Physics-informed CNN estimates steel fatigue life from micrographs

A new computer vision framework uses a 28-dimensional physics-informed feature extractor and CNN regression to predict steel fatigue life from optical micrographs. The model uses Gaussian negative log-likelihood loss to provide both fatigue life estimates and sample-specific uncertainty.