Predicting Steel Fatigue Life from Micrographs Using Physics-Informed Deep Learning

Short summary

A computer vision framework estimates fatigue life of lightweight alloy steels directly from optical micrographs, replacing mechanical testing that takes tens to hundreds of hours. The pipeline combines OpenCV preprocessing, a 28-dimensional physics-informed feature extractor, and a CNN regression model with Gaussian negative log-likelihood loss for uncertainty quantification. ResNet-50 achieves R²=0.93 and RMSE=0.18 log-cycles on synthetic micrographs, with GNLL reducing calibration error by 76%. The pipeline runs in under 65ms per image and is open-sourced, though validation on real field samples remains future work.

• •CV framework predicts steel fatigue life from micrographs in under 65ms, replacing hours-long mechanical testing
• •ResNet-50 achieves R²=0.93 on synthetic micrographs; GNLL loss reduces calibration error by 76%
• •Open-sourced pipeline and dataset generator; real field-sample validation is the next step

Generated with AI, which can make mistakes.