Physics-Informed Modeling of Transient Pulsating Microchannel Flow with Heat-Transfer and Entropy-Generation Trade-Off Analysis

Pulsating flow offers a practical means of modifying convective heat transfer in microchannels, but thermal enhancement may be accompanied by increased thermodynamic irreversibility. This study develops a physics-informed neural network (PINN) framework that couples transient thermo-fluid prediction with concurrent heat-transfer and entropy-generation assessment, rather than considering heat transfer alone. The continuity, momentum, and energy equations are embedded directly into the training process, with residual-based adaptive refinement used to improve solution resolution in high-residual regions. Independent verification of the steady reference solution gives relative discrepancies of 0.04% for the velocity profile against the analytical fully developed parabolic reference, 0.08% for the temperature profile against the finite-difference benchmark, and 3.51% for the downstream Nusselt number against the finite-difference benchmark. For the representative pulsating condition, the phase-resolved Nusselt number decreases from 5.0435 to 4.5413 (9.96%), while total entropy generation increases by 11.16% over the examined phases. The completed 27-case parameter sweep reveals a non-monotonic dependence on pulsation amplitude and frequency, indicating that stronger pulsation does not necessarily produce greater heat-transfer enhancement. Within the investigated parameter space, the condition A=0.2, f=20 Hz, and ϕ1=ϕ2=0.015 gives the highest mean Nusselt number of 7.1126, together with the lowest total entropy generation of 2.7496×10−4 W/K among the evaluated cases. These results demonstrate a physics-constrained framework for jointly evaluating thermal performance and thermodynamic cost, providing a more comprehensive basis for assessing pulsation conditions than heat-transfer enhancement alone.