Bonding Time Prediction of AA7075-T6 for Extrusion-Based Additive Manufacturing: Machine Learning and Mathematical Modelling

Sadettin Cem Altıparmak,
Bonding Time Prediction of AA7075-T6 for Extrusion-Based Additive Manufacturing: Machine Learning and Mathematical Modelling,
Additive Manufacturing Frontiers,
2025,
200261,
ISSN 2950-4317,
https://doi.org/10.1016/j.amf.2025.200261.
(https://www.sciencedirect.com/science/article/pii/S2950431725000711)
Abstract: ABSTRACT
Diffusion bonding and additive manufacturing (AM) are suitable joining and manufacturing techniques for producing aluminium alloy components even with highly intricate geometries. However, the generation of aluminium oxide (Al2O3) scaling occurring on the surfaces of successively deposited aluminium layers decreases the metal-to-metal contact in extrusion-based AM applications. Al2O3 scaling leads to poor quality of final fabricated parts and thereby achieving low bonding (joint) strength from bonded layers. In this regard, accurately predicting the minimum holding time required is essential for effectively mitigating Al2O3 scaling, and achieving high-quality aluminium parts. This consideration involves optimizing the duration of the diffusion bonding process and the consolidation pressure applied by print heads or consecutive rollers in extrusion-based AM process. There is currently neither mathematical model nor machine learning model developed for predicting the bonding time for similar or dissimilar aluminium and its alloys. Therefore, the current paper proposes both mathematical and supervised machine learning models using regression-based approach to predict the minimum holding time required to achieve sound AA7075-T6 joints at 450°C, 475°C and 500°C. The mathematical model explicitly incorporates the closure of micro-voids through plastic deformation coupled with power-law creep of both AA7075-T6 and Al2O3, as well as volume, grain boundary, and surface diffusion. Notably, the creep and plastic deformations of Al2O3 scaling are included for the first time in any diffusion bonding model for more realistically modelling void closure. Then, the bonding time was predicted using a machine learning code in which several techniques were incorporated to robust the model i.e. feature scaling, polynomial feature expansion, regularization and mean squared error technique. The proposed mathematical model exhibited an excellent agreement with experimental results, particularly at low pressure levels and all data points at 450°C, whereas the machine learning model predicted bonding time more accurately than the mathematical model particularly at 475°C.
Keywords: Bonding time prediction; Additive manufacturing; Machine learning; Mathematical modelling; Diffusion bonding; AA7075-T6