Numerical Simulation and Machine Learning Prediction of the Direct Chill Casting Process of Large-Scale Aluminum...

Area of Science:

• Materials Science
• Metallurgy
• Computational Modeling

Background:

• Large-scale 7xxx-series aluminum alloy ingots cast via direct chill (DC) casting are prone to defects like cracks and cold shut.
• Precise control of casting parameters is challenging, leading to significant foundry issues.

Purpose of the Study:

• To develop an integrated computational method combining numerical simulations and machine learning.
• To systematically estimate the evolution of multi-physical fields and grain structures during solidification.
• To establish an accurate predictive model for DC casting parameters to reduce defects.

Main Methods:

• Numerical simulations were employed to quantify the influence of casting parameters (pouring temperature, casting speed, cooling intensity, water flow rate) on mushy zone shape, heat transport, residual stress, and grain structure.
• Machine learning was utilized to establish a novel model linking casting parameters to solidification characteristics based on simulation data.

Main Results:

• The numerical simulations provided quantitative insights into the effects of key casting parameters.
• The machine learning model demonstrated reasonable accuracy in predicting sump profile, microstructure evolution, and solidification kinetics.
• The integrated approach successfully correlated casting parameters with solidification behavior.

Conclusions:

• The developed integrated computational method and predictive model offer an efficient and accurate approach for determining DC casting parameters.
• This methodology can significantly decrease casting defects in 7xxx-series aluminum alloys.
• The findings contribute to improved quality control in aluminum alloy ingot production.