<Continue>
2.3 Data Collection and History Log Processing
Though the data and model design are based on the past research results and the data collected from various sources, the existing mill data in most recent 2 years were collected for verification and reference. For each piece rolled, there is a set of data in log file(s) consisting of
- Slab Data
- Roll Data
- Scheduled Values
- Updated Values
- Measured Values
- Repredicted Values
- Long-Term Learning Values
As in most Chinese mills, the Level 2 history data are stored in the flat files (though there is an Oracle database in NISCO for accepting Level 3 data, etc.). The following flat files were collected:
- Log files for piece data in the most recent 2 years, totally over a million pieces. For each piece, one of the log files contains all the data showed in the above list.
- Existing grade files before the improvement. NISCO existing system uses about 815 model grades. The existing model grade list is available from the grade file list.
- Family files.
- Steel grades, with chemical composition, etc.
- Miscellaneous mill practice and mill operation data.
Computer programs have been developed to collect the above-mentioned data. All data were extracted from the flat files and stored in a relational database. It took 1.5 months in the fall 2009 to collect and consolidate the data, and to collect related information and diagnose the existing system. In this stage, quite amount of time was spent in consolidating steel grades, as will be discussed in the section "Steel Grade Consolidation".
2.4 Steel Grade Consolidation
At the beginning of the project, NISCO Plate/Coil Mill had about 590 sets of chemical compositions for the grades, consisting of two groups: that specified by the NISCO corporation and that controlled inside the Plate/Coil Mill, each 295 sets. Based on the model grade specification design for NISCO production practice, every grade can have up to 180 model grades. This would lead to an unacceptable and unnecessary number of model grades and flow stress models. Therefore, steel grades had to be redesigned. After consolidation, there were 129 grades, which means there were 129 sets of chemical compositions. This was dated to October 2009. Throughout the project, the number of steel grades has continuously increased. Up to the end of 2010, the number reached to about 150.
Before this project, NISCO classified its grade based on the purpose of steel application and the mechanical property. Many grades had quite similar compositions. From technical aspect, they should belong to the same grade though they are for different applications. To some extent, the steel grade consolidation stimulated the technical upgrade in NISCO.
2.5 Model Grade Design
NISCO tried very hard in controlled rolling. Besides, one of Metal Pass?
work focuses in model improvement is to make it more user-friendly for the controlled rolling. Therefore, the model grade design reflects detailed requirements for the controlled rolling. A model grade is generated based on the steel grade (150), product thickness (1, 2, 3, 4, 5) and Process Type (F, S, P, C), Slab Thickness (L, M, H) and Rolling Stage (1, 2, 3), with each category specified in the Table 2.
Table 2: Model Grade Classification
On this basis, a steel grade can have up to 180 possible model grades (5x4x3x3). Of course, not every combination can be used for the production. Certain grades are only for plate, some only for coil, and some thick products should only be rolled from thick slabs (for quality requirement), etc.
As mentioned above, NISCO initially only used 815 model grades before this improvement. There were very few categories (e.g., no Slab Thickness).
Initially, about 2000 model grades were planned. Many model grades were reserved for future use.
Composing the model grade list needs to follow mill抯 product mix, production practice and equipment specification. For example, for each grade, how many thicknesses (or widths) are to produce, what type of slabs (thickness, etc.) for each thickness is to roll; how many products are to roll with hold, two piece rolling, or with steckle furnace, etc. On the other hand, well-designed model grade classification also provides an extra guideline for production schedule creation and optimization, though it is initially for the Level 2 model. Products from various customer orders may use similar production procedure if they fall into the same model grade. As a result, the procedure can be continuously improved and well documented. Further, rolled steel property prediction may also follow this well-designed model grade classification.