Physics-Guided Deep Learning for Plate Permeability Estimation With Single to Multiple Frequency Transformation of Eddy-Current Testing
Zihan Xia, Tian Meng, Ruochen Huang, Adam Fletcher, Yuchun Shao, Mingyang Lu, Anthony Peyton, Wuliang Yin
Abstract
Eddy-current testing technique has been extensively explored for estimating the electromagnetic property of steel plates in various industrial applications. In this article, a physics-guided deep learning (DL) method is proposed to estimate the permeability of plate in high thickness with probe liftoff. A simplified analytical model is derived, which realises the single to multiple frequency inductance transformation and calculates the related physical properties of the measurement configuration. A constant is found, which is a fundamental coefficient describing the first-order nature of the sensor response to a plate and it is insensitive to plate properties and probe dimensions. The nonlinear mapping from physical information, derived from the simplified analytical model, to plate permeability is constructed by the DL model based on the modified ResNet18-1D. Numerical simulations and experiments have been performed to evaluate the proposed method for permeability estimation with various plate materials and probe liftoff. The method achieves real-time accurate estimation of plate permeability with a relative error lower than 3%.