Carbon fiber-reinforced polymer (CFRP) laminates are widely used in aerospace, automotive, and infrastructure industries due to their high strength-to-weight ratio. However, defect detection in CFRP remains challenging, particularly in low signal-to-noise ratio (SNR) conditions. Conventional segmentation methods often struggle with noise interference and signal variations, leading to reduced detection accuracy. In this study, we evaluate the impact of thermal image preprocessing on improving defect segmentation in CFRP laminates inspected via pulsed thermography. Polynomial approximations and first- and second-order derivatives were applied to refine thermographic signals, enhancing defect visibility and SNR. The U-Net architecture was used to assess segmentation performance on datasets with and without preprocessing. The results demonstrated that preprocessing significantly improved defect detection, achieving an Intersection over Union (IoU) of 95% and an F1-Score of 99%, outperforming approaches without preprocessing. These findings emphasize the importance of preprocessing in enhancing segmentation accuracy and reliability, highlighting its potential for advancing non-destructive testing techniques across various industries.

This study evaluates the effectiveness of thermographic signal reconstruction using polynomial approximations for preprocessing thermographic images obtained through Pulsed Thermography in the segmentation of defects in CFRP laminates. Neural networks were trained with both processed and unprocessed images and evaluated using IoU and F1-Score metrics. The preprocessing significantly increased defects' signal-to-noise ratio and detection accuracy, achieving an IoU of 92% and an F1-Score of 98%. The technique proved effective in enhancing defect segmentation and definition, particularly in scenarios with low signal-to-noise ratio.

Steel structures undergo loads and stresses during service life, subject to structural damages such as fatigue, corrosion, cracks, and plastic deformations. Therefore, to detect damage, the dynamic responses of the structures are used, comparing two states: with and without damage. These dynamic responses are obtained from a signal representing the structure's electromechanical impedance. Thus, these impedance signatures must accurately represent the analyzed structure. By comparing the impedance signatures of the low-cost device used at UFG/UFCAT with the SySHM system developed by LMEst/UFU, it can be observed that the low-cost equipment requires calibration in its impedance measurements. This work proposes a method based on the Least-Squares approach to determine a mathematical model to convert the signals acquired by the low-cost device into signals suitable for analysis. In conclusion, it was feasible to demonstrate the utilization potential of the cost-effective device under impedance-based monitoring conditions.

Electromechanical impedance-based structural health monitoring has been the subject of extensive research in recent decades. The method’s low cost and ability to detect minor structural damages make it an appealing alternative to other non-destructive techniques. Ongoing research on damage detection approaches continues to be a topic of interest in relation to the electromechanical impedance method. This work proposes the use of the K-Means, Decision Tree, and Random Forest algorithms to distinguish between four structural conditions in an aluminum beam. These techniques were applied to raw impedance data and a dataset reduced via principal components analysis. The findings revealed that the compressed dataset improved the accuracy of all models, except for the Random Forest approach, whose accuracy decreased by 2.9%. The K-Means algorithm was most affected by the reduction in dimensionality, with a 105.9% increase in accuracy. The Decision Tree and Random Forest methods yielded outstanding outcomes, comparable or superior to other state-of-the-art approaches. This makes them a compelling choice for detecting damage using electromechanical impedance data, even when using raw data as the input information.