In our Level 2 model upgrade (including new Level 2
System installation), for the extremely complex metallurgical process, offline
design was performed based on the base model, and then the results were
integrated into the online learning of the Level 2 System. In the same time,
learning logic defects were fixed, and the solution was put into the large
number of data models. Therefore, tens of thousands sets of models were
designed. See the tables below.
Steel grades |
Thick-ness |
Product types |
Rolling stages |
Slab thickness |
150 |
5 |
4 |
3 |
3 |
In
general rolling, one steel mill may be summarized into about 150 steel grades
based on chemical compositions. In the rolling there may be one of the
5*4*3*3=180 cases, totally, a rolling may be one of the 150*180=27,000 cases.
Every case should have high, medium and low temperature three zone, in which
each zone has a set of data models (flow stress, etc.). A design for those data
is needed in the Level 2 System. For each steel grade, there are 180 possible
data files, with each file usually contain several dozens numbers.
With difference between microalloyed steels and conventional carbon steels, each
data for a microalloyed steel can be totally different from that of the carbon
steels. A design for the new data in microalloyed steels is therefore critical
for the high accuracy of model prediction.
Due to large number of model calculation, a software package is created and thus
the design was made with software. With the software, calculation was made
through calculation of :
- carbon equivalence formula for hot strength (different from IIW)
- formula for hot strength based on chemical composition
- other procedures and related knowhow
These models have been successfully applied in Oregon Steel and China NISCO for
many years. Oregon Steel used to have product defects almost daily in the
hard-and-thin (5mm thickness) grades. After installation of new models, the
defects were totally eliminated.
The Level 2 System with metallurgical model, intelligent learning and advanced
software engineering, is referred as the New-Generation Level 2 System.
Currently, only the author’s team possesses this technology.
|
With the upgraded model, even rolling in a lower
temperature and a larger pass reduction, can still guarantee the plate/strip
shape, so there is a great room for mechanical properties improvement via
thermomechanical processing. Accurate parameter prediction also particularly
stabilizes the rolled steel properties. With increased and especially stablized
mechanical properties, there is an opportunity to reduce alloy consumption, so
the product cost would be greatly reduced and enterprise profit increased.
Those effects make it possible for the upgraded Level 2 System to achieve
extremely high parameter prediction accuracy, with an error usually only a
fraction of that before the upgrade. Level 2 System creates draft schedule based
on predicted parameters, which means the quality of the draft schedule is thus
greatly improved in this process.
|