Machine learning-assisted research on the correlation between surface bubbles and slip length

Micro/nanobubble-enabled drag reduction in microfluidics represents a cutting-edge frontier in interfacial fluid dynamics. This study employs both a conventional Fully Connected Neural Network (FCNN) and the recently proposed Kolmogorov–Arnold Network (KAN) to elucidate the complex interplay between bubble morphology and slip length. A comparative analysis demonstrates the decisive superiority of KAN in predicting the slip length based on key bubble parameters, including the protrusion angle, lateral size, and gas coverage ratio. The KAN model achieved substantial error reductions ranging from 34% to nearly 60% across most testing scenarios, with a remarkable 58.95% improvement on external experimental validation. While both models showed comparable performance in one specific case (a marginal 1.76% reduction), KAN consistently demonstrated superior robustness, achieving errors as low as 2.26%–4.74% on numerical benchmarks and 4.45% on experimental data—substantially lower, and often less than half, than the FCNN's corresponding errors. This study establishes KAN as a powerful analytical framework for unraveling the intricate physical mechanisms of bubble-assisted drag reduction, thereby providing critical theoretical guidance for the rational design of high-performance microfluidic devices.