Review of physics-informed machine learning (PIML) methods applications in subsurface engineering

In recent years, there has been a growing trend in leveraging state-of-the-art machine learning (ML) techniques to solve complex science and engineering problems. While ML serves as a powerful tool to simplify computations, fill in gaps, and handle missing data, it should not be treated as a complete black-box solution. Relying solely on a black-box ML approach poses a significant risk of generating unphysical or inaccurate results. To mitigate these risks and ensure that the model adheres to the governing physical principles, it is essential to integrate machine learning with physics-based constraints, forming what is known as a physics-informed machine learning (PIML) model. In the oil and gas industry , PIML has emerged as a prominent solution, blending the computational efficiency of ML with the rigor of physics-based methods. Hybrid models, which combine physics-driven frameworks with ML, address key limitations of purely physics-based approaches—such as the lack of sufficient inputs or the high computational expense required to achieve convergence—and the shortcomings of unguided, purely data-driven ML models. This paper offers a structured overview of application-centric areas in the oil and gas industry where physics-guided ML models and hybrid frameworks have been effectively deployed. It explores various use cases, methodologies for integrating physics into ML, the challenges faced in these implementations, and the advantages and disadvantages of these hybrid approaches compared to traditional physics-based methods.