vcard atul


Picture of Atul Agrawal

Atul Agrawal

Technical University of Munich

Associate Professorship of Data-driven Materials Modeling (Prof. Koutsourelakis)

Postal address

Postal:
Boltzmannstr. 15
85748 Garching b. München

Research Interests

  • Turbulence modeling
  • Inverse Problems
  • Physics Informed Machine Learning
  • Uncertainty Quantification
  • Reduced-order modeling
  • Scientific Machine Learning
  • Bayesian Statistics
  • Optimization under uncertainty

Publications

Atul Agrawal, Erik Tamsen, Phaedon-Stelios Koutsourelakis, Joerg F. Unger: From concrete mixture to structural design -- a holistic optimization procedure in the presence of uncertainties (Under Review)

Atul Agrawal, Phaedon-Stelios Koutsourelakis: A probabilistic, data-driven closure model for RANS simulations with aleatoric, model uncertainty (Under Review)

Atul Agrawal, Kislaya Ravi, P.S. Koutsourelakis, Hans-Joachim Bungartz: Multi-fidelity Constrained Optimization for Stochastic Black-Box Simulators (NeurIPS 2023, Machine learning for Physical Sciences)

Leon Riccius, Atul Agrawal, P.S. Koutsourelakis: Physics-Informed Tensor Basis Neural Network for Turbulence Closure Modeling (NeurIPS 2023, Machine learning for Physical Sciences)

Didier Lucor, Atul Agrawal, Anne Sergent: Simple computational strategies for more effective physics-informed neural networks modeling of turbulent natural convection  (Journal of Computational Physics)

Singh, D., Agrawal, A., and Roy Mahapatra, D., A Reduced-Order Modeling Framework for Simulating Signatures of Faults in a Bladed Disk, SAE Int. J. Aerosp.

Workshops and Seminars

Teaching

  • Uncertainty Modeling in Engineering (WS 2023 -)
  • Modeling in Structural Mechanics ( SS 2021 - )
  • Continuum Mechanics (SS 2020 - )

Supervised Student Projects

  • Jonas Eichelsdörfer (Master's Thesis) : Physics Informed Machine Learning and Hamiltonian Neural Networks ( supervised together with Sebastian Kaltenbach)
  • Leon Riccius (Master's Thesis): Machine Learning based approach for investigating Reynolds stress discrepancy based on DNS/LES data. Here

Open Student Projects :

  • Developing differentiable RANS solver in PyTorch for downstream UQ/ML tasks (details here)

Education

  • 2017-2019: M.Sc. Computational Mechanics, Ecole Centrale Nantes
    • Master's Thesis: Surrogate Modelling for Turbulent Thermal Convection Processes Based on Physics Informed Deep Learning ,(LIMSI-CNRS, Universite paris Sacalay, Orsay FRANCE)
  • 2010-2014: B.Tech. Mechanical Engineering, Indian Institute of Technology, Varanasi (IIT-BHU)