US 12,282,835 B1 · Methods and devices for constucting multi-dimensional coupled vibration identification model for roll system of strip mill

Abstract

Embodiments of the present disclosure provides a method and device for constructing a multi-dimensional coupled vibration identification model for a roll system of a strip mill, the method comprising: performing preprocessing, signal decomposition, and reconstruction of the multi-dimensional data for the roll system of the strip mill monitored through technical monitoring by multiple sensors during a working process of the strip mill, and performing feature extraction of the reconstructed data using a wavelet packet energy approach; obtaining an initial multi-dimensional coupled vibration feature local identification model for the roll system of the strip mill based on Support Vector Machine (SVM) by using a Gaussian radial basis function as a kernel function of the SVM model; and training the initial multi-dimensional coupled vibration feature local identification model for the roll system of the strip mill using a multi-source heterogeneous information fusion approach based on the feature vector to obtain the multi-dimensional coupled feature identification model for the roll system of the strip mill.

A method for constructing a multi-dimensional coupled vibration identification model for a roll system of a strip mill, comprising:

obtaining multi-dimensional data for the roll system of the strip mill during operation;
wherein the multi-dimensional data is monitored through technical monitoring by multiple sensors;
the multi-dimensional data includes a three-axis acceleration signal of a bearing seat, a torque signal of a drive shaft, a rolling force signal, a current signal of a servo valve, and a pressure signal of large-small cavities of a press-down cylinder;
obtaining preprocessed data by preprocessing the multi-dimensional data;
obtaining reconstructed data by decomposing and reconstructing the preprocessed data;
obtaining a feature vector by extracting a feature from the reconstructed data using a wavelet packet energy manner;
obtaining, based on using a Gaussian radial basis function as a kernel function of Support Vector Machine (SVM) model, an initial multi-dimensional coupled vibration feature local identification model for the roll system of the strip mill based on SVM;
training, based on the feature vector, the initial multi-dimensional coupled vibration feature local identification model for the roll system of the strip mill using a multi-source heterogeneous information fusion manner to obtain the multi-dimensional coupled feature identification model for the roll system of the strip mill;
wherein the training, based on the feature vector, the initial multi-dimensional coupled vibration feature local identification model for the roll system of the strip mill using the multi-source heterogeneous information fusion manner to obtain the multi-dimensional coupled feature identification model for the roll system of the strip mill includes:
constructing an identification framework for multi-dimensional coupling for the roll system of the strip mill;
determining a multi-dimensional coupled vibration parameter space for the roll system of the strip mill, collecting data for the roll system of the strip mill using multiple types of sensors and extracting feature parameters for categorization, and constructing a multi-dimensional coupled parameter space from a plurality of feature sub-spaces;
identifying, using the initial multi-dimensional coupled vibration feature local identification model for the roll system of the strip mill, for each of the feature sub-spaces in the multi-dimensional coupled parameter space, the plurality of feature sub-spaces, and determining a basic confidence assignment function set of a multi-dimensional coupled feature for the roll system of the strip mill;
calculating a confidence interval corresponding to each fault type in a recognition model framework for each feature space evidence body based on a basic confidence assignment function for each feature space evidence body in the basic confidence assignment function set;
constructing the multi-dimensional coupled feature identification model for the roll system of the strip mill based on the confidence interval corresponding to each fault type in the recognition model framework for each feature space evidence body, and an uncertainty parameter of an overall recognition model;
the obtaining the feature vector by extracting the feature from the reconstructed data using the wavelet packet energy manner includes:
denoting a frequency band distribution energy of the multi-dimensional data collected by a system after reconstruction at an i-th layer and j-th node of a wavelet packet by Mij(j=0, 1, 2, . . . , 2i−1), and using a formula:
M
ij
=
∑
k
=
1
n
❘
"\[LeftBracketingBar]"
d
j
,
k
❘
"\[RightBracketingBar]"
2
to extract the multi-dimensional coupled feature for the roll system of the strip mill;
wherein dj,k (j=0, 1, 2, . . . , 2i−1, k=1, 2, . . . , n) denotes wavelet packet decomposition node coefficients, and Mij denotes the multi-dimensional coupled feature for the roll system of the strip mill;
using a formula:
M
=
∑
j
=
0
2
i
-
1
M
ij
to determine a sum of energy M in each frequency band after reconstruction of the multi-dimensional signal for the roll system of the strip mill;
wherein the feature vector M of the multi-dimensional coupled feature signal energy distribution for the roll system of the strip mill extracted after reconstruction by wavelet packet decomposition is denoted as:
M
_
=
[
M
i
⁢
0
M
,
M
i
⁢
1
M
,
M
i
⁢
2
M
,
…
,
M
i
⁡
(
2
i
-
2
)
M
,
M
i
⁡
(
2
i
-
1
)
M
]
normalizing the feature vector M of the multi-dimensional coupled feature signal energy distribution for the roll system of the strip mill by a median normalization manner:
Y
′
=
Y
-
Y
m
⁢
i
⁢
d
Y
max
-
Y
min
obtaining a multi-dimensional coupled feature wavelet packet energy proportion feature vector Mn after wavelet three-layer decomposition reconstruction:
M
_
n
=
[
M
0
M
,
M
1
M
,
M
2
M
,
M
3
M
,
M
4
M
,
M
5
M
,
M
6
M
,
M
7
M
]
⁢
(
n
=
1
,
2
,
…
⁢
8
)
;
the obtaining, based on the Gaussian radial basis function as a kernel function of Support Vector Machine (SVM) model, the initial multi-dimensional coupled vibration feature local identification model for the roll system of the strip mill based on SVM further includes:
denoting the SVM-based the initial multi-dimensional coupled vibration feature local identification model for the roll system of the strip mill as:
f
⁡
(
y
,
α
)
=
sgn
⁢
{
∑
i
=
1
s
⁢
α
i
⁢
exp
[
-
❘
"\[LeftBracketingBar]"
y
-
y
i
❘
"\[RightBracketingBar]"
2
σ
2
]
-
b
}
;
wherein σ denotes the Gaussian radial basis kernel function, b denotes a classification threshold, ai being determined automatically according to the system, S denotes a coupling factor, and yi denotes an i-th set of multi-dimensional operational data collected by the system;
before the training, based on the feature vector, the initial multi-dimensional coupled vibration feature local identification model for the roll system of the strip mill using the multi-source heterogeneous information fusion manner, further comprising:
building a predetermined number of classifiers to diagnose a multi-dimensional coupled vibration type of the strip mill according to a multi-classification rule of the SVM;
wherein the multi-dimensional coupled vibration type of the strip mill includes machine-liquid coupling, machine-electrical coupling, electrical-liquid coupling, machine-electrical-liquid coupling, vertical-torsional coupling, vertical-horizontal coupling, torsional-horizontal coupling, and vertical-horizontal-torsional coupling.

206 more elements

Methods and devices for constucting multi-dimensional coupled vibration identification model for roll system of strip mill

Life of this patent

Methods and devices for constucting multi-dimensional coupled vibration identification model for roll system of strip mill

Life of this patent

Abstract

Claims7