Feature extraction for robust impact damage classification of CRFP plates using ultrasonic signals

In the context of ultrasonic non-destructive evaluation, we have developed a classification system oriented to damaged material testing, in particular, carbon fiber-reinforced composites which accurately identifies the severity of flaws. The proposed methodology consists of two main elements. Firstly, a sparse digital signal model that idealizes the ultrasound-material interactions is used to alternatively provide cepstral or discrete cosine transform features with a high discriminative power. Then, we can classify the signals according to the damage level of the considered specimen. In particular, we analyze the benefits of different feature reduction and classification techniques over the proposed system. According to the results, the proposed approach further enhances the classification performance while reduces the computational burden.