|
|
Level 2 Model Improvement
Case Study: Oregon Steel
<Continue>
In Table 5, a comparison between
the new and the old models is performed, with the pass count in various error
ranges. The new model has more passes in the low error range ("< 5%") and fewer
passes in the high error range (">15%").
Table 5:
Pass count in various error ranges
Grade, Slab |
New Model |
Old Model |
|
<5% |
5%-10% |
10%-15% |
>15% |
<5% |
5%-10% |
10%-15% |
>15% |
05012506CN1, NT2245A3 |
9 |
2 |
3 |
0 |
8 |
2 |
2 |
2 |
05012506CN1, NT2254A5 |
9 |
2 |
2 |
1 |
5 |
5 |
1 |
3 |
05010002SN1, NT2386A30 |
13 |
1 |
1 |
0 |
10 |
4 |
1 |
0 |
05010002SN1, NT2385A28 |
13 |
1 |
1 |
0 |
10 |
4 |
1 |
0 |
04010531CN1, NT2291A4 |
11 |
1 |
0 |
0 |
7 |
4 |
1 |
0 |
04010531CN1, NT2291A2 |
11 |
1 |
0 |
0 |
7 |
5 |
0 |
0 |
Further summary of the
data in Table 5 led to the result in the Table 6.
The high quality of the new model was demonstrated: 80% passes
were within 5% force error, 90% passes below 10%, and 99%
passes below 15% error. Error of over 10% for the new model
would later be attributed to the failure of flow stress
formula in the two-phase region and the problem of the flow
stress valid range (breakpoints). Continued work was conducted
later in the second improvement to further reduce the error.
Table 6:
Force error and quality level
Before improvement |
After 1st improvement |
- 57% passes: < 5%
- 87% passes: < 10%
- 94% passes: < 15%
- model failed for some grades (40% force error, bad shape)
|
- over 80% passes: < 5%
- over 90% passes: < 10%
- over 99% passes: < 15%
- No occurrence of quality problem found yet since use of new model
|
* The data here was based on the
troubled grades that encountered quality problems earlier.
Regular grades may have still better results.
In Table 7, the minimum, average
absolute and maximum errors for the new and old models were summarized. The
absolute values of the errors were used to calculate the average error. As the
average values for all the trial samples, the new model showed as low as 3.4%
for the absolute error, -6.44% for minimum and 9.5% for maximum error,
significantly improved from the old model.
To be claimed is that, the grades examined
here were all the troubled ones with high force prediction errors and frequent
defects in the past. Two of the three model grades tested were with
phase-transformation. For the regular grades, the errors are expected to be much
lower.
Table 7:
Minimal, average and maximal errors in the new and old models
Grade, Slab |
New Model (%) |
Old Model (%) |
|
Min |
Avg (abs) |
Max |
Min |
Avg (abs) |
Max |
05012506CN1, NT2245A3 |
-4.00 |
5.80 |
14.64 |
-4.96 |
7.33 |
32.4 |
05012506CN1, NT2254A5 |
-10.2 |
4.53 |
16.96 |
-6.78 |
9.02 |
32.6 |
05010002SN1, NT2386A30 |
-10.11 |
2.60 |
5.72 |
-5.84 |
4.33 |
12.93 |
05010002SN1, NT2385A28 |
-10.73 |
2.78 |
7.52 |
-6.00 |
4.88 |
14.78 |
04010531CN1, NT2291A4 |
-2.24 |
2.28 |
6.92 |
-2.40 |
4.91 |
10.10 |
04010531CN1, NT2291A2 |
-1.36 |
2.36 |
5.21 |
-1.46 |
4.91 |
8.97 |
Average |
-6.44 |
3.39 |
9.50 |
-4.57 |
5.90 |
18.63 |
After March?s trials,
decision was made by management to install all the new
coefficients in the Level 2 system. As more troubled grades
were identified on the mill, the newly designed coefficients
were activated and observed on a as-needed basis to prevent
disruption of production.
<To
Be Continued>
<Previous>
|
<Next>
|
Part1
|
Part2
|
Part3
|
Part4
|
Part5
|
Part6
|
Part7
|
Part8
|
Part9
|
|
|