Terahertz Nondestructive Testing Signal Recognition Based on PSO-BP Neural Network

Terahertz time domain spectroscopy technique was used to detect the defects of high temperature resistant composite materials with multi-bonded structures. In order to identify the debonding defects in both upper and lower layers at the same location, the terahertz signal waveforms in the non-defect area, upper and lower debonding areas were analyzed. The characteristic peak-to-peak, skewness, minimum value, peak-to-valley value, waveform factor and absolute mean value of signal amplitude were taken as the input of BP neural network.The initial weight and threshold value of BP neural network were optimized by Particle Swarm Optimization (PSO), which solved the problem that BP neural network was easy to fall into local optimum. The optimized PSO-BP neural network could realize the identification of the debonding defects of upper 100 mu m and lower 100 mu m, with the accuracy of 90.71% and 86.92%.