Výzvy v oblasti modelování plynulého odlévání oceli - velmi rychlý dynamický solidifikační model na GPU a jeho využítí pro řízení plynulého odlévání oceli

Challenges in numerical modelling of continuous steel casting - very fast GPU dynamic solidification model and its use in continuous casting control

conference paper

en

The paper describes the development of the parallel GPU dynamic solidification model with the use of CUDA computing architecture and NVIDIA Tesla GPU. Its computing performance is then compared with commonly used non-parallel models. Consequently, the GPU model is used for model-based predictive control with an emphasis on violent changes in the casting speed. This approach utilizes the GPU model for real-time prediction of the temperature field for various control strategies. The proper control strategy is then chosen according to the evaluation of the thermal behaviour of cast blank and other criteria. The results show that parallel computing and GPUs are great tools to enhance the computing performance of dynamic solidification models. Due to the significant speed up in order of tens, new opportunities for real-time control of casting have become available.

Článek se zabývá vývojem velmi rychlého modelu pro plynulé odlévání, který je určen pro běh na grafických kartách GPU. V článku je provedeno srovnání výpočetního výkonu vyvinutého modelu s komerčně dostupnými modely, které provádí výpočet pomocí CPU. Dále je v článku prezentován model dynamického řízení sekundárního chlazení, který využívá principu model-based predictive control a vyvinutého GPU modelu.

The paper describes the development of the parallel GPU dynamic solidification model with the use of CUDA computing architecture and NVIDIA Tesla GPU. Its computing performance is then compared with commonly used non-parallel models. Consequently, the GPU model is used for model-based predictive control with an emphasis on violent changes in the casting speed. This approach utilizes the GPU model for real-time prediction of the temperature field for various control strategies. The proper control strategy is then chosen according to the evaluation of the thermal behaviour of cast blank and other criteria. The results show that parallel computing and GPUs are great tools to enhance the computing performance of dynamic solidification models. Due to the significant speed up in order of tens, new opportunities for real-time control of casting have become available.

plynulé odlévání oceli, dynamický solidifikační model, GPU, výpočety GPGPU, paralelní výpočty, prediktivní řízení

continuous casting of steel, dynamic solidification model, GPU computing, parallel computing, model-based predictive control

2014

23.06.2014

Austrian Society for Metallurgy and Materials (ASMET)

Leoben, Austria

978-3-200-03664-2

Proceedings of the 8th European Continuous Casting Conference ECCC 2014

266–275

10