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

Abstract

The large-scale ingot of the 7xxx-series aluminum alloys fabricated by the direct chill (DC) casting is often suffered from foundry defects such as cracks and cold shut due to the formidable challenges in precise controlling of casting parameters. In this work, by using the integrated computational method combing numerical simulations with machine learning, we investigated systematically the evolution of multi-physical fields and grain structures during the solidification processes. The numerical simulation results quantified the influences of key casting parameters including pouring temperature, casting speed, primary cooling intensity, and secondary cooling water flow rate on the shape of mushy zone, heat transport, residual stress and grain structure of DC casting billets. Then, based on the data of numerical simulations, we established a novel model for the relationship between casting parameters and solidification characteristics through machine learning. By comparing with experimental measurements, the model showed reasonable accuracy in predicting the sump profile, microstructure evolution and solidification kinetics under the complicated influences of casting parameters. The integrated computational method and predicting model could be used to efficiently and accurately determine the DC casting parameters to decrease the casting defects.

Keywords: Direct chill casting, Solidification, Finite Element Analysis, machine learning, Process optimization, Aluminum alloys