Entropy production minimization in channel flows using computational fluid dynamics and physics-informed neural networks Available to Purchase

This study is based on two objective functions: minimizing heat transfer entropy production and minimizing viscous dissipation. The reverse-temperature equation and volumetric force sources are derived using variational methods for optimizing convective heat transfer in two-dimensional flows. Linear weights are adjusted to generate velocity and temperature fields corresponding to different performance metrics. The research demonstrates that the flow patterns determined through optimization effectively characterize optimal heat transfer performance under varying flow power consumption. Furthermore, compared to non-optimized flows, linear weights induce transitions in velocity and temperature fields from mild to highly perturbed states. Additionally, addressing the reverse-temperature equation with a negative diffusion coefficient that is challenging for traditional numerical methods, we utilize a physics-informed neural network strategy for solution. This approach significantly reduces the required grid resolution. The findings of this study can be applied to design passive techniques enhancing wall-to-fluid heat transfer and provide a novel approach for solving systems of mixed conventional and non-classical partial differential equations.