Vibration compensation controller with neural network band-pass filters for bearingless permanent magnet synchronous motor

Abstract

The controller comprises a displacement controller and a rotating speed controller. The displacement controller includes a vibration force compensation control module and a dead-time vibration compensation module. The vibration force compensation control module receives actual displacements and a rotor mechanical angle and outputs corresponding vibration compensation forces. The vibration force compensation control module comprises a first neural network band-pass filter, a second neural network band-pass filter, a third PID controller, and a fourth PID controller. The dead-time vibration compensation module receives a rotor electrical angle and an actual quadrature-direct axis currents and an actual direct axis current and outputs a quadrature-direct axis compensation voltages and a direct axis compensation voltage. The dead-time vibration compensation module consists of a third neural network band-pass filter in a direct axis direction, a fourth neural network band-pass filter in a quadrature axis direction, a sixth PI controller, and a seventh PI controller.

Description

CROSS-REFERENCE TO RELATED APPLICATION

This application is a 371 of international application of PCT application serial no. PCT/CN2021/082326, filed on Mar. 23, 2021, which claims the priority benefit of China application no. 202110195977.3, filed on Feb. 22, 2021. The entirety of each of the above mentioned patent applications is hereby incorporated by reference herein and made a part of this specification.

TECHNICAL FIELD

The present invention relates to the field of control of a bearingless motor, and in particular, to compensation control for vibration of a bearingless permanent magnet synchronous motor by using dead-time compensation control and rotor eccentricity control technologies for the bearingless permanent magnet synchronous motor.

BACKGROUND

Bearingless permanent magnet synchronous motors are a new type of special motors that have high speed and high precision and require no lubrication. Their application prospects in aerospace, chemical manufacturing, semiconductor industry, and other fields in need of special environments increasingly grow. The bearingless permanent magnet synchronous motor is used as a rotary drive motor. Due to problems such as material unevenness, processing errors, and assembly errors, a certain degree of rotor mass eccentricity inevitably exists and a centrifugal excitation force at the same frequency as a rotation speed is produced when the motor is in rotation. Meanwhile, in a control process of the bearingless permanent magnet synchronous motor, a dead time must be set to avoid a short circuit between upper and lower bridge arms of an inverter. The introduction of the dead time causes an increase in current harmonics, and the amplitude of an unbalanced force is further increased, which results in unbalanced vibration of a rotor and affects suspension control precision of the rotor.

Regarding control for unbalanced vibration of the rotor in the bearingless permanent magnet synchronous motor, compensation control is mainly performed against unbalanced vibration caused by rotor mass eccentricity in the prior art, while unbalanced vibration caused by a dead-time effect is rarely concerned. Chinese Patent Publication No. CN104659990A discloses a method for extracting an unbalanced vibration displacement of a bearingless motor through adaptive filtering, which paves the way for the primary condition of vibration compensation control of a bearingless motor. Chinese Patent Publication No. CN105048913A discloses a current compensation-based unbalanced vibration control system for a bearingless asynchronous motor, wherein compensation control for suspension vibration is realized by adjusting a compensation current. However, in these solutions, the vibration compensation control of the bearingless motor mainly focuses on detection and compensation for vibration caused by eccentricity and does not concern vibration caused by the dead-time effect. To improve the control precision of the unbalanced vibration displacement of the bearingless permanent magnet synchronous motor, compensation not only needs to be made for a rotor eccentricity displacement caused by rotor mass eccentricity, but also needs to be made for unbalanced vibration of the rotor caused by the dead-time effect, which is critical to the implementation of high-precision control of the bearingless permanent magnet synchronous motor.

SUMMARY

An objective of the present invention is to provide a vibration compensation controller with neural network band-pass filters for a bearingless permanent magnet synchronous motor. The controller performs vibration compensation to suppress vibration of the bearingless permanent magnet synchronous motor, thereby solving the problem in the prior art that compensation is only made for vibration caused by rotor mass eccentricity and vibration caused by a dead-time effect is ignored in the vibration compensation control of the bearingless permanent magnet synchronous motor. Therefore, stable suspension and efficient operation of a rotor of the motor are realized, the control precision of the motor is improved, and better application in an electric drive system is achieved.

The vibration compensation controller with neural network band-pass filters for a bearingless permanent magnet synchronous motor provided by the present invention adopts the following technical solution. The controller comprises a displacement controller and a rotating speed controller. The displacement controller includes a vibration force compensation control module and a dead-time vibration compensation module.

The vibration force compensation control module receives, as input, actual displacements x, y in x and y directions and a rotor mechanical angle θ_m and outputs corresponding vibration compensation forces F_xh and F_yh. The vibration force compensation control module comprises a first neural network band-pass filter, a second neural network band-pass filter, a third proportional-integral-derivative (PID) controller, and a fourth PID controller. The first neural network band-pass filter receives, as input, the actual displacement x in the x direction and the rotor mechanical angle θ_m and outputs a vibration displacement {circumflex over (x)}. A difference between a specified value 0 and the vibration displacement {circumflex over (x)} is input to the third PID controller, and the third PID controller outputs the vibration compensation force F_xh. The second neural network band-pass filter receives, as input, the actual displacement y in the y direction and the rotor mechanical angle θ_m and outputs a vibration displacement ŷ. A difference between the specified value 0 and the vibration displacement ŷ is input to the fourth PID controller, and the fourth PID controller outputs the vibration compensation force F_yh. A sum of the vibration compensation force F_xh and a specified force value F_x of a suspension winding in the x direction is input to a force/current conversion module, a sum of the vibration compensation force F_yh and a specified force value F_y of the suspension winding in the y direction is input to the force/current conversion module, and the current conversion module obtains a specified quadrature axis current value i*_Bq and a specified direct axis current value i*_Bd.

The dead-time vibration compensation module receives, as input, a rotor electrical angle θ_e, and an actual quadrature axis current i_Bq, and an actual direct axis current i_Bd and outputs a quadrature axis compensation voltage u_Bqh and a direct axis compensation voltage u_Bdh. The dead-time vibration compensation module comprises a third neural network band-pass filter in a direct axis direction, a fourth neural network band-pass filter in a quadrature axis direction, a sixth proportional-integral (PI) controller, and a seventh PI controller. The third neural network band-pass filter receives, as input, the actual current i_Bd in the direct axis direction and 6 times of the rotor electrical angle θ_e and obtains a harmonic current î_Bd in the direct axis direction. A difference between the specified value 0 and the harmonic current î_Bd is input to the sixth PI controller, and the sixth PI controller obtains the direct axis compensation voltage u_Bdh. A sum of a control voltage u_Bd in the direct axis direction and the direct axis compensation voltage u_Bdh serves as a direct axis command voltage u*_Bd. The fourth neural network band-pass filter receives, as input, the actual current i_Bq in the quadrature axis direction and 6 times of the rotor electrical angle θ_e and obtains a harmonic current Î_Bq in the direct axis direction. A difference between the specified value 0 and the harmonic current Î_Bq is input to the seventh PI controller, and the seventh PI controller obtains the quadrature axis compensation voltage u_Bqh. A sum of a control voltage u_Bq in the quadrature axis direction and the quadrature axis compensation voltage u_Bqh serves as a quadrature axis command voltage u*_Bq.

The present invention has the following beneficial effects:

1) By adopting dead-time vibration compensation control, the present invention not only compensates for the dead time, but also effectively suppresses vibration during the operation of the bearingless permanent magnet synchronous motor, thereby improving the suspension control precision.

2) The neural network band-pass filters adopted by the present invention have simple working principles and concise calculation processes and can obtain required signals according to real-time speeds of the motor.

3) The present invention adopts the PI controllers to regulate vibration. The controllers have simple principles, their coefficients can be adjusted conveniently, and they have strong robustness.

4) In the vibration compensation control of the bearingless permanent magnet synchronous motor, generally only the vibration caused by eccentricity is considered and compensation control is implemented, while the vibration caused by the dead-time effect is not concerned, which affects the entire suspension control precision. To achieve higher suspension control precision of the bearingless permanent magnet synchronous motor, the present invention not only performs analysis and compensation for vibration caused by eccentricity, but also performs compensation control for vibration caused by the dead-time effect, thereby effectively improving the suspension control precision.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to make the content of the present invention more obvious and understandable, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

FIG. 1 is a schematic block diagram of the structure of the present invention.

FIG. 2 is a schematic block diagram of the structure of a rotating speed controller 2 in FIG. 1.

FIG. 3 is a schematic block diagram of the structure of a displacement controller 1 in FIG. 1.

FIG. 4 is a schematic block diagram of a vibration force compensation module 5 in x and y directions in FIG. 3.

FIG. 5 is a schematic block diagram of a dead-time vibration compensation module 6 in direct and quadrature axis directions in FIG. 3.

FIG. 6 is a schematic block diagram of the internal structure of a first neural network band-pass filter 51 in FIG. 4.

FIG. 7 is a schematic block diagram of the internal structure of a second neural network band-pass filter 53 in FIG. 5.

FIG. 8 is a schematic block diagram of the internal structure of a third neural network band-pass filter 61 in FIG. 6.

FIG. 9 is a schematic block diagram of the internal structure of a fourth neural network band-pass filter 63 in FIG. 7.

FIG. 10 is a schematic block diagram of the overall structure of a motor vibration compensation controller according to the present invention.

DESCRIPTION OF THE EMBODIMENTS

The specific ideas and implementation steps of the present invention are illustrated below.

Referring to FIG. 1, the vibration compensation controller with neural network band-pass filters for a bearingless permanent magnet synchronous motor of the present invention comprises a displacement controller 1 and a rotating speed controller 2. Output ends of the displacement controller 1 and the rotating speed controller 2 are connected to a bearingless permanent magnet synchronous motor 3, so as to control the bearingless permanent magnet synchronous motor 3.

As shown in FIG. 2, the rotating speed controller 2 adopts double closed-loop control of speed and current, and comprises a first PI controller 21, a second PI controller 22, a third PI controller 23, a first coordinate transformation module 24, a second coordinate transformation module 25, a first SVPWM inverter 26, a coder 27, and a speed calculation module 28. An output end of the coder 27 is connected to the speed calculation module 28. The coder 27 collects speed pulse signals from a rotating shaft of the bearingless permanent magnet synchronous motor 3, performs an accumulate operation, and inputs an accumulation result ΔP into the speed calculation module 28. The speed calculation module 28 calculates an actual rotor speed n of the motor, and the speed n is calculated by the following formula:

$\begin{array}{cc}n=\frac{60\Delta P}{T_s{L}_e}& \left(1\right)\end{array}$

wherein T_s is an interrupt cycle of the rotating speed controller 2 and L_e is the number of lines of the coder.

A difference between the calculated actual speed n and a specified speed value n* serves as a speed error, and the error is input to the first PI controller 21. The first PI controller 21 makes adjustment to obtain a specified quadrature axis current value i*_Mq of a torque winding. Meanwhile, a current sensor collects torque currents i_2A and i_2C of the two-phase torque winding of the bearingless permanent magnet synchronous motor 3, and inputs the torque currents i_2A and i_2C to the second coordinate transformation module 25. The second coordinate transformation module 25 is configured for performing Clarke transform and Park transform. The second coordinate transformation module 25 transforms i_2A and i_2C into an actual quadrature axis current value i_Mq of the torque winding and an actual direct axis current value i_Md of the torque winding in a rotating reference frame. An error between the specified quadrature axis current value i*_Mq the torque winding and the actual quadrature axis current value i_Mq of the torque winding is input to the second PI controller 22 to obtain a specified quadrature axis voltage value u*_Mq of the torque winding. When a specified direct axis current value of the torque winding is i*_Md=0, an error between i*_Md and the actual direct axis current value i_Md of the torque winding is input to the third PI controller 23 to obtain a specified direct axis voltage value u*_Md of the torque winding. Output ends of the second PI controller 22 and the third PI controller 23 are both connected to an input end of the first coordinate transformation module 24. The first coordinate transformation module 24 is configured for performing inverse Park transform, through which the specified quadrature axis voltage value u*_Mq of the torque winding and the specified direct axis voltage value u*_Md of the torque winding can be transformed into voltages u_Mα and u_Mβ of the torque winding in a stationary reference frame. An output end of the first coordinate transformation module 24 is sequentially connected in series with the first SVPWM inverter 26 and the bearingless permanent magnet synchronous motor 3. The first coordinate transformation module 24 inputs the voltages u_Mα and u_Mβ to the first SVPWM inverter 26. An output of the first SVPWM inverter 26 is connected to an input of the bearingless permanent magnet synchronous motor 3. The first SVPWM inverter 26 obtains three-phase input voltages u_2A, u_2B, and u_2C of the bearingless permanent magnet synchronous motor 3.

As shown in FIG. 3, the displacement controller 1 adopts double closed-loop control of displacement and current, and comprises a first PID controller 11, a second PID controller 12, a vibration force compensation module 5, a force/current conversion module 13, a fourth PI controller 14, a fifth PI controller 15, a dead-time vibration compensation module 6, a third coordinate transformation module 16, an angle calculation module 17, a second SVPWM inverter 90, a fourth coordinate transformation module 91, a displacement calculation module 92, and the coder 27. A displacement sensor collects and inputs a rotor position of the bearingless permanent magnet synchronous motor 3 to the displacement calculation module 92. The displacement calculation module 92 converts a collected displacement signal into actual displacements in x and y directions, obtains a displacement error as a difference between the actual displacement x in the x direction and a specified value x* and inputs the error to the first PID controller 11. The first PID controller 11 makes adjustment to obtain a specified force value F_x of a suspension winding in the x direction. The displacement calculation module 92 obtains a displacement error as a difference between the actual displacement y in the y direction and a specified value y* and inputs the error to the second PID controller 82. The second PID controller 12 makes adjustment to obtain a specified force value F_y of the suspension winding in the y direction.

The output end of the coder 27 is further connected to the angle calculation module 17. A pulse signal output by the coder 27 is input to the angle calculation module 17 to obtain a rotor mechanical angle θ_m. The rotor mechanical angle at a moment k is calculated as follows:

$\begin{array}{cc}{\theta }_m\left(k\right)={\theta }_m\left(k-1\right)+\frac{60\Delta P}{9.55L_e}& \left(2\right)\end{array}$ wherein ΔP is the accumulation result of pulses output by the coder 27.

Output ends of the angle calculation module 17 and the displacement calculation module 92 are both connected to an input end of the vibration force compensation control module 5. The rotor mechanical angle θ_m output by the angle calculation module 17 and the actual rotor displacements x, y output by the displacement calculation module 92 are input to the vibration force compensation control module 5 to obtain compensation forces F_xh and F_yh.

As shown in FIG. 4, the vibration force compensation control module 5 comprises a first neural network band-pass filter 51, a second neural network band-pass filter 53, a third PID controller 52, and a fourth PID controller 54. The first neural network band-pass filter 51 receives the displacement in the x direction and the rotor mechanical angle θ_m and outputs a vibration displacement signal {circumflex over (x)}. FIG. 6 shows the specific structure of the first neural network band-pass filter 51 in the x direction, which includes a first weight adjustment module 5. A difference between the actual displacement x and the vibration displacement {circumflex over (x)} output by the first neural network band-pass filter 51 serves as an error signal e_x. The error signal e_x and sine and cosine values of the rotor mechanical angle θ_m are input to the first weight adjustment module 55 to obtain updated weights ω_{x_1} and ω_{x_2} in the x direction. The vibration displacement {circumflex over (x)} output by the first neural network band-pass filter 51 at the moment k is calculated by the following formula:{circumflex over (x)}(k)=ω_{x_1}(k)·cos θ_m(k)+ω_{x_2}(k)·sin θ_m(k)  (3).

The weights ω_{x_1} and ω_{x_2} are calculated by the following formulas:

$\begin{array}{cc}\{\begin{array}{c}{\omega }_{x\_1}\left(k+1\right)={\omega }_{x\_1}\left(k\right)+2{\mu }_1{e}_x\cos {\theta }_m\\ {\omega }_{x\_2}\left(k+1\right)={\omega }_{x\_2}\left(k\right)+2{\mu }_1{e}_x\sin {\theta }_m\end{array}& \left(4\right)\end{array}$

wherein e_x is a component in the x direction after harmonics are filtered out; ω_{x_1} and ω_{x_2} are updated weights in the x direction; μ₁ is a step factor.

Therefore, the vibration displacement {circumflex over (x)} in the x direction is obtained. As shown in FIG. 4, a displacement difference between a specified value 0 and the vibration displacement {circumflex over (x)} is input to the third PID controller 52, and the third PID controller 52 makes adjustment to obtain the vibration compensation force F_xh.

The second neural network band-pass filter 53 is identical to the first neural network band-pass filter 51 in structure and principle. Likewise, the displacement in the y direction and the rotor mechanical angle θ_m are input to the second neural network band-pass filter 53. FIG. 7 shows the specific structure of the second neural network band-pass filter 53 in the y direction. A difference between the actual displacement y and a vibration displacement ŷ output by the second neural network band-pass filter 53 serves as an error signal e_y. The error signal e_y and the sine and cosine values of the rotor mechanical angle θ_m are input to a second weight adjustment module 56 to obtain updated weights ω_{y_1} and ω_{y_2} in the y direction. The vibration displacement signal ŷ output by the second neural network band-pass filter 53 at the moment k is calculated by the following formula:ŷ(k)=ω_{y_1}(k)·cos θ_m(k)+ω_{y_2}(k)·sin θ_m(k)  (5).

The weights ω_{y_1} and ω_{y_2} are calculated by the following formulas:

$\begin{array}{cc}\{\begin{array}{c}{\omega }_{y\_1}\left(k+1\right)={\omega }_{y\_1}\left(k\right)+2{\mu }_1{e}_y\cos {\theta }_m\\ {\omega }_{y\_2}\left(k+1\right)={\omega }_{y\_2}\left(k\right)+2{\mu }_1{e}_y\sin {\theta }_m\end{array}& \left(6\right)\end{array}$

wherein e_y is a component in the y direction after harmonics are filtered out; ω_{y_1} and ω_{y_2} are updated weights in the y direction; μ₁ is the step factor.

Therefore, the vibration displacement signal ŷ in the y direction is obtained. As shown in FIG. 4, a displacement difference between the specified value 0 and the vibration displacement signal ŷ is input to the fourth PID controller 54, and the fourth PID controller 54 makes adjustment to obtain the vibration compensation force F_yh.

A sum of the force F_x in the x direction output by the first PID controller 11 and the vibration compensation force F_xh in the x direction output by the vibration force compensation module 5 and a sum of the force F_y in the y direction output by the second PID controller 12 and the vibration compensation force F_yh in the y direction output by the vibration force compensation module 5 are input to the force/current conversion module 13 to obtain a specified quadrature axis current value i*_Bq and a specified direct axis current value i*_Bd of the suspension winding.

Differences between the obtained a specified quadrature axis current value i*_Bq, and a specified direct axis current value i*_Bd and an actual quadrature axis current value i_Bq, an actual direct axis current value i_Bd of the suspension winding are obtained respectively. The current sensor collects currents i_1A and i_1C of the two-phase suspension winding of the bearingless permanent magnet synchronous motor 3 and inputs the collected currents to the fourth coordinate transformation module 91. The fourth coordinate transformation module 91 is configured for performing Clarke transform and Park transform. The fourth coordinate transformation module 91 processes i_1A and i_1C to obtain the actual quadrature axis current i_Bq and the actual direct axis current i_Bd of the suspension winding. The difference between i*_Bq and i_Bq is input to the fifth PI controller 15 to obtain a quadrature axis control voltage u_Bq of the suspension winding. The difference between i*_Bd and i_Bd is input to the fourth PI controller 14 to obtain a direct axis control voltage u_Bd of the suspension winding.

A rotor electrical angle θ_e, the actual quadrature axis current value i_Bq of the suspension winding, and the actual direct axis current value i_Bd of the suspension winding are input to the dead-time vibration compensation module 6 to obtain compensation voltages u_Bqh and u_Bdh. The angle calculation module 17 processes the pulse signal, collected by the coder 27, of the bearingless permanent magnet synchronous motor 3 to obtain the rotor electrical angle θ_e, which is calculated as follows:θ_e(k)=P_Mθ_m(k)  (7)

wherein θ_m(k) is the rotor mechanical angle at the moment k according to the formula (2) and P_M is the number of pole-pairs of the torque winding.

The obtained rotor electrical angle θ_e and the actual quadrature axis current value i_Bq and the actual direct axis current value i_Bd are input to the dead-time vibration compensation module 6. The dead-time vibration compensation module 6 comprises a third neural network band-pass filter 61 in the direct axis direction, a fourth neural network band-pass filter 63 in the quadrature axis direction, a sixth PI controller 62, and a seventh PI controller 64. In the dead-time vibration compensation module 6, compensations in the direct axis direction and the quadrature axis direction are shown in FIG. 5. The current i_Bd in the direct axis direction and 6 times of the rotor electrical angle θ_e are input to the third neural network band-pass filter 61 in the direct axis direction to obtain a harmonic current signal î_Bd in the direct axis direction. FIG. 8 is a schematic diagram of the internal structure of the third neural network band-pass filter 61 in the direct axis direction, which includes a third weight adjustment module 65. In FIG. 8, a difference between the current i_Bd in the direct axis direction and the harmonic current signal î_Bd output by the third neural network band-pass filter 61 serves as an error signal e_Bd; the error signal e_Bd and sine and cosine values of 6 times of the rotor electrical angle θ_e are input to the third weight adjustment module 65 to obtain updated weights ω_{d6_1} and ω_{d6_2} in the direct axis direction. The harmonic current î_Bd output by the third neural network band-pass filter 61 at the moment k is calculated by the following formula:î_Bd(k)=ω_{d6_1}(k)·cos 6θ_e(k)+ω_{d6_2}(k)·sin 6θ_e(k)  (8).

The weights ω_{d6_1} and ω_{d6_2} are calculated by the following formulas:

$\begin{array}{cc}\{\begin{array}{c}{\omega }_{d6\_1}\left(k+1\right)={\omega }_{d6\_1}\left(k\right)+2{\mu }_2{e}_{Bd}\cos 6{\theta }_e\\ {\omega }_{d6\_2}\left(k+1\right)={\omega }_{d6\_2}\left(k\right)+2{\mu }_2{e}_{Bd}\sin 6{\theta }_e\end{array}& \left(9\right)\end{array}$

wherein e_Bd is a component in the direct axis direction after harmonics are filtered out; ω_{d6_1} and ω_{d6_2} are updated sixth-harmonic weights in the direct axis direction; μ₂ is a step factor.

Therefore, the harmonic current signal î_Bd in the direct axis direction is obtained. As shown in FIG. 5, a difference between the specified value 0 and the harmonic current signal î_Bd is input to the sixth PI controller 62, and the sixth PI controller 62 makes adjustment to obtain the direct axis compensation voltage u_Bdh.

The fourth neural network band-pass filter 63 in the quadrature axis direction is identical to the third neural network band-pass filter 61 in structure. Likewise, the current i_Bq in the quadrature axis direction and 6 times of the rotor electrical angle θ_e are input to the fourth neural network band-pass filter 63 in the quadrature axis direction to obtain a harmonic current signal î_Bq in the direct axis direction. FIG. 9 is a schematic diagram of the internal structure of the fourth neural network band-pass filter 63 in the quadrature axis direction, which includes a fourth weight adjustment module 66. In FIG. 9, a difference between the current i_Bq in the quadrature axis direction and the harmonic current signal î_Bq output by the fourth neural network band-pass filter 63 serves as a current error signal e_Bq; the current error signal e_Bq and the sine and cosine values of 6 times of the rotor electrical angle θ_e are input to the fourth weight adjustment module 66 to obtain updated weights ω_{g6_1} and ω_{q6_2} in the direct axis direction. The harmonic current î_Bq output by the fourth neural network band-pass filter 63 at the moment k is calculated by the following formula:î_Bq(k)=ω_{q6_1}(k)·cos 6θ_e(k)+ω_{q6_2}(k)·sin 6θ_e(k)  (10).

The weights ω_{d6_1} and ω_{d6_2} are calculated by the following formulas:

$\begin{array}{cc}\{\begin{array}{c}{\omega }_{q6\_1}\left(k+1\right)={\omega }_{q6\_1}\left(k\right)+2{\mu }_2{e}_{Bq}\cos 6{\theta }_e\\ {\omega }_{q6\_2}\left(k+1\right)={\omega }_{q6\_2}\left(k\right)+2{\mu }_2{e}_{Bq}\sin 6{\theta }_e\end{array}& \left(11\right)\end{array}$

wherein e_Bq is a component in the quadrature axis direction after harmonics are filtered out; ω_{g6_1} and ω_{q6_2} are updated sixth-harmonic weights in the quadrature axis direction; μ₂ is the step factor.

Therefore, the harmonic current signal î_Bq in the direct axis direction is obtained. As shown in FIG. 5, a difference between the specified value 0 and the harmonic current signal î_Bq is input to the seventh PI controller 64, and the seventh PI controller 64 makes adjustment to obtain the quadrature axis compensation voltage u_Bqh.

A sum of the direct axis voltage u_Bd output by the fourth PI controller 14 and the direct axis compensation voltage u_Bdh output by the dead-time vibration compensation module serves as a direct axis command voltage u*_Bd. A sum of the quadrature axis voltage u_Bq output by the fifth PI controller 15 and the quadrature axis compensation voltage u_Bqh output by the dead-time vibration compensation module serves as a quadrature axis command voltage u*_Bq. The obtained u*_Bd and u*_Bq are input to the third coordinate transformation module 16. The third coordinate transformation module 16 is configured for performing inverse Park transform. The third coordinate transformation module 16 processes u*_Bd and u*_Bq to obtain voltages u_Bα and u_Bβ of the suspension winding in the stationary reference frame.

The voltages u_Bα and u_Bβ of the suspension winding are input to the second SVPWM inverter 90. An output of the second SVPWM inverter 90 is connected to the input of the bearingless permanent magnet synchronous motor 3. The second SVPWM inverter 90 obtains three-phase input voltages u_1A, u_1B, u_1C of the bearingless permanent magnet synchronous motor 3.

FIG. 10 is a schematic block diagram of the overall structure of the motor vibration compensation controller according to the present invention. Closed-loop control of speed and vibration compensation control are realized through design of the displacement controller 1, the rotating speed controller 2, and the modules therein as well as parameter adjustment of closed-loop regulators for speed and position. The rotating speed controller 2 performs speed regulation and control by using a commonly used vector control method with a direct axis command current of 0. The displacement controller 1 completes vector control through displacement adjustment, so that the rotor of the bearingless permanent magnet synchronous motor 3 maintains stable operation. The vibration force compensation module 5 performs compensation control on the eccentricity vibration signal in the displacement signals. Meanwhile, the dead-time vibration compensation control module 6 compensates for high-order harmonic signals in the current caused by the dead-time effect, thereby achieving more precise vibration compensation control.

The present invention can be implemented based on the above descriptions. Other changes and modifications made by persons skilled in the art without departing from the spirit and protection scope of the present invention still fall within the protection scope of the present invention.

Claims

1. A vibration compensation controller with neural network band-pass filters for a bearingless permanent magnet synchronous motor, comprising a displacement controller and a rotating speed controller, wherein the displacement controller comprises a vibration force compensation control module and a dead-time vibration compensation module;the vibration force compensation control module receives, as input, actual displacements x, y in x and y directions and a rotor mechanical angle θm and outputs corresponding vibration compensation forces Fxh and Fyh,the vibration force compensation control module comprises a first neural network band-pass filter, a second neural network band-pass filter, a third proportional-integral-derivative (PID) controller, and a fourth PID controller,wherein the first neural network band-pass filter receives, as input, the actual displacement x in the x direction and the rotor mechanical angle θm and outputs a vibration displacement {circumflex over (x)},a difference between a specified value 0 and the vibration displacement {circumflex over (x)} is input to the third PID controller, and the third PID controller outputs the vibration compensation force Fxh,the second neural network band-pass filter receives the actual displacement y in the y direction and the rotor mechanical angle θm and outputs a vibration displacement ŷ,a difference between the specified value 0 and the vibration displacement ŷ is input to the fourth PID controller, and the fourth PID controller outputs the vibration compensation force Fyh,a sum of the vibration compensation force Fxh and a specified force value Fx of a suspension winding in the x direction is input to a force/current conversion module,a sum of the vibration compensation force Fyh and a specified force value Fy of the suspension winding in the y direction is input to the force/current conversion module, andthe current conversion module obtains a specified quadrature axis current value i*Bq and a specified direct axis current value i*Bd;the dead-time vibration compensation module receives, as input, a rotor electrical angle θe, and an actual quadrature axis current iBq, and an actual direct axis currents iBd and outputs a quadrature axis compensation voltages uBqh and a direct axis compensation voltages uBdh,the dead-time vibration compensation module comprises a third neural network band-pass filter in a direct axis direction, a fourth neural network band-pass filter in a quadrature axis direction, a sixth proportional-integral (PI) controller, and a seventh PI controller,wherein the third neural network band-pass filter receives, as input, the actual direct axis current iBd in the direct axis direction and 6 times of the rotor electrical angle θe and obtains a harmonic current îBd in the direct axis direction,a difference between the specified value 0 and the harmonic current îBd is input to the sixth PI controller, and the sixth PI controller obtains the direct axis compensation voltage uBdh,a sum of a control voltage uBd in the direct axis direction and the direct axis compensation voltage uBdh serves as a direct axis command voltage u*Bd,the fourth neural network band-pass filter receives actual quadrature axis current iBq in the quadrature axis direction and 6 times of the rotor electrical angle θe, and obtains a harmonic current îBq in the quadrature axis direction,a difference between the specified value 0 and the harmonic current îBq is input to the seventh PI controller, and the seventh PI controller obtains the quadrature axis compensation voltage uBqh, anda sum of a control voltage uBq in the quadrature axis direction and the quadrature axis compensation voltage uBqh serves as a quadrature axis command voltage u*Bq.

2. The vibration compensation controller with the neural network band-pass filters for the bearingless permanent magnet synchronous motor according to claim 1, wherein the first neural network band-pass filter in the x direction comprises a first weight adjustment module,a difference between the actual displacement x and the vibration displacement {circumflex over (x)} serves as an error ex,the error ex and sine and cosine values of the rotor mechanical angle θm are input to the first weight adjustment module to obtain updated weights ωx_1 and ωx_2 in the x direction, andthe second neural network band-pass filter is identical to the first neural network band-pass filter.

3. The vibration compensation controller with the neural network band-pass filters for the bearingless permanent magnet synchronous motor according to claim 2, wherein the vibration displacement at a moment k is {circumflex over (x)}(k)=ωx_1(k)·cos θm(k)+ωx_2(k) sin θm (k), wherein ωx_1(k+1)=ωx_1(k)+2μ1ex cos θm, ωx_2(k+1)=ωx_2(k)+2μ1ex sin θm, ex is a component in the x direction after harmonics are filtered out, ωx_1 and ωx_2 are the updated weights in the x direction, and μ1 is a step factor.

4. The vibration compensation controller with the neural network band-pass filters for the bearingless permanent magnet synchronous motor according to claim 1, wherein the third neural network band-pass filter comprises a third weight adjustment module,a difference between the current iBd in the direct axis direction and the harmonic current signal îBd output by the third neural network band-pass filter serves as a current error eBd;the current error eBd and sine and cosine values of 6 times of the rotor electrical angle θe are input to the third weight adjustment module to obtain updated weights ωd6_1 and ωd6_2 in the direct axis direction,the third neural network band-pass filter outputs the harmonic current îBd, andthe fourth neural network band-pass filter is identical to the third neural network band-pass filter in structure and outputs the harmonic current îBq.

5. The vibration compensation controller with the neural network band-pass filters for the bearingless permanent magnet synchronous motor according to claim 4, wherein the harmonic current at the moment k is îBd(k)=ωd6_1(k)·cos 6θe(k)+ωd6_2(k)·sin 6θe(k), wherein ωd6_1(k+1)=ωd6_1(k)+2μ2eBd cos 6θe, ωd6_2(k+1)=ωd6_2(k)+2μ2eBd sin 6θe, eBd is a component in the direct axis direction after harmonics are filtered out, ωd6_1 and ωd6_2 are the updated sixth-harmonic weights in the direct axis direction, and, μ2 is a step factor.

6. The vibration compensation controller with the neural network band-pass filters for the bearingless permanent magnet synchronous motor according to claim 1, wherein a difference between the actual displacement x in the x direction and a specified value x* serves as a displacement error, and the displacement error is input to a first PID controller,the first PID controller makes adjustment to obtain the specified force value Fx of the suspension winding in the x direction,a difference between the actual displacement y in the y direction and a specified value y* serves as a displacement error, and the displacement error is input to a second PID controller, andthe second PID controller makes adjustment to obtain the specified force value Fy of the suspension winding in the y direction.

7. The vibration compensation controller with the neural network band-pass filters for the bearingless permanent magnet synchronous motor according to claim 1, wherein currents i1A and i1C of the two-phase suspension winding of the bearingless permanent magnet synchronous motor are collected and input to a fourth coordinate transformation module, and the fourth coordinate transformation module obtains the actual quadrature axis current iBq and the actual direct axis current iBd of the suspension winding;differences between the specified quadrature axis current value i*Bq, the specified direct axis current value i*Bd and the actual quadrature axis currents iBq, and the actual direct axis iBd are obtained and input to a fourth PI controller and a fifth PI controller, respectively,the fourth PI controller outputs the control voltage uBd in the direct axis direction, andthe fifth PI controller outputs the control voltage uBq in the quadrature axis direction.

8. The vibration compensation controller with the neural network band-pass filters for the bearingless permanent magnet synchronous motor according to claim 1, wherein the quadrature axis command voltage u*Bd and the direct axis command voltage u*Bq are input to a third coordinate transformation module,the third coordinate transformation module outputs voltages uBα, and uBβ of the suspension winding in a stationary reference frame,the voltages uBα, and uBβ of the suspension winding are input to a second space vector pulse width modulation inverter, andthe second SVPWM inverter obtains three-phase input voltages u1A, u1B, u1C of the bearingless permanent magnet synchronous motor.

9. The vibration compensation controller with the neural network band-pass filters for the bearingless permanent magnet synchronous motor according to claim 1, wherein a coder is adopted to collect pulse signals from the bearingless permanent magnet synchronous motor, and an angle calculation module is used to obtain the rotor mechanical angle at the moment

10. The vibration compensation controller with the neural network band-pass filters for the bearingless permanent magnet synchronous motor according to claim 9, wherein the rotor electrical angle at the moment k is θe(k)=PMθm(k), wherein θm(k) is the rotor mechanical angle at the moment k and PM is a number of pole-pairs of a torque winding.