Enhancing Data-Driven Physics Models through the Integration of Fundamental Theory and Physics-Informed Learning

Abstract

This study aims to analyze how the integration of fundamental physics theory and physics-informed learning enhances data-driven physics models. The study addresses the growing challenge of developing models that are not only accurate in prediction but also consistent with physical laws, especially when data are limited, noisy, or incomplete. A qualitative approach with an exploratory case study design was used. Data were collected through semi-structured interviews, limited observation, and documentation involving lecturers, researchers, postgraduate students, and scientific computing practitioners with experience in computational physics, machine learning, numerical simulation, or physics-informed learning. The data were analyzed using thematic analysis with inductive and deductive coding. The findings reveal five main themes: physics theory as a guide and constraint for model development, physics-informed learning as a response to data limitations, validation and interpretability challenges, the importance of documentation and transparency, and the need for interdisciplinary collaboration. These findings show that the quality of data-driven physics models depends not only on algorithmic sophistication, but also on physical validity, methodological reflection, and transparent research practice. The study contributes to scientific machine learning by positioning physics-informed learning as an epistemic bridge between theory and data. It also suggests the need for stronger scientific computing literacy, clearer model reporting standards, and further research across broader fields of applied physics and engineering.