Products

Our PIML tool leverages machine learning to accelerate physics-based simulations while ensuring all outputs honors the fundamental physical principles.

Dr. Birol Dindoruk

Utkarsh Sinha

In recent years, there has been a strong shift toward applying state-of-the-art machine learning (ML) methods to challenging science and engineering problems. While ML provides powerful capabilities for accelerating computations, estimating missing information, and improving predictive accuracy, it cannot be used as a complete black-box solution. Purely data-driven models risk producing results that violate known physical laws or generate behavior that is not realistic in the context of subsurface engineering. To address this, the tool developed here integrates ML with physics-based constraints, forming a physics-informed machine learning (PIML) framework that ensures predictions remain physically consistent while benefiting from modern data-driven techniques.

Within the oil and gas industry, such hybrid approaches have gained significant traction. By blending physics-driven models with ML components, hybrid frameworks overcome important limitations of traditional simulation techniques including the lack of complete inputs, numerical challenges in achieving convergence, and high computational cost while avoiding the weaknesses of unguided ML models that are not grounded in physics.

This work provides an application-focused overview of where physics-guided ML and hybrid modeling strategies have been successfully applied in oil and gas workflows. It highlights the methodologies used to embed physical principles into ML models, discusses common challenges encountered in implementing such frameworks, and evaluates the advantages and limitations of hybrid PIML techniques compared to purely physics-based approaches.

📄 Reference:

Sinha & Dindoruk (2025). Physics-Informed Machine Learning Framework for Subsurface Engineering Applications.