Control and Instrumentation Engineering

To implement a magneto-optic (MO) nondestructive inspection (MONDI) system for robot-based nondestructive inspections, quantitative evaluations of the presence, locations, shapes, and sizes of defects are required. This capability is essential for training autonomous nondestructive testing (NDT) devices to track material defects and evaluate their severity. This study aimed to support robotic assessment using the MONDI system by providing a deep learning algorithm to classify defect shapes from MO images. A dataset from 11 specimens with 72 magnetizer directions and 6 current variations was examined. A total of 4752 phenomena were captured using an MO sensor with a 0.6 mT magnetic field saturation and a 2 MP CMOS camera as the imager. A transfer learning method for a deep convolutional neural network (CNN) was adapted to classify defect shapes using five pretrained architectures. A multiclassifier technique using an ensemble and majority voting model was also trained to provide predictions for comparison. The ensemble model achieves the highest testing accuracy of 98.21% with an area under the curve (AUC) of 99.08% and a weighted F1 score of 0.982. The defect extraction dataset also indicates auspicious results by increasing the training time by up to 21%, which is beneficial for actual industrial inspections when considering fast and complex engineering systems.

Grain boundaries (GBs), which are among the mechanical properties of a material, are a microstructural aspect that contributes to the overall behavior of metal. A deep understanding of the behavior of the GBs' deformation, dislocation, and fracture will encourage the rapid development of new materials and lead to the better operation and maintenance of materials during their designed lifetimes. In this study, an integrated image processing toolset is proposed to provide an expeditious approach to extracting GBs, tracking their location, and identifying their internal deformation. This toolset consists of three integrated algorithms: image stitching, grain matching, and boundary extraction. The algorithms are designed to simultaneously integrate high and low spatial resolution images for gathering high-precision boundary coordinates and effectively reconstructing a view of the entire material surface for the tracing of the grain location. This significantly reduces the time needed to acquire the dataset owing to the ability of the low spatial resolution lens to capture wider areas as the base image. The high spatial resolution lens compensates for any weakness of the base image by capturing views of specific sections, thereby increasing the observation flexibility. One application successfully described in this paper is tracking the direction of the metal grain deformation in global coordinates by stacking a specific grain before and after the deformation. This allows observers to calculate the direction of the grain deformation by comparing the overlapping areas after the material experiences a load. Ultimately, this toolset is expected to lead to further applications in terms of observing fascinating phenomena in materials science and engineering.

Recent development of electromagnetic inspection proposed auspicious results for imaging 2-D static of leakage magnetic flux density (LMFD) on surface and sub-surface defects. This ability encourages an interest in implementing electromagnetic-based magnetic flux leakage (MFL) testing for nondestructive inspection (NDI). In recent NDI technologies, advanced magnetic inspection using magneto-optical Faraday effect (MOE) and magnetic camera dominated the pre-service and in-service inspection in critical engineering industries such as aerospace, steel manufacturing, marine, railroad, and nuclear facilities. Unfortunately, a compact comparison of these approaches has not been proposed yet. This study aimed to compare MOE and magnetic camera according to five primary analyses: the target object, magnetic source, configuration of the magnetic sensor, signal processing, and AI implementation. A literature study was conducted using 47 papers to analyze the performance of each device in capturing LMFD. This comparison study successfully obtained the advantages and limitations of each imaging technique, which is very beneficial as a compact reference for developing and selecting suitable nondestructive inspection devices for MFL testing in the future. In addition, artificial intelligence (AI) as an option to escalate the inspection confidence could increase the assessment efficiency and reliability considering fast and complex engineering systems.

In this study, theoretical models were proposed to explain the changes in self-magnetic flux density (SMFD) due to fatigue cracks in the presence and absence of external magnetic fields. Three theoretical models were proposed: rotation domain model (RDM), concentration domain model (CDM), and vertical domain model (VDM), considering the deformation and non-deformation possibilities. To prove the theoretical model, fatigue cracks with different depth angles were fabricated through fatigue testing and EDM processing on the CT specimens. In addition, tunnel magnetoresistance (TMR) sensors were used to evaluate the 3-axis distribution of SMFD. Comparing the simulation and experimental results, similar tendencies of the occurrence and depth angle of fatigue cracks and their effect on the distribution of SMFD were observed. According to the RDM, the distribution of SMFD occurs in the direction of the crack length (y-direction), while the CDM explains that the SMFD does not occur if the fatigue crack is in a direction perpendicular to the surface. In addition, the VDM shows that SMFDs occur in a direction perpendicular to the crack length (x-direction) and the specimen surface (z-direction). Interestingly, these trends agree with the experimental results, which confirms the validity of the theoretical model and thus can be used to estimate the depth direction of a fatigue crack