Nondestructive Localization of Subvisual Defects in Laser-Induced Graphene via Machine-Learning-Assisted Electrical Resistance Tomography

Abstract

Although conventional imaging techniques excel at capturing structural changes, they frequently overlook functional degradations that lack morphological signatures. Here, we demonstrate machine-learning-assisted electrical resistance tomography (ML-ERT) as a robust modality for the rapid, nondestructive localization of “subvisual” defects in porous laser-induced graphene (LIG). By employing masked O2 plasma irradiation, we introduced localized defects that exhibit a dramatic resistance surge up to 4 orders of magnitude while remaining indistinguishable under visual and electron microscopy. Our ML-ERT framework, powered by a one-dimensional convolutional neural network inverse solver, successfully pinpointed these hidden failures once the resistance contrast reached a threshold of R/R0 ≥ 6.71. Furthermore, 3D finite element analysis revealed that the tomographic contrast is driven by an effective conductive volume loss exceeding 30%, identifying the degradation of internal conductive pathways as the primary mechanism. These results establish ML-ERT as a high-sensitivity diagnostic tool capable of visualizing electrically critical but optically invisible failures, providing a definitive solution for the quality control of large-area carbon electronics.