Artificial Intelligence in Swimming Biomechanics and Fluid Dynamics: A Structured Narrative Review Towards Mechanistically Interpretable Models

Artificial intelligence is increasingly used in sport performance analysis, but its contribution to swimming biomechanics and fluid dynamics remains fragmented across wearable sensing, computer vision, predictive analytics and computational modelling. This structured narrative review synthesises current applications and evaluates how artificial intelligence can progress from describing movement to supporting mechanistically defensible interpretations of hydrodynamic propulsion, drag and performance. Evidence was identified through PubMed and Google Scholar (last search: 28 July 2026) and organised around inertial sensing, markerless pose estimation, training analytics, computational fluid dynamics, explainability and validation. Swimming cycle recognition, lap segmentation and temporal event detection are more extensively studied, but they lack adequate external validation, and direct validation of force- and flow-related outputs is weaker. Machine learning surrogates, physics-informed models and multimodal fusion offer promising routes to connect kinematics with hydrodynamic mechanisms, although small homogeneous datasets, data leakage and insufficient reporting constrain generalisability. The central contribution is the explicit distinction between kinematic observations, validated biomechanical estimates and hydrodynamic inferences. Artificial intelligence can extend biomechanical reasoning in swimming when predictions remain traceable to measurement quality, physical principles and the intended decision context.