Accelerated Method for Simulating the Solidification Microstructure of Continuous Casting Billets on GPUs

Area of Science:

• Materials Science
• Computational Materials Science
• Metallurgy

Background:

• Microstructure simulations are crucial for continuous casting billet production.
• Traditional Cellular Automaton (CA) models suffer from grid anisotropy, impacting dendrite morphology accuracy.
• Decentered Square Algorithm (DCSA) reduces anisotropy but is computationally expensive for large-scale simulations.

Purpose of the Study:

• To develop a high-performance GPU-accelerated CA-DCSA method for microstructure simulations.
• To overcome the computational limitations of existing models for continuous casting.
• To enhance the accuracy and efficiency of predicting solidification mechanisms.

Main Methods:

• Refactoring the CA-DCSA algorithm for CPU-GPU heterogeneous architecture.
• Implementing key optimizations for GPU utilization, including memory access and warp divergence reduction.
• Validating simulation results with industrial experiments on 65# and 60# steel.

Main Results:

• The GPU-accelerated CA-DCSA achieved a 1430× speedup compared to serial implementation using two GPUs.
• Experimental validation showed low relative errors: 2.5% (equiaxed crystal ratio) and 2.3% (dendrite arm spacing) for 65# steel.
• Accurate prediction of microstructure and temperature distribution was achieved, demonstrating the method's efficacy.

Conclusions:

• The proposed GPU-accelerated CA-DCSA method offers a powerful and efficient tool for microstructure simulation in continuous casting.
• This advancement enables detailed microstructure observation and effective process parameter optimization.
• The validated results confirm the method's reliability for industrial applications.