Laser ultrasonic damage identification of composites based on empirical mode decomposition and neural network

In order to ensure the safety and reliability of equipment during service, the implementation of non-destructive testing (NDT) plays a crucial role. Laser ultrasonic technology (LUT) offers several advantages over traditional ultrasonic technology, including non-contact operation, high frequency, wide frequency band. The utilization of LUT to evaluate the condition of composite materials has emerged as a significant approach to ensure safety. Nevertheless, the unsatisfactory surface optical quality of composite materials often results in indistinct imaging outcomes. Moreover, human factors frequently introduce unavoidable biases in the interpretation of the imaging results. As a solution, a neural network-based method for damage detection using LUT has been proposed. Initially, the empirical mode decomposition (EMD) is employed to decompose the signal and select the relevant intrinsic mode functions (IMFs) for signal reconstruction, which yields time-domain features. Subsequently, these features are inputted into the neural network to recognize the state of the material, and the recognition outcomes are incorporated into the pixel matrix. Lastly, C-scan images are created by stacking the pixel matrix and applying a threshold to reduce noise. The results demonstrate that the accuracy of delamination detection reaches 98.5%. Compared to conventional methods, the C-scan images obtained through neural network are more precise and easier to interpret, thereby effectively minimizing misinterpretations caused by manual intervention.