Magnetic bearing rotor vibration compensation method based on variable step-size LMS adaptive filter

Abstract

The invention discloses a magnetic bearing rotor vibration compensation method of a variable step-size LMS adaptive filter, which compensates for the imbalance of a six-pole radial hybrid magnetic bearing rotor, and makes a difference between a displacement difference at a current time n and a compensation signal value at a corresponding time n-1, respectively, to obtain instantaneous values ​​of errors in the X and Y directions at time n, and adopts a quasi-spike function model to calculate a step-size factor, and adaptively adjusts a weight coefficient based on the step-size factor to calculate the compensation signal values ​​at the current time n in the X and Y directions, and when the instantaneous error value is greater than or equal to an error threshold, makes a difference between the displacement difference and the compensation signal value at the current time n, to obtain a rotor vibration compensation difference; the invention adopts a variable step-size LMS algorithm, provides a longer step-size in the initial convergence stage of the algorithm, ensures that the algorithm has a faster convergence speed, and provides a shorter step-size in the convergence completion stage of the algorithm, so that the algorithm maintains better stability in the steady-state stage, ensures control accuracy, and makes the algorithm converge quickly and has low steady-state imbalance.

Description

Technical Field

[0001] The present invention relates to a technology for compensating the unbalanced vibration of a magnetic bearing rotor, and specifically uses an LMS (least mean square) adaptive filter to compensate for the unbalance of a six-pole radial hybrid magnetic bearing rotor. The invention belongs to the field of high-speed and ultra-high-speed electrical transmission and is suitable for aerospace, vacuum technology, mechanical industry, energy, transportation and other fields. Background Art

[0002] Magnetic bearing is the abbreviation of magnetic suspension bearing. It is a new type of bearing that relies on magnetic field force to offset the gravity of the rotor, so that it can be stably suspended in space, and there is no mechanical contact between the rotor and the stator. It is a device that uses electromagnetic force to suspend the rotor without contact and control the rotor. Because its speed is only limited by the rotor material, magnetic bearings have the advantages of no friction, high precision, fast speed, and low noise. It fundamentally changes the traditional support form and is particularly suitable for occasions with very high requirements such as high speed, ultra-cleanliness, and vacuum.

[0003] Magnetic bearings can make the rotor reach speeds of tens of thousands of revolutions, but the main problem caused by high-speed rotation is the imbalance of the rotor itself, that is, the periodic vibration problem. At present, for periodic vibration compensation, a certain control strategy is generally used to compensate for the periodic displacement or exciting force with the same frequency as the rotation speed generated when the rotor of the magnetic bearing system rotates. There are two different compensation schemes: 1. The strategy for compensation for displacement is called minimum displacement compensation, that is, let the rotor rotate around the geometric center; such compensation methods include influence coefficient method, open-loop feedforward method, iterative learning control method, etc., the purpose is to improve the rotor rotation accuracy, such as the application of magnetic suspension processing spindle, etc., but its control current is easy to saturate, making it difficult to increase the speed, hindering the improvement of processing efficiency. 2. Compensation for vibration force is called minimum inertial force compensation, that is, let the rotor rotate around its inertia center. Commonly used methods include notch filter method, frequency domain tracking method, force automatic balancing method, LMS adaptive filter method, etc., the purpose is to reduce the current fluctuation of the magnetic bearing coil, reduce the system unbalanced exciting force response, enhance system stability, and increase the rotor speed.

[0004] An adaptive filter based on a fixed-step-size LMS algorithm was proposed in the paper "Compensation Control of Rotor Mass Eccentricity of Bearingless Switched Reluctance Motor" published in the 21st issue of Volume 31 of the Proceedings of the CSEE on July 25, 2011. It can compensate for the displacement caused by the rotor unbalance vibration in both low-speed and high-speed states of the system, and effectively suppress the radial disturbance generated by the rotor when it rotates at high speed. However, when the step size of the LMS algorithm is fixed, its convergence speed and steady-state misalignment cannot be taken into account at the same time. Therefore, in order to allow the step size factor to be adaptively adjusted during the convergence process, a variable-step-size LMS adaptive filtering algorithm is required. Although the traditional variable-step-size LMS algorithm of the Sigmoid function takes into account the contradiction between the convergence speed and the steady-state error, in the convergence stage of the algorithm, a slight change in the error signal will cause a large change in the step size factor, thereby affecting the steady-state error. Summary of the invention

[0005] The purpose of the present invention is to provide a six-level radial hybrid magnetic bearing rotor vibration compensation method based on a variable step-size LMS adaptive filter, which has better tracking capability and smaller steady-state error, and improves the accuracy of rotor unbalance vibration compensation.

[0006] The technical solution adopted by the variable step size LMS adaptive filter magnetic bearing rotor vibration compensation method of the present invention is: the reference radial displacement x in the X and Y directions * ,y * The difference between the displacement in the X and Y directions is obtained by subtracting the corresponding actual radial displacement. x , d y The speed of the magnetic bearing rotor is input into the trap filter to generate the same frequency X-direction sine signal value x1 (nT), cosine signal value x2 (nT) and Y-direction sine signal value y1 (nT), cosine signal value y2 (nT). The X-direction sine signal value x1 (nT) and cosine signal value x2 (nT) are input into the X-direction variable step size LMS adaptive filter, and the Y-direction sine signal value y1 (nT) and cosine signal value y2 (nT) are input into the Y-direction variable step size LMS adaptive filter, and the following steps are also included:

[0007] Step 1): The displacement difference d at the current time n is x , d y The corresponding compensation signal value q at time n-1 x (n-1), q y (n-1) respectively, and obtain the instantaneous error value e in the X and Y directions at time n x (n), e y (n);

[0008] Step 2): The variable step size LMS adaptive filter in the X direction calculates the compensation signal value q in the X direction based on the sine signal value x1(nT) and cosine signal value x2(nT) in the X direction. x (n) = w1(n)x1(nT) + w2(n)x2(nT). Similarly, the variable step size LMS adaptive filter in the Y direction calculates the compensation signal value q in the Y direction based on the sine signal value y1(nT) and cosine signal value y2(nT) in the Y direction. y (n), w1(n) and w2(n) are weight coefficients;

[0009] Step 3): The variable step size LMS adaptive filters in the X and Y directions process the instantaneous error value e respectively. x (n), e y (n), the instantaneous value of the error e x (n), e y (n) is compared with the error threshold e. x (n), e y When (n) is greater than or equal to e, the displacement difference d x The compensation signal value q in the X direction at time n x (n) is subtracted to obtain the vibration compensation difference of the rotor in the X direction, and the displacement difference d y and the compensation signal value q in the Y direction at time n y (n) is subtracted to obtain the vibration compensation difference of the rotor in the Y direction.

[0010] Further, in step 3), when e x (n), e y When (n) is less than e, no compensation is performed and e x (n), e y (n) Direct output.

[0011] Further, in step 2), the step length factor is calculated using a quasi-spike function model Based on the step size factor μ(n), the weight coefficients w1(n) and w2(n) are adaptively adjusted as follows: e x (n-1) is the instantaneous value of the error at time n-1, E[e x (n)e x (n-1)] is e x (n) and e x (n-1); α is the shape parameter, which controls the shape of the function; β is the amplitude parameter, which controls the amplitude of the function; γ is the convergence rate parameter, which controls the convergence rate of the algorithm; w1(n-1) and w2(n-1) are the instantaneous values ​​of the weight coefficients at time n-1 respectively; x1(n-1) and x2(n-1) are the sine signal value and cosine signal value in the X direction at time n-1 respectively.

[0012] Furthermore, a first-order TS fuzzy inference engine is used to calculate the instantaneous value of the X-direction error e x (n) and the error variation de x (n) = e x (n)-e x (n-1), adjust the amplitude parameter β and the convergence speed parameter γ respectively.

[0013] The advantages of the present invention are:

[0014] 1. The present invention adopts an open-loop feedforward compensation control method. Compared with the closed-loop feedback compensation control, the advantage of this control method is that it does not increase the order of the transfer function of the original control system of the hybrid magnetic suspension bearing, has little impact on the stability of the entire system, and the PID parameters basically do not need to be changed, which can reduce the debugging time.

[0015] 2. The present invention adopts the minimum mean square filter, which has the ability to track periodic signals. It removes interference signals by setting its own notch angular frequency at the interference signal angular frequency position, thereby preserving the complete information of other frequency signals, and can ensure the accurate reliability of system operation. Compared with the traditional vector control method, it eliminates the cumbersome coordinate transformation link and reduces the complexity of system control.

[0016] 3. The present invention adopts a quasi-spider function and a first-order TS fuzzy inference engine, and uses the quasi-spider function to adaptively adjust the step size factor of the adaptive algorithm. Compared with the Sigmoid function and the quasi-spider function, this function has a faster convergence speed, better tracking ability, smaller steady-state error and stronger robustness; the first-order TS fuzzy inference engine is used to adjust the parameters in the quasi-spider function model.

[0017] 4. Adopt variable step size LMS algorithm, use the quasi-spike function to adaptively adjust the step size factor of the adaptive algorithm, provide a longer step size in the early stage of algorithm convergence, ensure the algorithm has a faster convergence speed, and enable the algorithm to quickly transition to the steady state stage; in the stage of algorithm convergence completion, provide a shorter step size, so that the algorithm maintains good stability in the steady state stage. While ensuring the control accuracy, the algorithm converges quickly and the steady-state imbalance is low. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is the equivalent structural block diagram of the composite controlled object;

[0019] Figure 2 It is the control block diagram of the magnetic bearing rotor vibration compensation system of the variable step size LMS adaptive filter;

[0020] In the figure, 1. Six-pole radial hybrid magnetic bearing; 2. Variable step size LMS adaptive filter in X direction; 3. Clark inverse transformation; 4. Current hysteresis three-phase power inverter; 5. Composite controlled object; 6, 7. PID controller; 8. Displacement sensor; 9. Force / current conversion module; 21. Displacement interface circuit module; 22. Variable step size LMS adaptive filter in Y direction; 23. LMS algorithm module; 24. Photoelectric encoder; 25. Notch filter; 26. High-speed motor. DETAILED DESCRIPTION

[0021] like Figure 1 As shown, the composite controlled object 5 is composed of Clark inverse transformation 3, current hysteresis three-phase power inverter 4 and six-pole radial hybrid magnetic bearing 1 connected in series in sequence. The power drive is realized by the current hysteresis three-phase power inverter 4, which is connected in series before the six-pole radial hybrid magnetic bearing 1. The coordinate transformation is realized by Clark inverse transformation 3, which is connected in series before the current hysteresis three-phase power inverter 4. Given radial control current of the six-pole radial hybrid magnetic bearing 1 The Clark inverse transformation 3 transforms it into the expected value of three-phase current Current hysteresis three-phase power inverter 4 tracks the expected value of three-phase current Output three-phase control current i u 、i v 、i w , driving the radial control coil of the six-pole radial hybrid magnetic bearing 1 to generate radial suspension force.

[0022] like Figure 2 As shown, the six-pole radial hybrid magnetic bearing 1 is driven to rotate by a high-speed motor 26. When the six-pole radial hybrid magnetic bearing 1 rotates, an unbalanced periodic vibration with the same frequency as the rotation speed is generated due to high-speed rotation. A displacement sensor 8 is used to detect the radial displacement of the rotor of the six-pole radial hybrid magnetic bearing 1 in the X and Y directions. The spatial rectangular coordinate system x, y, and z in the figure provide the direction in which the displacement sensor 8 detects the radial displacement of the rotor. The radial displacement in the X and Y directions is input to the displacement interface circuit module 21 to obtain the actual radial displacement in the X and Y directions. The actual radial displacement signal includes the vibration signal and displacement signal of the rotor. The reference radial displacement x in the X and Y directions is * ,y * Subtract the corresponding actual radial displacement to obtain the displacement difference d in the X and Y directions x , d y. At the same time, a photoelectric encoder 24 is used to detect the rotation speed ω0 of the rotor of the six-pole radial hybrid magnetic bearing 1, and the rotation speed ω0 is input into the trap 25. The trap 25 generates the X-direction sinusoidal signal value x1(nT) and the Y-direction sinusoidal signal value y1(nT) of the same frequency according to the rotation speed ω0. The X-direction sinusoidal signal value x1(nT) is phase-shifted by 90° to obtain the X-direction cosine signal value x2(nT), and the Y-direction sinusoidal signal value y1(nT) is phase-shifted by 90° to obtain the Y-direction cosine signal value y2(nT). Among them, the expression of the X-direction sinusoidal signal value x1(nT) is: The expression of the X-direction cosine signal value x2(nT) is: A is the amplitude of the vibration displacement with the same frequency in the X direction, n is the current moment, T is the sampling period, is the initial phase angle of the rotor. Similarly, the value of the sinusoidal signal in the Y direction is Cosine signal value in Y direction A1 is the amplitude of the vibration displacement with the same frequency in the Y direction.

[0023] The X-direction sine signal value x1(nT) and cosine signal value x2(nT) are input into the X-direction variable step size LMS adaptive filter 2, and the Y-direction sine signal value y1(nT) and cosine signal value y2(nT) are input into the Y-direction variable step size LMS adaptive filter 22. The X-direction variable step size LMS adaptive filter 2 outputs the X-direction compensation signal value q at time n. x (n), the Y direction variable step size LMS adaptive filter 22 outputs the Y direction compensation signal value q at time n y (n). X-direction compensation signal value q x (n) and Y direction compensation signal value q y (n) are used to compensate for the vibration of the magnetic bearing rotor in the X and Y directions respectively.

[0024] At the current time n, the displacement difference d in the X direction is x and the X-direction compensation signal value q at time n-1 stored in the X-direction variable step size LMS adaptive filter 2 x (n-1) to obtain the instantaneous value of the X-direction error at time n, e x (n). At the current moment n, the displacement difference d in the Y direction y and the Y direction compensation signal value q at time n-1 stored in the Y direction variable step size LMS adaptive filter 22 y (n-1) to obtain the instantaneous value of the Y direction error at time n, e y (n). Instantaneous value of the X-direction error e x (n) is input into the variable step size LMS adaptive filter 2 in the X direction, and the instantaneous error value in the Y direction is e y(n) is input into the Y-direction variable step size LMS adaptive filter 22. The X-direction variable step size LMS adaptive filter 2 and the Y-direction variable step size LMS adaptive filter 22 respectively input the corresponding instantaneous error value e x (n), e y (n) is processed to obtain the X-direction compensation signal value q x (n) and Y direction compensation signal value q y (n). The processing method is described as follows, taking the X-direction variable step size LMS adaptive filter 2 as an example:

[0025] The X-direction variable step-size LMS adaptive filter 2 has a built-in LMS algorithm module 23, and the instantaneous value of the X-direction error e x (n) is input into the LMS algorithm module 23, and the LMS algorithm module 23 converts the instantaneous value of the X-direction error e x (n) is compared with the pre-stored error threshold e. x (n) is greater than or equal to e, that is, when e x When (n)≥e, the X-direction compensation signal value q at time n is used x (n) performs compensation operation, when e x When (n) < e, no compensation operation is performed and the instantaneous value of the X-direction error e x (n) is directly input into the PID controller 6, which calculates the instantaneous value of the X-direction error e x (n) Calculate the force signal F x .

[0026] Similarly, the instantaneous value of the Y direction error e y (n) is input to the LMS adaptive filter 22, when e y When (n)≥e, the Y direction compensation signal value q at time n is used y (n) Perform compensation operation; when e y When (n)<e, the instantaneous value of the error in the Y direction is e y (n) is directly input into the PID controller 7, and the PID controller 7 outputs the force signal F x . The signal F x 、F y The two are input to the force / current conversion module 9 and converted into a given radial control current. Given radial control current Input into the composite controlled object 5.

[0027] Taking the variable step size LMS adaptive filter 2 in the X direction as an example, when the instantaneous value of the X direction error e x When (n)≥e, the method for compensation calculation is:

[0028] First, based on the instantaneous value of the X-direction error ex (n), the step size factor μ(n) is calculated by using the quasi-spike function model, and its expression is:

[0029]

[0030] In the formula, e x (n-1) is the instantaneous value of the error at the previous moment n-1 stored in the LMS algorithm module 23; E[e x (n)e x (n-1)] is e x (n) and e x (n-1) The expected value of the product; α is the shape parameter, which controls the shape of the function; β is the amplitude parameter, which controls the amplitude of the function; γ is the convergence speed parameter, which controls the convergence speed of the algorithm.

[0031] In the early stage of algorithm convergence, the instantaneous error value e x (n) in [1.2×10 - 7 m, 2×10 -7 m), keep the step size factor μ(n) in the range of (0.1~1] to ensure that the algorithm has a faster convergence speed and enables the algorithm to quickly transition to the steady-state stage; when the algorithm converges, the instantaneous error value e x (n) in (1×10 -7 m, 1.2×10 -7 m], and keep the step size factor μ(n) in the range of (0~0.1], so that the algorithm can maintain good stability in the steady state stage.

[0032] The LMS algorithm module 23 converts the instantaneous value of the X-direction error e x (n) is the instantaneous error value e at the previous moment stored in the LMS algorithm module 23 x (n-1) is subtracted to obtain the error change in the X direction de x (n), the calculation formula is:

[0033] de x (n) = e x (n)-e x (n-1).

[0034] The LMS algorithm module 23 has a built-in first-order TS fuzzy inference engine, which calculates the instantaneous value of the X-direction error e x (n) and the error variation de x (n), respectively adjust the amplitude parameter β and the convergence speed parameter γ. The first-order TS fuzzy inference engine consists of an input layer, a fuzzification layer, a fuzzy inference layer and an output layer. The instantaneous value of the X-direction error e x(n) and the error variation de x (n) are two inputs of the first-order TS fuzzy inference engine. The two inputs will have three fuzzy subsets in the fuzzification layer, corresponding to nine fuzzy rules, and each rule corresponds to one output. Due to its own characteristics, the first-order TS fuzzy inference engine omits the defuzzification step. The final output parameter is the weighted average of the outputs of each fuzzy rule according to its corresponding weight. Since the smaller the value of the amplitude parameter β, the smaller the step factor μ(n), the fuzzy control rules shown in Table 1 below of the first-order TS fuzzy inference engine obtain the amplitude parameter β and the convergence speed parameter γ:

[0035] Table 1

[0036]

[0037] When the instantaneous error value e x (n) in (1×10 -7 m, 1.1×10 -7 m) range, and the error change de x When (n) is zero or positive, or when the instantaneous value of the error e x (n) in [1.1×10 -7 m, 1.3×10 -7 m) range, and the error change de x When (n) is a negative number, the amplitude parameter β takes values ​​in the range of (0.005, 0.05), and the convergence speed parameter γ takes values ​​in the range of (0.05, 0.1). x (n) in [1.1×10 -7 m, 1.3×10 -7 m) range, and the error change de x When (n) is zero, the amplitude parameter β takes a value in the range of [0.05, 0.1), and the convergence speed parameter γ takes a value in the range of [0.1, 0.2). x (n) in [1.1×10 -7 m, 1.3×10 -7 m) range, and the error change de x When (n) is positive, the amplitude parameter β takes values ​​in the range of [0.1, 0.5), and the convergence speed parameter γ takes values ​​in the range of [0.1, 0.2). x (n) in [1.3×10 -7 m, 2×10 -7 m) range, and the error change de x When (n) is positive, zero or negative, the amplitude parameter β takes a value in the range of [0.1, 0.5), and the convergence rate parameter γ takes a value in the range of [0.2, 0.4).

[0038] When the first-order TS fuzzy inference engine adjusts the parameter β, the value range of the amplitude parameter β should be (0.005, 0.5). The larger the convergence speed parameter γ is, the faster the algorithm converges. However, when γ is too large, the steady-state error of the algorithm will increase. Therefore, a larger value is selected when a faster convergence speed is required, and a smaller value is selected when the convergence speed is not required. The value range of the convergence speed parameter γ is (0.05, 0.4). The larger the shape parameter α is, the larger the step size factor μ(n) provided by the algorithm at the initial convergence stage is, but the stability of the algorithm decreases in the convergence completion stage; the smaller the shape parameter α is, the more stable the algorithm is in the convergence completion stage, and the convergence speed of the algorithm is slower; considering the balance between stability and convergence speed, the shape parameter α is taken as 4.

[0039] The LMS algorithm adopted by the LMS algorithm module 23 is a gradient steepest descent algorithm, which continuously reduces the mean square error between the output signal and the expected signal by adjusting the weight coefficient. The present invention adopts square error instead of mean square error, reduces the inversion operation in the algorithm, and reduces the difficulty of gradient estimation. Based on the step factor μ(n), the weight coefficients w1(n) and w2(n) at the current moment n are adaptively adjusted through weight coefficient iteration, and the expressions of weight coefficients w1(n) and w2(n) are:

[0040]

[0041] Wherein, w1(n-1) and w2(n-1) are the instantaneous values ​​of the weight coefficients at the last moment stored in the LMS algorithm module 23; x1(n-1) and x2(n-1) respectively represent the X-direction sine signal value and cosine signal value of the reference input signal stored in the LMS algorithm module 23 at the last moment; e x (n-1) is the instantaneous value of the X-direction error at the previous moment stored in the LMS algorithm module 23; μ(n) is the step size factor at the current moment n.

[0042] The X-direction variable step size LMS adaptive filter 2 calculates the compensation signal value q based on the weight coefficients w1(n) and w2(n) and the X-direction sine signal value x1(nT) and cosine signal value x2(nT). x (n) and output:

[0043] q x (n)=w1(n)x1(nT)+w2(n)x2(nT).

[0044] The compensation operation method of the variable step size LMS adaptive filter 22 in the Y direction is similar to the compensation operation process of the variable step size LMS adaptive filter 2 in the X direction: based on the instantaneous value of the error in the Y direction e y(n), the step factor is calculated by using the quasi-spike function model. The amplitude parameter β and the convergence speed parameter γ in the step factor calculation formula are adjusted according to the built-in first-order TS fuzzy inference engine. The shape parameter α in the step factor calculation formula is 4. Based on the step factor, the weight coefficient of the current moment n is adaptively adjusted, and the compensation signal value q is calculated based on the adjusted weight coefficient and the Y-direction sine signal value y1(nT) and cosine signal value y2(nT) y (n) and output.

[0045] The displacement difference d in the X direction x The compensation signal value q in the X direction at time n x (n) is subtracted to obtain the vibration compensation difference of the rotor in the X direction, and the displacement difference d in the Y direction is y and the compensation signal value q in the Y direction at time n y (n) is subtracted to obtain the vibration compensation difference of the rotor in the Y direction, and finally the magnetic bearing rotor is stably suspended in the equilibrium position, thereby achieving the purpose of suppressing or even eliminating the unbalanced vibration of the rotor.

Claims

1. A variable step size LMS adaptive filter magnetic bearing rotor vibration compensation method, the reference radial displacement x in the X and Y directions * ,y * The difference between the displacement in the X and Y directions is obtained by subtracting the corresponding actual radial displacement. x , d y The speed of the magnetic bearing rotor is input into the trap filter to generate the same frequency X-direction sine signal value x1(nT), cosine signal value x2(nT) and Y-direction sine signal value y1(nT), cosine signal value y2(nT). The X-direction sine signal value x1(nT) and cosine signal value x2(nT) are input into the X-direction variable step size LMS adaptive filter, and the Y-direction sine signal value y1(nT) and cosine signal value y2(nT) are input into the Y-direction variable step size LMS adaptive filter, which is characterized by: The following steps are also included: Step 1): The displacement difference d at the current time n is x , d y The corresponding compensation signal value q at time n-1 x (n-1), q y (n-1) respectively, and obtain the instantaneous error value e in the X and Y directions at time n x (n), e y (n); Step 2): The variable step size LMS adaptive filter in the X direction calculates the compensation signal value q in the X direction based on the sine signal value x1(nT) and cosine signal value x2(nT) in the X direction. x (n) = w1(n)x1(nT) + w2(n)x2(nT). Similarly, the variable step size LMS adaptive filter in the Y direction calculates the compensation signal value q in the Y direction based on the sine signal value y1(nT) and cosine signal value y2(nT) in the Y direction. y (n), w1(n) and w2(n) are weight coefficients; The step length factor is calculated using a quasi-spike function model Based on the step size factor μ(n), the weight coefficients w1(n) and w2(n) are adaptively adjusted as follows: e x (n-1) is the instantaneous value of the error at time n-1, E[e x (n)e x (n-1)] is e x (n) and e x (n-1); α is the shape parameter, which controls the shape of the function; β is the amplitude parameter, which controls the amplitude of the function; γ is the convergence speed parameter, which controls the convergence speed of the algorithm; w1(n-1) and w2(n-1) are the instantaneous values ​​of the weight coefficients at time n-1; x1(n-1) and x2(n-1) are the sine signal value and cosine signal value in the X direction at time n-1; The first-order TS fuzzy inference engine is used to calculate the instantaneous value of the X-direction error e x (n) and the error change de x (n) = e x (n)-e x (n-1), adjust the amplitude parameter β and the convergence speed parameter γ respectively; The value range of the amplitude parameter β should be (0.005, 0.5), the value range of the convergence speed parameter γ should be (0.05, 0.4), and the shape parameter α=4; When the instantaneous error value e x (n) in (1×10 -7 m, 1.1×10 -7 m) and the error variation is de x When (n) is zero or positive, or when the instantaneous value of the error e x (n) in [1.1×10 -7 m, 1.3×10 -7 m) and the error variation is de x When (n) is negative, the amplitude parameter β takes values ​​in the range of (0.005, 0.05), and the convergence speed parameter γ takes values ​​in the range of (0.05, 0.1); when the instantaneous error value e x (n) in [1.1×10 -7 m, 1.3×10 -7 m) and the error variation is de x When (n) is zero, the amplitude parameter β takes a value in the range of [0.05, 0.1), and the convergence speed parameter γ takes a value in the range of [0.1, 0.2); when the instantaneous error value e x (n) in [1.1×10 -7 m, 1.3×10 -7 m) and the error variation is de x When (n) is positive, the amplitude parameter β takes values ​​in the range of [0.1, 0.5), and the convergence speed parameter γ takes values ​​in the range of [0.1, 0.2); when the instantaneous error value e x (n) in [1.3×10 -7 m, 2×10 -7 m) and the error variation is de x When (n) is positive, zero or negative, the amplitude parameter β takes a value in the range of [0.1, 0.5), and the convergence speed parameter γ takes a value in the range of [0.2, 0.4); Step 3): The variable step size LMS adaptive filters in the X and Y directions process the instantaneous error value e respectively. x (n), e y (n), the instantaneous value of the error e x (n), e y (n) is compared with the error threshold e. x (n), e y When (n) is greater than or equal to e, the displacement difference d x The compensation signal value q in the X direction at time n x (n) is subtracted to obtain the vibration compensation difference of the rotor in the X direction, and the displacement difference d y and the compensation signal value q in the Y direction at time n y (n) is subtracted to obtain the vibration compensation difference of the rotor in the Y direction.

2. The magnetic bearing rotor vibration compensation method of the variable step size LMS adaptive filter according to claim 1 is characterized in that: When the instantaneous error value e x (n) in [1.2×10 -7 m, 2×10 -7 m), keep the step factor μ(n) in the range of (0.1~1]; when the instantaneous error value e x (n) in (1×10 -7 m, 1.2×10 -7 m), keep the step size factor μ(n) in the range of (0~0.1].

3. The magnetic bearing rotor vibration compensation method of the variable step size LMS adaptive filter according to claim 1 is characterized in that: The first-order TS fuzzy inference engine is composed of an input layer, a fuzzification layer, a fuzzy inference layer and an output layer. x (n) and the error variation de x (n) As the two inputs of the first-order TS fuzzy inference engine, the two inputs have three fuzzy subsets in the fuzzification layer, corresponding to nine fuzzy rules, and each rule corresponds to one output.

4. The magnetic bearing rotor vibration compensation method of the variable step size LMS adaptive filter according to claim 1 is characterized in that: In step 3), when e x (n), e y When (n) is less than e, no compensation is performed and e x (n), e y (n) Direct output.

5. The magnetic bearing rotor vibration compensation method of the variable step size LMS adaptive filter according to claim 1 is characterized in that: Sinusoidal signal value A and A1 are the amplitudes of the same-frequency vibration displacements in the X and Y directions, respectively. n is the current moment, T is the sampling period, ω0 is the speed of the magnetic bearing rotor, is the initial phase angle of the rotor.

Images