Combination of 3D FEM model and neural network for analysis of conductive and ferromagnetic spots via non linear NDT method

Ferromagnetic rebars and steel strands are widely used to reinforce civil engineering structures but are prone to corrosion-induced defects. According to the World Corrosion Organization, the annual cost estimated of corrosion worldwide is about $2.4 trillion annually, roughly 3 % of world GDP. Assessing defects is vital for determining pipeline fitness-for-service and predicting failure pressures. In this work, we develop a physicsbased 3D finite element (FEM) simulation that captures both the nonlinear ferromagnetic behavior and the effects of magnetization frequency in a U-shaped magnetizing yoke. Tangential and normal magnetic fields are computed for conductive and magnetic defects over a range of geometries, lift-offs and excitation conditions. The resulting datasets are then fed into a neural network to accurately infer defect characteristics, such as size and material properties.