|Title:||Data-driven optimization of hot rolling processes|
|Abstract:||The rolling process is a complex real-world problem which requires adequate handling and analysis. During the production of strips, plates or coils, a huge amount of data is generated which can be used to optimize the process. The description of the process is done with so-called process models which are regular software programs. It is shown in this thesis that data-driven surrogate based optimization is an excellent method for improving even complex processes. The optimization procedure was tested with data of an aluminum roughing mill to optimize flow curve parameters for an unknown material. The tremendous improvement which was observed during the optimization was also validated with real process data. Residual prediction inaccuracies will still be observable for all kind of analytical models, even after optimization. To reduce these residuals, online algorithms are used. The most promising candidate for all rolling datasets was the online SVR algorithm, which was extended to fit the requirements for real-world processes. These extensions included strategies for handling infinite datasets, the management of the stored data and online parameter optimization. Furthermore, the thesis provides strategies for handling categorical variables. It is shown that all online algorithms outperform their offline variant and that the online SVR achieves the best results on these data. The application of the developed algorithms is not limited to the hot rolling process and can directly be applied to other real-world processes.|
Support vector regression
|Subject Headings (RSWK):||Walzen|
|Appears in Collections:||LS 11|
This item is protected by original copyright
All resources in the repository are protected by copyright.