5. plate mill application Examples
5.1 Evraz Oregon Steel (EOS)
Evraz Oregon Steel (EOS) Plate Mill Level 2 has experienced many years of continued improvements, including two years of commissioning after the installation, and five years of further development inside EOS until it runs smoothly. The most recent major improvements, from Metal Pass, have eliminated the shape defects for the hard and thin grades.
The work for the improvements
consists of :
- Correcting force learning logical errors for all the learning fittings
- Removing limitation of adaptive learning
- Refining the ranges of temperature regions
- Integrating metallurgical effects (e.g. retained strain) in the force prediction model
- Suggesting the modeling for the rolling in two-phase region
- Expanding formula valid range for passes
- Adding resume pass force prediction logics to reduce high errors caused by microstructure evolution during hold
- And so on
This round of improvements started in 2006 as the model frequently encountered product shape defects, especially in the hard and thin grades. Analysis indicated that the Level 2 model had significant force errors in the related passes. Because the draft schedules generated by the Level 2 model were primarily based on predicted forces, unreasonable draft schedules were created which led to bad finish shape . The early stage of the work was filled with confusion because the flow stress coefficients C3 (for strain) and C4 (for strain rate) were usually much higher than the theoretical values, and there was a fluctuation of C3 and C4 in a large range. The fluctuation was concluded to be due to a design logical problem on learning. Later, limitation of the adaptive learning was identified and issues of the metallurgical effects in the model were disclosed. Since then, the primary work was to integrate metallurgical effects in the Level 2 model, e.g. in the design of 6,000 sets of flow stress coefficients.
The Level 2 model improvements were divided into the two primary steps. The first step was mainly on learning and metallurgical issues, based on Guided Two-Parameter Fitting (FIT2G) discussed earlier in this paper. For each of the 2000 model grades in each temperature region (high, medium and low), a set of flow stress coefficients C1, C2, C3, and C4 were designed. The design took into account various influence factors, e.g. retained strain, controlled rolling practice, etc.). Mill tests were conducted after the first improvement step. Even for the troubled grades that had experienced phase transformation, the tests still showed an excellent prediction accuracy with errors usually below 5% or lower, that fully demonstrated the effectiveness of this learning procedure . With the issues and sources of errors revealed in the mill trials, the second improvement step was conducted to further reduce the model prediction error.