航空科学技术2025，Vol.36Issue(8)：43-51,9.DOI:10.19452/j.issn1007-5453.2025.08.006
基于深度学习的超声波特征提取与损伤定量化评估方法
Ultrasonic Feature Extraction and Damage Quantification Method Based on Deep Learning
摘要
Abstract
Carbon Fibre Reinforced Polymer(CFRP)is widely used in the aerospace field due to its characteristics of being lightweight and high-strength.However,CFRP exhibits weak impact resistance,and thus it is necessary to conduct quantitative nondestructive testing of its impact damage.Traditional detection methods have low accuracy in identifying CFRP impact damage.By utilizing the ultrasonic C-scan method in combination with artificial intelligence algorithms,the accuracy of CFRP damage identification can be significantly improved.Ultrasonic Amplitude(AMP)images were obtained through water-immersion ultrasonic experiments,and U-Net convolutional neural network framework was established to achieve quantitative identification of in-plane damage in CFRP.1D Convolutional Neural Network(1D-CNN)was used to learn the relationship between ultrasonic A-scan signals and damage depth,and 2D Convolutional Neural Network(2D-CNN)was built to correlate the time-frequency features of the A-scan signals with the damage location,enabling the evaluation of out-of-plane damage in CFRP.Finally,K-means clustering algorithm was proposed to cluster the Time of Flight(TOF)of ultrasonic signals,realizing quantitative tomographic research on CFRP damage.The results show that for in-plane damage evaluation of CFRP,the Intersection over Union(IoU),pixel accuracy,precision,and recall of the U-Net model on the test set were 91.67%,94.8%,96.15%,and 94.83%,respectively.For out-of-plane damage evaluation of CFRP,1D-CNN and 2D-CNN models could classify the location of internal CFRP damage,but could not effectively avoid the influence of spatial experimental noise.In contrast,the TOF clustering method based on the K-means clustering algorithm could realize layer-by-layer quantitative evaluation of internal damage in CFRP without the need for sample labels.A multi-dimensional and precise method for identifying CFRP impact damage is provided through the combination of ultrasonic technology and artificial intelligence algorithms.The development of intelligent nondestructive testing technology for aerospace structures is also promoted.关键词
复合材料/无损检测/深度学习/超声波/K均值Key words
composite material/non-destructive testing/deep learning/ultrasonic/K-means分类
航空航天引用本文复制引用
乙明国,赵金玲,常乐,张超,程经纬..基于深度学习的超声波特征提取与损伤定量化评估方法[J].航空科学技术,2025,36(8):43-51,9.基金项目
航空科学基金(2023Z061052001) （2023Z061052001）
江苏省研究生科研与实践创新计划项目(JXSS-040) Aeronautical Science Foundation of China(2023Z061052001) （JXSS-040）
Postgraduate Research&Practice Innovation Program of Jiangsu Province(JXSS-040) （JXSS-040）
