矩形通道内棒状纳米颗粒各向异性扩散的PINNs模拟

PINNs Simulation of Anisotropic Diffusion of Rod-Shaped Nanoparticles in Rectangular Channels

摘要: 本文旨在通过使用物理信息神经网络(PINNs)分析二维矩形通道内棒状纳米颗粒的各向异性取向行为。首先研究了含有棒状纳米颗粒的纳米流体在二维矩形通道中的瞬态流动，并表征浓度和颗粒取向的耦合演化。通过使用无数据的PINNs，将控制方程作为残差嵌入损失函数中，并通过在流动区域随机采样配点在整个区域中实施物理约束。PINNs模拟准确地预测了速度、取向分布和浓度的耦合场。结果分析表明，在近壁高剪切区域，纳米颗粒由于剪切而与流动对齐，而在通道核心，布朗旋转增强了更随机的取向分布。同时，较大的旋转Péclet数削弱了布朗旋转，增强了Jeffery取向排列，使颗粒更倾向于沿流动排列；平移Péclet数控制各向异性的平移扩散，决定了跨流线迁移的强度和方向。

Abstract: This article aims to investigate the anisotropic orientation behavior of rod-shaped nanoparticles within a two-dimensional rectangular channel by employing Physical Information Neural Networks (PINNs). Initially, the transient flow characteristics of nanofluids containing rod-shaped nanoparticles in such channels were examined, and the coupled evolution of concentration and particle orientation was delineated. By utilizing data-free PINNs, the governing equations are incorporated as residuals into the loss function, and physical constraints are enforced across the entire domain through random sampling and collocation within the flow region. The PINNs simulations accurately forecast the coupled fields of velocity, orientation distribution, and concentration. The analysis of results reveals that, in the high-shear regions adjacent to the walls, nanoparticles align with the flow direction due to shear forces, whereas in the channel core, Brownian rotation promotes a more random orientation distribution. Concurrently, a higher rotational Péclet number diminishes the effect of Brownian rotation, enhances Jeffery orientation alignment, and predisposes particles to align along the flow direction; the translational Péclet number governs anisotropic translational diffusion, dictating the magnitude and orientation of cross-streamline migration.