We welcome you to CDS-KIAC talk on 13th July 2026 (Monday). The details are as below:
Speaker : Prof. Kalyana B. Nakshatrala, University of Houston
Title : Adaptive Physics-Informed Neural Network Modeling Framework
for Dual-Porosity Flow
Date and Time : July 13, 2026: 04:00 PM
Venue : # 102, CDS Seminar Hall
ABSTRACT:
Porous materials are central to many scientific and technological systems, from fractured geological formations and tight shales to engineered foams, biological scaffolds, and plant vascular networks. Across these systems, flow is often governed by two hydraulically coupled pore networks: a highly permeable macro-network that provides preferential pathways, and a finer micro-network that controls storage and inter-network exchange. Accurately modeling such dual- network systems is essential for predictive simulations in subsurface energy, critical-mineral recovery, geological carbon and hydrogen storage, geothermal systems, and engineered porous materials. This talk presents an adaptive physics-informed machine learning framework for fluid flow in double-porosity/permeability media. The approach embeds the mixed-form governing equations for coupled pressure and velocity fields directly into a physics-informed neural network, enabling both forward prediction and inverse parameter identification. To resolve the multiscale features and sharp solution gradients typical of dual-network porous media, the framework combines a shared-trunk neural architecture with adaptive loss weighting and error-guided collocation-point refinement. Together, these elements balance competing physical constraints, focus learning in high-residual regions, and preserve coupling between macro- and micro-pore networks. Representative numerical studies show that this mesh-free formulation captures discontinuities across layered media, remains stable under large permeability contrasts, and avoids spurious oscillations often encountered in classical mixed finite element schemes. Beyond forward simulation, the framework provides a natural platform for assimilating sparse experimental, field, or sensor data to infer difficult-to-measure quantities such as permeability fields and inter-porosity transfer coefficients. By incrementally updating the model as new data become available, the approach enables continuously refined forecasting of porous media systems. These capabilities support critical-mineral extraction, tight-shale production, CO₂ and hydrogen storage-integrity assessment, and geothermal heat-exchange optimization. The talk concludes by discussing how adaptive scientific machine learning can complement—and, in some settings, transform— traditional computational mechanics workflows for porous media modeling.
BIOGRAPHY:
Professor Kalyana Nakshatrala is the Carl F. Gauss Professor and Department Associate Chair of Civil and Environmental Engineering at the University of Houston, with a courtesy appointment in Mechanical Engineering. His research and teaching have been recognized with several honors, including the Andrea Prosperetti Research Computing Faculty Award, the Kittinger Teaching Excellence Award, and the University of Houston Teaching Excellence Award. He previously served as a faculty associate at Caltech and completed postdoctoral work at UIUC in collaboration with Los Alamos and Pacific Northwest National Laboratories. He holds a Ph.D. in Civil Engineering, M.S. degrees in Civil Engineering and Applied Mathematics, and a certificate in Computational Science and Engineering from UIUC, as well as a bachelor’s degree from IIT Madras. He is an associate editor of ASCE’s Journal of Engineering Mechanics and serves on the EMSL User Executive Committee.
Host Faculty: Prof. Soumyendu Raha, CDS
ALL ARE WELCOME
