JP2013047868A - Periodic disturbance suppression device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To solve the problem that improvement of robustness against identification model errors is desired in torque ripple suppression control using a periodic disturbance observer because it is necessary to consider variation of a plant due to secular change, variation of plant characteristics, and the like.SOLUTION: A periodic disturbance suppression device includes: a phase correction amount operation unit which operates a phase of a vector locus described by a frequency component of periodic disturbance to operate a phase correction amount; a gain amount calculation unit which calculates an advance speed of the vector locus described by the frequency component of the period disturbance and calculates a gain correction amount while comparing the advance speed with a threshold; and a rotation vector calculation unit which multiplies the phase correction amount and the gain correction amount to calculate a system identification model correction amount and corrects a system identification model of a periodic disturbance observer unit on the basis of the correction amount.

Description

  The present invention relates to a periodic disturbance suppressing device that suppresses periodic disturbance generated in a controlled object.

A periodic disturbance may occur from a controlled object like a control device or a motor control device having a learning function.
For example, in the case of a motor, a pulsation called a torque ripple is generated in principle as a control object that generates periodic disturbances, causing various problems such as vibration, noise, deterioration of riding comfort, and mechanical resonance. . In particular, in an embedded magnet synchronous motor (hereinafter referred to as a PM motor) that has become popular in recent years, a cogging torque ripple and a reluctance torque ripple are generated in a composite manner. As a countermeasure, there is known a method of suppressing torque ripple by electrically providing a compensation signal for canceling pulsation.
The methods are roughly classified into a feed-forward compensation method that compensates mainly for torque ripple based on an approximate expression and an electromagnetic field analysis result, and a feedback method.

  As a feedback method, a method of learning using a torque meter, a method of estimating and suppressing torque ripple from a motor current ripple, a torque ripple disturbance observer method based on current / rotation speed detection values, and the like are known. These feedback methods can also respond to changes in ripple characteristics online, but there are approximation errors from current ripple to torque ripple estimated values, band limitation of disturbance observer filters in the high frequency band, and the like.

  Further, torque ripple is a periodic disturbance generated at the rotation order of the motor, and its high-order component tends to coincide with the mechanical resonance frequency even at low-speed rotation. Therefore, when learning control is applied to a resonant variable speed drive system, countermeasures against instability due to sudden amplitude change and phase inversion are indispensable, and these are generally complicated high-order models and appropriate control adjustments. Therefore, it is difficult to effectively suppress periodic disturbances such as torque ripple. Japanese Patent Application Laid-Open No. H10-228561 is known as a means for suppressing this periodic disturbance.

In Patent Document 1, as shown in FIG. 12, the torque command Tref is input to the command value conversion unit 1 and d-axis and q-axis current commands id ^* on rotational coordinates (crossing dq axes) synchronized with the rotation of the motor ^. , Iq ^* and the current command to the vector-controlled inverter 2. The inverter 2 generates an output based on the current commands id ^* and iq ^* and controls the PM motor 3 connected with a load via a shaft. The shaft torque detection value T _det detected by the torque detector attached to the shaft and the rotor phase angle θ detected by the position detector are input to the observer unit 4.

  In the observer unit 4, the periodic pulsation of the PM motor is detected as a direct current by the frequency component extraction means of Fourier transform, the periodic disturbance on the frequency component is estimated by the periodic disturbance observer compensation unit 4a, and this periodic disturbance is detected. It is configured to be added to the current command so as to suppress.

  The periodic disturbance observer unit 4a is one of the control methods for suppressing the periodic disturbance, but the basic configuration of the control is the same as that of a general disturbance observer, and the periodic disturbance components are individually controlled. By using a system identification model expressed as a complex vector for each frequency component in the inverse system model of the disturbance observer, the disturbance of the frequency to be controlled is directly estimated and compensated. Thereby, although it is a comparatively simple control configuration, a high control effect can be obtained for the target frequency regardless of the order.

International Publication WO2010 / 24195A1

FIG. 13 is a control block diagram of the periodic disturbance observer disclosed in Patent Document 1, and shows an n-order component in a simplified manner.
The definition of each symbol in the figure is as follows.

P _n: plant, P ^ _n: indicates that the n of the controlled object output subscript is n th component: system identification model r _n: control command d _n: periodic disturbance, d ^ _n: periodic disturbance estimated value y _n.

* All the above variables are complex vectors expressed as X _n = X _An + jX _Bn .

G _F (s): Low-pass filter
controlled object: Control target PDO: Periodic Disturbance Observer
First, system identification is performed in advance for the plant P _n to be controlled, and it is expressed as a formula (1) in the form of a one-dimensional complex vector.
P ^ _n = P ^ _AN + jP ^ _Bn (1)
However, P ^ _An : n-order component real part of the identification result, P ^ _Bn : n-order component imaginary part of the identification result.

  For example, when the system identification result of 1 to 1000 Hz is expressed as a complex vector every 1 Hz, a table composed of 1000 one-dimensional complex vector elements can be constructed. It is also possible to express the identification result by an approximate expression. In any method, the system model can always be expressed by a simple one-dimensional complex vector.

It should be noted, is not limited to the above-described system identification model, P in the following specification ^ _n, r _n,
d _n, d ^ _n, is a complex vector, denoted y _n be _{X n = X An + jX Bn} .

As a control method, a low-pass filter G _F (s) obtained by simplifying Fourier transform is passed through the plant output (control target output y _n ) to extract a frequency component to be controlled by the periodic disturbance. This multiplies the inverse system represented by the reciprocal P ^ _n ^-1 of the above extracted system identification model, the period by taking the difference between the control command value r _n generated from the upper controller in the adder A1 to estimate the disturbance d _n. The estimated period disturbance d ^ _n as compensation command value, subtracted from the control command value r _n in an adder A2, suppressing periodic disturbance d _n to thereby be added to the adder A3. The above flow is a control method for suppressing the periodic disturbance caused by the periodic disturbance observer.

  In this control method, the accuracy of the system identification model with respect to the true value is the basis of control and affects the control performance. In order to improve the ability to suppress periodic disturbances, more accurate system identification is required.

  However, it is difficult to obtain an identification model with high accuracy, and it is necessary to consider events such as plant fluctuations due to secular changes and the follow-up to frequent changes in plant characteristics. The error from the true value may increase the convergence time until the suppression is completed, or in the worst case, the suppression control itself becomes a disturbance component due to the phase error, which may make the control unstable. For this reason, improvement in robustness against identification model errors is required.

  The present invention solves the above-described problems, and an object of the present invention is to provide a periodic disturbance suppressing device capable of correcting a system identification model error.

Claim 1 of the present invention has a controller for generating a control command value at the upper level, and the frequency component of the periodic disturbance to be suppressed is the reciprocal of the system identification model in the output of the controlled object in which the periodic disturbance occurs. An observer unit that suppresses periodic disturbances by subtracting the periodic disturbance estimated by the periodic disturbance observer from the control command value as a compensation command value; ,
A phase correction amount calculation unit that calculates a phase correction amount by calculating a phase of a vector locus drawn by each frequency component of the periodic disturbance on a complex vector plane during the periodic disturbance suppression control by the observer unit;
A gain correction amount calculation unit that inputs the output of the control target and corrects the gain of the control target output;
A system identification model correction value is calculated by multiplying the phase correction amount from the phase correction amount calculation unit by the gain correction amount from the gain correction amount calculation unit, and the phase correction amount and gain correction amount from the phase correction amount calculation unit A system identification model correction value is calculated by multiplying the gain correction amount from the calculation unit, and a rotation vector calculation unit for correcting the system identification model of the periodic disturbance observer unit based on the correction value is provided. Is.

According to a second aspect of the present invention, the control object is a motor, a motor output is input, a suppression amount of occurrence of periodic disturbance in the motor is estimated by a periodic disturbance observer unit, and the estimated amount is subtracted from a torque command and rotated. In what generates d and q axis current commands on coordinates and controls the motor via an inverter,
A phase correction amount calculation unit that calculates a phase correction amount by calculating a phase of a vector locus drawn by a frequency component of a periodic disturbance by inputting the motor output;
A gain correction amount calculation unit that calculates the speed of vector trajectory drawn by the frequency component of the periodic disturbance by inputting the motor output and compares it with a threshold value, and calculates the gain correction amount by repeating this calculation and comparison a plurality of times;
A system identification model correction value is calculated by multiplying the phase correction amount from the phase correction amount calculation unit by the gain correction amount from the gain correction amount calculation unit, and the system identification model of the periodic disturbance observer unit based on the correction value A rotation vector calculation unit for correcting the above is provided.

According to a third aspect of the present invention, the phase correction amount θ ^ref _n in the phase correction amount calculation unit is obtained by adding the previously calculated phase correction amount θ ^ref _n−1 to the product of the weight a and the vector rotation angle θ. It is characterized by that.

  According to a fourth aspect of the present invention, a learning function unit is provided on the output side of the rotation vector calculation unit, the system identification model correction value obtained by the rotation vector calculation unit is stored in the learning function unit, and the correction amount for the system identification model error And learning value is input to the periodic disturbance observer unit and multiplied by the identification model correction value to obtain a new identification model correction value.

  According to a fifth aspect of the present invention, the learning function unit is provided with an identification model correction function unit, and the identification model correction function unit repeatedly performs an operation point movement while changing the operation point at the time of storing the system identification model correction value. The periodic disturbance suppression control is turned on upon completion, and the system identification model correction value is stored upon completion of the suppression control, and the periodic disturbance suppression control is turned off.

  Claim 6 of the present invention is characterized in that an interpolation correction means is provided in the learning function section so that an arbitrary frequency width is given to the system identification model correction value after learning.

  According to a seventh aspect of the present invention, the identification model correction means having the phase correction amount calculation unit, the gain correction amount calculation unit, and the rotation vector calculation unit includes n orders of frequency components of the periodic disturbance. It is a thing.

  As described above, according to the present invention, it is possible to adaptively correct the error of the system identification model used for the periodic disturbance observer and to improve the robustness against the identification model error.

The block diagram of the identification model correction | amendment part which shows embodiment of this invention. The flowchart at the time of gain correction determination. The block diagram of the identification model correction | amendment part which shows the other Example of this invention. Complex vector plane locus diagram. The system identification figure at the time of simulation. The simulation result figure when there is no learning function. The simulation result figure when there is a learning function. The simulation result figure at the time of a fluctuation model. Flow chart of erroneous learning prevention processing. The block diagram of multiple order simultaneous detection. Interpolation process explanatory drawing. The block diagram of the conventional disturbance suppression apparatus. The period disturbance observer control block diagram.

  The present invention is provided in an identification model correction means having a phase correction amount calculation unit, a gain correction amount calculation unit, and a rotation vector calculation unit, inputs a control target output, calculates a correction phase in the phase correction calculation unit, and gain The correction amount calculation unit calculates a correction gain, and based on these correction amounts, the rotation vector calculation unit corrects the system identification model error by determining the rotation vector. This will be described in detail below with reference to the drawings.

Prior to the description of the present invention, a correction method for the system identification model error will be described.
As an output to be controlled, for example, for each frequency component of the torque ripple, in the complex vector plane with the nth order torque pulsation extraction component (cosine coefficient) T _An as the real axis and the nth order torque pulsation extraction component (sine coefficient) _TBn as the imaginary axis. Pay attention to the locus drawn.

FIG. 4 shows a complex vector plane trajectory diagram, where 100 _n indicates the position (current detection position) at the elapsed time [t = t _n ] from the start of suppression, and 100 _n−1 indicates the elapsed time from the start of suppression. The position at time [t = t _n-1 ] (detected position one cycle before) is shown.

  If there is no true value or error in the system identification model, it goes straight with an optimal response time from the suppression start point to the origin, that is, in the absence of torque ripple (periodic disturbance). This trajectory draws a curve or a circular trajectory in the presence of an error, and the worst is to diverge and travel toward infinity.

In this embodiment, it is assumed that the torque ripple (periodic disturbance dn in FIG. 13) is canceled by setting the _n- th order compensation current command value to zero. However, when the command value is other than zero, the vector value of the compensation current command is set on the vector plane. The position of corresponds to the origin.

In the present invention, during the suppression control, the rotation vector P _n ^ref of the gain G ^ref and the phase θ ^{ref in} the following equation (2) is determined from the trajectory information, and multiplied by the identification model P ^ _{n as shown} in the equation (3). By doing so, the identification model is corrected to obtain a new P ′ _n identification model. P ′ _n is applied to the identification model of the inverse system in which this is used in the periodic disturbance observer PDO.
P _n ^ref = G ^ref · (cos θ ^ref + j sin θ ^ref ) (2)
P ′ _n = P ^ _n · P _n ^ref (3)
FIG. 1 shows a control block diagram of the first embodiment that realizes the expressions (2) and (3). FIG. 1 shows the configuration of FIG. 13 in a simplified manner with respect to frequency components, and the same components as those in FIG.

In the configuration diagram of a correction unit of system identification model shown in Figure 1, 10 is actual plant, vector locus 11 in the phase correction amount calculation unit, from the control target output y _n of the real plant 10, the frequency components of the periodic disturbance draw And a correction amount θ ^ref is calculated as will be described later. 12 is a gain correction amount calculation unit, from the control target output y _n, the rate of progression of vector locus which each frequency component of the periodic disturbance draws (current value of the gain) | v | is calculated and corrected on the basis of FIG.

Reference numeral 13 denotes a rotation vector calculation unit, which calculates a rotation vector P _n ^ref (correction command value for the system identification model) by multiplying the phase θ ^ref and the gain G ^ref based on the equation (2), and uses this rotation vector P _n ^ref . The system identification model of the periodic disturbance observer unit 14 is corrected. The cycle disturbance observer 14, the controlled object output y _n and the control command is input, obtains a period estimated disturbance value d ^ _n by performing the operation shown in FIG. 13, the cycle estimated disturbance value d ^ _n is in the adder A2 is subtracted from the control command r _n, suppressing periodic disturbance d _n to be added by the adder A3.

The phase correction amount θ ^ref based on the equation (2) is determined as follows.
As shown in FIG. 4, the vector in the origin direction at the position (T _An, T _Bn ) at the elapsed time [t = t _n ] from the start of suppression is P _t , and the time [t = t _n−1 ] to the time [ _Let v be the vector to t = t _n ]. The rotation angle of v viewed from P _t is θ. If there is no model error, θ is always zero. If the phase correction amount θ ^ref is determined in the opposite direction to the detected phase θ, the error of the system identification model can be corrected.
θ = tan- ¹ (P _t × v / P _t · v) (4)
In the equation (4), the [×] symbol represents an outer product, and the [•] symbol represents an inner product.

If it is a small model error that can be suppressed without correction, a periodic disturbance observer generates a compensation command in a direction where θ = 0. If there is a model error, the correction amount is determined by approximating θ as the phase error of the identification model. In order to reduce the effect of approximating the phase error, the phase correction amount calculation unit 11 in the present invention, as shown by the equation (5), the phase correction amount θ ^ref is the product of the weights a and θ and the previous phase correction. Request theta ^ref by performing an operation of adding an amount θ ^ref _n-1.
Note that the initial value of the phase correction amount θ ^ref _n is zero. In this way, it is possible to flexibly cope with sudden fluctuations by adaptively adjusting θ ^ref , and the model is corrected in a direction in which θ becomes zero.
θ ^ref _n = θ ^ref _n-1 −a · θ (5)
Next, the gain correction amount calculation unit 12 determines the correction gain G ^ref based on the flowchart shown in FIG. If the phase error is small but the gain error is large, there is a possibility that the vibrational behavior around the target point or the convergence time will be very long. By performing gain correction in addition to phase correction, it is possible to establish correction control with high robustness to model errors.

The correction gain G ^ref is determined from the position vector Pt and the velocity vector v in FIG. 4 based on the flowchart shown in FIG. From start to end is repeated periodically in parallel with the suppression control, and the gain correction amount is sequentially determined. As an adjustment direction of the correction amount, an increase / decrease in the correction gain G ^ref corresponds to a decrease / increase in the moving speed. The initial value of the correction gain G ^ref is 1.

The processing procedure of (A) to (E) in FIG. 2 will be described below.
(A) When the absolute value | r | of the position vector is equal to or less than the threshold value rth, it is determined as a converged state and is not adjusted.
(B) In order to prevent divergent behavior, the movement speed is suppressed at a relatively fast cycle (for example, every several + [msec]). The traveling speed | v 1 | is calculated from the speed vector, and the determination threshold is determined to be k · | r | in proportion to | r |. If the traveling speed | v 1 | is equal to or greater than the threshold value, it is determined that the gain is insufficient, and the correction gain G ^ref is increased by a1 [%]. Here, a1 and k are arbitrarily set parameters.
(C) The correction control cycle is divided into N times, and the traveling speed | v 2 | in this cycle (for example, every several [sec]) is calculated.
(D) | v 2 | the determined G ^ref and the gain excessively if the threshold vth less a2 [%] decreases. This prevents a convergence delay state due to an excessive gain and increases the moving speed. vth and a2 are arbitrarily set parameters.
(E) Limit the correction gain G ^ref and store the final value in memory.
As described above, the identification model is corrected by the equation (3) using the phase and gain correction method described above using the vector locus of (Y _AN , Y _BN ).

  Therefore, according to this embodiment, the system identification model error used for the periodic disturbance observer can be adaptively corrected, and the robustness against the system identification model error can be improved.

In the first embodiment, the means for adaptively correcting the system identification model of the periodic disturbance observer unit at a certain frequency is shown. In the second embodiment, a memory (recording) function is added by providing a memory for recording the obtained final correction value P _n ^ref .
FIG. 3 shows a control block diagram having this learning function. Reference numeral 15 denotes a memory, to which the output P ^ref _n of the rotation vector calculation unit 13 is input through the switch 16 and the rotation angular velocity nω of the PM motor is input. The memory 15 and the switch 16 constitute a learning function unit.

The switch 16 is turned on / off at a timing when the final corrected rotation vector P ^ref _n is taken in and stored in the memory 15, and the periodic disturbance is sufficiently suppressed by the correction process of the identification model. Time. The rotation vector P _n ^ref at that time is stored in the memory, and the P ^ref _n is output to the periodic disturbance observer 14 as P ^mem _n . The periodic disturbance observer 14 determines the corrected identification model Pn ′ by further multiplying P ^mem _n as shown in the expression (6) with respect to the calculation of the expression (3).
P ′ _n = P ^{^} _n · P ^ref _n · P ^mem _n (6)
5 to 8 show the present embodiment using the learning function and the simulation results when the learning function is not used. In the simulation, the rotation speed / torque constant condition state (42 [Hz]), (number of poles 4) and (30 [Nm]) are set, and the correction cycle is 20 [ms]. As a condition for torque ripple, a constant amount was given to the electrical frequency 1 and 2 frequency components (1f, 2f). In both the gain and phase diagrams shown in FIGS. 5A and 5B, the line a is nominal, the line b is the fluctuation model 1, and the line c is the fluctuation model 2.

As an error condition of the system identification model, the model shown in FIG. 5 was set as a nominal identification model of the inverse model, and the simulation was performed with the actual plant set as an error as the variation model 1. 6 shows a case where the learning function is disabled, FIG. 7 shows a case where the learning function is enabled, and in each figure, (a) is a vector locus diagram when the learning function is disabled, and (b) is ( T _an ), (T _bn ) from the origin to the time response of the distance | r |, and (c) shows the time response of the shaft torque. As is apparent from FIG. 7, since the nominal identification model that controls the phase error and the speed in the divergence direction immediately after the start of suppression is corrected, the suppression control does not diverge even in the case of 1f and 2f. It can be confirmed that torque ripple can be suppressed.

  FIG. 8 is a result of verification on the system sudden change and disturbance fluctuation. In a state where the suppression is completed in the fluctuation model 1 shown in FIG. 5, the system is suddenly changed to the fluctuation model 2, and at the same time, the torque ripple is fluctuated twice. As is apparent from FIG. 8, it can be confirmed that the divergence operation is prevented by sequentially correcting the identification model against sudden system changes and disturbance fluctuations, and torque ripple is suppressed.

  Therefore, according to this embodiment, in addition to the effect of the first embodiment, it is possible to further learn (store) the correction amount for the identification model error at a certain frequency, so that the correction is completed in the operation at the same frequency again. It is possible to shorten the learning time until it is necessary or to make learning unnecessary.

Note that the correction amount is stored in a memory in order to learn the identification model correction amount shown in FIG. The storage is repeatedly performed by changing the operating point in the entire operating range of the apparatus or in a certain amount of range, but at this time, there is a risk of erroneous learning.
In the third embodiment, when the operating point is moved, the suppression control is turned on / off to prevent erroneous learning during movement.

  FIG. 9 shows a processing sequence for that purpose. The operating point is moved in step S1, and when the movement is completed (S2), the suppression control is turned on in S3. When the suppression control is completed in S4, it is stored in the memory (S5), and the suppression control is turned off in S6. Finally, the identification model in the entire frequency range corrected and executed until all operating points of S7 are completed is completed. When the operation is performed at a certain amount of operating points, the shortcomings are interpolated by an arbitrary method.

  As a result, even when the system identification includes an error, the suppression control is completed by the model correction function, and a new identification model can be obtained at the same time.

In each of the above-described embodiments, it has been shown that the identification model error is estimated for a specific frequency component to enable correction. In Example 3, as shown in FIG. 10, the n-number identification model correcting means with observer unit 20 ₁ to 20 _N of the (order n number of frequency components of the periodic disturbance) provided, FIG. 1, the control system of FIG. 3 The configuration is such that estimation model error estimation and correction for each order to be suppressed are performed in parallel and simultaneously.

The observer units 20 _{1 to} 20 _N with identification model correction means have functions such as the periodic disturbance observer unit 14, the phase correction calculation unit 11, the gain correction calculation unit 12, and the rotation vector calculation unit 13 shown in FIGS. The estimation model error is estimated with respect to the nth-order control commands r _{1 to} r _N , and the periodic disturbance estimated values d ^ _{1 to} d ^ _N obtained as a result of correcting the system identification model based on the error are output. Is.

  According to this embodiment, in addition to the effects of Embodiments 1 and 2, it is possible to cope with a case where a plurality of periodic disturbance frequency components for which suppression and system identification model errors are to be estimated exist simultaneously. .

In each of the second and subsequent embodiments, a learning function is added, but the learning point is for one point frequency.
That is, it is assumed that the system identification model in the initial state has a frequency response graph regarding P _A (P _B ) as shown in FIG. In each of the above embodiments, the result of learning the identification model error is that, as shown in FIG. 11B, the periodic disturbance suppression control is performed only at a specified frequency point a or point b associated with model variation. The point at which the model error can be learned is only a specific frequency. For this reason, when there is even a slight shift from the learned frequency, the result value learned at the latest frequency cannot be applied. When the control target is a variable speed application such as a motor, relearning is required even with a minute frequency fluctuation.

In this embodiment, interpolation means is provided in the identification model correction means, and interpolation processing by the interpolation means is performed on the frequency regions around the learning points a and b as shown in FIG. An arbitrary frequency width is provided.
Therefore, according to this embodiment, it is possible to stabilize the suppression control and shorten the time required for learning even when the frequency changes, by influencing the learning result at a specific frequency up to the peripheral frequency. It is.

DESCRIPTION OF SYMBOLS 1 ... Command value conversion part 2 ... Inverter 3 ... PM motor 4 ... Observer part 10 ... Real plant (control object)
DESCRIPTION OF SYMBOLS 11 ... Phase correction amount calculating part 12 ... Gain correction amount calculating part 13 ... Rotation vector calculating part 14 ... Periodic disturbance observer part 15 ... Memory 16 ... Switch

Claims (7)

It has a controller that generates control command values at the top, and multiplies the frequency component of the periodic disturbance to be suppressed by the inverse system expressed by the reciprocal of the system identification model in the output of the controlled object where periodic disturbance occurs. A periodic disturbance observer that estimates the periodic disturbance, and an observer unit that suppresses the periodic disturbance by subtracting the periodic disturbance estimated by the periodic disturbance observer from the control command value as a compensation command value;
A phase correction amount calculation unit that calculates a phase correction amount by calculating a phase of a vector locus drawn by each frequency component of the periodic disturbance on a complex vector plane during the periodic disturbance suppression control by the observer unit;
A gain correction amount calculation unit that inputs the output of the control target and corrects the gain of the control target output;
A system identification model correction value is calculated by multiplying the phase correction amount from the phase correction amount calculation unit by the gain correction amount from the gain correction amount calculation unit, and the phase correction amount and gain correction amount from the phase correction amount calculation unit A rotation vector calculation unit is provided that calculates a system identification model correction value by multiplying the gain correction amount from the calculation unit, and corrects the system identification model of the periodic disturbance observer unit based on the correction value. Periodic disturbance suppression device. The object to be controlled is a motor, the motor output is input, the amount of suppression of periodic disturbance occurrence in the motor is estimated by the periodic disturbance observer unit, the estimated amount is subtracted from the torque command, and the d and q axes on the rotational coordinates are In what generates a current command and controls a motor via an inverter,
A phase correction amount calculation unit that calculates a phase correction amount by calculating a phase of a vector locus drawn by a frequency component of a periodic disturbance by inputting the motor output;
A gain correction amount calculation unit that calculates the speed of vector trajectory drawn by the frequency component of the periodic disturbance by inputting the motor output and compares it with a threshold value, and calculates the gain correction amount by repeating this calculation and comparison a plurality of times;
A system identification model correction value is calculated by multiplying the phase correction amount from the phase correction amount calculation unit by the gain correction amount from the gain correction amount calculation unit, and the system identification model of the periodic disturbance observer unit based on the correction value The period disturbance suppression device according to claim 1, further comprising a rotation vector calculation unit for correcting 2. The phase correction amount θ ^ref _n in the phase correction amount calculation unit is obtained by adding the previously calculated phase correction amount θ ^ref _n−1 to the product of the weight a and the vector rotation angle θ. Or the periodic disturbance suppression apparatus of 2 description. A learning function unit is provided on the output side of the rotation vector calculation unit, the system identification model correction value obtained by the rotation vector calculation unit is stored in the learning function unit, the correction amount for the system identification model error is learned, and the learning value The periodic disturbance suppressing device according to claim 1, wherein the identification disturbance correction unit is multiplied by the identification model correction value to obtain a new identification model correction value. An identification model correction function unit is provided in the learning function unit, and the identification model correction function unit repeatedly performs operation while changing the operating point at the time of storing the system identification model correction value, and performs periodic disturbance suppression control when the operation point movement is completed. 5. The periodic disturbance suppression device according to claim 4, wherein the periodic disturbance suppression control is turned off by storing the system identification model correction value when the suppression control is completed and turning off the periodic disturbance suppression control. 6. The periodic disturbance suppressing device according to claim 4, wherein an interpolation correction unit is provided in the learning function unit to give an arbitrary frequency width to the system identification model correction value after learning. 7. The identification model correction means having the phase correction amount calculation unit, the gain correction amount calculation unit, and the rotation vector calculation unit includes n orders of frequency components of periodic disturbances. One of the periodic disturbance suppression devices.

Images