Automated identification of front/rear surface cracks in ferromagnetic metals based on eddy current pulsed thermography

Abstract

Surface cracks may significantly degrade performance of metal components and may lead to failure of constructions. The present study develops a method for automated identification of front/rear surface cracks in ferromagnetic metals. The method employs eddy current pulsed thermography (ECPT) to induce eddy currents and Joule heat in the inspected metal parts. Front/rear surface cracks interfere both eddy current distribution and the following heat conduction, which results in abnormal thermal response on the metal surface. A local region-based strategy is developed to identify these crack regions from the ECPT inspection data through a supervised classification procedure. Due to this strategy, statistical features of local regions in the inspected surface are extracted and are used as the classification features to accomplish crack recognition. A multi-class support vector machine (SVM) classifier is constructed to segment the local regions into crack regions, sound regions and insufficiently-heated regions. To evaluate the proposed approach, experiments were implemented on steel plates with prefabricated notches of various sizes. Both front surface cracks and rear surface cracks were involved in these experiments. Experimental results verify the effectiveness of the approach for automated identification of front/rear surface cracks in metal parts.