Spectroscopy and Spectral Analysis, Volume. 45, Issue 2, 542(2025)
Infrared Thermography Detection of GFRP/NOMEX Honeycomb Sandwich Structure Defects Based on Convolutional Neural Network
Honeycomb sandwich structure is one of the important structural forms in the field of composite materials. Due to its complex preparation process and harsh service environment, it is easy to produce delamination, debonding, and other defects, which seriously affect the service life of materials. To ensure the performance and quality safety of related components, it is necessary to carry out regular quality monitoring and flaw detection of the honeycomb sandwich structure through appropriate non-destructive testing technology. Therefore, quantitatively detecting defects is the fundamental way to prevent and solve such problems. Based on infrared thermal imaging technology to GFRP/NOMEX honeycomb sandwich structure specimens containing prefabricated delamination and debonding defects as the object of study for pulsed infrared thermal wave nondestructive testing experimental research, the acquisition of several frames of the specimen surface temperature distribution thermograms, take several defective areas and the healthy region of the pixel temperature signals to construct a sample dataset, and randomly divided into a training set and a validation set, take the fourth row of defective The center horizontal line area is taken as the test set data. Combined with convolutional neural network technology, GFRP/NOMEX honeycomb sandwich structure defect detection and depth prediction is completed. Analyze the -one-dimensional convolutional neural network structure, introduce multi-scale dilated convolution, residual module, and attention mechanism to build a one-dimensional convolutional neural network prediction model, and use the constructed temperature signal data set to train the network model. The training results show that the Loss and RMSE trends of the validation and training sets are consistent. The final Loss of the validation set is 1.67×10-5, the RMSE is 0.005 8, and there is no over-fitting problem. The test set data is input into the trained network. The results show that the constructed network can effectively identify the defects and the depth prediction error at the defect center is controlled within 2%. It can be seen that the combination of convolutional neural network and infrared thermal imaging detection technology can realize the reliable detection of GFRP/NOMEX honeycomb sandwich structure defects and the stable prediction of defect burial depth and also provide a reference for the identification and quantitative detection of defects in other composite materials.
Get Citation
Copy Citation Text
TANG Qing-ju, GU Zhuo-yan, BU Hong-ru, XU Gui-peng, TAN Xin-jie, XIE Rui. Infrared Thermography Detection of GFRP/NOMEX Honeycomb Sandwich Structure Defects Based on Convolutional Neural Network[J]. Spectroscopy and Spectral Analysis, 2025, 45(2): 542
Received: Feb. 24, 2024
Accepted: May. 21, 2025
Published Online: May. 21, 2025
The Author Email: