In modern applications, high-fidelity computational models are often impractical due to their slow performance and also lack information about the certainty of their predictions. Hence, we introduce innovative deep learning surrogate frameworks that are scalable, robust, require minimum hyper-parameter tuning, are fast at the inference stage, and are accurate in forecasting non-linear deformation responses of solid objects. These surrogate frameworks are constructed using various deep learning techniques under deterministic as well as Bayesian settings. Bayesian frameworks enable us to capture uncertainties and provide a means to trust the predictions of the neural network approaches.
References:
Saurabh Deshpande, Stéphane P.A. Bordas and Jakub Lengiewicz. MAgNET: A graph U-Net architecture for mesh-based Simulations. arXiv 2023. https://arxiv.org/abs/2211.00713
Saurabh Deshpande, Raùl Ian Sosa, Stéphane P.A. Bordas and Jakub Lengiewicz. Convolution, aggregation and attention based deep neural networks for accelerating simulations in mechanics. Frontiers in Materials 2023. https://doi.org/10.3389/fmats.2023.1128954
Saurabh Deshpande, Jakub Lengiewicz and Stéphane P.A. Bordas. Probabilistic Deep Learning for Real-Time Large Deformation Simulations. Computer Methods in Applied Mechanics and Engineering (CMAME) 2022. https://doi.org/10.1016/j.cma.2022.115307
Codes and datasets:
Source codes and implementation for Multi-channel Aggregation Network (MAgNET):
Github repository: https://github.com/saurabhdeshpande93/MAgNET
Dataset repository: https://doi.org/10.5281/zenodo.7784804
Implementation of convolutional-aggregation and attention based neural networks:
Github repository: https://github.com/saurabhdeshpande93/convolution-aggregation-attention
Dataset repository: https://doi.org/10.5281/zenodo.7585319
In addition to meeting functional requirements, all materials must satisfy structural adequacy criteria, which heavily rely on properties such as Elastic modulus and fracture toughness, among others. The microstructure of a material plays a pivotal role in determining its structural integrity. Given that fracture can manifest through various modes, including intra-granular and inter-granular interfaces, characterizing and representing the stochastic nature of microstructure become crucial for accurate estimation of mechanical properties. Establishing such complex relationship between material properties and statistically equivalent microstructures necessitates sophisticated finite element simulations.
The continuous evolution of manufacturing technologies and the increasing emphasis on sustainability have led to the emergence of several new materials, spanning alloys, ceramics, and polymers. This diversity of materials and the stochastic nature of microstructures presents a vast spectrum of possibilities. Consequently, our research endeavors to develop machine learning-aided simulation tools, aiming accurate prediction of fracture responses. This collaborative effort involves the Tech startup EmTDLab Space Division, along with partnerships with The European Space Agency and the Luxembourg Space Agency.
Team:
1. Prof. Stephane Bordas (Head, Legato Team and Professor, Comtutional Mechanics)
2. Dr. ir. Lars Beex (Senior Research Scientist)
3. Dr. Surendran Murugesan (Postdoctoral Researcher)
Duration:
September 2023 - Present
Buildings consume large amounts of energy from the electrical grid that often originates from coal and fossil fuel power plants, increasing greenhouse gas footprint. Moreover, today’s buildings use more energy then they produce. On the other hand, buildings are still viewed as shelters, a combination of bricks, steel and concrete. If we could shift this archetype toward a building as a fully independent entity, society would experience tremendous benefits. The realisation of this vision seems to be an urgent matter, for two reasons: acceleration of population growth and urbanisation of the world. Namely, the current world population is expected to increase by 29% by 2050, while at the same time, the projected urban population will increase by 11% according to a recent United Nations report.
The general aim of the DATA4WIND project is to develop a novel hybrid approach to evaluate the wind energy harvesting potential in urban areas through data assimilation and machine learning based on computational and experimental wind engineering techniques. The long-term vision of DATA4WIND project is to reach a point of energy self-sustainable buildings. That would enable each building, each home, to become its clean energy power plant, doing its part to energize world around us.
DATA4WIND project was funded by Luxembourg National Research Fund (FNR) and was also granted with H2020-MSCA-IF-2019.
References:
Anina Šarkić Glumac, Onkar Jadhav, Vladimir Despotović, Bert Blocken, Stephane P.A. Bordas. A multi-fidelity wind surface pressure assessment via machine learning: A high-rise building case. Building and Environment. 2023. https://doi.org/10.1016/j.buildenv.2023.110135
Lucas Villanueva, Miguel Martínez Valero, Anina Šarkić Glumac, Marcello Meldi, Augmented state estimation of urban settings using on-the-fly sequential Data Assimilation, Computers & Fluids, 106118, 2023. https://doi.org/10.1016/j.compfluid.2023.106118
Kristina Kostadinović Vranešević, Stanko Ćorić, Anina Šarkić Glumac. LES study on the urban wind energy resources above the roof of buildings in generic cluster arrangments: Impact of building position. Journal of Wind Engineering and Industrial Aerodynamics 240, 105503. 2023. https://doi.org/10.1016/j.jweia.2023.105503
Kristina Kostadinović Vranešević, Gulio Vita, Stephane P.A. Bordas, Anina Šarkić Glumac. Furthering knowledge on the flow pattern around high-rise buildings: LES investigation of the wind energy potential. Journal of wind engineering and industrial aerodynamics. 226. 105029. 2022. https://doi.org/10.1016/j.jweia.2022.105029
Gulio Vita, Anina Šarkić Glumac, Hassan Hemida, Simone Salvatori, Charalampos Baniotopoulos. On the Wind Energy Resource above High-Rise Buildings. Energies. 13. 3641. 2020. https://doi.org/10.3390/en13143641
Hassan Hemida, Anina Šarkić Glumac, Gulio Vita, Kristina Kostadinović Vranešević, Rüdiger Höffer. On the flow over high-rise building for wind energy harvesting: An experimental investigation of wind speed and surface pressure. Applied Sciences, 10, 5283, 2020. https://doi.org/10.3390/app10155283