Transient heat flux evolution in pool boiling on inhomogeneous surfaces: A data-driven lattice Boltzmann study

Capturing the highly nonlinear, transient temporal dynamics of nucleate boiling heat flux imposes severe computational burdens, particularly when involving complex inhomogeneous surfaces where intense liquid–vapor interactions induce chaotic thermal fluctuations. To deeply reveal these underlying phase-change mechanisms, this study establishes a coupled framework integrating the hybrid pseudopotential lattice Boltzmann method with deep learning architectures. Systematic analysis reveals that the transient surface heat flux exhibits pronounced high-frequency oscillations, which are fundamentally driven by the periodic physical cycles of vapor blanketing, bubble departure, and rapid surface rewetting. While conventional neural networks (such as multilayer perceptrons and convolutional neural networks) structurally disrupt the temporal continuity of this thermal history, the recurrent neural network (RNN) uniquely aligns with the physical periodicity of the boiling cycle. The internal memory cells of the RNN function analogously to the thermal inertia of the heating wall, successfully capturing the cyclic destruction and reconstruction of the thermal boundary layer and yielding superior predictive accuracy. Furthermore, on inhomogeneous surfaces with isosceles trapezoidal microstructures, the fundamental periodicity of boiling is physically disrupted. Intense contact line pinning and premature lateral bubble coalescence drastically prolong localized vapor blanketing, shifting the multiphase heat transfer from a stable cyclic process to a highly chaotic, nonlinear dynamic regime. The proposed framework effectively tracks these complex transitions, proving that structural micro-confinement profoundly modulates the localized thermal fluctuations.