PHYSICS-INFORMED NEURAL NETWORKS FOR ACCURATE SHEAR STRESS MAPPING AT CELL-SUBSTRATE INTERFACES IN ORGAN-ON-A-CHIP SYSTEMS

Emīls Šmits1, Jānis Cīmurs1, Viesturs Šints1

1 MMML lab, Department of Physics, Faculty of Science and Technology, University of Latvia

[email protected]

Accurate measurement of shear stress at the cell–substrate interface is crucial for understanding cell behavior in organ-on-a-chip (OOC) systems, where mechanical forces regulate processes such as cell alignment and differentiation. Precision of traditional methods heavily depend on the spatial resolution of Particle Image Velocimetry (PIV) measurements and are further compromised by noise inherent in PIV data, which amplifies uncertainties in local shear stress estimation [1].

In this work, we address the inverse problem of reconstructing a continuous flow field from sparse 2D PIV data using Physics-Informed Neural Networks (PINNs) [2]. By embedding the Navier–Stokes equations and incorporating a power-law formulation of the medium’s non-Newtonian rheology directly into the loss function, our approach captures the complex fluid behavior while enforcing physical constraints during neural network training. The main goal of our research is to define the cell–substrate boundary and compute local shear stress with high spatial resolution and precision.

Using well-defined simulations with known ground truth, we demonstrate that our technique provides improved boundary detection compared to conventional image processing methods and generates shear stress maps that accurately capture the medium’s detailed rheological behavior. Moreover, our PINN-based approach exhibits robustness to noise in PIV measurements, resulting in more reliable flow reconstructions. These refined maps support more informed microfluidic device design and enhance the evaluation of in vitro models for biomedical research.

Figure 1
Fig. 1. Comparison between a simulated sparse and noisy PIV dataset (left) and the PINN‐predicted continuous velocity field (right). Colors indicate velocity magnitude (in m/s)


[1] V. Šints et al., ‘Physical model of serum supplemented medium flow in organ-on-a-chip systems’, 2024, arXiv. doi: 10.48550/ARXIV.2409.13650.

[2] M. Mahmoudabadbozchelou, G. Em. Karniadakis, and S. Jamali, ‘nn-PINNs: Non-Newtonian physics-informed neural networks for complex fluid modeling’, Soft Matter, vol. 18, no. 1. Royal Society of Chemistry (RSC), pp. 172–185, 2022. doi: 10.1039/d1sm01298c.