Reconstruction of three-dimensional fluid stress field via photoelasticity using physics-informed convolutional encoder-decoder
Daichi Igarashi, Shunsuke Kumagai, Yuto Yokoyama, Yee Jingzu, Masanobu Horie, Yoshiyuki Tagawa · 2025
Read Reconstruction of three-dimensional fluid stress field via photoelasticity using physics-informed convolutional encoder-decoder by Daichi Igarashi, Shunsuke Kumagai, Yuto Yokoyama, Yee Jingzu, Masanobu Horie, Yoshiyuki Tagawa (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-15952.txt.
Introduction
Daichi Igarashi and five co-authors rebuild 3D fluid stress fields via photoelasticity in two thousand twenty-five. Stress maps matter in engineering, medicine, and bioengineering alike. Classic routes infer stress from velocity, but fail for complex fluids. Their constitutive laws are unknown in those hard cases.
Photoelastic reads show stress inside a flow without touch. A model named PICED learns to turn those reads into full fields. It joins a convolutional encoder decoder with physics informed learning. Tests on a rectangular channel flow hit high accuracy. Many interpolated points in the volume are predicted at once.
Catalog record
| CREATOR | Daichi Igarashi, Shunsuke Kumagai, Yuto Yokoyama, Yee Jingzu, Masanobu Horie, Yoshiyuki Tagawa |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-15952
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Reconstruction of three-dimensional fluid stress field via photoelasticity using physics-informed convolutional encoder-decoder Daichi Igarashi, Shunsuke Kumagai, Yuto Yokoyama, Yee Jingzu, Masanobu Horie, Yoshiyuki Tagawa Abstract Measuring stress fields in fluids and soft materials is crucial in various fields such as mechanical engineering, medicine, and bioengineering. However, conventional methods that calculate stress fields from velocity fields struggle to measure complex fluids where the stress constitutive equation is unknown. To address this, we propose a novel approach that combines photoelastic measurements—which can non-invasively visualize internal stresses—with machine learning to measure stress fields. The machine learning model, which we named physics-informed convolutional encoder-decoder (PICED), integrates a convolutional neural network (CNN)-based encoder-decoder model with a physics-informed neural network (PINN). Using this approach, three-dimensional stress fields can be predicted with high accuracy for multiple interpolated data points in a rectangular channel flow. Keywords— machine learning, data-driven modeling, physics-informed neural networks, photoelasticity, 1 Introduction The measurement of the nonlinear dynamics of fluids and soft materials has wide-ranging applications in various fields, such as mechanical engineering, medicine, and bioengineering. Nonlinear dynamics involve complex relationships between deformation and stress, and stress measurement provides insights into phenomena such as cerebral aneurysm formation [ 1 , 2 ] , cellular mechanical responses [ 3 , 4 ] , and the mechanical basis of biochemical changes in cancer cells [ 5 , 6 ] . In engineering, understanding stress distribution is critical for advancing needle-free injectors [ 7 , 8 ] and improving the structural analysis of soft robots [ 9 , 10 ] . In fluid dynamic interactions such as those described above, the stress field is typically derived from velocity measurements using a constitutive equation or pressure sensors. However, these methods face limitations, particularly for complex fluids lacking known constitutive equations or in situations when it is crucial to avoid disturbances to the flow. To address such limitations, this study focuses on photoelastic measurement, a non-contact method for directly measuring stress fields within materials. Photoelastic measurement is an optical technique that utilizes the property of transparent materials, where the refractive index changes when stress is applied [ 11 , 12 , 13 ] . While this technique has been widely applied to solid materials, such as glass, recent studies have extended its use to fluids and soft materials [ 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 ] . When using a photoelastic fluid containing high-aspect-ratio particles, the particles align in the direction of stress and exhibit an optical phase difference \Delta , which corresponds to the magnitude of the stress. This phase difference and particle orientation \phi are related to the difference and direction of the principal stresses [ 22 , 23 ] , enabling the direct, non-contact measurement of stress fields in fluids. To determine the actual stress field in a fluid, it is necessary to reconstruct it from the photoelastic parameters (phase difference and orientation), which are obtained via a polarization camera. A challenge arises because stress is a tensor with up to six independent components at each point, while the photoelastic parameters provide only two values, which are integrated along the optical axis. This discrepancy makes reconstruction difficult. Previous photoelastic measurements have assumed axial symmetry to simplify the process by reducing the number of stress components [ 24 , 25 ] . Although several studies have calculated phase difference and orientation from a 3D stress field [ 26 , 27 , 28 ] , to the authors’ knowledge, no research has reconstructed a 3D fluid stress field from these parameters without assuming axial symmetry. As mentioned above, reconstructing the 3D stress tensor field in a fluid from phase difference and orientation data requires complex matrix calculations; however, this involves a nonlinear problem with no theoretical solution. Therefore, the present study focuses on machine learning, which can handle nonlinear problems. In recent years, machine learning has been integrated with fluid mechanics, especially in physics-informed neural networks (PINNs) [ 29 , 30 , 31 ] , which satisfy the governing physical equations of fluid mechanics. These approaches have been used to predict pressure distributions [ 32 ] and fluid behavior in complex fluids [ 33 ] , but none have focused on predicting stress fields. Some studies have also enhanced model generalization by incorporating numerical differentiation [ 34 , 35 , 36 ] . This study’s goal is to reconstruct 3D stress fields from 2D phase difference and orientation images using a machine learning model called physics-informed convolutional encoder-decoder (PICED), which combines a CNN-based encoder-decoder and a PINN. To verify the model before applying it to more complex flow paths, this study focuses on a simpler system—a rectangular channel. Section 2 explains the machine learning architecture and the method for obtaining the data used for the input and output images. Section 3 describes the prediction results of the three-dimensional stress field, the principal stress distribution, and the element analysis. Finally, in Section 4, the conclusions of this study are presented, and future prospects are discussed. 2 Methodology This section describes the method for obtaining 3D stress distributions from integrated images of phase difference and orientation, which are acquired through polarization measurements. In this study, we reconstructed the stress field for a Newtonian fluid in laminar flow within a rectangular channel. The input images were obtained using the experimental methods described in Sections 2.1.1 and 2.1.2, while the output images were derived from the theoretical solutions detailed in Section 2.2. Finally, Section 2.3 illustrates the structure of the machine learning model—specifically, a convolutional neural network (CNN) which is generally used and the PICED method developed in this study—and provides a comprehensive explanation of how the three-dimensional stress field is obtained. 2.1 Input: Two-dimensional integrated images of phase difference and orientation This section explains the methodology for acquiring input images from experiments. 2.1.1 Measurement principle The optical system is illustrated in Figure 1 . A photoelastic material is positioned along the optical axis between a polarization camera and a circularly polarized light source. The refractive index varies according to the direction of polarization when the photoelastic material is subjected to stress. Since the speed of light is inversely proportional to the refractive index, an optical phase difference occurs when light passes through two orthogonal polarizers [ 37 ] . This phenomenon, whereby two refracted rays appear when light passes through a material, is known as birefringence. Optically anisotropic particles, such as cellulose nanocrystal (CNC), exhibit birefringent properties [ 20 , 21 , 15 ] . These particles are randomly oriented when no stress is applied but are aligned in a uniform direction under stress. However, since the distortion caused by stress is an elastic deformation, birefringence disappears once the external force is removed. Figure 1: Schematic drawing of the theory of the photoelastic method. The monochromatic light generated by the light source is converted to circularly polarized light by passing it through a linear polarizer and a quarter-wave plate. Through circular polarization, the light exhibits uniform vibrations in all directions perpendicular to its propagation. The transmitted light—oriented because of the birefringence of the photoelastic object—passes through a linear polarizer positioned between the camera and the photoelastic object. This linear polarizer is rotated