TY - JOUR
T1 - Characterizing drilling-induced damage in CFRP/Ti stacks based on Unsupervised Pixel-level Image Segmentation Labelling Method and Attention Assisted U-Net
T2 - A novel architecture: Characterizing the Damage in CFRP/Ti via UPISL and AU-Net
AU - YAO, Hang
AU - ZHANG, Kaifu
AU - LUO, Bin
AU - CAO, Sipeng
AU - LI, Haolian
AU - LIU, Chenyu
N1 - Publisher Copyright:
© 2025 The Author(s)
PY - 2026/11
Y1 - 2026/11
N2 - Drilling-induced damage in Carbon Fiber Reinforced Polymer/Titanium (CFRP/Ti) stacks exhibits complex characteristics due to thermo-mechanical coupling effects, severely compromising mechanical properties. Recognizing and characterizing the damage with high accuracy and efficiency play a vital role for further optimizing the process parameters. However, the undamaged CFRP regions exhibit similar features compared to the drilling-induced damage, which decreases the accuracy and reliability of identifying the target damage. This paper proposes a novel architecture for characterizing the drilling-induced damage in CFRP/Ti based on industrial Computed Tomography (CT). The architecture consists of an Unsupervised Pixel-level Image Segmentation Labelling Method (UPISL) and Attention Assisted U-Net (AAU-Net) to segment drilling-induced damage. Based on the experiment of drilling CFRP/Ti with typical parameters, 200 CT images are scanned along the axial direction to establish the recognizing model. The average precision of the proposed model reaches 0.80. The segmented results are evaluated and validated by experimental analysis of the drilling thermo-mechanical effects. The damage formation mechanism is revealed in a further step by Scanning Electron Microscope (SEM). The precision score of the proposed method turns out to have a better performance on recognizing the drilling-induced damage than the traditional semantic segmentation models such as E-Net, U-Net, VGG16, and the residual U-Net with 0.63, 1.88, 1.90 and 3.17 times higher respectively.
AB - Drilling-induced damage in Carbon Fiber Reinforced Polymer/Titanium (CFRP/Ti) stacks exhibits complex characteristics due to thermo-mechanical coupling effects, severely compromising mechanical properties. Recognizing and characterizing the damage with high accuracy and efficiency play a vital role for further optimizing the process parameters. However, the undamaged CFRP regions exhibit similar features compared to the drilling-induced damage, which decreases the accuracy and reliability of identifying the target damage. This paper proposes a novel architecture for characterizing the drilling-induced damage in CFRP/Ti based on industrial Computed Tomography (CT). The architecture consists of an Unsupervised Pixel-level Image Segmentation Labelling Method (UPISL) and Attention Assisted U-Net (AAU-Net) to segment drilling-induced damage. Based on the experiment of drilling CFRP/Ti with typical parameters, 200 CT images are scanned along the axial direction to establish the recognizing model. The average precision of the proposed model reaches 0.80. The segmented results are evaluated and validated by experimental analysis of the drilling thermo-mechanical effects. The damage formation mechanism is revealed in a further step by Scanning Electron Microscope (SEM). The precision score of the proposed method turns out to have a better performance on recognizing the drilling-induced damage than the traditional semantic segmentation models such as E-Net, U-Net, VGG16, and the residual U-Net with 0.63, 1.88, 1.90 and 3.17 times higher respectively.
KW - Attention Assisted U-Net
KW - CFRP/Ti stack drilling
KW - Damage evaluation
KW - Deep learning
KW - Unsupervised Pixel-level Image Segmentation Labelling
UR - https://www.scopus.com/pages/publications/105051712398
U2 - 10.1016/j.cja.2025.103966
DO - 10.1016/j.cja.2025.103966
M3 - 文章
AN - SCOPUS:105051712398
SN - 1000-9361
VL - 39
JO - Chinese Journal of Aeronautics
JF - Chinese Journal of Aeronautics
IS - 11
M1 - 103966
ER -