Laboratoire des Écoulements Géophysiques et Industriels

Nos tutelles


Nos partenaires


Accueil > Actualités > Séminaires > Séminaires 2022

Mardi 6 Septembre 16h00 - LEGI Salle K118 VIRTUAL

Soledad le Clainche, Universidad Politecnica de Madrid

Flow patterns and machine learning and to develop predictive reduced order models

Modeling turbulent flows and solving the engineering problem mentioned, is a highly complex task that requires a large amount of computational resources. The alternative is developing Reduced order models (ROMs) using (among others) : (i) modal decompositions (i.e., singular value decomposition - SVD [1], higher-order dynamic mode decomposition - HODMD [2]), (ii) clustering based methods (i.e. principal component analysis - PCA and local PCA [3]) and (iii) machine learning tools [4,5].

The main goal of this work is to apply these techniques to solve several engineering problems with applications in aerospace engineering, presenting new strategies to develop efficient and accurate ROMs. More specifically, HODMD is used to identify the main patterns and to develop a ROM in an axisymmetric, time-varying, non-premixed co-flow flame and, PCA and LPCA are applied to develop a ROM in a synthetic jet. Finally, machine learning tools (artificial neural networks) are combined with modal decompositions (SVD) to develop a novel and efficient ROM [4].

[1] Sirovich, L. Turbulence and the dynamics of coherent structures. Parts I-III, Quart. Appl. Math., 45(3):561-571, 1987.
[2] Le Clainche S., Vega J.M., Higher order dynamic mode decomposition, SIAM J. Appl. Dyn. Sys., 16(2):882-925, 2017.
[3] Parente, A., Sutherland, J.C. Principal component analysis of turbulent combustion data : Data pre-processing and manifold sensitivity. Comb. Flame, 160:340-350, 2013.
[4] Abadia-Heredia, R., Lopez-Martin, M., Carro, B., Arribas, J.I., Perez, J.M., Le Clainche, S., A predictive hybrid reduced order model based on proper orthogonal decomposition combined with deep learning architectures. Exp. Syst. Appl. (in press), 2021.
[5] Lopez-Martin, M., Le Clainche, S., Carro, B. Model-free short-term fluid dynamics estimator with a deep 3D-convolutional neural network. Exp. Syst. Appl., 177:114924, 2021.

Dr. Soledad Le Clainche is an Associate Professor in the Dept. of Applied Mathematics of the School of Aerospace Engineering at the Polytechnic University of Madrid (UPM). In 2013 she completed her PhD thesis at the Dept. of Fluid Dynamics of the same University. Since 2011, she has published more than 80 scientific contributions, written two books, and presented her work in more than 40 international conferences, and invited seminars. She is an associate editor of the international journal Results in Engineering (from Elsevier JCR Q1), is a member of the Women and Mathematics committee of the Spanish Royal Mathematical Society and has participated in several national, international and industry collaborative projects. In 2022 she was awarded with the international Young Scientist Award by SARES (Sustainable Aviation Research Society).
Currently, Dr. Le Clainche leads the research line focused on the development and application of new mathematical models with applications in engineering, where she supervises 6 doctoral theses and has already supervised 2 more. She is PI of several national and European projects, whose main goal is to develop new mathematical models to improve energy efficiency in airplanes, reduce air pollution in cities and improve the efficiency of combustion systems.

Contact Martin Obligado for more information.