High-fidelity computational fluid dynamics driven by exploratory active incremental learning: Characterizing oscillating cylinders from subcritical to critical Reynolds numbers

The operating Reynolds number of offshore risers often transitions from the subcritical to the critical Reynolds number range at engineering application, which results in distinctly different vortex-induced vibration (VIV) responses. Consequently, such complexities challenge both full-scale computational fluid dynamics (CFD) and high-precision experiments, emphasizing the construction of a comprehensive database that integrates CFD and experimental data to enable rapid prediction. This fidelity-adaptive strategy successfully overcomes severe small-sample challenges under high-dimensional parameter spaces. For critical Reynolds number, the excitation region becomes shorter and wider and the positive range of added mass coefficient C m y expands significantly while its sensitivity to the amplitude-to-frequency ratio decreases markedly. The drag crisis is diminished when both the amplitude ratio and frequency ratio are high, reflecting the laminar separation bubbles(LSBs) being disrupted. The lift and drag signals exhibit multifrequency characteristics and amplitude modulation, and the lock-in range shifts towards lower frequency ratios. The present framework resolves high- R e hydrodynamic database construction bottlenecks, providing a reliable tool for deepsea riser VIV prediction and a instructive methodology for high-dimensional fluid mechanics parameter exploration.