# US 2019/0377841 A1 · Classification optimization method for steel grade family of steel rolling model

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2019/0377841 A1
- Document ID: US20190377841A1
- Status: Published application
- Kind: A1
- Application number: 16/261,527
- Filed: 2019-01-29
- Published: 2019-12-12
- Earliest claimed: 2018-06-06
- Assignees: Wuhan University of Science and Technology
- Applicants: Wuhan University of Science and Technology
- Inventors: Weigang LI; Wei MA; Wensheng XU; Yuntao ZHAO
- Web page: https://patentgrep.com/patent/US20190377841A1

## CPC

- G06F 30/20: Design optimisation, verification or simulation (optimisation, verification or simulation of circuit designs G06F30/30)
- G06F 17/10: Complex mathematical operations {(function generation by table look-up G06F1/03; evaluation of elementary functions by calculation G06F7/544)}
- G06N 20/00: Machine learning
- G06F 2111/10: Numerical modelling
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Provided is a classification optimization method for steel grade family of a steel rolling model, including: step 1: calculating the deformation rate and deformation resistance of a rolled piece according to rolling process data; step 2: reversing the “equivalently measured” deformation resistance according to the measured roll force; step 3: drawing a scatter plot of data on a graph, and judging whether layer division is reasonable according to distribution characteristics; and step 4: for the unreasonable division of steel grade family, optimizing the existing classification of steel grade family using a new algorithm called line clustering, determining steel grade codes that need to be divided into new family, searching for the shortest distance from a sample point to the curve of each cluster centre based on a Nelder-Mead algorithm, dividing all data points into k subsets according to the principle of proximity, fitting a power curve to each data subset using a least square method, repeating the above steps, and stopping iterating when cluster centrelines obtained by two successive iterations are consistent. The present invention improves the precision of the division of steel grade family of a hot rolled model.

## Claims

### Claim 1

A classification optimization method for steel grade family of a steel rolling model, characterized by comprising:
step 1: for each coil of strip in steel grade family, calculating a deformation rate e and deformation resistance KmPOS of a rolled piece according to latest rolling process measured data obtained during model adaptive post-calculation process,
calculation formulas for the deformation rate being as follows:
ɛ
=
ln

(
1
1
-
r
)
=
ln

(
H
h
)
ɛ
.
=
ɛ
·
v
R
R
′
·
(
H
-
h
)
where ε is the deformation degree of the rolled piece, {dot over (ε)} is the deformation rate of the rolled piece, r is a reduction rate of a stand, H and h are the entry thickness and exit thickness of the rolled piece on a stand respectively, vR is a measured speed of rolls, and R′ is a roll flattening radius,
a calculation formula for the deformation resistance being as follows:
Km
POS
=
(
a
0
+
∑
i
=
1
n

a
i
,
x
i
)
·
exp

(
b
1
T
+
273
)
·
ɛ
b


2
·
ɛ
.
b


3
where KmPOS is a calculated value of the deformation resistance of the rolled piece, T is a rolling deformation temperature, xi is the content of each chemical component of strip, and ai and bi are model parameters to be determined;
step 2: for each coil of hot rolled strip, substituting a measurement value of roll force according to roll force model calculation formulas, and reversely obtaining the deformation resistance KmACT of the rolled piece, the calculation formulas being as follows:
Km
ACT
=
F
ACT
W
ACT
·
L
d
ACT
·
Q
p
ACT
+
α
b
·
T
b
ACT
+
α
f
·
T
f
ACT
L
d
ACT
=
R
′

Δ


h
Q
p
ACT
=
(
a
F


1
+
a
F


2
·
r
)
+
(
a
F


3
+
a
F


4
·
r
+
a
F


5
·
r
2
)
·
R
′
h
R
′
=
(
1
+
16
·
(
1
-
v
2
)
π
·
E
·
F
act
·
1000
W
ACT
·
Δ


h
)
·
R
where KmACT is a “measurement” value of the deformation resistance of the rolled piece, FACT is a measured roll force, WACT is a measured strip width, LdACT is a roll flattening arc length, QpCT is a stress state coefficient, TbACT is a measured backward tension, TfACT is a measured forward tension, αb and αf are tension influence coefficients, αFi is a model parameter, v is a Poisson's coefficient, E is an elasticity modulus, Δh is a rolling reduction, and R is a roll original radius;
step 3: drawing a scatter plot of data ({dot over (ε)}, KmPOS) and ({dot over (ε)}, KmACT) on the same graph, according to strip distribution characteristics of data scatters: if the data scatters are concentrated within a strip-shaped range, layer division is more reasonable; if they are scattered into several strip-shaped distributions segmented obviously, the layer division is unreasonable, and the layer needs to be further subdivided; and
step 4: for the unreasonable division of steel grade family, optimizing the existing classification of steel grade family using a new algorithm called line clustering, and determining steel grade code that need to be divided into a new family;
(1) randomly initializing k curves as cluster centres within the range of a data set, wherein randomly initializing k curves refers to randomly selecting k groups of C and m coefficient values according to a relationship Km=C·{dot over (ε)}m existed in a deformation resistance data set;
(2) calculating the distance from all sample points to the k cluster centrelines Kmcenj=Cj·{dot over (ε)}mj, for each sample point, finding a cluster centreline closest to the sample point, adding the sample point into a cluster corresponding to the cluster centreline closest to the sample point, and dividing the whole data set into k clusters after traversing all the sample points; and
(3) fitting power curves to data subsets of the k clusters using a least square method respectively, updating the C and m coefficient values to update all the cluster centrelines, repeating step (2), and stopping iterating when the cluster centrelines obtained by two successive iterations are consistent to complete clustering.

### Claim 2 (depends on 1)

The classification optimization method for steel grade family of a steel rolling model of claim 1, characterized in that:
the steps of calculating the shortest distance from any sample Xi(Kmi, {dot over (ε)}i); to each cluster centreline in step 4 are as follows:
1) calculating a distance function Dij from Xi to any point on the jth cluster centre curve Kmcenj=Cj·{dot over (ε)}mi: Dij=√{square root over ((Kmi−Kmcenj)2+({dot over (ε)}i−{dot over (ε)})2)}
2) based on a nonlinear optimized algorithm: searching for a minimum value minDij of Dij using a Nelder-Mead simplex algorithm, namely the shortest distance from the sample Xi point to the jth cluster centre curve,
finding a basic feasible solution firstly to judge whether it is an optimal solution; if not, transforming to another improved superior basic feasible solution according to a predetermined rule, and then identifying; if still not, re-transforming, and repeating based on this, wherein since the number of basic feasible solutions is limited, the optimal solution of the problem must be obtained after a finite number of transformations.
