Abstract:Photoelasticity enables full-field stress analysis in transparent objects through stress-induced birefringence. Existing techniques are limited to 2D slices and require destructively slicing the object. Recovering the internal 3D stress distribution of the entire object is challenging as it involves solving a tensor tomography problem and handling phase wrapping ambiguities. We introduce NeST, an analysis-by-synthesis approach for reconstructing 3D stress tensor fields as neural implicit representations from polarization measurements. Our key insight is to jointly handle phase unwrapping and tensor tomography using a differentiable forward model based on Jones calculus. Our non-linear model faithfully matches real captures, unlike prior linear approximations. We develop an experimental multi-axis polariscope setup to capture 3D photoelasticity and experimentally demonstrate that NeST reconstructs the internal stress distribution for objects with varying shape and force conditions. Additionally, we showcase novel applications in stress analysis, such as visualizing photoelastic fringes by virtually slicing the object and viewing photoelastic fringes from unseen viewpoints. NeST paves the way for scalable non-destructive 3D photoelastic analysis.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **How to non - destructively reconstruct the three - dimensional stress distribution inside transparent objects from 3D photoelastic measurements?** Specifically, traditional 3D photoelastic techniques usually require slicing the object to obtain 2D stress distributions. This method is both expensive and time - consuming and is also destructive. However, this paper proposes a new method - NeST (Neural Stress Tensor Tomography), which uses a multi - axis polarimeter setup and neural implicit representations to directly reconstruct the complete 3D stress tensor field from the captured polarization measurement data. Through this method, high - precision, non - destructive reconstruction of the internal stress distribution can be achieved without destroying the object, and photoelastic fringes from different viewing angles can be visualized. ### Main problems 1. **Limitations of existing 3D photoelastic techniques**: - Traditional methods rely on physical slicing, which is not only expensive and time - consuming but also destructive. - Existing techniques face challenges in dealing with phase wrapping and tensor tomography, resulting in less accurate reconstruction results. 2. **How to efficiently and accurately reconstruct the 3D stress tensor field**: - This paper proposes a differentiable non - linear forward model, combining the Jones matrix and neural implicit representations, to solve the problems of phase wrapping and tensor tomography. - Through the multi - axis polarimeter experimental setup, the effectiveness of this method is verified, and its applications under various shapes and loading conditions are demonstrated. ### Solutions - **NeST method**: Use neural implicit representations and analysis - synthesis methods to reconstruct the internal stress distribution from the captured polarization measurement data. - **Multi - axis polarimeter setup**: By rotating the sample and polarizing elements, obtain polarization measurement data from multiple viewing angles to ensure sufficient measurement diversity. - **Differentiable forward model**: A non - linear model based on the Jones matrix can more accurately interpret the actually captured data and has higher accuracy compared to the linear approximation model. ### Applications and contributions - Propose a differentiable non - linear forward model that can faithfully match the captured polarization measurement data. - Develop an analysis - synthesis method based on neural implicit representations to directly reconstruct the internal stress distribution from the captured polarization measurement data. - Verify the proposed stress tensor tomography method through the multi - axis polarimeter experimental setup. - Demonstrate the generation of novel stress visualizations from the learned neural stress fields, such as virtual slices and viewing photoelastic fringes from unseen viewing angles. In summary, this paper aims to overcome the limitations of existing 3D photoelastic techniques by introducing the NeST method and provide an efficient, accurate, and non - destructive 3D stress reconstruction method.

NeST: Neural Stress Tensor Tomography by leveraging 3D Photoelasticity

Photoelastic stress response of complex 3D-printed particle shapes

High-speed photoelastic tomography for axisymmetric stress fields in a soft material: Temporal evolution of all stress components

Photoelastic Stress Field Recovery Using Deep Convolutional Neural Network

Spatially resolved stress measurements in materials with polarization-sensitive optical coherence tomography: image acquisition and processing aspects

Full stress tensor measurement using colour centres in diamond

Advanced Ground Truth Multimodal Imaging Using Time Reversal (TR) Based Nonlinear Elastic Wave Spectroscopy (NEWS): Medical Imaging Trends Versus Non-destructive Testing Applications

POLARIZED SPECKLE-PHOTOELASTICITY FOR 3-D STRESS ANALYSIS

Integrated photoelasticity in a soft material: phase retardation, azimuthal angle and stress-optic coefficient

Energy Resolved Neutron Imaging for Strain Reconstruction using the Finite Element Method

Direct inversion of the Longitudinal Ray Transform for 2D residual elastic strain fields

Elasticity imaging using physics-informed neural networks: Spatial discovery of elastic modulus and Poisson's ratio

Non-line-of-Sight Imaging via Neural Transient Fields

Cartesian Neural Network Constitutive Models for Data-driven Elasticity Imaging

Application of Neural Networks in Rock Mass Stress Assessment by Photoelasticity

NeRSP: Neural 3D Reconstruction for Reflective Objects with Sparse Polarized Images

Three-dimensional polarimetric ptychography

NeRS: Neural Reflectance Surfaces for Sparse-view 3D Reconstruction in the Wild

Physics‐Informed Deep‐Learning For Elasticity: Forward, Inverse, and Mixed Problems