Prediction of creep mechanical properties and bonded ratio curves in diffusion bonding using finite element-based void closure models and optimized deep learning techniques

Abstract

A void closure analytical model for the diffusion bonding of titanium TC4 alloy is developed in this study, in which an FEA-based deformation mechanism was coupled with the numerical analysis for diffusion. The focus was on evaluating the effect of pressure and temperature on the bonded ratio. By increasing the value of bonding pressure or bonding temperature, the bonding time decreases. The dependence of the bonded ratio on time was modeled as an exponential curve. Furthermore, deep learning models were employed to convert the simulation videos of bonded ratio over time into their respective graphs using the encoder-decoder and U-Net models. The U-Net model predicted the value of bonded ratio with 95% accuracy, while the encoder-decoder model predicted it with an accuracy of 94%. The prediction of the material properties was done by a convolutional 2D model, and the loss and Mean Absolute Error (MAE) for the model were 0.0303 and 0.12, respectively, for the test results.