US 10,794,790 B2 · Large-scale high-speed rotary equipment measuring and neural network learning regulation and control method and device based on rigidity vector space projection maximization

Abstract

The present invention provides a large-scale high-speed rotary equipment measuring and neural network learning regulation and control method and device based on rigidity vector space projection maximization, belonging to the technical field of mechanical assembly. The method utilizes an envelope filter principle, a two-dimensional point set S, a least square method and a learning neural network to realize large-scale high-speed rotary equipment measuring and regulation and control. The device comprises a base, an air flotation shaft system, an aligning and tilt regulating workbench, precise force sensors, a static balance measuring platform, a left upright column, a right upright column, a left lower transverse measuring rod, a left lower telescopic inductive sensor, a left upper transverse measuring rod, a left upper telescopic inductive sensor, a right lower transverse measuring rod, a right lower lever type inductive sensor, a right upper transverse measuring rod and a right upper lever type inductive sensor. The method and the device can perform effective measuring and accurate regulation and control on large-scale high-speed rotary equipment.

A large-scale high-speed rotary equipment measuring and neural network learning regulation and control method based on rigidity vector space projection maximization, comprising:

step 1:
obtaining a morphological filter based on non-equal interval sampling angles by utilizing an envelope filter principle, and extracting a rotor circular profile from a functional perspective;
taking any point P1 in a two-dimensional point set S, and enabling points which take the point P1 as a start point and are less than 2α from the point P1 to constitute a subset S1, wherein S is a two-dimensional space coordinate point set of circular profile sampling points, and α is an alpha disc radius;
step 2:
taking any point P2 in the subset S1 so that two alpha discs which have a radius α and pass through the point P1 and the point P2 exist, wherein the locus equation of the centers of inner and outer circles of the alpha discs is
{
ρ
0
=
x
0
2
+
y
0
2
φ
0
=
arc
⁢
⁢
tan
⁢
⁢
y
0
x
0
⁢
⁢
or
⁢
⁢
{
ρ
0
′
=
x
0
′
2
+
y
0
′
2
φ
0
′
=
arc
⁢
⁢
tan
⁢
⁢
y
0
′
x
0
′
wherein P0 and P0′ are the centers of circles of the two alpha discs respectively;
and
{
x
0
=
ρ
1
⁢
cos
⁢
⁢
φ
1
+
0
.
5
⁢
(
ρ
2
⁢
cos
⁢
⁢
φ
2
-
ρ
1
⁢
cos
⁢
⁢
φ
1
)
+
H
⁡
(
ρ
2
⁢
sin
⁢
⁢
φ
2
-
ρ
1
⁢
sin
⁢
⁢
φ
1
)
y
0
=
ρ
1
⁢
sin
⁢
⁢
φ
1
+
0
.
5
⁢
(
ρ
2
⁢
sin
⁢
⁢
φ
2
-
ρ
1
⁢
sin
⁢
⁢
φ
1
)
+
H
⁡
(
ρ
1
⁢
cos
⁢
⁢
φ
1
-
ρ
2
⁢
cos
⁢
⁢
φ
2
)
⁢
⁢
{
x
0
′
=
ρ
1
⁢
cos
⁢
⁢
φ
1
+
0
.
5
⁢
(
ρ
2
⁢
cos
⁢
⁢
φ
2
-
ρ
1
⁢
cos
⁢
⁢
φ
1
)
-
H
⁡
(
ρ
2
⁢
sin
⁢
⁢
φ
2
-
ρ
1
⁢
sin
⁢
⁢
φ
1
)
y
0
′
=
ρ
1
⁢
sin
⁢
⁢
φ
1
+
0
.
5
⁢
(
ρ
2
⁢
sin
⁢
⁢
φ
2
-
ρ
1
⁢
sin
⁢
⁢
φ
1
)
-
H
⁡
(
ρ
1
⁢
cos
⁢
⁢
φ
1
-
ρ
2
⁢
cos
⁢
⁢
φ
2
)
⁢
⁢
⁢
H
=
α
2
(
ρ
1
⁢
cos
⁢
⁢
φ
1
-
ρ
2
⁢
cos
⁢
⁢
φ
2
)
2
+
(
ρ
1
⁢
sin
⁢
⁢
φ
1
-
ρ
2
⁢
sin
⁢
⁢
φ
2
)
2
-
1
4
⁢
,
wherein, ρ0, φ0, ρ0′, φ0′, ρ1, φ1, ρ2, and φ2 are polar diameters and polar angles of points P0, P0′, P1 and P2 in polar coordinates, respectively;
x0, y0, x0′ and y0′ are coordinates of points P0 and P0′ in an X-axis direction and a Y-axis direction, respectively;
step 3:
obtaining a relationship between an alpha envelope boundary ∂Hα(S) and sampling point polar coordinates (ρi,φi) by utilizing the locus equation in the step 2, wherein the relationship between the alpha envelope boundary ∂Hα(S) and the sampling point polar coordinates (ρi,φi) is expressed as
∂
H
α
⁡
(
S
)
=
∑
i
=
1
n
-
1
⁢
F
⁡
(
ρ
i
,
ρ
i
+
1
,
φ
i
,
φ
i
+
1
)
,
wherein n is the number of the circular profile sampling points, and ρi and φi are respectively the polar diameter and the polar angle of a point Pi in the polar coordinates;
F is a non-equal interval morphological filter design rule based on an alpha shape theory;
step 4:
performing validity processing on circular profile sampling data by the non-equal interval filter to obtain valid circular profile data;
then, according to the valid circular profile data, fitting the rotor circular profile by utilizing a least square method, and evaluating the rotor offset to obtain single-stage rotor offset;
step 5:
according to the single-stage rotor offset, determining the accumulated offset of a kth stage rotor after an nth stage rotor assembly by utilizing a multistage rotor vector stack projection theory, wherein the accumulated offset is expressed as
[
d
⁢
x
0
-
k
d
⁢
y
0
-
k
]
=
[
1
0
0
0
1
0
]
·
∑
i
=
1
k
⁢
(
∏
j
=
2
i
⁢
S
rj
-
1
⁢
S
xj
-
1
⁢
S
yj
-
1
)
⁢
S
ri
⁡
(
p
i
+
dp
i
)
,
wherein, dx0-k is the accumulated offset of the circle center of a measuring surface of the kth stage rotor in the X-axis direction after nth stage rotor assembly;
dy0-k is the accumulated offset of the circle center of a measuring surface of the nth stage rotor in the Y-axis direction after nth stage rotor assembly;
Sxj-1 is a rotation matrix of a (j-1)th stage rotor/stator rotating for a θxj-1 angle around the X-axis;
Syj-1 is a rotation matrix of the (j-1)th stage rotor/stator rotating for a θyj-1 angle around the Y-axis;
pi is an ideal location vector of the circle center of a radial measuring surface of an ith stage rotor;
dpi is a processing error vector of the circle center of the radial measuring surface of the ith stage rotor;
Srj-1 is a rotation matrix of the (j-1)th stage rotor/stator rotating for a θrj-1 angle around the Z-axis;
and Sr1 is a unit matrix;
step 6:
according to the ISO standard definition of coaxiality, determining the expression of the coaxiality of the nth stage rotor after assembly, wherein the expression of the cross-sectional area S of the contact surface between rotors after assembly and the expression of the cross-sectional inertia moment I of the assembly contact surface between two stages of rotors after assembly are respectively:
coaxiality=max{2√{square root over ((dx0-k)2+(dy0-k)2)},k=1,2, . . . ,n}
S=π*(R2−r2)/4−2*∫0de∫0dθπ*(R2−r2)/4dedθ
I=π*(R4−r4)/64−2*∫0de∫0dθπ*(R4−r4)/64dedθ,
wherein, R is the outer diameter of contact surface, r is a contact surface inner diameter, bending rigidity is EI;
eccentricity de=√{square root over ((dx0-n)2+(dy0-n)2)}, eccentric angle dθ=arctan(dy0-n/dx0-n), multistage rotor tensile rigidity is ES, and E is elastic modulus of the material;
step 7:
determining the amount of unbalance of an nth rotor caused by the offset of each of stage rotor, wherein the expression of the amount of unbalance of the nth rotor is:
[
U
⁢
x
0
-
n
U
⁢
y
0
-
n
]
=
[
m
0
-
n
0
0
0
m
0
-
n
0
]
·
∑
i
=
1
n
⁢
(
∏
j
=
2
i
⁢
S
rj
-
1
⁢
S
xj
-
1
⁢
S
yj
-
1
)
⁢
S
ri
⁡
(
p
i
+
dp
i
)
,
wherein, Ux0-n is the unbalance of a measuring surface of the nth stage rotor in the X-axis direction after assembly;
Uy0-n is the unbalance of the measuring surface of the nth stage rotor in the Y-axis direction after assembly;
m0-n is the mass of the nth stage rotor after assembly;
pi is an ideal location vector of the circle center of a radial measuring surface of the ith stage rotor;
dpi is a processing error vector of the circle center of the radial measuring surface of the ith stage rotor;
step 8, enabling the unbalance of the single stage rotor and the unbalance introduced by the offset of each stage of rotor in the assembly process to be subjected to vector addition to obtain the unbalance of any stage of rotor after multistage rotor assembly;
then, respectively projecting the unbalance of each stage of rotor to two correction surfaces, synthesizing the unbalance according to a dynamic balance formula, and establishing a prediction model of the unbalance of multistage rotor;
and
step 9, establishing an intelligent learning neural network by utilizing the prediction model obtained in step 8, and combining with assembly environment influence factors such as temperature and humidity and simultaneously combining with assembly process influence factors such as the tightening torque and tightening sequence of bolts as well as the elastic modulus, height and contact surface radius of the equipment material;
and finally, combining the prediction model with the intelligent learning neural network to complete large-scale high-speed rotary equipment measuring and neural network learning regulation and control.

558 more elements

Large-scale high-speed rotary equipment measuring and neural network learning regulation and control method and device based on rigidity vector space projection maximization

Life of this patent

Large-scale high-speed rotary equipment measuring and neural network learning regulation and control method and device based on rigidity vector space projection maximization

Life of this patent

Abstract

Claims6