Prevention of defects during radial-axial rolling of rings based on online data analysis

所有行业的许多机器和系统都需要具有高动态负载能力和高可变性等要求的无缝环形部件。径向-轴向环件轧制（RARR）是制造此类部件的最重要工艺，能够生产外径为100 mm至16 m、环件高度为4 m、部件重量为300 t的环件。在已完成的“Fehlervermeidung in Radial-Axial Ringwalzprozessen durch Online-Analyse der Zustandsdaten“项目中，机器学习模型被训练并用于预测各种形状缺陷（不圆度和波纹度）。这种预测是在过程之后的第一步中完成的，因此是离线的，但也可以在项目过程中在线完成。这样的在线预测可以用于优化轧制过程，使得可以防止而不是仅仅预测形状缺陷，其中在轧制时，机器学习模型可以通过人类无法解释的轧制数据中的模式来检测和学习形状缺陷。以前的工作集中在可用的环数据的分类，并将进一步扩展到回归问题。通过这种回归方法，不仅可以对出现的形状缺陷进行分类，而且还可以预测它们的表达程度（“1.212 mm不圆度”）。回归使得能够更好地实施轧制过程的控制，因为可以更好地估计合适的反轧制措施的干预强度。除此之外，之前的研究项目还确定了在RARR中使用半监督学习数据和合成数据生成方面的研究差距。因此，在继续申请中，正是这些研究领域将致力于迄今为止获得的数据和发现的基础上。为了进一步提高研究的质量，还将包括来自合作环件轧制公司（蒂森克虏伯rothe erde德国GmbH和Schmiedewerke Gröditz）的工业轧制米尔斯的额外数据。即使是小型环轧机，也将在研究所自己的环轧机上进行大范围轧制。所附的蒂森克虏伯rothe erde德国有限公司和Schmiedewerke Gröditz的意向书突出说明了拟议专题的工业相关性。基础研究的相关性已经在以前的项目中通过大量的成果和研究差距的出版物得到了证明。

Seamless, ring-shaped components with requirement profiles such as high dynamic load capacity and high variability are needed for many machines and systems in all branches of industry. Radial-axial ring rolling (RARR) is the most important process for the manufacture of such components and enables the production of rings in the order of 100 mm up to 16 m outer diameter, up to 4 m ring height and a component weight of up to 300 t. Within the completed project on " Fehlervermeidung in Radial-Axial Ringwalzprozessen durch Online-Analyse der Zustandsdaten ", machine learning models were trained and used to predict various shape defects (non-circularity and waviness). This prediction is done in a first step following the process and thus off-line, but is also done on-line during the process in the course of the project. Such on-line prediction can be used to optimize the rolling process to the extent that shape defects, which at the time of rolling are detectable and learnable by machine learning models through patterns in the rolling data that cannot be interpreted by humans, can be prevented instead of merely predicted. The previous work focuses on a classification of the available ring data and will be further extended to a regression problem. By such a regression approach, occurring shape defects can not only be classified, but also predicted in their degree of expression ("1.212 mm out-of-roundness"). Regression enables a better implementation of the control of the rolling process, since the intervention strength of suitable counter-rolling measures can be better estimated. In addition to this, the previous research project identified a research gap in the use of semi-supervised learning data and synthetic data generation in RARR. For this reason, within the continuation application, precisely these research fields are to be dedicated on the basis of the data and findings obtained so far. In order to further increase the quality of the research, additional data from industrial rolling mills at cooperating ring rolling companies (thyssenkrupp rothe erde Germany GmbH and Schmiedewerke Gröditz) are to be included. To cover even small ring mill dimensions, extensive rolling will also be carried out on the institute's own ring rolling mill. The industrial relevance of the proposed topic is highlighted by the attached letters of intent from thyssenkrupp rothe erde Germany GmbH and Schmiedewerke Gröditz. The basic research relevance has been demonstrated within the previous project by numerous publications of the results and research gaps.