| description abstract | Abstract. Tribology has traditionally functioned as an empirical discipline, heavily reliant on physical experimentation rather than unified first principles. To address the resulting epistemic bottlenecks, the paradigm of “triboinformatics” has emerged, leveraging machine learning (ML) to extract structured knowledge from high-entropy, multiscale datasets. Moving beyond conventional systematic reviews that primarily catalog bibliographic trends, this article presents a conceptual framework that deconstructs triboinformatics into its fundamental building blocks. We identify three major pillars: the machine learning component (the data-driven computational engine encompassing data sources, algorithms, and task modes), the Physics Component (the domain context defined by system scale and physical laws), and the dynamic Interface between them. This article critically analyzes this bidirectional Interface, conceptualizing two primary pathways of innovation. Direction A explores how ML algorithms drive epistemic gain through surrogate modeling, inverse materials design, and explainable AI. Conversely, Direction B examines how the strict laws of tribology compel methodological innovation, resulting in physics-informed machine learning (PIML) architectures that constrain algorithms within thermodynamically and mechanically plausible boundaries. Ultimately, this framework demonstrates that the future of triboinformatics lies not in replacing physical experiments with black-box algorithms but in their symbiotic integration, highlighting the transition from statistical correlation to causal discovery as the next critical frontier. | |
