Comparative Assessment of Thermal NDT Data Processing Techniques for Carbon Fiber Reinforced Polymers

The comparison and ranking of advanced data processing techniques of thermographic nondestructive testing (NDT) seems decisive for end users who are required to optimize their NDT tools. This is a difficult task, which explains why there are so few articles dealing with this challenge. Furthermore, these works are often incomplete, since the choice of metrics is too limited and/or inappropriate to an accurate, unbiased ranking. Most of them are based on the number of detected discontinuities among a series of discontinuities of various sizes and depths, and the only assessed metric is the contrast-to-noise ratio of the discontinuity signatures. This observation justifies the need for a more thorough comparison, based on the evaluation of several unbiased parameters. In this article, three metrics are considered: not only the contrast-to-noise ratio, but also the sharpness of the discontinuity edges and the accuracy of the identified characteristic dimension of the discontinuity using the contrast full-width at half-maximum. They are assessed for a trio of thermal images of a composite sample with 12 embedded calibrated discontinuities, provided by four different pulse thermographic NDT data processing techniques: thermographic signal reconstruction polynomial coefficient (PC-TSR) technique, pulse phase thermography (PPT), principal component thermography (PCT), and higher order statistics (HOS). The relative comparison of the relevance of each of these techniques is then based on 432 discontinuity signatures (4 techniques × 3 images × 12 discontinuities ×3 metrics). In order to simplify the final ranking procedure, it is also shown that this high number of data can be drastically reduced to four numbers.