Wednesday 09 April 2025
Scientists have developed a new approach to predicting the dissolution of minerals in porous media, which is crucial for understanding and mitigating environmental hazards such as carbon capture and storage.
Porous media, like soil or rock formations, are complex systems that can affect the flow of fluids and the behavior of chemicals. One key process that occurs within these systems is reactive dissolution, where minerals break down into smaller particles in response to changes in temperature, pressure, or chemical composition.
Predicting this process accurately is essential for understanding how pollutants like carbon dioxide might move through subsurface environments, but traditional numerical methods are often too computationally expensive and inaccurate. This has led researchers to explore the use of machine learning algorithms to speed up simulations and improve their accuracy.
The new approach, developed by a team of scientists from Heriot-Watt University, uses a deep-learning-based method that incorporates both spatial and temporal information to predict the future states of reactive dissolution at a fixed time-step horizon. The team trained their model on an ensemble of numerical simulation models, which were created using various pore structures and fluid trajectories.
The results show that the new approach can accurately predict the evolution of porosity and permeability in porous media over time, with errors decreasing as the model is iteratively refined. This has significant implications for predicting the behavior of pollutants like carbon dioxide in subsurface environments, where understanding how they move and interact with minerals is critical for effective storage.
The team’s approach also offers a potential speedup of around 10^4 compared to traditional numerical methods, making it a promising tool for simulating complex environmental processes. By combining machine learning algorithms with physical laws, scientists may be able to better understand and predict the behavior of pollutants in porous media, ultimately informing more effective strategies for mitigating environmental hazards.
The researchers plan to continue refining their approach by exploring ways to improve the accuracy of their predictions and to extend their model to larger-scale domains. As the team’s work continues to evolve, it may lead to new insights into the complex interactions between pollutants, minerals, and porous media, ultimately informing more effective strategies for managing environmental risks.
Cite this article: “Revolutionizing Reactive Dissolution Modeling: A Deep Learning Approach to Predicting Porous Media Dynamics”, The Science Archive, 2025.
Machine Learning, Porous Media, Reactive Dissolution, Carbon Capture, Storage, Environmental Hazards, Pollutants, Subsurface Environments, Numerical Simulations, Deep-Learning-Based Method
