Automatic defect detection in fiber-reinforced polymer matrix composites using thermographic vision data

The detection of internal defects, not visible to the naked eye from the outside of materials, using non-destructive testing (NDT) are increasingly requested by industrial processes. This study proposes a novel methodology for acquisition and processing of images from a thermographic camera using computer vision methods to test composite materials made of a polymer matrix reinforced with glass, carbon, and kevlar fibers. The image is acquired while cooling the sample, following a suggested procedure. The processing methodology is divided into three steps, image pre-processing, image processing, and data post-processing. In image preprocessing, filters are applied to improve image quality, and methods are proposed to segment and identify the region of interest. In image processing, a blob analysis method is suggested for defect identification, isolation and characterization. A data analysis method is proposed for the post-processing step to characterize the defects identified in the previous step. Samples with known defects in terms of size, geometry, and location were used to test the developed system. The system showed high performance, achieving 98% accuracy, and suitability for defect detection larger than 0.5 mm in thickness and 600 mm2 in area. The experimental results showed that the algorithm did not detect any false positives, and that the type of reinforcement used in the analyzed samples had no influence on the results. On the other hand, the depth of the delaminations had an influence on the pixel intensity contrast of the defect region, and its instant of maximum contrast. The lesser the depth of the defects detected, the higher the value of their intensity and the shorter the instant of maximum contrast.