Mohammad Sarabian: Unlocking the Potential of AI-Driven CFD

Throughout my career, I have been driven by a passion for advancing scientific knowledge and pushing the boundaries of innovation in Computational Fluid Dynamics (CFD), Artificial Intelligence (AI), and biomedical sciences. My research projects and publications showcase the depth and breadth of my expertise, highlighting my contributions to these rapidly evolving fields. Below, you will find a selection of my most impactful publications. For a comprehensive list of my research works, please visit my Google Scholar profile.

Publications:

  • FV-FluidAttentionNet: A Label-Free Physics-Informed Autoencoder with Finite-Volume Discretization for Rapid Navier-Stokes Solutions: 

We introduce FV-FluidAttentionNet, an innovative physics-informed autoencoder with an attention mechanism that revolutionizes computational fluid dynamics (CFD) simulations. This advanced surrogate model seamlessly integrates fundamental fluid mechanics principles with cutting-edge AI technologies and finite-volume discretization, all while operating without the need for labeled data. FV-FluidAttentionNet employs finite-volume discretization-based schemes to approximate spatio-temporal partial derivatives of the Navier-Stokes equations, a crucial strategy for achieving superior accuracy, stability, and convergence in these highly non-linear, coupled PDEs. Our novel approach incorporates finite-volume discretization schemes within the computational graph, enabling rapid GPU-based calculation of Navier-Stokes residuals. This optimization facilitates efficient weight updates in the autoencoder, resulting in swift convergence to accurate solutions. We demonstrate FV-FluidAttentionNet’s capabilities by solving steady, incompressible Navier-Stokes equations for various scenarios, including lid-driven cavity flow and flow past a cylinder at different Reynolds numbers (see the following movie). The results are remarkable: our physics-informed network reduces computational time by a factor of 1,000 compared to conventional Navier-Stokes solvers, while maintaining high fidelity and eliminating the need for labeled training data. FV-FluidAttentionNet represents a paradigm shift in CFD modeling, offering unprecedented speed and efficiency in solving Navier-Stokes equations without compromising physical accuracy. This breakthrough has far-reaching implications for fluid dynamics research, engineering design optimization, and real-time flow prediction across various industries, paving the way for a new era in computational fluid dynamics.

Stay tuned for the upcoming paper, which promises to reveal groundbreaking insights and exciting developments in the field!

  • Ashtiani, S. Z., Sarabian, M., Laksari, K., & Babaee, H. (2024). Reconstructing Blood Flow in Data-Poor Regimes: A Vasculature Network Kernel for Gaussian Process Regression. In Press, [Journal of the Royal Society Interface].

In this groundbreaking study, we propose a novel approach for reconstructing blood flow in data-poor regimes using a vasculature network kernel for Gaussian process regression. By leveraging the inherent structure of vascular networks and incorporating prior knowledge of hemodynamics, our method enables accurate and efficient estimation of blood flow in scenarios where traditional measurement techniques fall short. This research has the potential to revolutionize our understanding of blood flow dynamics in complex vascular systems and pave the way for improved diagnostic and therapeutic strategies in cardiovascular medicine. Please find the paper here.

  • Kamali, A., Sarabian, M., & Laksari, K. (2023). Elasticity Imaging Using Physics-informed Neural Networks: Spatial Discovery of Elastic Modulus and Poisson’s Ratio. Acta Biomaterialia 1742-7061 (*Equal contribution).

In this pioneering study, where I served as the first co-author, we introduce a novel physics-informed neural network (PINN) framework for elasticity imaging, enabling the spatial discovery of elastic modulus and Poisson’s ratio in biological tissues. By integrating the governing equations of elasticity into the neural network architecture, our approach successfully estimates the space-dependent distribution of mechanical properties from limited and noisy displacement measurements. This research opens new avenues for non-invasive characterization of tissue mechanics, with potential applications in disease diagnosis, treatment planning, and biomaterial design. To delve deeper into the details of our innovative approach and its implications, I invite you to read the full paper here.

  • Sarabian, M., Babaee, B., & Laksari, K. (2022). Physics-informed neural networks for brain hemodynamic predictions using medical imaging. IEEE Transactions on Medical Imaging 41(9) 2285-2303. doi: 10.1109/TMI.2022.3161653. 

In this seminal work, we introduce a cutting-edge physics-informed neural network (PINN) framework for brain hemodynamic predictions using medical imaging. By seamlessly integrating medical imaging data with the governing equations of fluid dynamics, our approach enables accurate and high-resolution predictions of cerebral blood flow, pressure, and vessel diameters. This research showcases the immense potential of AI-driven technologies in advancing our understanding of cerebrovascular physiology and facilitating personalized diagnosis and treatment of neurological disorders. For a comprehensive understanding of our methodology and its clinical implications, please refer to the full paper here.

  • Sarabian, M., Rosti, M.E., Brandt, L., & Hormozi, S. (2020). Numerical simulations of a sphere settling in simple shear flows of yield stress fluids. Journal of Fluid Mechanics 896, A17. 

In this computational study, we employ high-fidelity numerical simulations to investigate the settling behavior of a sphere in simple shear flows of yield stress fluids. By capturing the intricate coupling between the particle dynamics and the non-Newtonian fluid rheology, our simulations shed light on the complex flow patterns and force balance governing particle motion in yield stress fluids. This research contributes to a better understanding of particle-fluid interactions in complex fluids, with applications ranging from materials processing to environmental engineering. To explore the intricacies of our numerical approach and the insights gained, I encourage you to read the full paper here.

  • Sarabian, M., Firouznia, M., Metzger, B., & Hormozi, S. (2019). Fully developed and transient concentration profiles of particulate suspensions sheared in a cylindrical Couette cell. Journal of Fluid Mechanics 862, 659-671.

In this experimental investigation, we study the fully developed and transient concentration profiles of particulate suspensions sheared in a cylindrical Couette cell. Using advanced flow visualization and image analysis techniques, we capture the spatio-temporal evolution of particle concentration fields under various shear conditions. Our findings provide new insights into the migration and segregation phenomena in sheared suspensions, with implications for the design and optimization of industrial mixing and transport processes. For further details, please see the paper here.

  • [Cover] Izbassarov, D., Rosti, M. E., Ardekani, M. N., Sarabian, M., Hormozi, S., Brandt, L., & Tammisola, O. (2018). Computational modeling of multiphase viscoelastic and elastoviscoplastic flows. International Journal for Numerical Methods in Fluids, 88(12), 521-543.

In this collaborative work, featured on the cover of the International Journal for Numerical Methods in Fluids, we present a comprehensive computational framework for modeling multiphase viscoelastic and elastoviscoplastic flows. By incorporating advanced constitutive models and numerical techniques, our framework enables accurate and efficient simulations of complex fluid phenomena, such as polymer extrusion, emulsification, and biological flows. This research demonstrates the power of computational modeling in advancing our understanding of multiphase flows and guides the development of innovative processing strategies in various industrial sectors. For an in-depth exploration of our computational framework and its applications, please refer to the full paper here.

Talks and Presentations:

  • Physics-Inspired Artificial Intelligence in Biomedical Engineering, University of Arizona, department of Biomedical Engineering (BME) Seminar, Feb 2022. Please find more details here.
  • Sarabian, M., Rosti, M.E., & Brandt, L. (2023, November). Particle-resolved simulations of the wall effect on the sedimentation of a single sphere in yield-stress fluids. The 76th Annual Meeting of the American Physical Society’s Division of Fluid Dynamics (APS-DFD), Washington, DC.
  • Sarabian, M., Mataran, P., & Torrado, R. (2022, November). Finite PINN Net: Physics-informed deep convolutional neural networks for learning 3D transient Darcy flows in heterogeneous porous media. The 75th Annual Meeting of the American Physical Society’s Division of Fluid Dynamics (APS-DFD), Indianapolis, Indiana.
  • Zamani Ashtiani S., Sarabian, M., Laksari K., & Babaee H. (2022, November). Blood flow predictions in data-poor regimes: A physics-informed Bayesian approach. The 75th Annual Meeting of the American Physical Society’s Division of Fluid Dynamics (APS-DFD), Indianapolis, Indiana.
  • Sarabian, M., Babaee, H., & Laksari, K. (2021, November). Brain hemodynamic predictions from noninvasive Transcranial Doppler ultrasound and angiography data Using Physics-Informed Neural Networks. The 74th Annual Meeting of the American Physical Society’s Division of Fluid Dynamics (APS-DFD), Phoenix, Arizona.
  • Sarabian, M., Rosti, M.E., Brandt, L., & Hormozi, S. (2019, October). Shear-induced sedimentation of a sphere in yield stress fluids: A computational study. The 91th Annual Meeting of The Society of Rheology (SOR), Raleigh, North Carolina.
  • Sarabian, M., Rosti, M.E., Brandt, L., & Hormozi, S. (2019, November). Particle resolved simulations of a sphere settling in simple shear flows of yield-stress fluids. The 72nd Annual Meeting of the American Physical Society’s Division of Fluid Dynamics (APS-DFD), Seattle, Washington.
  • Sarabian, M., Rosti, M.E., Brandt, L., & Hormozi, S. (2018, November). Interface-resolved simulations of a sphere settling in simple shear flows of elastoviscoplastic fluids. The 71nd Annual Meeting of the American Physical Society’s Division of Fluid Dynamics (APS-DFD), Atlanta, Georgia.
  • Rashedi, A.R., Sarabian, M., Ovarlez, G., & Hormozi, S. (2018, November). An experimental study on fracturing flows of Newtonian fluids. The 71nd Annual Meeting of the American Physical Society’s Division of Fluid Dynamics (APS-DFD), Atlanta, Georgia.
  • Sarabian, M., Metzger, B., & Hormozi, S. (2017, November). Dispersion and layering of solid particles in cylindrical Couette flows. The 70nd Annual Meeting of the American Physical Society’s Division of Fluid Dynamics (APS-DFD), Denver, Colorado.

These selected publications and presentations demonstrate my commitment to advancing the frontiers of knowledge in mathematical modeling, CFD, AI, and biomedical sciences. By combining rigorous experimentation, cutting-edge computational modeling, and innovative AI techniques, I strive to develop transformative solutions that address complex challenges across diverse engineering disciplines. To explore more of my research works, please visit my Google Scholar profile, where you will find a comprehensive list of my publications, citations, and research impact.

 

As I continue to explore new avenues of research and collaboration, I remain dedicated to pushing the boundaries of what is possible and driving meaningful impact in the scientific community and beyond. Feel free to reach out if you have any questions or would like to discuss potential collaborations.