Data Source Preparation
Data sources for the work were from both
the Level 2 log files and Level 3 business database. Data collection from the
Level 3 Informix database was performed with SQL scripts. After data was
transferred into the local SQL Server 2000, stored procedures were used to
select and process data, for example, to select hard and thin grades, and to
round the data.
Flow stress coefficients C1, C2,
C3 and C4 for all grades were retrieved by solving the
matrices of the intermediate values files through long term learning regression
stored in thousands of grade.
In the initial data collection, the data
from all passes rolled in the year 2006 were used. In later studies, the data
from the first half of 2007 were also included. This involved about 1,500,000
records with data size over 1 gigabyte (GB). Besides the stored procedures,
Microsoft .Net was used as a programming environment to process the data, so for
example, a tool was developed to create drawings automatically, in order to
display influences of major rolling process parameters on the force errors.
Error Source Survey
Though the learning logic was considered
the primary source of errors in the early stage of the project, it was still
necessary to exclude other potential sources of error. For this reason, an error
survey was carried out. General rule to track an influence factor on a parameter
is to isolate the influence factor. In other words, while the selected influence
factor varies in its range, all the remaining factors should remain unchanged.
This was, however, impossible to accomplish due to the shear amount of actual
production data. Instead, to isolate a rolling process factor, such as
temperature, the data was rounded to reduce the scatter; so for example, data
points within 1590?F and 1610?F were rounded to 1600?F.
To view influence factors of various parameters on the
force error, graphs were drawn. Because it was too time-consuming to draw
thousands of graphs manually, a tool was developed to create drawings. Fig. 1
shows the temperature influence on the force error. It is interesting that for
some grades and in some temperature ranges, the temperature had an effect on the
error, while in others the effect was unnoticeable. This phenomenon was
explained as follows: the different metallurgical effects for different grades
in different temperature ranges are different.
Fig. 1:
Temperature influence on the force error
Further error surveys confirmed that the
force learning process should be the primary focus, at least in the initial
stage of the project. In fact, after the learning problem was resolved, other
sources of error were then easily identified, such as the error in the resume
passes and in the passes with small draft below 10%.
<To
Be Continued>
<Previous>
|
<Next>
|
Part1
|
Part2
|
Part3
|
Part4
|
Part5
|
Part6
|
Part7
|
Part8
|
Part9
|