Deep CNNs and Adversarial Regularization for Fatigue Damage Failure Prediction of Concrete Anchors

Abstract

Fatigue experiments present with large scatter even for identical specimens tested under controlled laboratory conditions. It has long been known that variations in the mechanical behavior of fatigued metals occur, such as hardening or softening and variations on the shape of hysteresis curves. In order to incorporate measurements in the fatigue life prediction of concrete anchors, a fully data-driven model, using load-displacement and acceleration data, was trained to predict directly the remaining cycles to failure for anchors embedded in cracked concrete loaded in variable amplitude fatigue.