Steckle mill Level 2 model from former
Tippins has some learning logics issues in the force prediction. In addition, adaptive learning also has some limitations such as the dependence between the strain coefficient and the strain rate coefficient. Beyond this, in today抯 plate mill production, metallurgical aspect such as retained strain, metallurgical transformation during hold, etc., has significant impact in the force prediction accuracy. Force model improvement to handle those common problems is of great importance to every Level 2, especially the Tippins package.
This project was initially created to provide some quick improvement to the Tippins Level 2 force prediction logics for the NISCO Plate/Coil Mill. During the development it has been evolved into a full scale system infrastructure upgrade including redesign of grade family, consolidation for the steel grades, and regeneration of Level 2 model grade list, automatic design for large number of flow stress coefficients, as well as handling in continuously increased steel grades and model grades, and so on. Total number of model grades, and so the number of flow stress models, has reached three times of the initial plan. Since this mill is relatively new compared with similar US mills, many technical issues, such as some draft scheduling logic problems which were already solved in the US mills, still remain to be resolved (in a different, ongoing project). As extension to the draft schedule, shape control and controlled rolling are on the way. Metal Pass, with its three specialized areas, the force prediction, the metal deformation and the microstructure modeling, is considered to be the right fit.
In this paper, development background, process, results, experience and lessons, together with the further development, etc., in the NISCO Level 2 force model improvement, are introduced.
In order to fix the learning logic problem and to remove the limitation of adaptive learning, and in order to include metallurgical principle in the Level 2, the Guided Two-Parameter Fitting (FIT2G) is applied. FIT2G uses only material coefficient and temperature coefficient for learning, while strain coefficient and strain rate coefficient are carefully designed and remain unchanged during the learning. Related work includes, at first, redesigning grade family list based on chemical composition, and consolidation for steel grades with similar composition. Then, a large number of model grades have been created based on product thickness, production process, rolling stage and slab thickness, that are used in NISCO production practice. Core development exists in the automatic design and development for the large number of flow stress coefficients for every model grades in three temperature regions. Grade files for each model grade were also created programmatically, accompanied with the redesign of all the related model data. Solution to the new grade and new model grade were also provided. Further improvements include modification for the model in the resume pass and extension to flow stress formula valid range, etc. In the model development, great attention was paid in applying controlled rolling in the mill operation.
High accuracy has been achieved for the force prediction, with the error mere one half of that in the old model. During the test, other weaknesses of the system were identified, e.g., draft scheduling logics, which has to be improved before applying the new model. Though the project of improving draft scheduling logics is ongoing and has made good progress, it is beyond the scope of the force prediction and is not covered in this paper. A list of further developments have been planned. In addition to the draft scheduling logics, it also includes the shape improvement, improvement of rolled steel properties, and software engineering advancement, etc.
 B. Li, D. Cyr and P. Bothma: Level 2 Model Improvements at Evraz Oregon Steel Mills. AISTech 2009. May 4-7, 2009, St. Louis, Mo., USA.
 B. Li & J. Nauman: Metallurgical, Modeling and Software Engineering Issues in the Further Development of the Steel Mill Level 2 Models. AISTech 2008. Pittsburgh, PA, USA. May 5-8, 2008.
 B. Li & J. Nauman: Significance and Development of a Next-Generation Level 2 Model as a Metallurgical System. MS&T Conference, 2008. Pittsburgh, PA., USA. October 5-9, 2008.
 B. Li: Mill Level 2 Model in Improvement of Product Quality and Productivity. AISTech 2010. Pittsburgh, PA, USA. May 3-6, 2010.
 B. Li: Flow Stress. Online at www.metalpass.com/flowstress. Metal Pass LLC, Pittsburgh, PA, USA. Accessed in February 2011.
 B. Li: High Temperature Properties. Online at www.metalpass.com/hit. Metal Pass LLC, Pittsburgh, PA, USA. Accessed in February 2011.
 B. Li: General Properties. Online at www.metalpass.com/general. Metal Pass LLC, Pittsburgh, PA, USA. Accessed in February 2011.