specificities in terms of resolution and reliability, and merging them to produce the overall model of the aircraft design is crucial for the later stages of the design, for example, the structural design to withstand mechanical loads. In addition, numerical simulations and wind tunnel tests are often supplemented with other data from flight tests or the aircraft's previous design state.

Aerodynamic models are then produced by merging this data, where information from these multiple sources is aggregated to produce a better quality estimate of the quantity of interest. The use of artificial intelligence, and more precisely of deep learning in this data fusion step is decisive in making it possible to effectively model complex data such as parietal pressure fields. Techniques based on convolutional neural networks have thus been shown to be very effective. Nevertheless, these algorithms modeling directly and only the surface fields, do not consider the geometry at all. This has the consequence of having to perform the data fusion on the same geometry regardless of the data source. This proves to be extremely penalizing in a multidisciplinary process, where the geometry evolves rapidly, and where the availability of data sources for a certain geometry is not guaranteed (e.g. wind tunnel tests which require several months of preparation). The need to be able to manage different geometric shapes with deep learning therefore takes on its full meaning.

The Role

Geometric Deep Learning (GDL) is a particular branch of artificial intelligence in which graph theory is used to extend the applicability of neural networks to non-Euclidean data. While classical algorithms such as convolutional neural networks (CNNs) are generally applied to data distributed on Cartesian grids such as the pixels of an image, geometric deep learning aims to precisely apply CNNs to non-Euclidean data described according to graph theory.

In this PhD, algorithms based on geodesic convolutions will be defined to enable the generation of aerodynamic models. The fusion of wing pressure distribution data will be performed with the aim of obtaining high-dimensional output with maximum quantification of precision and uncertainty from various input sources (CFD, wind tunnel tests, flight tests). To achieve this goal, geodesic convolution techniques will be studied and adapted to the specific case. Furthermore, the candidate will explore the possibilities offered by recent advances in DL techniques in terms of surrogate modeling to apply this approach in a global manner. The end result will be a methodology to systematically evaluate and select the best approach for data fusion in each technical setup. The resulting data fusion methodology will contribute to the development of future Airbus aircraft, from the early design phases to the analysis and identification of flight tests.

This PhD will be conducted in collaboration with the Aerodynamics Department at ISAE-Supaero.

Required skills

As the successful candidate, you will be able to demonstrate the following skills and competencies:

• Advanced Deep Learning (Graph Neural Network, Convolutional Neural Nets, GAN, VAE, Diffusion Model)

• Advanced Mathematics (Linear Algebra, Probability)

• Advanced knowledge of the main Deep Learning libraries (PyTorch or Tensorflow)

• General Aerodynamics skills

• Ability to read & apply state of the art papers in Deep Learning (Computer Vision, Geometric Neural Nets, Generative Models…)

• Python programming

• Communication

- Language skills:

• English: negotiation

• French: would be an advantage

This job requires an awareness of any potential compliance risks and a commitment to act with integrity, as the foundation for the Company’s success, reputation and sustainable growth.

Company:

Airbus Operations SAS

Employment Type:

PHD, Research

-------

Classe Emploi (France): Classe F11

Experience Level:

Student

Job Family:

Flight & Space Physics

By submitting your CV or application you are consenting to Airbus using and storing information about you for monitoring purposes relating to your application or future employment. This information will only be used by Airbus.
Airbus is committed to achieving workforce diversity and creating an inclusive working environment. We welcome all applications irrespective of social and cultural background, age, gender, disability, sexual orientation or religious belief.

Airbus is, and always has been, committed to equal opportunities for all. As such, we will never ask for any type of monetary exchange in the frame of a recruitment process. Any impersonation of Airbus to do so should be reported to emsom@airbus.com.

At Airbus, we support you to work, connect and collaborate more easily and flexibly. Wherever possible, we foster flexible working arrangements to stimulate innovative thinking.