JP2008045687A - Magnetic bearing device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a magnetic bearing device capable of suppressing the degradation of stability under the influence of a gyro effect by carrying out linearization without using a bias current for the purpose of reducing power consumption. <P>SOLUTION: The magnetic bearing device carrying out zero bias linear control for the purpose of reducing power consumption includes a linear parameter variation system adding a term depending on rotation frequency, to an equation of motion in order to suppress instability under the influence of the gyro effect. A plurality of controllers to which convex interpolation is performed, are prepared to carry out gain schedule control switching the controllers by rotational frequency. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a magnetic bearing device that is used in a flywheel type power storage device and the like, and a rotating body is supported by a magnetic bearing in a non-contact manner and rotated by an electric motor.

  As this type of magnetic bearing device, the machine body and the controller are connected by a cable, and the machine body is a control-type axial that supports the rotating body in a non-contact manner in the axial and radial directions by the magnetic attractive force of the rotating body and electromagnet. A magnetic bearing and a radial magnetic bearing, a displacement sensor for detecting displacement in the axial direction and radial direction of the rotating body, an electric motor for rotating the rotating body, and a rotation sensor for detecting the rotational speed of the rotating body, A touchdown protective bearing is provided, and a controller controls an electromagnet of a magnetic bearing based on an output of a displacement sensor and controls an electric motor based on an output of a rotation sensor.

  Since magnetic bearings have a strong non-linearity, a linearization method has been used in which linear approximation is performed by Taylor expansion in the vicinity of the equilibrium point in order to perform control by applying linear theory. However, in this linearization method, a bias current must always be supplied in order to constantly float the rotating body (rotor). Further, in the method of supplying the bias current, the rotating body always crosses the magnetic flux during rotation, and the relative magnetic flux fluctuation of the rotating body is generated. For this reason, hysteresis loss and eddy current loss inside the magnetic material increase.

  Also, in the case of a flywheel power storage device, a magnetic bearing device in which a flywheel is attached to a rotating body generates a forward swinging motion and a backward swinging motion due to the gyro effect as the rotational frequency changes. A petal-like whirling occurs in the trajectory of the body, impairing stability.

  The prior art of the zero-bias control magnetic bearing was modeling that did not consider the gyro effect in order to reduce the order of the controller. Therefore, when actually rotating, the gyro effect increases as the speed increases, and the stability of the system deteriorates.

  An object of the present invention is to provide a magnetic bearing device that solves the above problems, performs linearization without using a bias current for the purpose of reducing power consumption, and can suppress deterioration of stability due to the effect of the gyro effect. is there.

  SUMMARY OF THE INVENTION An object of the present invention is to provide a magnetic bearing device that solves the above-mentioned problem, considers the gyro effect from the model base, schedules the gyro compensation gain of the controller according to the rotational speed, and improves the stability of the rotating body. There is to do.

  The magnetic bearing device according to the present invention adds a term depending on the rotational frequency to the equation of motion in order to suppress instability due to the effect of the gyro effect in a magnetic bearing device that performs zero bias linear control for the purpose of reducing power consumption. A linear parameter variation system is included, and a plurality of convexly interpolated controllers are prepared in advance, and gain schedule control for switching the controllers according to the rotation frequency is performed.

  The magnetic bearing has non-linearity is a relational expression between the attractive force and the electric current supplied to the electromagnet. By considering the attractive force as a control input, the magnetic bearing can be regarded as a linear control object. In other words, when the rotating body approaches one electromagnet, the control current of the approaching electromagnet is set to zero, and the control current is supplied only to the opposite electromagnet. Is not to flow.

  Furthermore, the gain schedule controller is designed considering that the model is a linear parameter dependent model in which the system varies depending on the rotation frequency. When implementing the gain schedule controller in the experimental system, a controller is prepared for each rotation speed in advance for the convex interpolation controller for the problem that the calculation process is delayed for discretization for each sampling. A method of switching the controller according to the scheduling parameter is proposed.

  Considering the attractive force as a control input, the magnetic bearing allows only one of the currents to flow through the electromagnets arranged opposite to each other using a linearization method that is a linear control target.

  By adding a term that depends on the rotational frequency to the equation of motion and constructing a linear parameter fluctuation system, the effect of the gyro effect is added to the model.

  When applying gain schedule control to an experiment, several controllers with convex interpolation are prepared in advance, and a method of switching the controllers according to the rotation frequency is applied.

  Zero bias control was realized by switching the current. Furthermore, we proposed gain schedule control using a table for the flywheel / magnetic bearing system. Such a trajectory can be sufficiently suppressed by control.

  The magnetic bearing device according to the present invention is a linear bearing in which a gyro effect term is added to the equation of motion in order to suppress instability due to the effect of the gyro effect in a magnetic bearing device that performs zero bias linear control for the purpose of reducing power consumption. A parameter fluctuation system is included, and gain schedule control for gyro compensation is performed to make the state quantity follow a hyperplane that constantly fluctuates according to the fluctuation parameter.

  A linear parameter fluctuation system was constructed by adding a term of the gyro effect to the equation of motion. In order to control, a gyro-compensated gain-scheduled sliding mode control law was devised to make the state quantity follow a hyperplane that always fluctuates according to the fluctuation parameter.

  Zero bias control was realized by switching the current. Furthermore, when viewed from the root locus, stability was maintained with almost no movement regardless of the rotational speed.

  According to the magnetic bearing device of the present invention, it is possible to prevent deterioration of stability due to the influence of the gyro effect.

  According to the magnetic bearing device of the present invention, the stability of the rotating body can be improved.

  Hereinafter, an embodiment in which the present invention is applied to a five-axis control type magnetic bearing device used in a flywheel power storage device will be described with reference to the drawings.

  FIG. 1 is a block diagram schematically showing the overall configuration of the magnetic bearing device, FIG. 2 is a longitudinal sectional view showing the main part of the mechanical part of the magnetic bearing device, and FIG. 3 is a perspective view showing the main part of FIG. is there.

  The magnetic bearing device includes a machine body (1) and a controller (control means) (2) connected by a cable.

  The magnetic bearing device is a vertical type in which a vertical shaft-like rotating body (rotor) (4) rotates inside a vertical cylindrical casing (3), and the rotating body (4) protruding from the casing (3) A flywheel (4a) is fixed to the upper end of the. In the following description, the control axis (axial control axis) in the vertical axis direction (axial direction) of the rotating body (4) is Z axis, two horizontal radial directions (radial direction) orthogonal to the Z axis and orthogonal to each other. These control axes (radial control axes) are defined as an X axis and a Y axis.

  The machine body (1) has a set of control type axial magnetic bearings (5) that support the rotating body (4) in a non-contact manner in the axial direction, and two sets of upper and lower that support the rotating body (4) in a non-contact manner in the radial direction. Control type radial magnetic bearing (6) (7), displacement detector (8) for detecting axial and radial displacement of the rotating body (4), built-in type for rotating the rotating body (4) at high speed The electric motor (9), the rotation sensor (10) for detecting the number of rotations of the rotating body (4), and the rotating body (4) by restricting the movable range in the axial direction and the radial direction of the rotating body (4) Two sets of upper and lower touch-down protection bearings (11) and (12) that mechanically support the rotating body (4) when not supported by the magnetic bearings (5), (6), and (7) are provided. .

  The controller (2) is provided with a sensor circuit (13) (14), an electromagnet drive circuit (15), an inverter (16) and a DSP board (17), and the DSP board (17) can be software-programmed. A DSP (18), ROM (19) as a digital processing means, a RAM (20) as a nonvolatile storage device, AD converters (21) (22), and DA converters (23) (24) are provided. .

  The displacement detector (8) includes an axial displacement sensor (25) for detecting the axial displacement of the rotating body (4) and an upper and lower position for detecting the radial displacement of the rotating body (4). Two sets of radial displacement sensor units (26) and (27) are provided.

  The axial magnetic bearing (5) is a pair of axial electromagnets (28a) (28b) arranged so as to sandwich a flange portion (4b) formed integrally with the lower part of the rotating body (4) from both sides in the Z-axis direction. It has. The axial electromagnet is generically designated by reference numeral (28).

  The axial displacement sensor (25) is disposed so as to face the lower end surface of the rotating body (4) from the lower side in the Z-axis direction, and is a distance signal proportional to the distance (gap) from the lower end surface of the rotating body (4). Is output.

  The two sets of radial magnetic bearings (6) and (7) are arranged at a predetermined distance in the vertical direction on the upper side of the axial magnetic bearing (5), and the motor (9) is arranged therebetween. . The upper radial magnetic bearing (6) includes a pair of radial electromagnets (29a) (29b) arranged so as to sandwich the rotating body (4) from both sides in the X-axis direction, and the rotating body (4) in the Y-axis direction. A pair of radial electromagnets (29c) and (29d) arranged so as to be sandwiched from both sides of the same. These radial electromagnets are collectively referred to by reference numeral (29). Similarly, the lower radial magnetic bearing (7) also includes two pairs of radial electromagnets (30a) (30b) (30c) (30d). These electromagnets are also collectively referred to by reference numeral (30).

  The upper radial displacement sensor unit (26) is disposed in the vicinity of the upper radial magnetic bearing (6), and is rotated from both sides in the X-axis direction in the vicinity of the electromagnets (29a) and (29b) in the X-axis direction. 4) A pair of radial displacement sensors (31a) and (31b) arranged to sandwich the rotating body (4) from both sides in the Y-axis direction in the vicinity of the Y-axis electromagnets (29c) and (29d) A pair of radial displacement sensors (31c) and (31d) arranged as described above are provided. These radial displacement sensors are collectively referred to by reference numeral (31). Similarly, the lower radial displacement sensor unit (27) is disposed in the vicinity of the lower radial magnetic bearing (7), and two pairs of radial displacement sensors (32a) (32b) (32c) (32d) It has. These radial displacement sensors are also collectively referred to by reference numeral (32). Each radial displacement sensor (31) (32) outputs a distance signal proportional to the distance from the outer peripheral surface of the rotating body (4).

  The electromagnets (28) (29) (30) and the displacement sensors (25) (31) (32) are fixed to the casing (3).

  The protective bearings (11) and (12) are rolling bearings such as angular ball bearings, the outer rings of the respective protective bearings (11) and (12) are fixed to the casing (3), and the inner rings are arranged around the rotating body (4). It is arranged with a gap. The two sets of protective bearings (11) and (12) can both be supported in the radial direction, and at least one set can also be supported in the axial direction. In this example, the upper protective bearing (11) only supports radial support, and the lower protective bearing (12) supports radial support and axial support.

  The ROM (19) of the controller (2) stores a processing program for the DSP (18). The RAM (20) stores control parameters for the magnetic bearing.

  The sensor circuit (13) drives the displacement sensors (25), (31), and (32) of the displacement detector (8), and based on the outputs of the displacement sensors (25), (31), and (32), the rotating body The displacement in the axial direction of (4) and the displacement in the X-axis direction and the Y-axis direction in the upper and lower radial displacement sensor units (26) and (27) are calculated, and the displacement signal as the calculation result is converted into an AD converter ( 21) to the DSP (18). The displacement detector (8) and the sensor circuit (13) constitute a displacement detector, the radial displacement sensor unit (26) (27) of the displacement detector (8) and the radial displacement sensor (31) of the sensor circuit (13). The portion related to (32) constitutes a radial displacement detector.

  The sensor circuit (14) drives the rotation sensor (10), converts the output of the rotation sensor (10) into a rotation speed signal corresponding to the rotation speed of the rotating body (4), and converts this to an AD converter (22) To the DSP (18).

  Based on the displacement signal input from the AD converter (21), the DSP (18) calculates the control current value for the electromagnets (28), (29), (30) of the magnetic bearings (5), (6), and (7). The excitation current signal corresponding to the control current value is obtained and output to the magnetic bearing drive circuit (15) via the DA converter (21). The drive circuit (15) supplies the excitation current based on the excitation current signal from the DSP (18) to the electromagnets (28), (29), (30) of the corresponding magnetic bearings (5), (6), (7). Thereby, the rotating body (4) is brought into non-contact with the predetermined target position. The DSP (18) also converts the current control signal for the motor (9) to the inverter via the DA converter (24) based on the rotation speed signal from the rotation sensor (10) input from the AD converter (22). Based on this signal, the inverter (16) controls the current of the motor (9) to control the rotation speed. As a result, the rotating body (4) is rotated at high speed by the motor (9) while being supported in a non-contact manner at the target position by the magnetic bearings (5), (6) and (7).

  Next, control of the upper and lower radial electromagnets (29) and (30) by the DSP (18) will be described.

Gain schedule H∞ control of flywheel / magnetic bearing system Since magnetic bearings are unstable systems, stabilization by feedback control is an issue. Furthermore, since magnetic bearings have strong non-linearity, in order to perform control by applying linear theory, a linearization method has been adopted in which linear approximation is performed by Taylor expansion near the equilibrium point. . However, this linearization method must always supply a bias current in order to always float the rotor. Further, in the method of supplying a bias current, the rotor always crosses the magnetic flux during rotation, and a relative rotor magnetic flux fluctuation is generated. For this reason, hysteresis loss and eddy current loss inside the magnetic material increase. For this reason, it has the fault that power consumption and a rotation loss are large from a viewpoint of the energy balance in the flywheel for energy storage.

  Therefore, in recent years, a control system that does not use a bias current has been actively studied with the aim of minimizing power consumption in magnetic bearings. In one study, the conventional current is considered as the control input and zero power control is performed, but the control current has a steady deviation. Conventionally, in a magnetic bearing, the control input has been considered as a current. However, magnetic bearings have non-linearity because of the relational expression between the attractive force and the current supplied to the electromagnet. By considering the attractive force as a control input, the magnetic bearing can be regarded as a linear control object. it can.

  In previous studies, robust control theory was applied and its effectiveness was confirmed. In addition, we proposed a zero bias method that considers the attractive force as a control input for the characteristics of nonlinear control type magnetic bearings, and succeeded in reducing power consumption.

  However, in the experiment, the trajectory of the rotor is disturbed, and touchdown occurs near 100 Hz. This is because the flywheel is attached to the rotor for the purpose of increasing the inertial force, so that the forward and rearward touching movements due to the gyro effect and the change in their natural frequencies rotate at high speed. It is thought that this is caused by obstructing.

  Therefore, in this paper, we first modeled the characteristics of the gyro effect. Considering the rotation frequency ω as a constant in the A matrix of the plant, the performance of the H∞ fixed controller is verified by simulation and experiment.

  Next, considering that the system is a linear parameter dependent model (LPV) in which the system fluctuates depending on the rotation frequency, a gain schedule (GS) controller is designed and compared with the H∞ fixed controller by simulation and eigenvalue analysis. When considering the implementation of the GS controller in the experimental system, a controller is prepared in advance for each rotation speed against the controller with convex interpolation for the problem that the discretization for each sampling delays the calculation process. A method of switching the controller according to the GS parameter is proposed. In addition, it is verified from the initial response whether the closed loop stability of the GS controller is impaired by using the table.

  Finally, we verify using a reduced-order model including a bending primary elastic back mode, which is thought to have deteriorated control performance in the experiment. The simulation is considered to be due to the specifications of the actuator for the experimental inconsistency problem, and a closed-loop eigenvalue analysis is performed when an unknown parameter coefficient of α is added to the GS controller.

  This paper discusses the effectiveness of LMI-based GS control for a 0.5 kWh class magnetic bearing flywheel system.

  The controlled object is the magnetic bearing device shown in FIGS.

3. Control Theory 3.1 Concept In H∞ control, a feedback system shown in FIG. 4 is used so that various control problems can be handled in a unified framework. In FIG. 4, G is called a plant and is defined as a transfer function having an input / output signal expressed by the following equation.

The state space realization of G is defined as follows.

Here, w∈R ^m1 is referred to as an external input, and represents an input applied to the control system from the outside such as a reference signal, disturbance, sensor noise, etc., z∈R ^p1 is referred to as a controlled variable, and control deviation, control input, etc. , Represents the amount you want to reduce by control. Further, uεR ^m2 and yεR ^p2 are quantities that become the output and input from the compensator, respectively, in the control input and the observation output. Now, compensator for generalized plant G u = Ky (3.3)
When the feedback control is performed using, the transfer function from w to z is as follows.

z = G _zw w
However,
G _zw : = G ₁₁ + G ₁₂ K (A−G ₂₂ K) ⁻¹ G ₂₁ (3.4)
It is. The control purpose is to keep the control amount z as small as possible with respect to the external input w. Therefore, a compensator K that reduces the size of the transfer function G _zw in any sense may be designed. In H∞ control, an H∞ norm is used as a measure of the magnitude of G _zw . When G _zw is stable, its H∞ norm is written as ‖G _zw ‖∞ and defined as follows.

Here, σ represents the maximum singular value of the matrix and is defined by the following equation.

Where λ _max is the maximum eigenvalue of the matrix,

Is the complex conjugate transpose of the matrix G _zw (jw). At this time, the H∞ norm means the maximum value of the input / output energy ratio.

Based on the above, the H∞ control problem stabilizes the closed loop by performing feedback control with u = Ky on the generalized plant G in FIG. ‖G _zw (s) ‖∞ <γ (3.7)
The compensator K that satisfies As described above, the H∞ control problem is formulated as a very simple one. By changing the configuration of G, that is, by selecting the external input w and the control amount z according to the control purpose, robust control, disturbance It can be applied to various problems such as suppression control, sensitivity optimization, target value tracking, and closed-loop shaping problems. Also, under the definition of equation (3.2) and the following assumptions, the H∞ control problem is called a standard H∞ control problem.

Assumption 1 (A, B ₂ ) is stable, and (C ₂ , A) is detectable assumption 2 D ₁₂ is vertical column full rank, and D ₂₁ is horizontal row full rank assumption 3 G ₁₂ is invariant on the imaginary axis No zeros, ie for all w

Is full rank assumption 4 G ₂₁ has no invariant zero on the imaginary axis, ie, for all w,

Full rank 3.2 Disturbance suppression control and H∞ control Feedback control is not only aimed at stabilizing the controlled object, but also aimed at achieving suppression problems such as disturbance suppression characteristics and target value tracking characteristics. In the H∞ control problem in the previous section, if w is an actual disturbance and z is a physical quantity for which the influence of the disturbance is to be suppressed, the disturbance suppression control itself is obtained. Now, as shown in FIG. 5A, consider the problem of suppressing the disturbance w applied to the input end to be controlled at the output end z. z and u are respectively z = y = P (u + w) and u = Ky (3.8)
It can be expressed. From this, the transfer function from w to z is z = (A−PK) ⁻¹ Pw (3.9)
It is obtained. That is, the disturbance suppression problem in FIG. 5 (a) is as small as possible positive number γ.
‖ (A-PK) ⁻¹ P‖∞ <γ (3.10)
This is a problem of finding a compensator K that satisfies the above. However, in the evaluation of formula (3.10), it means that the gain of the transfer function from w to z (maximum singular value in the case of multiple input / output systems) is kept below γ over all frequency bands from the definition of H∞ norm. In general, this is a strict requirement. However, in practice, there is no disturbance that the frequency spectrum is large in all frequency bands. Therefore, it is sufficient if (A-PK) ⁻¹ P can be reduced only in a frequency band in which the frequency spectrum of the disturbance w is large. Therefore, a transfer function W _M having a sufficiently large gain is introduced only in the frequency band to be reduced, and instead of the equation (3.10),
‖W _M (A-PK) ^-1 P‖∞ <γ (3.11)
And now,
_Mr : = (A-PK) ^-1P (3.12)
When M _r is 1 input / output, from the definition of H∞ norm, equation (3.11) is | W _M (jw) || M _r (jw) | <γ, ∀w (3.13)
Thus, it can be seen that | M _r (jw) | can be made particularly small in a frequency band where | W _M (jw) | is large.

3.3 Target Value Tracking Control and H∞ Control Next, if w is a target value and z is an output signal, target value tracking control itself is obtained. In FIG. 6, the transfer function from w to z is z = (A−PK) ⁻¹ w (3.14)
It is obtained. That is, the target value tracking problem in FIG. 6A is as small as possible positive number γ.
‖ (A-PK) ^-1 ‖∞ <γ (3.15)
This is a problem of finding a compensator K that satisfies the above. However, as in the previous section, in the evaluation of equation (3.15), it is generally a strict requirement to keep the gain of the transfer function from w to z (maximum singular value for multiple input / output systems) below γ over all frequency bands. Become. Therefore, a transfer function W _M having a sufficiently large gain is introduced only in the frequency band to be reduced, and instead of the equation (3.15),
‖W _M (A-PK) ^-1 ‖∞ <γ (3.16)
And now,
M _t : = (A−PK) ⁻¹ (3.17)
When M _t is 1 input / output, the equation (3.16) is expressed as | W _M (jw) || M _t (jw) | <γ, ∀w (3.18)
Thus, it can be seen that | M _t (jw) | can be made particularly small in a frequency band where | W _M (jw) | is large. In general, equation (3.17) is called a sensitivity function.

3.4 Robust Stabilization Problem H∞ control can handle the robust stabilization problem explicitly, because the small gain theorem used to guarantee robust stability is described in the H∞ norm condition. is there.

Theorem 3.1 (Small Gain Theorem)
In FIG. 7, A (s) and B (s) are stable and proper transfer functions. At this time,
‖A (s) B (s) ‖∞ <1 (3.19)
If the condition is satisfied, the closed loop of FIG. 7 becomes stable.

3.5 Robust stabilization against multiplicative errors In Fig. 8 (a), consider the robust stabilization problem of the feedback system. The actual control object P is obtained by using the nominal model P and the multiplicative error Δ for it. P = (I + Δ) P (3.20)
It is expressed. Here, Δ is assumed to be stable for simplicity. Further, in FIG. 8A, when the transfer function from the point a to the point b is obtained, T _m : = (I−PK) ⁻¹ PK (3.21)
It becomes. Here, Equation (3.21) is called a complementary sensitivity function. At this time, since FIG. 8 (a) and FIG. 8 (b) are equivalent, the condition for stability when applying the small gain theorem ‖ΔT _m ‖∞ <1 (3.22)
Get. However, it is impossible to accurately estimate the modeling error. Therefore,
σ {Δ (jw)} <| w _m (jw) |, ∀w (3.23)
By using a known scalar transfer function w _m that satisfies ‖w _m T _m ‖∞ <1 (3.24)
This is a condition for robust stability. Therefore, what is necessary is just to obtain | require the compensator K which comprises the generalized plant of FIG.8 (c) and makes Hinfinity norm from w to z less than one.

3.6 Mixing sensitivity problem Originally, the purpose of feedback control is to achieve robust stabilization while achieving some kind of control performance. In other words, even if only the control performance is considered, the existing modeling error is not guaranteed. Even if only robust stabilization is achieved, it is only a sufficient condition. After all, the design must consider both control performance and robust stabilization. Now consider the block diagram of FIG. Here, w ₁ and w ₂ are external inputs, and z ₁ and z ₂ are control amounts. The transfer function G _z1w2 from w ₂ to z ₁ is G _z1w2 = W _s (s) M _t (s) (3.25)
here,
M _t (s) = (I + PK) ⁻¹ (3.26)
The target value followability is achieved, and equation (3.26) is also a weighted sensitivity function. Decreasing this becomes a problem of sensitivity reduction. In general, the sensitivity is often lowered by weighting the low frequency range. Therefore, the sensitivity reduction problem becomes an evaluation function of the following equation, and becomes a problem of searching for the smallest possible γ.

‖G z1w2 _‖∞ <γ (3.27)
In order to achieve the disturbance suppression characteristic, the transfer function G _z1w1 from w ₁ to z _{1 is used.}
G _z1w1 = W _s (s) M _r (s) (3.28)
here,
M _r (s) = (I + PK) ⁻¹ P (3.29)
However, there is a problem of searching for the smallest possible γ.

‖G z1w1 _‖∞ <γ (3.30)
On the other hand, in order to ensure robust stability, the transfer function G _z2w1 from w ₁ to z ₂ is _expressed as G _z2w1 = W _t (s) T _m (s) (3.31)
here,
T _m (s) = (I + PK) ⁻¹ K (3.32)
So
‖G z2w1 _‖∞ <1 (3.33)
Must be designed to meet Furthermore, the transfer function G _z2w2 from w ₂ to z ₂
G _z2w2 = W _t (s) T _a (s) (3.34)
here,
T _a (s) = (I + PK) ⁻¹ K (3.35)
Even in ‖G _z2w2 ＜ ∞ <1 (3.36)
Must be designed to meet Equation (3.35) is generally called a quasi-complementary sensitivity function. It is a conflicting requirement to satisfy Equation (3.27), Equation (3.30), Equation (3.33), and Equation (3.36) at the same time, and a trade-off must be considered. Actually, the weight W _s (jw) of S may be large in the low frequency range, and W _t (jw) reflecting the uncertainty of the control target may be large in the high frequency range. Therefore, a design guideline for reducing M _t (s) and M _r (s) in the low frequency region and T _m (s) and T _a (s) in the high frequency region is derived. Such a problem is called a mixed sensitivity problem. In combination of formula (3.27), formula (3.30), formula (3.33), formula (3.36)

Thus, a problem arises in designing a compensator K (s) that simultaneously satisfies equation (3.37). Now, if we define the variables w = [w ₁ w ₂ ] ^T and [z ₁ z ₂ ] ^T , the transfer function from w to z is

And instead of equation (3.37)

This is a problem of finding a compensator K (s) that satisfies the above.

3.7 Gain Schedule H∞ Control Based on LMI The formulation of LMI-based GSH∞ control is described in detail in the literature. The purpose of this study is to design a control system based on this formulation. The state space representation of the plant to be controlled and the controller is expressed by the following equation.

Here, x and xk _represent plant and controller state vectors, respectively, and z and y represent controlled variables and measured output vectors. u is a control input, and w is a disturbance input vector. Combining the two equations gives the following closed-loop system.

The closed loop matrices A _cl , B _cl , C _cl , D _cl are

here

Note that the controller matrix is combined into a unit matrix Ω. If the Lyapunov function V (x) = x ^T Px, P> 0 is defined for the closed-loop system of Equation (3.42), global asymptotic stability can be guaranteed. Now, _{L 2} induction norm from w of LTI system to z is ‖z‖ ₂ <γ‖w‖ ₂ (3.45)
And given. In the end, there is a positive Lyapunov function V (x) = x ^T Px that satisfies the following equation.

The H∞ suboptimal control problem is equivalent to the existence of the following inequality solution for X _cl > 0.

Furthermore, the solution of the LMI equation (3.47) is equivalent to the existence of two symmetric matrices R and S that satisfy the following inequality.

Here, _{N R} and _{N S} are each

And the null space of (C ₂ , D ₁₂ ).

The above H∞ control problem is effective only for LTI systems, but can be extended to LPV systems. An LPV plant is represented by a state space model

  Here, x, y, and u represent a state vector, an observation output vector, and a control input vector, respectively. p is a parameter vector of the time-varying plant, and as a result, the plant to be controlled is a function of p. Actually, p is a time-dependent physical parameter such as speed, damping, and rigidity, and is given by the minimum value and the maximum value of the fluctuation range as follows.

_{p min ≦ p i (t)} ≦ p max (3.52)
When p (t) undergoes large fluctuations during control, high performance control performance cannot be obtained with a fixed robust LTI controller.

If the actual value of p (t) is obtained in real time during control, the dynamic characteristics of the controller can be changed in real time depending on the actual value p (t). At this time, the controller is expressed by the following equation.

  Here, y is an observation vector and u is a control input. If the time-varying parameter can be measured accurately, global stability and good control performance can be realized by continuously adjusting the controller so as to follow changes in plant dynamics.

Convex decomposition of current parameter value p (t)

, The values of A _k (p), B _k (p), C _k (p), and D _k (p) are

By the end of the parameter box,

Derived from the value of. In other words, the state space matrix of the controller at the operating point p (t) is the LTI endpoint controller.

Is obtained by convex interpolation. As a result, the controller matrix can be smoothly scheduled based on the actual measurement value p (t) of the time-varying parameter.

Now, the state space representation of the LPV plant is considered in general terms.

The matrices A (•), B (•), C (•), D (•) are functions of the variation parameter p. The B ₂ , C ₂ , D ₁₂ , and D ₂₁ matrices are not affected by p.

For the controller, consider the synthesis problem for the interconnection of FIG. The solution of the H∞ control problem for the LPV plant has the same form as the LTI plant as shown in the following equation.

Here, A _i , B _1i , C _1i , and D _11i are parameter values of A (p), B ₁ (p), C ₁ (p), and D ₁₁ (p) in the polytope parameter p = P _i. is there. Inequalities (3.57) to (3.59) can be solved with a toolbox using a convex optimization algorithm. The structure of the controller matrix Ω obtained from the matrices R and S can also be obtained by the same convex program.

4). Modeling 4.1 Gyro Moment Gyro moment is an action that occurs when a high-speed rotating disk tilts, and is a moment that occurs in a direction perpendicular to the disk tilt. The gyro moment is obtained by the following equation.

  Due to this effect, the natural frequency, which is one when stationary, is divided into a forward swing mode and a backward swing mode, and the variation width is determined by the magnitude of the gyro effect.

  Due to the gyro effect, the rotor becomes a controlled object in which the vibration characteristics change with the rotation speed and the XY directions are coupled to each other for the magnetic bearing, which is a major obstacle in determining the control characteristics of the magnetic bearing. Therefore, it is necessary to accurately obtain in advance how the vibration characteristics of the rotor change with respect to the rotation frequency.

  FIG. 11 shows a natural frequency diagram in which the imaginary part of the polar locus, that is, the change in natural frequency is on the vertical axis and the rotation frequency is on the horizontal axis. From equation (4.1), the gyro moment increases as the rotation speed and the tilt angular velocity increase, and the gyro moment is larger in the secondary mode, which is considered to be the cause of touchdown in the experiment in the previous chapter. Therefore, in order to exceed the critical speed at the intersection of the rigid secondary mode and the transport line, it is necessary to design in consideration of the gyro effect.

4.2 Modeling in consideration of gyro effect The magnetic bearing targeted by this research is a five-axis control type flywheel as shown in FIG. 12, and its parameters are shown in Table 4.1. Since the axial direction is subjected to PID control, modeling is considered only for the radial direction.

The symbols in FIGS. 12 and 13 are:
G: Rotor center-of-gravity position θ _x , θ _y : Rotational angle L _u , L _l around each axis: Distance to upper and lower electromagnets L ₁ , L ₂ : Distance to upper and lower sensors i _{1 to} i ₈ : Control current x _g , x _u , x _l in each electromagnet: displacement at the rotor center of gravity, upper electromagnet, and lower electromagnet position in the XZ plane. Gyro effect

Equation (4.2) is the equation of motion in the translational and rotational directions with respect to the rotor center of gravity.

The output y is

So state variables, inputs and outputs

Then the state space realization is

It is.

5. Control system design and performance verification 1
5.1 H∞ Controller Design In the design, the generalized plant used the modified mixed sensitivity problem of FIG. A virtual disturbance is introduced immediately before P in order to obtain a disturbance suppression characteristic, and a virtual disturbance is introduced immediately after P in order to obtain a target value tracking characteristic. Note that ε is also used to regard the virtual disturbance as observation noise, and is for satisfying the solvability condition of the standard H∞ control problem. The weight function was designed separately because the upper and lower mass proportions were different. The frequency response of the weight function used for the design is shown in FIG. The ε value introduced to satisfy the rank condition was set to 0.04. The γ value is 0.89.

W _t is a frequency weighting function for robustness, and reduces the influence of the control input in the high frequency region. Ws is a frequency weighting function for control performance, and designed to be in the form of a servo system in the low frequency region with the control guideline of fixing the rotor to the origin. The frequency weighting functions for the upper and lower portions of the Y axis are the same as those for the X axis.

  The compensator obtained in the previous section cannot be applied to an experimental machine as it is because the control input is a suction force. That is, conversion from force control input to control current is required based on the switching conditions. In this study, this switching is performed only by positive / negative of the force control input. That is, the two pairs of electromagnets are switched depending on whether the force control input is positive or negative, and the current that reproduces the attractive force f obtained by the controller is calculated backward. Since switching on the upper side and the lower side is considered independently, the upper side will be described in detail.

1) When F _xu > 0 In FIG. 15C, the electromagnet 1 may be attracted. The relational expression between the force and current i ₁ at that time is set as i ₃ = 0

When calculated backward from, the following equation is obtained.

2) When F _xu <0 In FIG. 15 (d), the electromagnet 3 may be attracted. At this time, the relational expression of the current i ₃ is set to i ₁ = 0.

When calculated backward from, the following equation is obtained.

Similarly for the lower side
1) When F _xl > 0

2) When F _xl <0

  Control is performed by converting the force control input into the control current of each electromagnet using this switching condition. The YZ plane is the same as the XZ plane and is as follows.

Upper side
1) When F _yu > 0

2) When F _yu <0

Lower
1) When F _yl > 0

2) When F _yl <0

5.2 Control Current Switching Method The basic method of zero power switching control is to consider the rotor part of the flywheel system as a rigid body, and when the rotor approaches one electromagnet, the control current of the approaching electromagnet is set to zero. When the control current is supplied only to the opposite side and the rotor is completely at the equilibrium point, no current (power) flows. Further, when designing the controller, the control input is assumed as the attractive force between the opposing electromagnets. The actual control current is obtained from the obtained suction force by a switching method.

The resultant force of the magnetic attractive force generated by the two pairs of electromagnets of the upper magnetic bearing and the lower magnetic bearing is _defined as _Fxu and _Fxl , respectively. Since the resultant force of the magnetic attractive force of the electromagnet has non-linearity, it can be treated as a linear control object by considering the attractive force as a control input as in the following equation.

Here, k: attractive force constant i: input current x _g : nominal air gap x: rotor displacement. Looking at equation (5.2), the magnetic attractive force is proportional to the square of the input current and inversely proportional to the square of the distance between the electromagnet and the rotor, and has a very strong nonlinearity. In the conventional method, by applying a bias current to both opposing electromagnets, the relationship between the attractive force and the current can be linearized in a small displacement region. First, the relationship between the input current and the attractive force is as shown by the dotted lines in FIGS. However, F ₁ and F ₂ are the attractive force characteristics of the opposing electromagnets. When a bias current is allowed to flow through both electromagnets, the difference between the attraction forces of the electromagnets becomes linear as shown by the solid line in FIG.

  It becomes easy to design a controller by applying a linear control method based on this linearization. However, even if the rotor floats stably, it is necessary to keep a bias current flowing constantly. The eddy current loss due to the bias current cannot be ignored, causing problems such as rotational resistance and heat generation, resulting in increased power consumption. In addition, since this linearization method performs a first order approximation of Taylor expansion in the vicinity of a general equilibrium point in modeling, a desired control performance can be obtained in the vicinity of the equilibrium point, but the control performance deteriorates when there is a large displacement. There is also a problem.

5.3 Simulation As a preliminary step of the experiment, a simulation using MATLAB Lab Simulink was performed to verify the closed-loop stability of the controller. The simulation was configured by Simulink as shown in FIG. In order to make it closer to the experimental machine system, the force control input is converted to the control current once based on the switching condition and converted back to the force control input. Since the model obtained from the physical parameters regards the control input as an attractive force, it is converted again from the control current to the force control input. The control current is provided with a 1.5 A limiter as in the actual experiment. The sampling period was set to 10 kHz as in the actual experiment, and the compensator was discretized by bilinear transformation. A step target value of 10 μm is added to the upper side of the X axis from 0.01 s. The initial value response at this time is shown in FIG. 18, and the control current of each electromagnet is shown in FIG.

  In FIG. 18A, it can be seen that the overshoot is small, the settling time is fast, and good results are obtained. Moreover, from FIG. 19, when the control input fluctuates positively and negatively, the current of the electromagnet is also switched firmly. Furthermore, no current flows after settling, and it can be seen that zero power control is being performed, which is a good result.

5.4 Experiment The controller used in the simulation was applied to an actual magnetic bearing flywheel, and a rotation experiment up to 93 Hz (5580 rpm) was performed. The experimental system is configured as shown in FIG. 20, the position information is acquired by the sensor, taken into the DSP through the A / D board, the force control input is calculated, and then converted into the control current of each electromagnet based on the switching condition. It is structured to output to the electromagnet through the D / A board. Sampling time is 0.1 ms. Further, information can be acquired from the MATLAB to the DSP in real time, and each sensor information can be displayed on the screen by the figure as shown in FIG. A speed controller manufactured by MyWay Giken is used for the inverter, and a constant acceleration of 512 rpm / s is applied.

  FIGS. 23 to 42 show the locus of the rotor on the upper side and the lower side at the rotation frequency of 0 to 90 Hz and the control current of each electromagnet. FIG. 22 shows an enlarged control current of the upper pair of electromagnets.

  From FIG. 22, it can be confirmed that the control current is switched in the control current. It was also found that a half-wave wave synchronized with the rotation frequency was seen in the control current. This is presumably because a kind of harmonic disturbance is applied by rotation, and the compensator supplies a control current synchronized with the rotation frequency in order to stabilize it.

  The track of the lower rotor is larger than the upper track. This is presumably because the distance from the center of gravity to the lower magnetic bearing is about 1.8 times longer than the upper side, and because of interference with the axial PID compensator.

  The system, which was stable when ascending, increased the lower side swing at a rotation frequency of 93 Hz, resulting in a touchdown. In this model, the locus of the rotor like a petal can be confirmed at rotation frequencies of 60 Hz and 80 Hz. There are second-order forward and backward runouts in the model stiffness mode. From this result, even if coupling is considered in the model, it is considered that the effect of reducing the effect of the gyro effect is small only by the robustness of the fixed controller.

6). Control system design and performance verification 2
6.1 Linear Parameter Dependent Model In the experiment, the gyro effect is remarkably generated near the rotational frequency of 80 Hz. Therefore, it is considered that the system is a linear parameter dependent model (LPV) in which the system varies depending on the rotational frequency. In designing a GS controller using the MATLAB LMI control toolbox, a system matrix A that varies according to linear parameters is decomposed into a time-varying system and a linear time-invariant system (LTI) as follows.

  Here, A22 is a parameter variation matrix including a gyro moment term. It is assumed that the rotation frequency is a scheduling parameter and changes in the following range.

p = f pε [0.1,100] (6.4)
6.2 Gain schedule design design Design the control system for the obtained LPV model using LMI control theory. The optimum controller corresponding to the scheduling parameter is obtained by convex interpolation of the LTI end point controllers (K _max , K _min ) corresponding to the rotation frequencies of 0 Hz and 100 Hz by the equation (6.5).

The frequency response of the GS controller is shown in FIG. The model has 4 inputs and 4 outputs, and there are 16 frequency responses due to consideration of coupling. Here, as an example, the frequency response to the input from the upper side of the X axis is shown. Further, FIG. 43 shows the weighting function used in the design. The center of gravity of the rotor exists on the upper side of the magnetic bearing, and different weights are used on the upper side and the lower side as shown in Equation (6.6) and Equation (6.7) for balance.

  From FIG. 44 (a) and FIG. 44 (b), it can be confirmed that the gain of the controller with respect to the Y-axis output is increased. This means that as the rotational frequency increases, the control input for the influence in the coupling direction increases.

6.3 Simulation 6.3.1 Comparison with H∞ controller In order to verify the closed-loop stability of the GS controller, a simulation was performed with a model that takes into account the gyro effect. The performance comparison of the H∞ controller and the GS controller was performed with the initial value response. In order to make it closer to the experimental machine system, the force control input is once converted to the control current based on the switching condition, and then converted again to the force control input.

  The control current was provided with a 1.5 A limiter as in the experiment, the sampling frequency was 10 kHz as in the actual experiment, and the compensator was discretized by bilinear transformation. The rotational frequency is set to 0 Hz during ascent and 80 Hz (4800 rpm) at which the gyro effect is remarkably generated in order to take a response to the variation of the parameters. FIG. 45 shows a comparison of output displacements based on an initial value response at 0 Hz when ascending, and FIG. 46 shows control inputs at that time. Further, FIG. 47 shows the output displacement due to the initial value response at 80 Hz (4800 rpm) at which the gyro effect is remarkably generated, and FIG. 48 shows the control input at that time. As a simulation condition, a step target value of 10 μm is added from 0.01 s above the X axis.

45 and 47, both the H∞ controller and the GS controller are stabilized, but the convergence speed is different. In particular, in the GS controller, due to the effect of taking the gyro effect into consideration, the response on the lower side in the Y-axis direction has little vibration and good results can be obtained. Also in FIG. 45, a good result of following the target value quickly was obtained as compared with the H∞ controller. This is probably because the end point controller (K _min ) of the GS controller is optimally designed at 0 Hz compared to the H∞ controller that requires robustness.

6.3.2 GS control using a table When considering implementation in an experimental system, in addition to being a very short sampling time to control the high-frequency region, and being a 4-input 4-output system, rigidity Since it is necessary to control the secondary mode, the model cannot be reduced in dimension, and there is a problem that the calculation process is delayed in the discretization for each sampling because of the use of a controller having a large degree. Therefore, a method is proposed in which eleven controllers are obtained in advance from 0 Hz to 100 Hz at intervals of 10 Hz by convex interpolation, and the controllers that are discretized offline are used as a table and switched according to scheduling parameters (FIG. 139). reference). The number of controllers to be mounted was selected based on the DSP memory capacity.

  It was verified by the initial response whether the closed loop stability of the GS controller was impaired by using the table. The simulation conditions are the same as in the previous section. 49 and 50 show input and output at 0 Hz when ascending. In addition, in order to compare the response at the time of switching the controller, a controller obtained by convex interpolation at 70 Hz (4200 rpm) and 80 Hz (4800 rpm) is prepared, and the controller is switched at the 0.1 s point. The input / output at this time is shown in FIGS. Although there was a slight difference in the vicinity of 80 Hz where the influence of the gyroscope was strong, there was no problem in stability and good results could be obtained.

6.4 Experiment A rotation experiment was performed using a GS controller using a table. Conditions such as the sampling frequency are the same as those of the H∞ controller. The rotor trajectories and control inputs on the upper and lower sides at a rotation frequency of 0 to 90 Hz (5400 rpm) are shown in FIGS.

  The control input has a waveform in which the half-wave synchronized with the harmonic disturbance is broken. This is considered to be because not only a synchronization signal depending on the rotation frequency but also an input for the coupling direction is included.

  It was confirmed by experiments that the rotor was fixed near the origin not only during ascent but also in the rotation experiment, and that the effect of the gyro moment could be controlled sufficiently. Also, no petal-like trajectory is generated due to the backward swing motion of the rigid mode and the elastic mode. From the above, the effectiveness of the GS controller was shown against the influence of the gyro effect.

  However, since the result of touchdown is the same as that of the H∞ controller at a rotation frequency of 95 Hz (5700 rpm), it is considered that there is a cause of deterioration in control performance in addition to the gyro effect.

6.5 Eigenvalue analysis In order to examine the experimental results, a closed-loop system eigenvalue analysis was performed for parameter fluctuations for the GS controller used in the experiment. FIG. 53 (a) shows the complex plane for the H∞ controller when the rotation frequency is increased from 0 Hz to 100 Hz by 5 Hz, and FIG. 53 (b) shows the complex plane for the GS controller considering the gyro effect.

  Both the GS controller and the H∞ controller are located on the left half of the complex plane, and the system is stable. However, it is confirmed that some eigenvalues have moved to the right plane, whereas the poles of the GS controller are almost fixed due to plant parameter variations. Comparing FIG. 53 (a) and FIG. 53 (b), it can be confirmed that the GS controller compensates the stability against the parameter fluctuation. However, since the system remains stable even when the rotational frequency is increased to 200 Hz, it is necessary to model the cause of the deterioration of the control performance in the experiment.

7). 7.1 Verification by elastic model Consideration of closed-loop system stability In the stiffness model, the closed loop was stable regardless of whether the controller that ignored the gyro effect or the controller that considered it was used. However, in the rotation experiment, it is unstable near 100 Hz. This inconsistency problem will be considered in consideration of the elastic bending mode. The elastic model is derived by the finite element method. The rotor is divided into 8 elements as shown in FIG. 72 using the flywheel as the concentrated mass. An electromagnet is disposed between the elements 2 and 3 and the elements 7 and 8. Here, damping and unbalanced vibration are not considered. In the modeling of FIG. 72, the equation of motion of the rotor considering the gyro effect is as follows.

M is a mass matrix, G is a gyro antisymmetric matrix, K is a rigidity matrix, U is a control input vector, T is a matrix (M, G, KεR ^{36 × 36} , TinR ^{36 × 4} ) representing an installation location of an electromagnet, q is a state vector (qinR ^{36 × 1} ). First, FIG. 73 shows the relationship between the natural frequency and the shaft rotational speed for the free-free rotor. It was found that the forward swing mode and the backward swing mode of the bending primary mode branch greatly after rotation. For this reason, the backward swing mode of the bending primary elastic mode is strongly influenced by the gyro effect and is lowered to around 100 Hz. As a result, it was found that the crossing frequency between the transport line and the backward swing mode is around 110 Hz, and the rotor is approaching the backward bending resonance point when the rotor rotates beyond 100 Hz.

7.2 Mode Separation and Reduced Dimension In order to achieve high-speed rotation exceeding 100 Hz, the influence of the gyroscopic effect of the stiffness mode and the bending primary mode of the FEM model is important. Here, since an antisymmetric gyro matrix G exists in the equation (7.1), the standard mode analysis method cannot be used easily. New vector for mode separation

When is introduced, the following equation is obtained.

The eigenvalue problem of equation (7.2) is considered by the following equation.

Equation (7.4) is similar to the standard form, but G ^* is still an antisymmetric matrix due to the characteristics of K and G. However, the solution of such eigenvalue problem is n sets of conjugate pure imaginary numbers.

And corresponding n eigenvectors of conjugate complex numbers

I know I have Then, s = iw and z = u + iu are substituted into equation (7.4) to separate the real part and the imaginary part. Finally, the equations related to u and v can be obtained in the following two eigenvalue problem forms.

Here, K ^* is a positive definite symmetric matrix. Equation (7.5) has n sets of important λ _r , and it can be seen that u _r and v _r correspond to the same λ _r . Then, when the Cholesky decomposition on ^{M * (M * = RR -T} ), become the next standard eigenvalue problem.

Here, Φ is a symmetric positive definite matrix. The eigenvector corresponding to λ _r is P = [u _r v _r ]. Set the coordinate conversion formula z = R ^-T Pξ, is substituted into equation (7.2), the mode separated equation (7.7) is obtained. ξ = | η _r ζ _r | is a mode coordinate. Θ is a block diagonal matrix as represented by Equation (7.7).

Based on the equation (7.8), the model reduction including the rigid mode and the bending primary elastic back mode becomes the following equation.

Here, A _r εR ^{10 × 10} , B _r εR ^{10 × 4} , C _r εR ^{4 × 10} , D _r εR ^{4 × 4} , and U are control input vectors (UεR ^{4 × 1} ). A Bode diagram of the mode before and after the reduction in dimension is shown in FIG.

7.3 Simulation Verify whether the reduced-order model including the stiffness mode and the bending primary elastic back mode created in the previous section is consistent with the experiment. As the controller, a GS controller designed in consideration of only the rigid mode was used. FIG. 75 shows a comparison between an initial value response simulation and an experiment when the sampling frequency is 10 kHz and a step target value of 10 μm is given on the upper side of the X axis.

  By using the elastic model, the same high frequency as the experimental result was excited in the simulation. Since it coincides with the natural frequency of the elastic mode at the time of ascent, it can be confirmed from the initial value response that the reduced-order model includes the bending primary elastic back mode. However, it can be said that the low frequency component and the convergence time are significantly different, and the model is easier to stabilize than the experimental system. The cause is considered to be due to the specifications of the actuator such as inductance and eddy current loss, and the control current obtained by the GS controller is multiplied by a coefficient α as shown in FIG. FIG. 77 shows a closed-loop simulation with α added. Although the low-frequency components do not match completely, the overshoot amplitude and the amplitude at the time of rise are effective results that approach the experimental results. The value of α was set to 0.22 by trial and error.

7.4 Eigenvalue Analysis Stability comparison between when the coefficient α was added and when the coefficient α was not added was performed by closed-loop eigenvalue analysis shown in FIG. As shown in FIG. 78, when α is not added, the eigenvalue exists at a point far from the real axis, and is stable. However, although the experimental system is stable, the pole exists near the real axis. This phenomenon is considered to be one of the causes of the mismatch between simulation and experiment. By adding α, the pole moved to the right, and the pole arrangement was close to the experimental system. In the simulation results in the previous section, the rigid secondary mode is controlled and not excited by the time series response. Again, in the pole analysis when the coefficient α is added, the pole of the rigid secondary mode is stabilized and agrees with the experimental result. FIG. 79 shows an initial value response when the frequency is set to 100 Hz to be touched down in the experiment and a step target value of 10 μm is given on the upper side of the X axis. Despite the use of a GS controller that was stable even at 100 Hz in the previous simulations, the results were divergent. From the above, it can be confirmed that by adding the coefficient α, the reduced-order elastic model shows a response close to that of the actual system.

8). Conclusion In this paper, we studied the control of a magnetic bearing type power storage flywheel. Because the flywheel is attached to the rotor for the purpose of increasing the inertial force, forward and backward whirling movements due to the gyro effect and changes in their natural vibrations prevent high-speed rotation. .

  Therefore, in this study, we first modeled the characteristics of the gyro effect. The H∞ controller was designed considering the rotational frequency ω as a constant in the A matrix of the plant. The system, which was stable when ascending, increased the lower side swing at a rotation frequency of 93 Hz, resulting in a touchdown. The reason why the lower side is touched down first is that the distance from the center of gravity to the lower magnetic bearing is about 1.8 times longer than the upper side, so that the track of the lower rotor is larger than the upper side. Moreover, the rotor drew a locus like a petal at rotation frequencies of 60 Hz and 80 Hz. It was confirmed that there were second-order forward and backward runouts in the stiffness mode of the model. From this result, it is considered that only the robustness of the fixed controller in consideration of coupling with the model has little effect of reducing the influence of the gyro effect.

Next, considering a linear parameter dependent model (LPV) in which the system fluctuates depending on the rotational frequency, a GS controller was designed and simulated. At the time of ascent, both the H∞ controller and the GS controller were stabilized, but the convergence speed was different. In particular, in the GS controller, due to the effect of taking the gyro effect into account, there was little vibration in the lower response in the Y-axis direction, and a good result that quickly followed the target value was obtained. This is probably because the end point controller (K _min ) of the GS controller is optimally designed at 0 Hz compared to the H∞ controller that requires robustness.

  When considering implementation in an experimental system, it is necessary to control the rigid secondary mode in addition to being a very short sampling time to control the high-frequency region and being a 4-input 4-output system, Since the model cannot be reduced in dimension and a controller with a large degree is used, there is a problem that the calculation process is delayed in the discretization for each sampling. Therefore, a method has been proposed in which controllers are obtained beforehand by convex interpolation at intervals of 10 Hz from 0 Hz to 100 Hz, and the controllers that are discretized offline are used as a table and switched according to scheduling parameters. The number of controllers to be mounted was selected because of DSP memory capacity constraints. It was verified by the initial response whether the closed loop stability of the GS controller was impaired by using the table. In the switching near 80 Hz where the influence of the gyro is strong, there was a slight difference in the convergence speed, but there was no problem in stability and good results could be obtained.

  In the rotation experiment as well as during the ascent, the experiment confirmed that the rotor was fixed near the origin and the influence of the gyro moment could be controlled sufficiently. Also, no petal-like trajectory is generated due to the backward swing motion of the rigid mode and the elastic mode. From the above, the effectiveness of the GS controller was shown against the influence of the gyro effect. However, at a rotational frequency of 95 Hz (5700 rpm), the result was a touchdown similar to the H∞ controller.

  In order to examine the experimental results, a closed-loop eigenvalue analysis was performed with respect to parameter fluctuations for the GS controller used in the experiment. The poles of the GS controller and H∞ controller are both located on the left half of the complex plane, and the system is stable. However, when the plant parameters fluctuated, the poles of the GS controller were fixed, whereas in the H∞ controller, it was confirmed that some poles moved to the right plane. However, even if the rotational frequency of the rigid model considering the gyro effect is increased to 200 Hz, the stability of the closed loop is guaranteed. Therefore, the experiment was verified using a reduced-order model including a bending primary elastic back mode, which is considered to deteriorate the control performance in the experiment. By using the elastic model, the same high frequency as the experimental result was excited in the simulation. In addition, since it matches the natural frequency of the elastic mode at the time of ascent, it can be confirmed from the initial value response that the bending primary elastic back mode is included in the low-dimensional model. However, it can be said that the low frequency component and the convergence time are significantly different, and the model is easier to stabilize than the experimental system. The cause was considered to be due to the specifications of the actuator, and the simulation was performed again by multiplying the control current obtained by the GS controller by a coefficient α. Although the low-frequency components do not match completely, the overshoot amplitude and the amplitude at the time of rise are effective results that approach the experimental results.

  A comparison of the stability when the coefficient α was added and when it was not added was performed by closed-loop eigenvalue analysis. When α was not added, it was stable because the eigenvalue exists at a point far from the real axis. However, although the experimental system is stable, the pole exists near the real axis. This phenomenon is considered to be one of the causes of mismatch between simulation and experiment. By adding α, the pole moved to the right, and the pole arrangement was close to the experimental system. In the simulation results in the previous section, the rigid secondary mode is controlled and not excited by the time series response. Also in this case, in the pole analysis when the coefficient α is added, the pole of the rigid secondary mode is stabilized and agrees with the experimental result. In addition, when the initial value response was performed, the result was divergent even though the GS controller that was stable at 100 Hz in the simulation so far was used. From the above, it can be said that the reduced-order elastic model to which the coefficient α is added is a model close to a real system.

  For future work, it is necessary to design a controller with better performance using a model obtained by adding a coefficient α to an elastic model.

Zero Power Control of Magnetic Bearing System Considering Gyro Effect 1.3 Purpose of this Study The purpose of this study is zero power switching control method based on nonlinear control theory without passing bias current and maintaining nonlinearity. By designing a controller that can stabilize the rotor and reduce power consumption, the rotor can rotate at high speed, and verify its effectiveness.

1.3.1 Modeling simplification In a magnetic bearing system, when obtaining an equation of motion, it is common to obtain a state equation relating to the center of gravity in consideration of the rotor system. However, in the experimental apparatus, the observation point of the sensor is different from the position of the center of gravity, and is placed on the rotor near the lower electromagnet. Therefore, when designing the controller, since the feedback state quantity from the sensor must be converted to the one related to the center of gravity, it takes time to calculate and the accuracy is also reduced. For these disadvantages, the rotor displacement at the position of the center of gravity and the rotor displacement at the point corresponding to the electromagnet, the relationship between the rotation angle and displacement, the electromagnetic force and rotation applied to the rotor center of gravity for all electromagnets If the moment is considered separately on the rotor by the lower electromagnet, the state equation obtained at the end is simplified. All state quantities are related only to the displacement signal from the sensor.

1.3.2 Zero power switching control method The basic method of zero power switching control is that the rotor part of the flywheel system is regarded as a rigid body, and when the rotor approaches one electromagnet, the control current of the approaching electromagnet is set to zero. Thus, if the control current is supplied only to the opposite side and the rotor is completely at the equilibrium point, no current (power) flows. Further, when designing the controller, the control input is assumed as the attractive force between the opposing electromagnets. The actual control current is obtained from the obtained suction force by a switching method.

1.3.3 Design and performance verification of nonlinear controller This paper designs a controller for the 0.5 kWh class magnetic bearing flywheel system based on the control theory aimed at optimizing the evaluation function using the LQR method. And verify with experiments. After designing a regulator controller for the LQR method, an LQI controller with a servo system added is designed to eliminate the steady-state deviation. From the experimental results, when the servo system was added, the rotational speed reached around 100 Hz. However, when the gyro effect is taken into account, the system becomes an LPV system that depends on a fluctuation parameter such as a rotational speed, and therefore cannot be strictly compensated by an ordinary LTI controller. Therefore, we propose gain-scheduled sliding control. As a result of the experiment, the influence of the gyro effect was compensated, but it touched down at the same place as the LQI controller. In other words, it was found that if a certain frequency was reached, the system would become unstable without considering the elastic mode. Therefore, the relationship between the rotation frequency and the natural frequency is obtained by an excitation experiment in a floating state. From this result, a large peak occurs at a rotation frequency of around 100 Hz. That is, 100 Hz is recognized as the backward resonance point of the elastic mode. In this paper, because the order of the elastic model is high, the speed was increased to 130 Hz by non-model-based PID control. However, there is a problem that the simulation results of the elastic model of PID control and the experimental results do not match, which will be considered in the future.

  The controlled object is the magnetic bearing device shown in FIGS.

3. Control Theory In this magnetic bearing system, since the bias current can be avoided completely, the controller is designed based on the nonlinear theory while maintaining the nonlinear relationship between the electromagnetic force and the displacement.

3.1 Control system design based on linear quadratic evaluation model 3.1.1 Optimal regulator A design method for determining an optimal control input by state feedback so as to minimize the evaluation function is described. Controllable linear time-invariant system

Therefore, in order for a control law that minimizes a certain evaluation function to be a linear state feedback control law, the evaluation function must be in a quadratic form as follows.

Here, the weight matrix Q for control purposes is a non-negative definite symmetric matrix, and the control input weight matrix R is a positive definite symmetric matrix. Evaluation function of equation (3.3) with end time t _f → ∞ in quadratic evaluation function equation (3.2)

The optimal control input that minimizes

Given in. Here, P is a positive definite symmetric matrix of the Riccati-type algebraic equation (3.5), and is uniquely determined.

  As can be seen from equation (3.3), as a qualitative design guideline for the weight matrix, if Q is increased, control performance increases, and if R is increased, control input can be reduced. There is a trade-off between the objective function and the control input.

What is used as the objective function of control depends on the control specifications, but in the case of a vibration system control problem, there are cases where position energy: (1/2) kx ² and kinetic energy: (1/2) mx ² are adopted. is there. For example, when the purpose is to suppress acceleration, the acceleration is generally y = Cx + Du (3.6)
And including the direct terms of the control input,

Must be included in the evaluation function. In this case, a quadratic evaluation function is adopted after adding a cross-term term.

The optimum control input in this case is given as follows.

P is a unique positive definite symmetric matrix of the following Riccati equation.

  A control system that constitutes a closed loop system that derives a linear state feedback control law using a quadratic form for the evaluation function and returns the state to the origin is called a linear quadratic form regulator.

3.2 Lyapunov's theory of stability 3.2.1 Definition of internal stability The behavior of a nonlinear time-varying system is not represented by an exponential function like a linear time-invariant system, and is generally complex. Therefore, if a precise definition of stability is made, various concepts can be considered. As in the case of continuous systems, Lyapunov's concept of stability is applied to discrete value systems. Now consider the following system. Here are some of them.

x = f (x, t) (3.11)
Here, x is an n-dimensional state vector, and f is a continuous function with respect to times t and x. Equation (3.11) has a unique solution over all times for any initial time t ₀ and initial state x ₀ . A value at t ≧ 0 is represented as x (t, x ₀ , t ₀ ). However, x (t ₀ , x ₀ , t ₀ ) = x ₀ . In this system _{x (t, x 0, t} e) ≡x e (3.12)
A state x _{e in} which is established is called an equilibrium state. Since the equilibrium state is a constant solution of equation (3.11), f (t, x _e ) = 0 (3.13) at all t.
It satisfies. Since this is a nonlinear time-varying equation, it does not always have a constant solution. Also, even if you have it, it is not necessarily unique. When considering the behavior around the equilibrium state x _e of the system, generality is not lost even if x _e is the origin x = 0 of the state space. If x _e ≠ 0, the coordinate system of the state space is moved, the differential equation for the new f-function has the origin x = 0 as the equilibrium state, and the behavior of the solution is the same as in equation (3.11).

Stability Stability is the concept that the solution x (t, t ₀ , x ₀ ) can be stopped near x = 0 by taking the initial state x ₀ close to the equilibrium state x = 0.

Definition 1: For any positive number ε and initial time t ₀ , there exists a certain positive number δ (t ₀ , ε), and all times t ≧ t for the initial state x ₀ where ‖x ₀ ‖ ≦ δ. _{When 0} and ‖x (t, t ₀ , x ₀ ) ‖ ≦ ε hold, the equilibrium state x = 0 is stable. Reference is made to FIG.

Uniform stability In the definition of stability, in order for the solution x (t) to stay close to the equilibrium state, the distance that the initial state must be close to the equilibrium state depends on the initial state, and the initial time Regardless of whether or not it is almost the same.

Definition 2: In Definition 1, when an independent number δ (ε) exists at the initial time as a positive number δ, the equilibrium state x = 0 is said to be uniformly stable. If the differential equation (3.11) is time invariant, that is, the function f on the right side is time independent and the nonlinear function f (x), the solution does not exist at the initial time. Therefore, the stability and the uniform stability at this time are equivalent.

Uniform asymptotic stability In addition to uniform stability, when a solution converges to equilibrium with time, it is called uniform asymptotic stability.

Definition 3: With uniform stability, there exists a time T (μ, r) for a certain positive number r and an arbitrarily small positive number μ, and all initial states x ₀ satisfying ‖x ₀ ‖ ≦ r For any initial time t ₀ , で x (t, t ₀ , x ₀ ) ‖ ≦ μ (3.14) at all times t ≧ t ₀ + T
When is established, it is said to be uniformly asymptotically stable. Reference is made to FIG.

Global uniform stability When a solution converges to an equilibrium state, the set of initial states from which the converged solution starts is called an absorption region. The global state indicates that this absorption region is the entire state space. Reference is made to FIG.

Definition 4: The equilibrium state x = 0 is uniformly asymptotically stable even when considering the set ‖x ₀ ‖ ≦ r of the initial state x ₀ for any large positive number r.

There exists a positive number ξ (η) for any positive number η, and for any initial state x ₀ and all initial times t ₀ where ‖x ₀ ‖ ≦ η, at all times t ≧ t ₀ ,
‖X (t, t ₀ , x ₀ ) ‖ ≦ ξ (3.15)
When is established, it is said to be globally uniform asymptotically stable.

  The definition of global uniform asymptotic stability is not only made the absorption region of uniform asymptotic stability global, but also added a condition. This condition is called uniform boundedness. Uniform boundedness is similar to uniform stability, but with uniform stability, the range of the initial state relative to the existence range of the solution is considered. This is the point of looking at the range. Therefore, there are equilibrium states that are uniformly bounded and not uniformly stable, and vice versa.

3.2.2 Lyapunov stability discrimination method There is a method called Lyapunov stability discrimination method as a method for examining which of the stability properties described in the previous section the equilibrium state of the differential equation has. The distance from the equilibrium state of the solution is evaluated using a scalar function of the state, and the stability can be examined analytically without obtaining a solution. This scalar function can be regarded as an extension of the concept of distance from the equilibrium state, or, if the state is a physical variable, an extension of the concept of potential energy possessed by the element corresponding to the variable.

  Consider a scalar function (t, x) with time t and state x as variables. Continuous differentiation is possible with respect to t and x, and V (t, 0) = 0 at all t. The following properties are defined for this function.

α (σ) is a univariate scalar function that is continuous and non-decreasing with respect to σ, and α (σ)> 0 for α (0) = 0 and σ> 0. For all t, x α (‖x‖) ≦ V (t, x) (3.16)
V (t, x) is said to be positive definite when there exists α (σ) that holds.

Α (σ) that satisfies equation (3.16)

V (t, x) is said to be unbounded in the radial direction.

There exists a function β (σ) having properties similar to α (σ), and V (t, x) ≦ β (‖x‖) (3.18) for all t, x.
Is established, V (t, x) is said to be dexcent. This means that V (t, x) decreases uniformly as ‖x‖ decreases.

Next, assuming that the state is x at time t, consider the time derivative V (t, x) of V (t, x) along the solution of equation (3.11) at that point. V (t, x) is

And defines the following properties:

For all t, x V (t, x) ≤0 (3.20)
V (t, x) is quasi-negative.

There exists a function γ (σ) having properties similar to α (σ), and V (t, x) ≦ −γ (‖x‖) for all t, x (3.21)
When V is satisfied, V (t, x) is said to be negative.

  Based on the above preparation, the stability theorem based on Lyapunov's idea is described.

  If a function V (t, x) having the following properties exists, the following can be said with respect to the equilibrium state x = 0 in the equation (3.11).

  If V (t, x) is positive and V (t, x) is quasi-negative, it is stable.

  If V (t, x) is positive and dexcent, and V (t, x) is quasi-negative, it is uniformly stable.

  If V (t, x) is positive and dexcent and V (t, x) is negative, it is uniformly asymptotically stable.

  If V (t, x) is positive, decent, unbounded in the radial direction, and V (t, x) is negative, it is globally uniform asymptotically stable.

  As is clear from the definition, stability, uniform stability, and uniform asymptotic stability are properties of the solution near the equilibrium state. Therefore, it is sufficient that the condition of this theorem is established near the equilibrium state. This theorem is important in that it can know the equilibrium state without solving the nonlinear differential equation. A function that satisfies the conditions of these theorems is called a Lyapunov function.

3.2.3 Control Lyapunov function (CLF)
Next, the control Lyapunov function, which is an extension of the Lyapunov function in the previous section, will be described.

  Consider the following time-invariant system:

x = f (x, u), x∈R ⁿ , u∈R, f (0,0) = 0 (3.22)
Here, a closed loop system obtained by substituting the feedback control law α for the control input u x = f (x, α (x)) (3.23)
Consider designing α (x) such that is globally asymptotically stable at equilibrium point x = 0. Assuming that V (x) is a Lyapunov function candidate, V (x) ≦ −W (x) needs to be satisfied in order for Equation (3.23) to be stable. Here, W (x) is a positive definite function. For all x∈R ⁿ

This results in the problem of finding α that satisfies. However, this problem is a difficult problem. This is because there may be a control law that stabilizes equation (3.22), but equation (3.24) may not be satisfied depending on how V (x) and W (x) are selected. Thus, a system that can select the desired V (x) or W (x) is said to “own the control Lyapunov function”. Clarify the concept below.

Definition 2.4 A continuous positive definite function that is unbounded in the radial direction V: When R ⁿ → R ₊ satisfies the following equation, V (x) is called a control Lyapunov function.

Further, a system represented by the following formula: x = f (x) + g (x) u, f (0) = 0 (3.26)
In the case of, the inequality of the control Lyapunov function is

If V (x) is the control Lyapunov function of equation (3.26), then the continuous stabilization control law α (x) is unique for all x ≠ 0 by Sontag's formula as follows: Given to.

Where equation (3.27) is

Note that the condition is met only when In this case, equation (3.28) is

It becomes.

Further, as a feature of the stabilization control law for the equation (3.26), α (x) is continuous at x = 0 only when the control Lyapunov function V (x) satisfies the following condition. When x ≠ 0 for any ε> 0, there exists δ (ε)> 0 satisfying | x | <δ, and when u satisfying | u | <ε satisfies the following expression

The main drawback of the control Lyapunov function when using control system design is that the control Lyapunov function is unknown for most nonlinear systems. It is also difficult to design a stabilization control law as much as that. However, if the backstepping method is used, the two problems of finding the control Lyapunov function and designing the stabilization control law can be solved simultaneously. In starting this method, at least first, it must be possible to find the control Lyapunov function V (x) and the stabilization control law α (x) for the scalar system. Fortunately, always reasonable control Lyapunov function ^{V (x) = x 2/} 2 for the scalar system, easily satisfy inequality (3.27).

3.3 Sliding mode method The theory of changing the structure of the control system is called variable structure control theory, and the most theoretically structured among them is the sliding mode control theory. One of the features is that an excellent robust control system can be constructed, and the theory has advanced to the point where it can be applied to various control purposes such as system stabilization and servo systems.

3.3.1 Presence condition of sliding mode In sliding mode control, the behavior in the phase space of the system is divided into two modes, reaching mode and sliding mode, and the state is directed toward the hyperplane and reaches the hyperplane. It is called a condition reaching condition. If this arrival condition is satisfied, the arrival mode reaches the hyperplane for a finite time and is robust against disturbances that satisfy the matching condition. For these reasons, the following procedure is necessary to realize the sliding mode control.

(1) Design a switching surface to represent the desired system dynamics of a lower order than the given plant, generally (system dimension: n) (input dimension: m).

(2) Design a variable structure control input u (t) so that any state quantity outside the switching surface reaches the switching surface in a finite time.

  These are none other than “designing a sliding mode controller”. Various approaches have been proposed to satisfy the sliding mode arrival condition. In this study, the Lyapunov function method is applied. Now, select Lyapunov function candidates as follows.

V = σ ^T・ σ / 2 (3.32)
At this time, a comprehensive sliding mode reaching condition is given by the following equation.

V = σ ^T・ σ ≦ 0 (3.33)
It is important to determine the variable structure control input to satisfy this reaching law.

3.3.2 Analysis by equivalent control method When the sliding mode exists, the system is extremely difficult to analyze because it has the most nonlinear switching input. Replacing this switching input with a continuous input improves the prospects for analysis and design. The system is x = Ax + Bu
σ = Sx (3.34)
It is defined as, in A∈R ^n, and Pururanku the B∈R ^m and S∈R ^m. In equation (3.35), if there is a sliding mode, σ = 0 (3.35)
The relationship of equation (3.36) is obtained.

σ = Sx = S (Ax + Bu) = 0 (3.36)
If det (SB) ≠ 0, the equivalent control input is obtained as follows.

u _eq =-(SB) ^-1 SAx (3.37)
This equivalent control input u _eq is an ideal continuous control input of the system, and by substituting u _eq into the equation (3.34) of the original system, the following system reduced in number by the number of inputs can be obtained. can get. This system is called an equivalent control system.

x = {IB (SB) ^-1S } Ax (3.38)
The eigenvalue of this closed loop is composed of (n−m) zeros and m origin poles of the transfer function S (sI−A) ⁻¹ B. In order to show the idea of sliding mode control, consider a control object of a linear time-invariant system that can be expressed as: For simplicity, it is assumed that the control target is one input (m = 1), the order of the system is n-th order, and is converted to a canonical form.

The structure of the sliding mode control system is governed by a switching function σ as shown below.

Here, since the regular matrix s ₂ is arbitrary, s ₂ = Im. The switching function divides the phase plane so that the signs of σ are different. By switching the control input according to the sign of this switching function, the system state is constrained on the switching hyperplane, and a desirable system reduced in number by the number of inputs is constructed. A system whose state is constrained on the switching hyperplane is called an equivalent control system. By considering this system, the dynamic characteristics of the system in the sliding mode can be analyzed.

Since the switching function σ shown in the equation (3.40) is 0 on the hyperplane,
x ₂ (t) = − s ₁ x ₁ (t) (3.41)
Next, _{x 2} is the input to the subsystem of _{x 1.} That is, the equation (3.39) is expressed as follows: x ₁ (t) = (A ₁₁ −A ₁₂ s ₁ ) x ₁ (t) σ
= S ₁ x ₁ (t) + s ₂ x ₂ (t) = 0 (3.42)
Thus, the system is separated from the low-dimensional slow x ₁ system and the fast system σ = 0, which is not affected by parameter fluctuations or disturbances that satisfy the matching condition. As shown in the above equation, the dynamic characteristics of the sliding mode are determined by the matrix S.

3.3.3 Sliding Mode Achieving Conditions and Chattering Reduction Here, a control input for obtaining the condition of the system on the hyperplane and restraining it, that is, a control input satisfying the sliding mode reaching condition is obtained. As shown earlier, the Lyapunov function is set as follows.

^{V = σ 2/2 (3.43} )
At this time, the sliding mode arrival condition may satisfy the following equation.

v = σσ <0 (3.44)
The switching function can be expressed as:

However, z (t) is obtained by calculating an error integral value with respect to the output with the target value as r. At this time, the control input u (t) must satisfy the following condition in order to satisfy the sliding mode arrival condition of the equation (3.44).

The control input has the form

Here, it is a sign function relating to sqn (σ). At this time, if the state of the system on the hyperplane, that is, the response and disturbance of the equivalent control system are known, it is possible to always satisfy the equation (3.40). However, it is often impossible to actually observe disturbances and the like, and therefore, the switching gain k is set to an appropriate constant, but it is full. If the switching width is too large, chattering may occur in the control input. Further, when used in an actual continuous plant, there is some fear in the analog switching device. In addition, when digitally used, the sampling period cannot be made infinitely small, and there is a kind of delay. If such a non-ideal device is used, it is impossible to increase the switching frequency infinitely, and the actual sliding mode does not slide on the switching surface but chatters in the vicinity thereof. The chattering of the control input excites a high-frequency region that is ignored when modeling the controlled object, which may cause spillover. In order to reduce such chattering, this study introduces the following smoothing constants to the control input.

By smoothing the switching control input, chattering can be reduced, but the sliding mode reaching condition cannot be satisfied near the hyperplane, and a deviation from the switching hyperplane occurs. At this time, formula (3.50) is

It becomes. In the above equation, when there is a steady disturbance or the like, when the value of the switching function σ is σ → ∞, the right side of the equation (3.50) → 0, and when the switching width k is a constant, the left side = the right side Will be balanced by σ. As a result, the characteristic of the sliding mode control that is insensitive to system parameter fluctuations and disturbances that satisfy the matching condition is lost, so the smoothing constant δ should be set to the minimum necessary size. At this time, it is closely related to the sampling time, and it seems necessary to design a discrete time system.

4. Modeling 4.1 Modeling when the gyro effect is not taken into account The experimental apparatus used is a 5-axis control type as described in the left figure of FIG. 81, but the flying in the z direction is controlled by PID control. Only the remaining 4 degrees of freedom in the radial direction are considered to be controlled in this study. The right figure shows only the X-Z plane because the movement modes of the rotor in the X- and Y-axis directions are exactly the same. Equation (4.1) is an equation of motion with respect to the center of gravity of the rotor.

Here, F _xu , F _xl , F _yu , and F _yl are attractive force differences between the electromagnets facing the upper and lower portions in the X and Y axis directions of the rotor. Each is as shown in Equation (4.2).

The state space model can be obtained from the equation of motion for the center of gravity, but the state quantity is the rotor displacement and rotation angle at the center of gravity position. However, in the experiment, the observation point of the sensor is different from the position of the center of gravity. It is placed near the electromagnet. To solve the problem, simplify the model. First, referring to the diagram on the right side of FIG. 81, the rotor displacement at the center of gravity position, the rotor displacement at the point corresponding to the electromagnet, and the relationship between the respective rotation angles and displacements are as shown in equation (4.3).

Furthermore, considering the electromagnet added to the center of gravity of the rotor and the rotational moment for all electromagnets separated by the electromagnets at the top and bottom of the rotor, substituting equation (4.3) into equation (4.1), and finally the equation of motion is It becomes simple like Formula (4.4).

The coefficients a _u and a _l are given by equation (4.5).

The equation of state is expressed by the following equation (4.4).

Here, x ₁ and x ₂ are displacement and speed state quantities, and the control input U is a difference in attractive force between the opposing electromagnets.

Each parameter is shown in Table 4.1.

5. Nonlinear controller design and performance verification 1
5.1 LQI controller design 5.1.1 Regulator controller design First, according to the backstepping control theory described in Chapter 3, the state displacement is set to x ₁ and the speed is set to x ₂ . It becomes such a secondary system.

x ₁ = x ₂
x ₂ = B ₂ U (5.1)
Where x ₁ = (x _u x _l y _u y _l ) ^T , x ₂ = (x _u x _l y _u y _l ) ^T
_{U = (F xu F xl F} yu F yl) T
The evaluation function is expressed by equation (5.2).

here,

Set as follows. Furthermore, a servo system that follows the target value input without steady deviation is designed. Here, when r is a constant target value, the control problem is to stabilize the closed loop system and to match the output y with the target value. In this case, to obtain an enlarged state equation as including an integral x _r of the error e = r-y of the target value and the output equation (5.3).

Here, when x _a = [x ^T x _r ^t ], the evaluation function is expressed by equation (5.4).

Consider state feedback control that minimizes the evaluation function. The state feedback control input is given by equation (5.5).

  A block diagram is shown in FIG.

5.1.2 Control current switching method In the experiment, the control input is not the attractive force between the opposing electromagnets, but the current flowing in the end electromagnet. The current is calculated backward based on the switching method of the two pairs of electromagnets. Here, a method for switching the electromagnet in the X-axis direction of the rotor will be described in detail.

1) When F _xu ≧ 0 When F _xu ≧ 0, the electromagnet 1 may be attracted in FIG. Relationship between force and current i ₁ at that time (assuming i ₃ = 0)

Therefore, i ₁ is shown as follows.

2) When F _xu <0 When F _xu <0, the electromagnet 3 may be attracted in FIG. Relationship between force and current i ₃ at that time (assuming i ₁ = 0)

From this, i ₃ is shown as follows.

  The same applies to the lower part.

1) When F _xl ≧ 0 When F _xl ≧ 0, the electromagnet 5 may be attracted in FIG. Relationship between force and current i ₅ at that time (assuming i ₇ = 0)

From shows _{i 5} as follows.

2) When F _xl <0 When F _xl <0, the electromagnet 7 may be attracted in FIG. Relationship between force and current i ₇ at that time (assuming i ₅ = 0)

From this, i ₇ is shown as follows.

  Control is performed by converting the attraction force into the current of each electromagnet according to this switching condition and using it as the output of the electromagnet.

5.1.3 Simulation FIG. 83 (a) shows the initial value from the touchdown to the equilibrium point and the impulse response when a disturbance of 50N is applied at 0.05 seconds. From this result, it can be seen that the stability of the system is fast, the rotor goes to the center and the stability is good. FIG. 83 (b) shows a virtual force control input when designing the controller. From this figure, when the rotor is at the equilibrium point, the electromagnet between the opposing electromagnets becomes zero, and then when an impulse disturbance is applied instantaneously, an electromagnetic force equivalent to the disturbance is generated simultaneously, canceling the influence, It was found to have good robustness. FIG. 84 shows that the actual control current is exchanged from the force control input based on the switching condition. Here, i ₁ and i ₃ , i ₅ and i ₇ are currents in the electromagnet facing the top of the rotor. i ₂ and i ₄ , i ₆ and i ₈ are the lower electromagnet currents. This shows that the current suddenly occurred due to the influence of the impulse disturbance at 0.05 seconds and becomes zero again, so that the power consumption of the system finally becomes zero. 85 to 88 are simulation results of respective controllers corresponding to the case where the rotation speed is set to 50 Hz and 100 Hz.

5.1.4 Experimental results In the experiment, the controller used is exactly the same as the simulation. Compile the controller so that the sampling time is 0.125 ms. Using the DSP, the position information of the magnetic bearing is obtained from the four displacement sensors through the AD converter, converted into eight control currents according to the switching conditions, and after entering the amplifier and driving the actuator, the experiment I do. FIG. 89 illustrates the relationship between each apparatus in the experiment and the method to be performed.

In the experiment, assuming that the Z-axis direction has already been PID-controlled, the rotor performs a floating and rotation experiment from the touchdown state. FIG. 90 shows a situation where the rotor floats. The top is a Lissajous waveform, and the middle is the phase power spectrum. Below is the control current in the X-axis direction. From this figure, it can be seen that both the upper side and the lower side converge to a central point, and there is no steady-state deviation, so that it rises stably. FIGS. 91 to 97 show the trajectory at each rotation stage, the power spectrum of the displacement, and the control current when the rotor is rotating. From the results, the upper and lower rotor trajectories went to the center, returned and repeated, but reached a rotational speed of 5700 rpm exceeding the rotational frequency of 85 Hz. Furthermore, when viewed from the control currents of FIGS. 94 and 95, the valley of the current i ₁ is just the top of the opposite i ₃ wave. That is, it can be seen that the current is switched properly. However, although the step response is slow and the locus of the lower rotor becomes larger, the gain parameter is adjusted and the lower control input is increased. However, the gyro effect starts to appear around 50 Hz, and the stability deteriorates.

5.2 Analysis of Experimental Results From the results shown in FIGS. 90 to 97, it can be seen that the rotor can be converged to the center equilibrium point without steady deviation, and the current is properly switched. However, during rotation, the upper and lower rotor trajectories went back and forth to the center and touched down at 97 Hz. When viewed from the trajectory of FIG. 97, a trajectory like a petal is generated, which is considered to be a backward precession due to the gyro effect. In other words, it seems that the gyro effect is strengthened and the stability is lost due to the increase in the rotation speed, which is the cause of touchdown. In order to solve the problem, we made a model considering the gyro effect.

5.3 Gyro effect Since many rotors using magnetic bearings rotate at a relatively high speed, the gyro effect is a problem. The gyro effect is an action that occurs when the high-speed flywheel performs a tilting motion. This effect causes the following characteristic properties.

-When the motor is stopped, the natural frequency, which is one, is divided into a forward swing mode (forward mode) and a backward swing mode (backward mode), and the range of change is determined by the magnitude of the gyro effect.

・ The vibrations of the X and Y axes of the rotor are coupled, and the swing trajectory becomes close to a circle. Due to this gyro effect, the rotor system has a vibration characteristic that changes with the number of rotations for the magnetic bearing, and becomes a controlled object in which the X and Y directions are coupled, which is a major obstacle in determining the control characteristics of the magnetic bearing. .

  Here, the gyro moment is analyzed in detail. First, as shown in FIG. 98, when the flywheel bends and tilts toward the X axis (Y axis), it rotates along its own Z axis. Is to occur. This effect becomes stronger as the rotational speed increases. Eventually, the flywheel will have a forward natural vibration and a backward natural vibration.

According to the new coordinate system, the angular momentum component is expressed by the following equation.

_{Here, θ x, θ y, θ} z is the angular displacement of the flywheel, _{I r} is the moment of inertia about the X-axis, Y-axis, the _{I z} is a polar moment of inertia about the Z axis. Since the equation of motion is established for the coordinate system (x, y, z) fixed in space, the component of angular momentum is projected from the fixed coordinate system according to FIG.

It becomes. However, all angles were minute. When the equation (5.7) according to the law of angular momentum is differentiated with respect to time, and θ _z = −ω is further substituted, rotational moments about the X axis and the Y axis are obtained as follows.

Here, the terms -ωI _z θ _y and ωI _z θ _x indicate the influence of the so-called gyro effect.

5.4 Modeling in Consideration of Gyro Effect The equation of motion for the center of gravity of the rotor in consideration of the gyro moment is given by Equation (5.9).

As described in the previous section, the model is simplified, and the final model is expressed by equation (5.10).

The equation of state is expressed as follows by equation (5.10).

Here, x ₁ and x ₂ are displacement and speed state quantities, and the control input U is a difference in attractive force between the opposing electromagnets.

  Then, the current LQI controller is applied to modeling in consideration of the gyro effect, and the root locus is shown in FIG.

  From this figure, it can be seen that as the rotational speed increases, the closed-loop pole becomes closer to the imaginary axis. That is, the eigenvalue of the closed-loop A matrix is an LPV system (Linear Parameter Varying system) that changes depending on the rotational speed.

6). Nonlinear controller design and performance verification 2
6.1 Design of Gain Schedule Type Sliding Mode Control System The modeling obtained in the previous chapter is an LPV system. We propose gain-scheduled sliding mode control for this LPV system.

6.1.1 Design of Gain Schedule Type H∞ Control Hyperplane Consider the following canonical system plant model.

  In particular, if the control system is regarded as a servo system, the error signal can be expressed as follows.

e = r-Cx ₁ (6.2)
In this control system, the hyperplane is defined as:

φ = S (x ₁ ) + x ₂ (6.3)
Here, S (x ₁ ) is a linear operator of x ₁ . The defined switching function has a new state variable z dynamics compared to the conventional switching function.

z = F (w) z + G (w) e (6.4)
S (x ₁ , z) = − H (w) −L (w) e (6.5)
Then, by combining Equation (6.1) and Equation (6.5), the expansion system is obtained as follows.

Further, from the equation (6.3), x ₂ = Hz−LCx ₁ + Lr (6.8)
Can be obtained. Therefore, on the hyperplane, the state equation of the system is reduced as follows.

  A control system using time-varying parameters is as shown in FIG.

However, the gyro term in real systems because it entered the A _22, can not be applied as it is. Therefore, we will use the following filter.

1 / (0.001s + 1) (6.10)
Combining the plant state-space model with the filter gives the following extended system:

  FIG. 100 is as shown in FIG.

  The Lyapunov function V is defined as in the following equation for the hyperplane of equation (6.7).

V = φ ^T φ / 2 (6.12)
In order to stabilize the closed loop system, the following equation must be satisfied.

V = φ ^T φ <0 (6.13)
If the state is constrained on the hyperplane, φ = φ = 0 (6.14)
From the above equation, the control input u _eq is given as follows.

In this study, L (ω) is set to zero. Furthermore, according to equation (6.3)

A control input that satisfies the stability condition given by equation (6.16) is selected as follows.

Here, the nonlinear input is selected as follows.

6.1.2 Controller calculation H (ω), F (ω), G (ω), and L (ω) in equation (6.15), ie, the controller, was designed using MATLAB's LMI control toolbox. . A Bode diagram of the LPV system is shown in FIG. Here, only the Bode diagram of the first input pair and the four outputs is shown, but the others are almost the same and are omitted. The broken line is the Bode diagram of the plant with the minimum variation parameter, and the solid line is the Bode diagram with the maximum variation parameter.

  Since there is a peak in the vicinity of 200 Hz from the Bode diagram of the plant having the maximum variation parameter, the corresponding weight function and the Bode diagram of the controller are shown in FIGS.

6.2 Simulation results 6.2.1 Real-time discrete system by trapezoidal approximation In order to discretize the controller, it is necessary to approximate the controller dynamics in continuous time with the dynamics in discrete time. The controller was designed in the preceding paragraph, the sampling time is T, the state at time kT will be represented as x (kT) = x _k. Further, it is assumed that the scheduling parameter α (t) and the observation amount y (t) can be approximated by values at the time kT at the time [kT, (k + 1) T]. That is, at kT ≦ t <(k + 1) T

Assuming that At this time, the following trapezoidal approximation theorem holds.

To prove this theorem, from the assumption, for the state x _{k + 1 at} time (k + 1) T

Holds. If this equation is approximated to a trapezoidal shape,

It becomes. Furthermore, when divided into a term of time kT and a term of (k + 1) T,

It becomes. here,

Then, equation (6.24) becomes

And transformed. Well now

Is established. Therefore,

Is present, the equation (6.26) becomes

Here, applying equation (6.25),

It becomes. Than this,

Is guided.

Also, equation (6.21)

Can be derived in the same manner using the relationship of equation (6.25). FIG. 105 shows the initial value response from the touchdown to the equilibrium point and the impulse response when a 50N disturbance is applied for 0.05 seconds. Further, a step response in which a target value of 0.02 mm in the X-axis direction and 0.01 mm in the Y-axis direction of the upper part of the rotor is entered in 0.1 second as in the case of the backstepping controller. As a result, it has almost the same performance as the LQI controller, and the step response is fast. FIG. 106 shows that the four switching functions become zero and the state quantity is constrained to the hyperplane and constrained to the equilibrium point. It was clarified that the switching function goes up when there is a disturbance. FIG. 107 shows a virtual force control input, and FIGS. 108 and 109 show control currents calculated backward from the force control input by the switching method. The 0.05 second and 0.1 second current peaks are caused by respective impulse disturbances and step signals.

6.3 Experimental Results To further simplify the calculation, Equation (6.30) is

To change. However, if the order is high, the sampling time is short, so that the DSP calculation is not in time and the experiment cannot be performed. To solve this problem, a method is proposed in which convex interpolation is performed after discretization with MATLAB.

  As in the previous simulation, FIG. 110 shows the initial value response from the touchdown to the equilibrium point and the impulse response when a 50N disturbance is applied for 0.05 seconds. Further, a step response is shown in which a target value of 0.02 mm in the X-axis direction and 0.01 mm in the Y-axis direction is entered in the upper portion of the rotor in 0.1 second as in the backstepping controller. Comparing this with FIG. 105, the performance of the method of convex interpolation after discretization was somewhat worse, and the step response was particularly delayed. However, it still seems to be acceptable. In the experiment, the same controller is used as in the simulation. In the experiment, assuming that the Z-axis direction has already been PID-controlled, the rotor performs a floating and rotation experiment from the touchdown state. FIG. 111 shows a situation in which the rotor floats. The top is a Lissajous waveform, and the middle is the phase power spectrum. Below is the control current in the X-axis direction. From this figure, it can be seen that both the upper side and the lower side converge to a central point, and there is no steady-state deviation, so that it rises stably. FIGS. 112 to 120 show the trajectory at each rotation stage, the power spectrum of the displacement, and the control current when the rotor is rotating.

6.4 Analysis of the experimental results Certainly switching is done in a positive and negative manner like a triangular wave. From the control current result calculated backward from the force control input, it can be seen that the current is clearly turned on and off, and the circular locus is clean, as compared with the LQI controller, at any rotational frequency. Further, comparing the current root locus shown in FIG. 121 and FIG. 99, it can be seen that the root movement is smaller in the gain schedule control than in the fixed controller.

  From the above results, it was found that the stability of the gain schedule type sliding mode control is better than that of the LQI controller. However, as with the LQI control, the stability deteriorated from 90 Hz and touched down near 100 Hz. The problem of not exceeding 100 Hz that had been left before was not solved.

7). Elastic Model and PID Control 7.1 Introduction of Elastic Modeling In the stiffness model, the closed loop system is stable regardless of whether the compensator that ignores the gyro effect or the compensator that considers it. However, in the rotation experiment, it is unstable near 100 Hz. This inconsistency problem will be considered in consideration of the elastic bending mode. The elastic model is derived by the finite element method. Using the flywheel as a concentrated mass, the rotor is divided into eight elements as shown in FIG. An electromagnet is disposed between the elements 2 and 3 and the elements 7 and 8. Here, damping and unbalanced vibration are not considered.

In the modeling of FIG. 122, the equation of motion of the rotor in consideration of the gyro effect is as follows.

M is a mass matrix, G is a gyro antisymmetric matrix, K is a stiffness matrix, U is a control input vector, and T is a matrix indicating an installation location of an electromagnet (M, G, K∈R ^{36 × 36} , T∈R ^{36 × 4} ), Q is a state vector (qεR ^{36 × 1} ).

  First, FIG. 123 shows the relationship between the natural frequency and the shaft rotational speed for the free-free rotor. It was found that the forward and backward bending primary modes branch greatly after rotation. From this, the backward eigenvalue of the bending primary elastic mode is strongly influenced by the gyro effect and is lowered to around 100 Hz. As a result, it was found that the crossing frequency between the transport line and the backward mode is near 100 Hz, and when the rotor rotates beyond 100 Hz, it approaches the backward bending primary resonance point.

7.2 PID Control The obtained elastic modeling is 72nd order, and model-based control cannot be realized. Therefore, we propose non-model based PID control. A block diagram is shown in FIG.

  PID1 is a controller that aims to stabilize the system. That is, the state of rising from the touchdown state is controlled by the control of PID1. Considering the gyro effect, the X and Y directions interfere with each other. PID2 is a cross-feedback controller that crosses and feeds back signals in the X and Y directions. That is, signals that interfere with each other are cross-feedbacked to compensate for the gyro effect, which stabilizes the system. A Bode diagram of the PID controller used is shown in FIG.

7.2.1 PID Control Experimental Results FIG. 126 shows the situation where the rotor is rotating at 90 Hz. The top is a Lissajous waveform, and the middle is the phase power spectrum. Below is the control current in the X-axis direction. From this figure, it can be seen that both the upper side and the lower side converge to one point in the center, and there is no steady deviation, and the surface rises stably. 127 to 133 are the locus, the power spectrum of the displacement, and the control current at each rotation stage when the rotor is rotating.

7.2.2 Analysis of experimental results Looking at the results of PID control, although it is a fixed controller, the circular trajectory and the shape of the current are not clean, but the rotational speed has increased by about 30 Hz from the previous. It was found that the cross-feedback controller compensated for the effect of the elastic backward-facing primary mode to some extent. FIG. 134 shows a comparison between simulation and experiment of a reduced-order elastic model using a PID controller capable of increasing the speed to 130 Hz. The solid line is the simulation of the elastic model in which the rotation speed is set to 0 Hz, and the broken line is the experimental data when flying with the PID controller. Both have a step signal of 0.1 * 10 ⁻⁴ m at 0.037 seconds and see the response.

  Furthermore, the lower right figure is enlarged and shown in FIG.

  The high-frequency vibration shown in FIG. 135 seems to be an elastic mode, which is about 170 Hz when counted, which is almost the same as the initial value of the uppermost elastic bending primary mode in FIG. 123 at 0 Hz. However, the result of the step response does not match the simulation and the experimental result, and this is also considered. FIG. 136 shows a simulation result when the current input is multiplied by 0.2.

  From this figure, it can be seen that the experimental results almost coincide with each other when the control input is increased by a factor of 0.2. Further, FIG. 137 and FIG. 138 show the root locus from 0 Hz to 150 Hz of the closed loop in which the total input and the control input are 0.2 times.

  Comparing the two figures, the rigid primary mode is 65 Hz and the rigid secondary mode is 80 Hz for all inputs. The root corresponding to the elastic primary bending primary mode approaches the imaginary axis, but is a completely stable region. That is where the experimental results do not agree. Conversely, when the input is 0.2 times, the rigid primary mode is about 25 Hz, and the rigid secondary mode cannot be seen. It agrees with the experimental data of FIG. Furthermore, at 140 Hz, the root corresponding to the elastic primary bending mode is too close to the imaginary axis, and stability is deteriorated. That is also consistent with the experiment. However, it is still under consideration why the working input is only 0.2 times the actual input. Probably a hardware or electrical problem.

8). Summary In this paper, we studied the control of a magnetic bearing supported power storage flywheel. The system is extremely unstable due to the large flywheel. In the conventional method, even if the rotor reaches the equilibrium point, a bias current must be supplied, so that current consumption increases. There is a drawback that the control performance is deteriorated when the displacement is large. In order to solve this problem, this study designed a controller based on the nonlinear theory of modern control and verified its effectiveness, without using bias current and maintaining nonlinearity.

  In this research, we first designed an optimal regulator controller with servo based on linear quadratic evaluation criteria, and obtained good results from simulations. It can be seen that the current is switched properly. However, during rotation, the upper and lower rotor trajectories went back and forth to the center and touched down at 93 Hz. Seen from the locus in FIG. 97, a locus like a petal came out. It is considered as a backward precession trajectory due to the gyro effect. In other words, it seems that the gyro effect is strengthened and the stability is lost due to the increase in the rotation speed, which is the cause of touchdown. In order to solve the problem, we made a model considering the gyro effect. Then, by applying the current LQI controller to modeling that takes into account the gyro effect, it can be seen that the pole of the closed loop becomes closer to the imaginary axis as the rotational speed increases from the root locus shown in FIG. . That is, it can be seen that the eigenvalue of the closed-loop A matrix is an LPV system (Linear Parameter Varying system) that changes depending on the rotation speed, and the fixed controller cannot compensate. We propose gain-scheduled sliding mode control for this LPV system. In this method, the sliding mode hyperplane schedules and converges the state quantity to the constantly changing hyperplane to achieve stability. Comparing the experimental results with the LQI control, it has a clean trajectory and the asynchronous component disappears. The current signal is also a regular signal. It can be seen that the stability is improved by compensating the gyro effect. However, like LQI, the stability deteriorated near 100 Hz, and the touchdown occurred. In this system, an elastic mode resonance point appears at 100 Hz. The LQI controller and the gain schedule type sliding mode controller that do not consider the elastic mode collide with the resonance point, and the stability is lost and the touchdown occurs. Although elastic modeling was performed, model-based control cannot be realized due to the high degree. Therefore, although the speed can be increased up to 130 Hz by non-model-based PID control, there is a problem that the simulation result of the closed loop system of the PID controller and the simulation of the reduced-order elastic model do not coincide with the experimental results. When the control input was increased to 0.2 times the original value, it was in good agreement with the experiment, but we are currently investigating why this is the case. It seems to be a hardware and electrical problem. Future issues are as follows.

1. Tune the PID controller more. The balance between the top and bottom is improved, and the speed is increased to 200 Hz by switching according to the frequency.

2. Investigate the cause of discrepancy between experimental results and simulation. A gain schedule type sliding mode control system is designed by modifying the reduced elastic model.

  The overall configuration of the magnetic bearing device, the configuration of each part, the control method, and the like are not limited to those of the above-described embodiment, and can be changed as appropriate.

FIG. 1 is a block diagram schematically showing the overall configuration of a magnetic bearing device according to an embodiment of the present invention. FIG. 2 is a longitudinal sectional view showing the main part of the mechanical part of the magnetic bearing device of FIG. FIG. 3 is a perspective view of the main part of FIG. FIG. 4 is a block diagram showing a generalized plant. FIG. 5 is a block diagram of H∞ control. FIG. 6 is a block diagram of H∞ control. FIG. 7 is a block diagram of a closed loop system. FIG. 8 is a block diagram of a feedback system representing the robust stabilization problem. FIG. 9 is a block diagram of a feedback system representing the mixed sensitivity problem. FIG. 10 is a block diagram of a closed loop. FIG. 11 is a natural vibration diagram. FIG. 12 is a perspective view showing a model of a flywheel magnetic bearing. FIG. 13 is a diagram in the XZ plane of FIG. FIG. 14 is a block diagram showing a generalized plant. FIG. 15 is an explanatory view showing an upper magnetic bearing. FIG. 16 is a graph showing the attractive force of the electromagnet. FIG. 17 is a block diagram of a closed loop system. FIG. 18 is a diagram showing an initial response. FIG. 19 is a diagram illustrating the control current. FIG. 20 is a block diagram showing an experimental apparatus. FIG. 21 is a diagram showing a monitor of the experimental apparatus. FIG. 22 is a diagram showing the control current of the electromagnet. FIG. 23 is a diagram showing experimental results. FIG. 24 is a diagram showing the input current. FIG. 23 is a diagram showing experimental results. FIG. 26 is a diagram illustrating an input current. FIG. 27 is a diagram showing experimental results. FIG. 28 is a diagram showing an input current. FIG. 29 is a diagram showing experimental results. FIG. 30 is a diagram illustrating the input current. FIG. 31 is a diagram showing experimental results. FIG. 32 is a diagram showing an input current. FIG. 33 is a diagram showing experimental results. FIG. 34 is a diagram showing the input current. FIG. 35 is a diagram showing experimental results. FIG. 36 is a diagram showing an input current. FIG. 37 is a diagram showing experimental results. FIG. 38 is a diagram showing the input current. FIG. 39 is a diagram showing experimental results. FIG. 40 is a diagram showing the input current. FIG. 41 is a diagram showing experimental results. FIG. 42 is a diagram showing the input current. FIG. 43 is a diagram illustrating a weight function. FIG. 44 is a diagram showing the frequency response of the LTI controller. FIG. 45 is a diagram showing an initial value response. FIG. 46 is a diagram showing the input current. FIG. 47 is a diagram showing an initial value response. FIG. 48 is a diagram showing the input current. FIG. 49 is a diagram showing an initial value response. FIG. 50 is a diagram showing the input current. FIG. 51 is a diagram showing an initial value response. FIG. 52 is a diagram showing the input current. FIG. 53 is a diagram of a closed-loop root locus. FIG. 54 is a diagram showing experimental results. FIG. 55 is a diagram showing an input current. FIG. 56 is a diagram showing experimental results. FIG. 57 is a diagram showing the input current. FIG. 58 is a diagram showing experimental results. FIG. 59 is a diagram showing an input current. FIG. 60 is a diagram showing experimental results. FIG. 61 is a diagram showing an input current. FIG. 62 is a diagram showing experimental results. FIG. 63 is a diagram showing the input current. FIG. 64 is a diagram showing experimental results. FIG. 65 is a diagram showing an input current. FIG. 66 is a diagram showing experimental results. FIG. 67 is a diagram showing an input current. FIG. 68 is a diagram showing experimental results. FIG. 69 is a diagram showing an input current. FIG. 70 is a diagram showing experimental results. FIG. 71 is a diagram showing an input current. FIG. 72 is a diagram in which the rotor of the magnetic bearing device is divided into eight elements. FIG. 73 is a diagram showing the relationship between the natural frequency and the shaft rotation speed. FIG. 74 is a Bode diagram. FIG. 75 is a diagram illustrating an initial value response of the reduced-order model. FIG. 76 is a block diagram of a closed loop system. FIG. 77 is a diagram illustrating an initial value response of the reduced-order model. FIG. 78 is a diagram illustrating closed-loop eigenvalue analysis. FIG. 79 is a diagram illustrating an initial value response of the reduced-order model. FIG. 80 is a diagram illustrating the concept of stability. FIG. 81 is a diagram showing a model of a 5-axis control type magnetic bearing device. FIG. 82 is a block diagram of the LQI control system. FIG. 83 is a diagram showing an initial response. FIG. 84 is a diagram showing the control current of the electromagnet. FIG. 85 is a diagram showing an initial response. FIG. 86 is a diagram showing the control current of the electromagnet. FIG. 88 is a diagram showing an initial response. FIG. 88 is a diagram showing the control current of the electromagnet. FIG. 89 is a block diagram of the experimental apparatus. FIG. 90 is a diagram showing experimental results. FIG. 91 is a diagram showing experimental results. FIG. 92 is a diagram showing experimental results. FIG. 93 is a diagram showing experimental results. FIG. 94 is a diagram showing experimental results. FIG. 95 is a diagram showing experimental results. FIG. 96 is a diagram showing experimental results. FIG. 97 is a diagram showing experimental results. FIG. 98 is an explanatory diagram of the gyro moment. FIG. 99 is a diagram illustrating a root locus of a closed loop. FIG. 100 is a block diagram of the LPV reduction system. FIG. 101 is a block diagram of an LPV reduction system with a filter. FIG. 102 is a Bode diagram of the LPV system. FIG. 103 is a diagram of a weight function. FIG. 104 is a diagram showing a Bode diagram of the controller. FIG. 105 is a diagram showing an initial value response, an impulse response, and a step response. FIG. 106 is a diagram of a switching function. FIG. 107 is a diagram of force control input. FIG. 108 is a diagram illustrating the control current of the X axis. FIG. 109 is a diagram illustrating a Y-axis control current. FIG. 110 is a diagram illustrating an initial value response, an impulse response, and a step response. FIG. 111 is a diagram showing experimental results. FIG. 112 is a diagram showing experimental results. FIG. 113 is a diagram showing experimental results. FIG. 114 is a diagram showing experimental results. FIG. 115 is a diagram showing experimental results. FIG. 116 is a diagram showing experimental results. FIG. 117 is a diagram showing experimental results. FIG. 118 is a diagram showing experimental results. FIG. 119 is a diagram showing experimental results. FIG. 120 is a diagram showing experimental results. FIG. 121 is a diagram illustrating a root locus of a closed loop. FIG. 122 is a diagram in which the rotor of the magnetic bearing device is divided into eight elements. FIG. 123 is a diagram showing the relationship between the natural frequency and the shaft rotation speed. FIG. 124 is a block diagram of the closed loop of the PID controller. FIG. 125 is a Bode diagram of the PID controller. FIG. 126 is a diagram showing experimental results. FIG. 127 is a diagram showing experimental results. FIG. 128 is a diagram showing experimental results. FIG. 129 is a diagram illustrating experimental results. FIG. 130 is a diagram showing experimental results. FIG. 131 is a diagram showing experimental results. FIG. 132 is a diagram showing experimental results. FIG. 133 is a diagram showing experimental results. FIG. 134 is a diagram showing a comparison result between simulation and experiment. FIG. 135 is an enlarged view of the lower right diagram of FIG. FIG. 136 is a diagram showing a comparison result between simulation and experiment. FIG. 137 is a diagram illustrating a root locus of a closed loop. FIG. 138 is a diagram illustrating a root locus of a closed loop. FIG. 139 is a diagram illustrating GS control using a table.

Explanation of symbols

(1) Machine body
(2) Controller
(4) Rotating body
(4a) Flywheel
(5) Axial magnetic bearing
(6) (7) Radial magnetic bearing
(8) Displacement detector
(9) Electric motor
(10) Rotation sensor
(11) (12) Protective bearing
(25) Axial displacement sensor
(26) (27) Radial displacement sensor unit
(28a) (28b) Axial electromagnet
(29a) (29b) (29c) (29d) (30a) (30b) (30c) (30d) Radial electromagnet
(31a) (31b) (31c) (31d) (32a) (32b) (32c) (32d) Radial displacement sensor

Claims (2)

  In order to suppress instability due to the effect of the gyro effect in a magnetic bearing device that performs zero-bias linear control for the purpose of reducing power consumption, it includes a linear parameter fluctuation system that adds a term that depends on the rotational frequency to the equation of motion. A magnetic bearing device is characterized in that a plurality of controllers with convex interpolation are prepared in advance, and gain schedule control for switching the controllers according to the rotation frequency is performed.   In order to suppress instability due to the effect of the gyro effect in the magnetic bearing device that performs zero-bias linear control for the purpose of reducing power consumption, it includes a linear parameter fluctuation system that adds a gyro effect term to the equation of motion. A magnetic bearing device that performs gain schedule control for gyro compensation in which a state quantity follows a hyperplane that constantly varies according to a variation parameter.

Images