Roughness effects on two-dimensional turbulent convection: heat transport, flow reversals, and machine learning analysis

In two-dimensional turbulent convection, the effects of roughness configurations on heat transport and flow reversal are examined in this work. The impact of five distinct rough models on the Nusselt number (Nu) as a function of Rayleigh number (Ra) is investigated and analysed. All the rough models show reduced heat transport at low Ra; the model with locally compact roughness elements shows the most significant reduction in heat transport. As Ra increases, the normalized Nu generally increases, with differences observed between models with sparsely distributed and locally compact roughness. Flow reversals in 2DRB convection are also explored, with the presence or absence of reversals categorized among the rough models. Flow reversal processes are identified using angular momentum analysis. The study reveals chaotic oscillations in the flow field and Nu for certain models, indicating the influence of roughness on the Large-Scale Circulation (LSC). Sparse models with widely spaced rough elements exhibit more active correlations between the cavity’s fluid and LSC, leading to enhanced heat transfer. The scaling relationship between Nu and Ra is investigated, showing distinct scaling regimes for different Ra ranges. The distribution of roughness elements and the relative contributions of the majority of the surface and boundary-layer areas to thermal dissipation influence scalar behaviour. Machine learning techniques, including Convolutional Auto-encoders (CAEs) and Gated Recurrent Units (GRUs), are employed to compress and predict snapshots of turbulent convection data. These techniques offer a promising approach to analyse complex turbulence data and facilitate sequence analysis and prediction. Overall, this work delivers valuable insights into the role of roughness configurations in two-dimensional turbulent convection, shedding light on heat transport, flow reversals, and scaling relationships. The use of machine learning models enhances the understanding and prediction of complex turbulence behaviour.