Tensor-based ultrasonic data analysis for defect detection in fiber reinforced polymer (FRP) composites

Non-destructive testing (NDT) is an important tool for defect detection in composite materials. Compared to other NDT methods, ultrasonic testing (UT) has the principal advantages of high penetrating power and high detection sensitivity. To better identify the locations and depths of defective regions, various ultrasonic signal processing methods have been adopted to enhance defect signals. However, most of the existing methods cannot deal with the entire third-order tensor of UT data in an efficient manner. In order to solve this problem, a tensor based ultrasonic data analysis method is proposed based on Tucker3 decomposition. After decomposition, the defect information is extracted by a small number of factors, which is further summarized by three leverage vectors. The candidate defective regions are then identified from the leverages in the second and third modes, facilitating the following clustering step for finding the locations and the shapes of defects. Moreover, the defect depths are estimated from the peaks in the leverages in the first mode. The proposed method was applied to detecting defects in fiber reinforced polymer (FRP) composites. The experimental results illustrated the feasibility of the proposed method.