This paper outlines the work completed and suggestions to further improve Evraz Oregon Steel?s Level 2 models, along with the presentation of the trial results to validate the theory. The sources of the model errors were identified to be initiated from the adaptation function, ignorance of the metallurgical effects, and the valid range of the formula, etc.

The Level 2 model improvements were carried out in the following two primary areas:

First area of improvement (for learning and metallurgical issues)

• Learning logic - applied the Guided Two-Parameter Learning (FIT2G)

• Metallurgical interaction - considered effect of retained strain, etc., into the FIT2G.

Second area of improvement (defining the valid range of the formula and further metallurgical issues e.g. in resume passes)

• Flow stress formula - reduced errors in passes with draft below 10% or over 30%

• Resume pass force - reduced errors caused by microstructure evolution

• Ranges of the temperature regions and rolling in the two-phase region, etc.

One of the problems in the adaptation function was allowing the use of zeros instead of medium values for the C₃ and C₄ parameters. Other learning issues were such as, the scatter of the C₃ and C₄ due to their dependence on each other. This dependence indicated that the blind adaptive learning, even the Four-Parameter Learning (FIT4), could only reach limited accuracy. Including the C₃-C₄ dependence in the learning logics would greatly increase the prediction accuracy. This project demonstrated that it is very critical for a Level 2 model to consider metallurgical effects. The project calls for an expansion of Level 2 models from current mechanical system to a combined metallurgical and mechanical system, in order to satisfy the new production practices. Current Level 2 systems in the market were usually designed as a mechanical system without sufficient metallurgical principles taken into account. With increasing applications of metallurgical processes such as controlled rolling in the steel mills, traditional Level 2 models without metallurgical consideration could only reach a limited level of accuracy.

The Guided Two-Parameter Learning (FIT2G), by using carefully designed medium values for C₃ and C₄, and performing the adaptation by adjusting C₁ and C₂, would be a simple but very effective solution, especially for improving existing Level 2 models. This procedure can, not only, remove the limitation of the adaptive learning, but also include the metallurgical effects into a Level 2 model. The large number of flow stress coefficients, over 6000 sets of C₁, C₂, C₃ and C₄, is the result of all the solutions for the learning logic and metallurgical effects. In addition, it only requires very limited modifications to the Level 2 source code and needs a very small temperature range (number of points) to perform the regression. Even for the troubled grades that experienced phase transformation, the first improvement, by applying the Guided Two-Parameter Learning, still led to over 80% passes below 5% error, over 90 passes below 10% and over 99% passes below 15% errors. This result, confirmed by the trials, fully demonstrated the effectiveness of this learning procedure.

With the issues and sources of error revealed in the mill trials, particularly those 1% of passes that were still with error of 15% or higher, the second improvement was conducted and theoretically showed to further reduce the model prediction error. However this solution as yet to be implemented. The temperature regions were divided based on the metallurgical, dividing points and the range of the low-temperature region was narrowed. Modification to the strain was done to extend the valid range of the flow stress formula for small strain (below 0.1) and large strain. Predicted flow stress for the resume pass was scaled down (or up) using the temperature-dependent empirical factor, in order to compensate for the error caused by microstructure evolution during holds. In addition, issues on entry into the two-phase region was identified and a possible solution was suggested.

|  Part1  |  Part2  |  Part3  |  Part4  |  Part5  |  Part6  |  Part7  |  Part8  |  Part9  |