The study of transport phenomena, using nondestructive imaging technologies and simulations of fluid flow through complex geometries, has become increasingly vital, particularly in the biomedical domain. Such approaches provide valuable insight into critical processes, including blood perfusion of tissues, drug delivery, and extracellular transport, contributing to our understanding of physiology and the design of medical interventions. However, significant challenges persist in extracting geometry from imaging to reconstruct flow domains for use in direct simulations. Recent advancements in machine learning offer a promising solution by reconstructing domain geometry based on relatively sparse sampling of flow velocity. Motivated by this, we apply deep learning techniques to predict the shape and location of arbitrary bluff obstacles placed in a two-dimensional (2D) channel flow, utilizing convolutional neural networks (CNNs). A CNN is trained on the velocity or concentration field acquired from 2D Lattice Boltzmann simulations. We investigated various scenarios by adjusting CNN input and sparsity, focusing first on the $x$-component of the velocity field, then showing that the CNN performs similarly when applied to the concentration field. Future work will extend this methodology to experimental data to aid in the development of quantitatively accurate predictive models of transport in complex geometries, with numerous applications in biomedical transport and beyond.

Close

PrevNext

View figure in article

• Channel flow
• Laminar flows
• Convolutional neural networks
• Lattice-Boltzmann methods