What is it about?
When you squeeze a clear plastic spoon or bend a phone cover, colorful rainbow bands appear if you look through polarized glasses. Those colors come from photoelasticity—a change in how light travels through the material when it is under stress. Scientists have used this trick for a century to measure stress, but classic methods only work on thin slices or flat sheets; engineers often have to cut the object apart to see inside. Our study introduces NeST (Neural Stress Tensor Tomography), a new combination of polarized‑light imaging and artificial intelligence that maps the full 3‑D stress field inside a solid object without cutting or damaging it. We photograph the object from many directions in a special multi‑axis polariscope, then feed the images into a physics‑guided neural network that learns how every point inside must be stressed in order to create the observed color patterns. The network’s built‑in “light physics” (Jones calculus) lets it unwrap confusing fringe bands and reconstruct the complete six‑component stress tensor at each voxel. Finally, we can “virtually slice” the finished model or view it from any angle to see where dangerous stress concentrations hide.
Featured Image
Photo by Mulyadi on Unsplash
Why is it important?
Hidden residual stresses introduced during molding, 3‑D printing, or tempering are a leading cause of sudden failures in everything from shattering smartphone glass and swelling lithium‑ion batteries to cracking aerospace composites, where they initiate fracture long before external loads reach design limits. Conventional inspection tools such as hole‑drilling, X‑ray diffraction, ultrasonic testing, or thin‑slice photoelasticity either damage the part, probe only surface layers, or require expensive X‑ray/neutron facilities that few factories can afford for everyday quality control. NeST bridges this gap by fusing classic birefringence imaging with a physics‑guided neural network to recover the full six‑component 3‑D stress tensor from simple polarized‑light photographs, removing the need for destructive sectioning and correcting the phase‑wrapping errors that have limited earlier 2‑D or linear tomographic approaches. The resulting digital stress volume lets engineers virtually slice the model to pinpoint hotspots, cross‑check simulations, and iteratively refine lightweight 3‑D‑printed or injection‑molded designs before they reach customers. By turning a manual fringe‑counting lab exercise into an automated differentiable inversion pipeline, NeST paves the way for real‑time in‑line quality assurance that can prevent recalls, extend product lifetimes, and make consumer and industrial systems safer and more sustainable.
Perspectives
Working at the intersection of optics and AI, I have always dreamed of “x‑raying” stress the way CT scans reveal bones. Building NeST was a breakthrough moment: a simple set‑up with off‑the‑shelf polarizers suddenly let us peer into the mechanical soul of a plastic clip or a tempered‑glass phone screen. I hope this work inspires engineers to add polarization cameras to their toolkits and researchers to keep fusing physical laws with neural networks—the union is far more powerful than either alone.
Akshat Dave
Massachusetts Institute of Technology
Read the Original
This page is a summary of: NeST: Neural Stress Tensor Tomography by leveraging 3D Photoelasticity, ACM Transactions on Graphics, April 2025, ACM (Association for Computing Machinery),
DOI: 10.1145/3723873.
You can read the full text:
Contributors
The following have contributed to this page
