A rivet-free adhesive riveted joint mechanical property prediction method based on CNN-Transformer

Abstract

The present application belongs to the technical field of rivetless bonding, and discloses a rivetless bonding joint mechanical property prediction method based on CNN-Transformer, aiming to solve the problems of destructive test, low prediction efficiency and insufficient prediction accuracy in the mechanical property detection of the existing rivetless bonding joint. First, the micron and nanometer double-scale topography images of the upper and lower plates of the rivetless bonding joint are collected by a white light interference microscope and an atomic force microscope, and standardized preprocessing is completed. Then, the feature re-labeling is completed by a double-plate cross-scale interactive attention enhancement module, and the global correlation is established by a four-branch CNN to extract deep features and a cross-plate interactive Transformer. Finally, the maximum peak load and failure displacement under the tensile shear working condition of the joint are predicted by a multi-regression head optimized by homoscedastic uncertainty loss. The present application has the advantages of low prediction cost, high prediction efficiency and accurate accuracy.

Images