Ehsan Haghighat

Ehsan Haghighat

Researcher in Scientific Computations, Stochastic Modeling, and Deep Learning

University of British Columbia

Massachusetts Institute of Technology

Biography

Dr. Ehsan Haghighat is a researcher and practitioner in the areas of Scientific Machine Learning, Computational Mechanics, and Mechanics of Solids and Porous Media. His research interests include computational solid- and poro-mechanics and multiphase flow in porous media, stochastic modeling and uncertainty quantification of engineering systems, and the use of artificial intelligence tools, including deep learning, in engineering analysis. Currently, he collaborates with multiple groups on the development and application of machine learning tools for engineering applications. Additionally, he provides consulting services in these areas. For more information, please contact him through the form below.

Interests

  • Scientific Machine Learning
  • Stochastic Modeling and Uncertainty Quantification
  • Numerical Methods
  • Mechanics of Solids and Porous Media

Education

  • Postdoctoral Fellow, 2020-2021

    University of British Columbia

  • Postdoctoral Associate, 2017-2019

    Massachusetts Institute of Technology

  • PhD in Computational Mechanics, 2011-2015

    McMaster University

Experience

 
 
 
 
 

Simulation Consultant

Seismix Reservoir Management, LLC

Jan 2020 – Present Cambridge MA, USA

Research activities include:

  • Stochastic modeling
  • Uncertainty quantification
  • Multiphase flow simulations
 
 
 
 
 

Postdoctoral Fellow

University of British Columbia

Jan 2020 – May 2021 Vancouver BC, CA

Research activities include:

  • Deep learning for engineering
  • Stochastic modeling
  • Uncertainty quantification
 
 
 
 
 

Postdoctoral Associate

Massachusetts Institute of Technology

Jan 2017 – Dec 2019 Cambridge MA, US

Research activities include:

  • Assessment of induced seismicity through multiphase flow and geomechanics simulations
  • Development of MATLAB’s vFEMLab for geological modeling using implicit interface methods
  • Development of SciANN library for Physics-Informed Deep Learning
  • Stochastic modeling of gas leakage to the surface
 
 
 
 
 

Lead Mechanics

Forming Technologies Inc.

Oct 2014 – Dec 2016 Burlington ON, Canada

Research and development activities included:

  • Development of a new implicit-incremental FEM solver using thick shell theory with large-deformation and contact considerations.
  • Development and exploration of various linear system solvers.
 
 
 
 
 

Research Assistant and Graduate Student

McMaster University

Jul 2011 – Dec 2014 Hamilton ON, Canada

Study and research activities included:

  • Constitutive modeling
  • FEM, XFEM, Meshfree
  • Scientific computations and linear algebra
  • Programming, C++, FORTRAN, Python, MATLAB.

Recent Publications

Quickly discover relevant content by filtering publications.
(2021). PINNtomo: Seismic tomography using physics-informed neural networks. Submitted.

Preprint PDF Source Document

(2021). A holistic approach to computing first-arrival traveltimes using neural networks. Submitted.

Preprint PDF Source Document

(2020). Machine learning for accelerating 2D flood models: Potential and challenges. Hydrological Processes.

PDF Source Document

Research Background and Interests

Deep Learning in Engineering

Use of deep learning in engineering analysis and inversion.

Mechanics of Contact and Impact

Contact description, friction (Coulomb and Rate-and-State), and contact search.

Constitutive Modeling

Constitutive modeling of isotropic and anisotropic materials including nonlinear elasticity, plasticity, and heterogeneity.

Modeling of Fracturing and Localization

Cohesive fraction propagation, localization and shear band formation, and hydraulic fracking.

Numerical Methods

FEM, XFEM, Meshfree, FVM, IGA, etc.

Projects

SciML

ML for engineering and science applications.

SciANN

A Keras wrapper for scientific computations and physics-informed deep learning using artificial neural networks

vFEMLab

A vectorized FEM MATLAB library using the implicit interface methods

NextMeal

A proposal for a recommendation system for learning taste profile (3rd place winner).

Contact