huawei's big prediction model was first used in real-time steel production control. baosteel: will continue to expand scenarios
2024-09-15
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
following the announcement of the full launch of its ai strategy in april this year, the ai cooperation between baosteel group (600019.sh), the leading listed central steel enterprise, and huawei has made new progress.
baosteel co., ltd. recently disclosed that the hot rolling natural width prediction model developed by the company and huawei based on huawei's pangu prediction big model has been officially put into the hot rolling 1880 production line for online control, completing the first closed-loop control technology verification of huawei's prediction big model in the steel manufacturing field.
the project expert team, which consists of huawei, data ai department, equipment department, hot rolling mill, central research institute, baosight software and other departments, tracked and verified the strip width control after the large model was put into use. the expert team unanimously agreed that during the rolling process of 348 strips, the prediction accuracy and delay response of huawei's large model met the target requirements, the real-time control of strip width was normal, and the verification was a complete success.
on april 26 this year, baosteel announced that it had formulated a new round of digital transformation strategy marked by "ai+", fully launched the ai strategy, and promoted the digital transformation and intelligent upgrading of the steel industry. the company signed a full-field cooperation agreement with huawei on the same day, indicating that the two companies will join forces to carry out in-depth cooperation in all fields of ai.
on may 10, baosteel and huawei officially unveiled the "steel + ai" joint innovation center. according to the introduction at that time, the two parties established the center with the purpose of creating an "incubator" for the innovative research and development of large steel models and an "accelerator" for the iterative upgrade of application + ai scenarios, so as to achieve complementary advantages and win-win cooperation. the joint innovation center will become a key platform for the two parties to jointly research and implement ai-enabled steel industry scenarios.
the two sides will also jointly research and build a systematic methodology for large models in the steel industry. xiao su, director of the data ai department of baosteel co., ltd., said at the time, "this year we plan to conduct research on large models in five areas, including blast furnaces, converters, surface inspection, new steel grade research and development, and scheduling. we will also set goals to solve difficult and painful problems in professional fields and carry out the best development practices of scenario application models. the joint innovation center will also ensure that these topics can be effectively promoted in terms of mechanisms, processes, resources and personnel."
baosteel co., ltd. recently mentioned whether the big model technology can be truly applied to the steel manufacturing process? can it really improve the efficiency of development and iteration? is the control accuracy better than the "small model" currently used on site? can the delay of inference calculation based on big data analysis meet the requirements of high-speed and real-time production control on site? the team has been thinking about such questions before.
the team decided to first select scenarios with high control accuracy requirements and fast production pace, that is, the natural width expansion control link of hot rolling, and carry out large-scale model application technology verification.
it is reported that after more than half a year of exploration, the team has overcome difficulties such as inaccurate data collection and processing, prolonged average time of large model output inference results, and interface development between the process control end and the large model inference end through multiple rounds of technical solution demonstration, data collection, feature engineering, model comparison modeling, model online deployment, and parallel verification. the team has completed three version iterations and creatively designed and developed a high-reliability deployment and operation plan for the large model at the process control end.
regarding the verification results, baosteel co., ltd. stated that this is the first time that the predictive big model based on the huawei platform has been officially put into real-time production control. it also marks that the "ai technology" characterized by big data analysis has officially kicked off its application in baosteel's production process control.
next, baosteel will continue to cooperate with huawei to expand the practical experience of large-scale model applications to similar production line control scenarios.
the paper reporter he liping
(this article is from the paper. for more original information, please download the "the paper" app)
