Investigation of the accuracy of multi-input models using deep learning to estimate the tensile shear strength of adhesive joints

Abstract

The correlation between the fracture surface images and residual tensile shear strength of adhesive joints immersed in water was investigated using deep learning. Multi-input models were used to investigate the effects of input parameters on the estimated strength. Specifically, the immersion time and temperature were added as inputs to the fracture surface image to identify the effects of multiple inputs and the conditions that contribute to the improved estimation accuracy. The estimation results showed that the highest accuracy was obtained when both immersion time and immersion temperature were added as input parameters. The variance in the estimated strength tends to be smaller for each additional parameter. To improve the accuracy of estimation using a multi-input model, approximate trends should be estimated for each input parameter alone. Conversely, the use of input parameters for which the trend cannot be estimated may reduce the estimation accuracy.