Scholarlink

<p>The production chain for aluminium beverage can production typically encompasses multiple thermo-mechanical processing steps, including casting, hot and cold rolling, heat treatment, and deep drawing. Each of these steps uniquely influences the material’s mechanical properties, resulting in anisotropy that can cause earing in the drawn beverage can body. Advancements in digitalization and Industry 4.0 are enabling companies to acquire vast amounts of production data. In combination with fast simulation tools to generate material quality parameters, the combined data contain valuable process knowledge. While process experts are utilizing these data to optimize operations, there remains significant untapped potential to achieve further benefits, such as data-driven process monitoring for anomaly detection using advanced data analytics methods. Moreover, despite efforts to run industrial processes at optimal efficiency, deviations in alloy composition or ingot temperature can still occur. Such deviations may propagate through the process chain, leading to undesirable effects on the mechanical properties of the final product. To address this challenge, data-based intelligent systems are essential for effectively responding to the deviations and minimizing their negative impacts. In this work, we propose a flexible and scalable adaptive optimization approach that integrates a genetic algorithm with a conditional generative adversarial network, alongside a data-driven XGBoost surrogate model of the process chain. This integration enables the determination of modified process variables in response to the deviations. The proposed approach is applied to an industrial use-case focussed on minimizing earing formation in cold-rolled aluminium strips (3xxx alloy) used for beverage can production. By intentionally introducing deviations, we demonstrate the effectiveness of the proposed method in suggesting countermeasures. Furthermore, this research highlights the suitability of a generative model in anomaly detection during production. The results reveal significant potential to enable process monitoring and optimization, thereby improving production planning and strip quality.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Data-driven adaptive approach integrating genetic algorithm and conditional generative adversarial network for minimizing earing formation in industrially rolled aluminium strips

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月2日

查看全文
  • Nilesh Thakare,
  • Kai Karhausen,
  • Holger Brüggemann,
  • Emad Scharifi,
  • Junhe Lian

摘要

The production chain for aluminium beverage can production typically encompasses multiple thermo-mechanical processing steps, including casting, hot and cold rolling, heat treatment, and deep drawing. Each of these steps uniquely influences the material’s mechanical properties, resulting in anisotropy that can cause earing in the drawn beverage can body. Advancements in digitalization and Industry 4.0 are enabling companies to acquire vast amounts of production data. In combination with fast simulation tools to generate material quality parameters, the combined data contain valuable process knowledge. While process experts are utilizing these data to optimize operations, there remains significant untapped potential to achieve further benefits, such as data-driven process monitoring for anomaly detection using advanced data analytics methods. Moreover, despite efforts to run industrial processes at optimal efficiency, deviations in alloy composition or ingot temperature can still occur. Such deviations may propagate through the process chain, leading to undesirable effects on the mechanical properties of the final product. To address this challenge, data-based intelligent systems are essential for effectively responding to the deviations and minimizing their negative impacts. In this work, we propose a flexible and scalable adaptive optimization approach that integrates a genetic algorithm with a conditional generative adversarial network, alongside a data-driven XGBoost surrogate model of the process chain. This integration enables the determination of modified process variables in response to the deviations. The proposed approach is applied to an industrial use-case focussed on minimizing earing formation in cold-rolled aluminium strips (3xxx alloy) used for beverage can production. By intentionally introducing deviations, we demonstrate the effectiveness of the proposed method in suggesting countermeasures. Furthermore, this research highlights the suitability of a generative model in anomaly detection during production. The results reveal significant potential to enable process monitoring and optimization, thereby improving production planning and strip quality.