Physics-Informed Deep Learning Predicts Steel Fatigue Life from Micrographs
A new computer vision framework, CV, estimates the fatigue life of lightweight alloy steels directly from optical micrographs, bypassing traditional mechanical testing that can take tens to hundreds of hours. The method, detailed in a preprint on arXiv (2607.28695), uses a seven-stage OpenCV preprocessing routine to remove artifacts, followed by a 28-dimensional physics-informed feature extractor that quantifies crack morphology, grain structure, porosity, and texture. A CNN regression model, trained with a Gaussian negative log-likelihood (GNLL) loss, jointly predicts the fatigue life (log N_f) and sample-specific uncertainty. The study evaluates three architectures, including SE-CNN, to determine the most effective model. This approach could enable rapid quality control in steel manufacturing, significantly reducing the time and cost associated with fatigue testing.
Key facts
- The framework is named CV.
- It estimates fatigue life (log N_f) of lightweight alloy steels.
- It uses optical micrographs instead of physical testing.
- The pipeline includes a seven-stage OpenCV preprocessing routine.
- A 28-dimensional physics-informed feature extractor is used.
- The CNN regression model uses Gaussian negative log-likelihood (GNLL) loss.
- It predicts sample-specific uncertainty (sigma hat).
- Three architectures are evaluated, including SE-CNN.
- The research is available on arXiv with ID 2607.28695.
Entities
—