Uncertainty-Aware Decision Support for Thermowettability in Graphene Channels: Traceable Translation under Uncertainty

Abstract

Thermowettability in graphene channels links temperature gradients and surface interactions to transport behavior, but decisions based on that relationship must cross uncertain boundaries between simulation, experiment, and device operation. This review examines probabilistic AI as a translation tool rather than a substitute for mechanism. Gaussian-process surrogates and physics-informed learning are used conceptually to connect temperature, wettability, channel geometry, flux, and selectivity while preserving prediction intervals and sensitivity to assumptions. The synthesis distinguishes interpolation within a characterized regime from extrapolation to new surfaces, scales, or water chemistries. It proposes active measurement when uncertainty is reducible, abstention when the target lies outside support, and traceable records linking each prediction to source data, preprocessing, model version, and physical boundary conditions. Decision thresholds are tied to qualification or control consequences instead of model accuracy alone. No new graphene-channel experiment is reported. The contribution is an uncertainty-aware AI framework that makes cross-domain translation inspectable and identifies which additional evidence is required before a thermowettability inference can support engineering action.