JP5644203B2 - Torque ripple suppression control device and control method for rotating electrical machine - Google Patents

Description

  The present invention relates to a control device and a control method for automatically suppressing torque ripple (torque pulsation) of a rotary electric machine in a torque control device of a rotary electric machine such as a motor, and more particularly to periodic disturbance by compensated current Fourier coefficient learning control. It relates to suppression control.

  In principle, the motor generates torque ripple (torque pulsation), causing various problems such as vibration / noise, adverse effects on riding comfort, and electrical / mechanical resonance. In particular, an embedded magnet PM motor (IPMSM), which has become popular in recent years, generates a combination of cogging torque ripple and reluctance torque ripple. As a countermeasure, various methods for superimposing a compensation current for canceling the torque ripple on the torque command have been studied.

  However, for example, in the method of performing feedforward compensation using a mathematical analysis model, there is a concern about the influence of analysis errors. Further, in the method of storing the feedback learning control result at the steady operating point and performing feedforward compensation, it takes time to appropriately adjust the control parameter for each operating point, and online compensation becomes difficult. Further, the method for reducing the current ripple is not always optimally suppressed from the viewpoint of torque ripple. In addition, torque ripple observer compensation methods have been studied, but the characteristics during variable speed operation and on-line feedback suppression verification have been insufficient.

  The inventors of the present application have already proposed a feedback suppression control method using an axial torque meter for the purpose of highly accurately suppressing torque ripple, which is a cause of electrical / mechanical resonance, in response to the above problems (for example, Non-Patent Document 1). reference). This control method focuses on the periodicity of torque ripple and constructs a control system that compensates for each pulsation frequency component, and also has a function that automatically adjusts parameters to respond quickly to changes in the operating state using the system identification result. The details of this control method will be described below.

(1) Basic Configuration of Torque Ripple Suppression Control Device FIG. 4 is a basic configuration diagram of a conventional torque ripple suppression control device. The figure is applied to a system in which a load is torque-excited by a motor.

A motor 1 that is a torque ripple generation source and some load device 2 are coupled by a shaft 3, and its shaft torque is measured by a torque meter 4 and input to a torque ripple suppression device 5. Moreover, the rotor position (phase) information of a motor is input using rotational position sensors 6, such as a rotary encoder. The torque ripple suppression device (controller) 5 is equipped with torque pulsation suppression means, and a command value obtained by adding a torque pulsation compensation current to the current command value generated based on the torque command value (or speed command value) is supplied to the inverter 7. give. In the example of FIG. 4, in consideration of the current vector control by the inverter 7, the d-axis and q-axis current command values id ^* and iq ^* on the rotation coordinates (orthogonal dq axes) synchronized with the rotation of the motor are given. Yes.

It is known that torque ripple (torque pulsation) periodically occurs according to the rotor position due to the structure of the motor. Therefore, means for extracting a torque ripple frequency component in synchronism with the motor rotation is used to convert the torque ripple of arbitrary order n into cosine and sine coefficients T _An and T _Bn [Nm]. The exact measurement means for torque ripple frequency components includes Fourier transform, but if importance is placed on the ease of operation, a low-pass filter can be passed through a single-phase harmonic rotation coordinate system based on the rotation phase θ [rad]. A torque ripple frequency component can be extracted.

The torque ripple suppression device 5 performs torque ripple suppression control using the cosine and sine coefficients T _An and T _Bn described above, and the cosine / sine coefficients I _An and I _Bn [of the torque pulsation compensation current iqc ^* [A] of an arbitrary frequency component. A] is generated. Conversion to the torque ripple compensation current iqc ^* is determined using the same rotational phase θ at the time of conversion, it performs a normal vector control the torque pulsation compensation current iqc ^* is superimposed on the q-axis current command value.

In FIG. 5, the cosine and sine coefficients T _An and T _Bn of the torque ripple are obtained from the shaft torque T _det and the rotational phase θ, and the compensation current cosine coefficient I _An and the sine coefficient I _Bn are obtained from these and the rotational speed ω. It is a control block diagram of the torque ripple suppression apparatus which calculates | requires the compensation current of torque periodic external suppression. The symbols in the figure have the following meanings.

T ^* : Torque command value, T _det : Torque detection value, T _An : n-th order torque pulsation extraction component (cosine coefficient), T _Bn : n-th order torque pulsation extraction component (sine coefficient), ω: Rotation speed detection value, θ : Rotation phase detection value, iqc ^* : torque ripple compensation current, id: d-axis current detection value, id ^* : d-axis current command value, iq: q-axis current detection value, iq ^* : q-axis current command value, iu. iv. iw: u, v, w-phase current, iqo ^* : q-axis current command value (before compensation current superposition), I _An : n-order compensation current cosine coefficient, l _Bn : n-order compensation current sine coefficient, abz: rotation sensor signal . Note that the subscript n is an nth-order component torque ripple.

In FIG. 5, the command value conversion unit 11 converts the torque command value T ^* into d-axis and q-axis current command values Id ^* and Iqo ^* in the rotation dq coordinate system in vector control. In general, conversion formulas and tables that realize maximum torque / current control are used. Current vector control unit 12 suppresses the torque ripple those superimposed torque pulsation compensation current iqc ^* the q-axis current command value Iqo ^* as q-axis current command value iq ^*. In the example of FIG. 5, the compensation current i _qc is superimposed on the q-axis current command value, but it may be applied to the d-axis current or both the d-axis and the q-axis. Alternatively, in a system where interference of dq axis current does not become a problem, a torque pulsation compensation signal may be directly superimposed on the torque command value.

  The current vector control unit 12 drives the load device 14 by performing a current vector control operation on a general orthogonal rotating coordinate system d-axis q-axis and driving the motor (IPMSM) 13 by vector control. The coordinate conversion unit 15 converts the three-phase alternating currents iu, iv, iw and the motor rotation phase θ detected by the current sensor 16 into currents id, iq in a dq axis orthogonal rotation coordinate system synchronized with the motor rotation coordinates. The rotation phase / speed detection unit 17 converts the rotation sensor signal abz of the rotation position sensor 18 such as an encoder into information on the rotation speed ω and the rotation phase θ.

The torque pulsation frequency component extraction unit 19 extracts torque periodic disturbance for each pulsation frequency component from the shaft torque detection value T _det detected by the shaft torque meter 20 and the rotation phase θ. A typical extraction means is Fourier transform. The method of extracting the pulsation component is arbitrary, but the emphasis is placed on ease of calculation, and the shaft torque detection value T _det [Nm] is multiplied by the nth-order cosine wave and sine wave based on the rotational phase θ, and each is passed through a low-frequency range. By applying the filter, the approximate Fourier transform shown in the equations (1) to (3) is performed. This is called torque ripple synchronous coordinate transformation.

£: Laplace transform, G _F : Pulsation extraction filter, ω _f : Pulsation extraction low-pass filter cut-off frequency [rad / s], s: Laplace operator The torque ripple suppression control unit 21 uses the expressions (1) to (3) _Torque ripple suppression control is performed with each order component using the converted T _An and T _Bn, and the cosine coefficient I _An ^* [A] and the sine of the n-th order frequency component i _qcn ^* [A] of the compensation current i _qc ^* [A]. A coefficient I _Bn ^* [A] is generated. The conversion to the n-th order compensation current i _qcn ^* is calculated from the equation (4) using the same rotation phase θ as that at the time of torque ripple synchronous coordinate conversion.

The torque command T ^* [Nm] is converted into a dq-axis current command value i _d ^* , i _q0 ^* that realizes maximum torque / current control, and a combined value i _qc ^* of each primary compensation current generated by the equation (4) ^. _Is superposed on i _q0 ^* to perform vector control. Basically, the contents of calculation processing performed by the torque ripple suppression control device are extraction of shaft torque pulsation component, torque ripple suppression, and compensation current signal generation, and other processing is performed by a general inverter.

  As a representative form of the torque ripple suppression control unit 21, a periodic disturbance observer compensation method or a compensation current Fourier coefficient learning control method described later can be used.

The compensation current generator 22 generates a torque pulsation compensation current command value iqc ^* by the above equation (4) and superimposes it on the q-axis current command value i _q0 ^* . In FIG. 5, the torque pulsation compensation current command value i _qc ^* is superimposed on the q-axis current command value i _q0 ^*. However, the d-axis current command value, the current command values for both the d-axis and the q-axis, the torque command value, etc. It is also possible to replace with.

(2) System Identification The system configuration shown in FIG. 4 is a multi-inertia torsional resonance system due to the moment of inertia of the PM motor 1, the load device 2, the torque meter 4, the coupling 3, and the like. When the detected value of the shaft torque is fed back, since there are a plurality of resonance / anti-resonance frequencies, the suppression control parameter must be appropriately determined according to the operation state. If the learning time of the control parameter is long, there is a risk of increasing the electric / mechanical resonance phenomenon, so a quick automatic adjustment function is necessary.

Therefore, in Non-Patent Document 1, in order to derive a variable nominal control parameter adapted to the rotational speed change, the system transfer characteristic from the output value to the input value of the torque ripple suppression device 5 in FIG. 4, that is, the torque pulsation compensation in FIG. The frequency transfer characteristic from the current command value iqc ^* to the detected value T _det of the shaft torque detector 4 is identified. Since the system identification method is a general technique, an arbitrary method can be used.

(3) Compensation current Fourier coefficient learning control method This is a torque ripple suppression control method described as method 1 in Non-Patent Document 1 described above. The torque ripple is obtained as a Fourier coefficient of a frequency component of the torque ripple, and from the above equation (4), The torque pulsation compensation current iqc ^* is obtained by calculation. In this control method, a system transfer function of a frequency component synchronized with the torque ripple frequency is expressed by a one-dimensional complex vector, and a real part and an imaginary part of the torque ripple of an arbitrary frequency component are extracted by Fourier transform or the like. By applying the cosine and sine Fourier coefficients to the real and imaginary parts of the complex vector, a feedback suppression control system is constructed.

  The torque pulsation compensation current coefficient is obtained by the IP (proportional / integral) learning control method. The proportional / integral gain is determined by the model matching method so that the closed loop characteristic of the IP suppression control system matches the pole arrangement of an arbitrary standard system reference model. In addition, since the parameters are automatically adapted using the system identification result and the rotational speed information, the mounting on the multi-inertia resonance system is facilitated (see, for example, Patent Document 1).

Kanno et al., “Examination of torque ripple suppression control method focusing on periodic disturbance of PM motor”, 2009 IEEJ Industrial Application Division Conference, I-615-618, Date: August 31-September 2, 2009 Sun, venue: Mie University

International Publication Number WO2010 / 024194A1

  In general, in a device that detects torque ripple and suppresses torque ripple by feedback control over the entire frequency band, the feedback control response decreases due to the influence of computation dead time in the high frequency band, and disturbance suppression performance in the high frequency band decreases. It becomes difficult to generate a desired compensation current.

  On the other hand, the periodic learning method obtained by the above-described IP (proportional / integral) learning control method generates a sine wave / cosine wave having the same frequency as that of the periodic disturbance, and then generates the sine / cosine coefficient (amplitude / cosine coefficient). (Equivalent to phase). Therefore, this method is easier to deal with high frequency disturbances than a controller that collectively generates compensation currents in all frequency bands, and can be expected to improve periodic disturbance suppression performance. Further, if a suppression controller is configured in parallel for each target order, torque ripple suppression of a plurality of orders can be handled at the same time.

Further, in FIG. 5, in order to construct a control system with coordinates synchronized with the torque ripple frequency, the system transfer characteristic from the torque ripple compensation current command value iqc ^* to the shaft torque detection value T _det is identified. From this system identification result, only an arbitrary frequency transfer characteristic is extracted and feedback controlled. In the steady state, the amplitude / phase transfer characteristic of the system synchronized with the torque ripple frequency can be expressed by a one-dimensional complex vector. Can be defined as follows. In addition, in the vector amount in the equation (5) and in the following equations, the symbol “→” is attached to the top of P, but in the specification, the arrow is omitted.

P _A (ω): Real part of real system, P _B (ω): Imaginary part of real system In Non-Patent Document 1, in order to construct a control system with torque ripple synchronous coordinates, an arbitrary n-th order component is obtained from Equation (5). Only the frequency transfer characteristic is extracted, and the actual system of torque ripple synchronous coordinates is expressed by equation (6). That is, it is noted that the amplitude / phase characteristics of an arbitrary n-order component system can be expressed by a simple one-dimensional complex vector P _n . As in the equations (1) to (4), the real part / imaginary part axes are defined with reference to the rotational phase θ, the cosine coefficient corresponds to the real part component, and the sine coefficient corresponds to the imaginary part component.

P _An : Real system n-order component real part, P _Bn : Real system n-order component imaginary part The system identification result in the control system of FIG. 5 can be similarly defined as a formula (7) with a one-dimensional complex vector. it can. In addition, although the estimated value in the equation (7) and the following equations are shown with the “^” symbol added to the top of P, it is written as “P ^” in the specification.

P ^ _An : n-order component real part estimated value of identification result, P ^ _Bn : n-order component imaginary part estimated value of identification result For example, when the system identification result from 1 to 1000 Hz is expressed by a complex vector for each IHz, A table composed of 1000 one-dimensional complex vector elements can be constructed. In Patent Document 1, this is divided into a real part and an imaginary part component to be look-up tables LUT _A and LUT _B relating to the rotational speed ω. The real part P ^ _An and the imaginary part P ^ _Bn of the identification system of the equation (7) synchronized with the torque ripple n-order component are referred to from the system identification tables LUT _A and LUT _B using linear interpolation.

  From the above, even in a complex system, the system model is always a simple one-dimensional complex vector, and equation (7) can be changed instantaneously according to changes in motor rotation speed ω (torque ripple frequency change). It can easily handle variable speed operation.

By the way, in Patent Document 1, the pulsation extraction filter G _F (s) in the expression (3) is limited to a primary low-pass filter to construct a control system. In addition, a model matching method is used to calculate the closed loop response, and a method in which the user arbitrarily designates the learning control parameter has been adopted.

  However, in practice, it is possible to increase the order of the low-pass filter, and in that case, it is necessary to reconsider the closed-loop response. For example, when a high-order low-pass filter is used as the pulsation extraction filter in Patent Document 1, perfect model matching cannot be performed, and a desired suppression effect cannot be obtained.

  An object of the present invention is to provide a torque ripple suppression control device and a control method capable of automatically adjusting a learning control parameter to obtain a desired suppression effect even when the filter order of a torque pulsation extraction filter is increased. .

  In order to solve the above-mentioned problem, the present invention provides a representative pole arrangement of a model approximately by using a partial model matching method for periodic learning control of torque ripple suppression using a high-order low-pass filter as a torque pulsation extraction filter. And is characterized by the following apparatus and method.

(Invention of torque ripple suppression control device)
(1) A torque pulsation component included in the torque detection value Tdet of the driven rotating electrical machine is detected in the form of two n-th order complex Fourier coefficients TAn and TBn through a torque pulsation extraction filter, and the complex Fourier coefficient is detected by a learning control system. The control parameters are automatically adjusted to suppress TAn and TBn to 0, and the torque ripple of the rotating electrical machine is suppressed by using the nth-order compensation currents IAn * and IBn * of the torque pulsation obtained by the learning control system. In the torque ripple suppression control device,
The torque pulsation extraction filter is a high-order low-pass filter configured to detect two high-order complex Fourier coefficients TAn and TBn,
The learning control system is a part for matching only the low-order poles that govern the response of the rotating electrical machine among the n-order compensation currents IAn * and IBn * obtained by the learning control using the high-order complex Fourier coefficients TAn and TBn. The control parameters are automatically adjusted by a dynamic model matching method, and configured by an I-PD (proportional / integral / derivative) learning control method, and the three control parameters can be selected from any torque pulsation extraction filter (regardless of order and filter type). (1) is provided .

(2) Torque pulsation components included in the detected torque value Tdet of the rotating electric machine to be driven are detected in the form of two n-th order complex Fourier coefficients TAn and TBn through a torque pulsation extraction filter, and the complex Fourier coefficients are detected by a learning control system. The control parameters are automatically adjusted to suppress TAn and TBn to 0, and the torque ripple of the rotating electrical machine is suppressed by using the nth-order compensation currents IAn * and IBn * of the torque pulsation obtained by the learning control system. In the torque ripple suppression control device,
The torque pulsation extraction filter is a high-order low-pass filter configured to detect two high-order complex Fourier coefficients TAn and TBn,
The learning control system is a part for matching only the low-order poles that govern the response of the rotating electrical machine among the n-order compensation currents IAn * and IBn * obtained by the learning control using the high-order complex Fourier coefficients TAn and TBn. Automatic adjustment of the control parameters by a dynamic model matching method,
A learning control response frequency ωC of an arbitrary standard model (a reference model serving as a reference) is also used as a control parameter, and means for performing model matching is provided.

(Invention of torque ripple suppression control method)
(3) Torque pulsation components included in the detected torque value Tdet of the rotating electric machine to be driven are detected in the form of two n-order complex Fourier coefficients TAn and TBn through a torque pulsation extraction filter, and the complex Fourier coefficients are detected by a learning control system. The control parameters are automatically adjusted to suppress TAn and TBn to 0, and the torque ripple of the rotating electrical machine is suppressed by using the nth-order compensation currents IAn * and IBn * of the torque pulsation obtained by the learning control system. In the torque ripple suppression control method,
The torque pulsation extraction filter is a high-order low-pass filter and detects in the form of two high-order complex Fourier coefficients TAn and TBn,
The learning control system is configured by an I-PD (proportional / integral / derivative) learning control system, and the three control parameters are associated with an arbitrary torque pulsation extraction filter (regardless of order and filter type). And

ただし、Ｐ^AN：同定結果のｎ次成分実部推定値、Ｐ^ＢＮ：同定結果のｎ次成分虚部推定値、ωｃ：学習制御応答周波数［rad／ｓ］、ωｆ：フーリエ変換応答周波数、 (4) The learning control system performs a calculation of the following expression 1-1 in order to make the three control parameters Kp, Ki, Kd correspond to the torque pulsation extraction filter .

However, P ^ AN: n-order component real part estimated value of identification result, P ^ BN: n-order component imaginary part estimated value of identification result, ωc: learning control response frequency [rad / s], ωf: Fourier transform response frequency ,

(5) Torque pulsation components included in the detected torque value Tdet of the rotating electric machine to be driven are detected in the form of two nth-order complex Fourier coefficients TAn and TBn through a torque pulsation extraction filter, and the complex Fourier coefficients are detected by a learning control system. The control parameters are automatically adjusted to suppress TAn and TBn to 0, and the torque ripple of the rotating electrical machine is suppressed by using the nth-order compensation currents IAn * and IBn * of the torque pulsation obtained by the learning control system. In the torque ripple suppression control method,
The torque pulsation extraction filter is a high-order low-pass filter and detects in the form of two high-order complex Fourier coefficients TAn and TBn,
The learning control system is a part for matching only the low-order poles that govern the response of the rotating electrical machine among the n-order compensation currents IAn * and IBn * obtained by the learning control using the high-order complex Fourier coefficients TAn and TBn. Automatic adjustment of the control parameters by a dynamic model matching method,
The model matching is performed using the learning control response frequency ωC of an arbitrary standard system model (reference model as a reference) as a control parameter.

  As described above, according to the present invention, the representative pole arrangement of the model is approximately maintained using the partial model matching method for the periodic learning control of torque ripple suppression using the high-order low-pass filter as the torque pulsation extraction filter. Therefore, even when the filter order of the torque pulsation extraction filter is increased, the learning control parameter can be automatically adjusted to obtain a desired suppression effect.

FIG. 4 is a calculation block diagram of Fourier coefficient learning control (first embodiment). FIG. 4 is a calculation block diagram of Fourier coefficient learning control (second embodiment). The example of the experimental result of the torque ripple suppression control by this invention. The basic block diagram of the conventional torque ripple suppression control apparatus. The control block diagram of the conventional torque ripple suppression control apparatus.

(Embodiment 1)
FIG. 1 is a calculation block diagram of Fourier coefficient learning control in the present embodiment. In the figure, a torque ripple frequency component extraction unit (19 in FIG. 5) 34 is configured as a calculation block according to the above equations (1) to (3). The torque ripple suppression control unit (21 in FIG. 5) includes an IP control unit 31 and a non-interacting unit 32 in torque ripple synchronous coordinates, and a cosine coefficient (real part) T _An ^* and a sine coefficient (imaginary part) of the torque ripple command value. Part) 0 is input to T _Bn ^* for suppression. Reference numeral 33 denotes an actual system.

Here, when the n-order compensation current i _qcn ^* shown in the above equation (4) is expressed by a complex vector and given to the system expressed by the above equation (6), the n-th order of the torque detection value T _det The complex Fourier coefficient of the component T _detn can be expanded as shown in equation (8) using Euler's formula (however, the periodic disturbance component is excluded).

When the real part component T _An and the imaginary part component T _Bn of the expression (8) are compared with each other for the positive order (n> 0) complex Fourier coefficients, they can be expressed as the following expression (9).

As can be seen from this equation (9), it includes a real part I _An ^* and an imaginary part I _Bn ^* of compensation current to each of the real and imaginary parts of the n th component T _DETN of the torque detection value, interferes I understand that. In order to cancel this interference, the non-interacting unit 32 according to the equation (10) is configured.

  If the identification result is correct and formula (10) is substituted into formula (9) to make it non-interfering, it is replaced as formula (11), and the compensation current for the real part and the imaginary part can be controlled independently and identically. .

In conventional Patent Document 1, control design is performed by limiting the pulsation extraction filter G _F (s) to the primary low-pass filter. However, in the present embodiment, the torque pulsation extraction filter G _F (s) has a second or higher order. A proportional preceding type IP suppression control system is configured using a low-pass filter.

  Although the order and type of the low-pass filter are arbitrary, a case where the second-order binomial coefficient form shown in the equation (12) is used as an example will be described.

Substituting the equation (12) into the equation (11) and _obtaining the closed-loop transfer function from the torque target values T _An ^* and T _Bn ^* to the detected values T _An and T _Bn in FIG.

  A model matching method is a method for matching the response of the control target with respect to the target value to an arbitrary reference model pole arrangement. In the IP control, there are two control parameters. on the other hand. As can be seen from the equation (13), the characteristic polynomial of the closed-loop transfer function is a cubic equation, so there are three poles.

  In this case, the response of the reference model is approximated using a partial model matching method that matches only the representative pole (that is, the low-order pole that dominates the response of the controlled object). For example, here, the pole placement is approximated using a third-order binomial coefficient standard system reference model shown in the near (14) equation.

ω _C : Learning control response frequency [rad / s]
Regarding the equations (13) and (14), the coefficients are compared with polynomials obtained by dividing the denominator by the numerator.

However, as described above, since the IP control system has two control parameters, K _p and K _i , the order of coefficient comparison with respect to s is limited to the first and second orders. As a result of the coefficient comparison, equation (16) is derived.

From the equation (16), it can be seen that the control parameters K _p and K _i are variable gains of a speed adaptive type by P ^ _An and P ^ _Bn read from the system identification table.

  Therefore, by using the variable proportional / integral gain of the equation (16) designed by the partial model matching method, the parameters can be automatically adjusted so as to always approximate the pole arrangement of the desired reference model even if there is a speed change. . Thereby, the adjustment of the torque ripple suppression control at various operating points can be saved.

  In the present embodiment, the filter order / type is a second-order binomial coefficient form as a representative example, but other filter forms such as a third-order order or a Butterworth form can be similarly designed.

(Embodiment 2)
In this embodiment, the first embodiment is expanded, and the IP control unit of the Fourier coefficient learning control is configured by proportional differential preceding type I-PD (proportional / integral / differential) control. This configuration diagram is shown in FIG. 2 and the description of the basic method is the same as that of the first embodiment. However, by adding a differential control element (Kd · s), the response waveform can be improved, or a part in the first embodiment. The degree of approximation (accuracy) of the dynamic model matching method.

For example, when the second-order binomial coefficient form expressed by the equation (12) is applied to the pulsation extraction filter G _F (s) as in the first embodiment, the detected value T _{An is obtained} from the target values T _An ^* and T _Bn ^*. , T _Bn , the closed-loop transfer function is obtained as follows.

  Similarly to the first embodiment, when the coefficients of the reference model are compared with the polynomials obtained by dividing the denominator by the numerator with respect to the expressions (14) and (17) as the expression (14), the following expression (18) is obtained.

  In the case of I-PD control, since there are three control parameters, Kp, Ki, and Kd, it is possible to match coefficients of all orders related to s in the equation (18). As a result of the coefficient comparison, the following equation (19) can be derived.

From equation (19), it can be seen that the control parameters Kp, Ki, and Kd are speed adaptive variable gains by P ^ _An and P ^ _Bn read from the system identification table.

  Therefore, by using the variable proportional / integral / differential gain of the equation (16) designed by the partial model matching method, the parameter can always be automatically adjusted to the pole arrangement of the desired reference model even if there is a speed change. Thereby, adjustment of torque ripple suppression control at various operating points can be saved.

  Further, although the first embodiment is a partial model matching, the second-order filter of the present embodiment realizes a complete model matching and can eliminate a response error due to approximation. As for the third and higher orders, the closed loop transfer function of the suppression control system can be approximated to a desired response by partially matching from the lower order as in the first embodiment.

(Embodiment 3)
The learning control response frequency ω _C in the first and second embodiments is a parameter that is arbitrarily determined according to the suppression response characteristic desired by the user. However, since the partial model matching method has insufficient parameters, the learning control response frequency ω _C can also be treated as a parameter that determines the pole arrangement of the closed-loop transfer function, and the approximation accuracy can be improved. In the present embodiment, the pole arrangement is determined by actively using the learning control response frequency ω _C.

Here, as a specific example, a calculation example in the case of configuring proportional precedence type IP control using a quaternary binomial coefficient filter of equation (20) for the pulsation extraction filter G _F (s) is shown.

  As in the first and second embodiments, when a closed loop transfer function from the target value is obtained, equation (21) is obtained.

  From the partial model matching method, this target value response is approximated to the pole arrangement of the fifth-order binomial coefficient standard system model shown in Equation (22).

ω _C : Learning control response frequency [rad / s]
The IP control parameter has two variables, but the learning control response frequency ω _C that determines the learning response is also regarded as a control parameter and is set to three variables, and the denominator is divided by the numerator in each of equations (21) and (22). Approximate up to the low order 3rd order. That is, the proportional gain Kp, integral gain Ki, and learning control response frequency ω _C of equation (24) are derived from the coefficient comparison of equation (23) below.

From the equation (15), the gains Kp and Ki are speed adaptive variable gains by P ^ _An and P ^ _Bn read from the system identification table. In addition, the learning control response frequency ω _C treated as an arbitrary parameter in the first and second embodiments is designated, and there are no undetermined parameters other than the pulsation extraction filter cutoff frequency ω _f .

  According to the present embodiment, not only the approximation accuracy to the reference model response of Equation (22) can be improved, but also the desired response can be realized by determining the design guideline of the undetermined parameter.

  It should be noted that the control gain parameter can be similarly designed for other orders and filter types. Although the present embodiment has been described with the IP control type, other PID control types can be similarly realized.

  FIG. 3 shows the result when the experiment was performed by applying the parameter of the equation (24). The waveform shows from the top the shaft torque detection waveform, the Fourier coefficient of the sixth-order component torque detection value, the Fourier coefficient of the sixth-order component compensation current, and the torque pulsation amplitude value. In FIG. 3, it can be seen that when the suppression starts at 0 seconds, the resonance component of the shaft torque is suppressed and the pulsation component is reduced with a good response. Note that the pulsation extraction filter at this time is quaternary and has a cutoff frequency of 1 Hz.

DESCRIPTION OF SYMBOLS 11 Command value conversion part 12 Current vector control part 13 Motor 14 Load apparatus 15 Coordinate conversion part 16 Current sensor 17 Rotation phase / speed detection part 18 Rotation position sensor 19 Torque pulsation frequency component extraction part 20 Shaft torque detector 21 Torque ripple suppression control part 22 Compensation current generation unit 31 IP control unit 32 Decoupling unit 33 Real system 34 Torque ripple extraction unit

Claims (5)

Torque pulsation components included in the detected torque value Tdet of the rotating electric machine to be driven are detected in the form of two n-order complex Fourier coefficients TAn and TBn through a torque pulsation extraction filter, and the complex Fourier coefficients TAn and TBn are detected by a learning control system. Control parameter of the rotating electrical machine, which automatically adjusts the control parameter so as to suppress the torque to 0 and suppresses the torque ripple of the rotating electrical machine using the n-th order compensation currents IAn * and IBn * of the torque pulsation obtained by this learning control system In the device
The torque pulsation extraction filter is a high-order low-pass filter configured to detect two high-order complex Fourier coefficients TAn and TBn,
The learning control system is a part for matching only the low-order poles that govern the response of the rotating electrical machine among the n-order compensation currents IAn * and IBn * obtained by the learning control using the high-order complex Fourier coefficients TAn and TBn. The control parameters are automatically adjusted by a dynamic model matching method, and configured by an I-PD (proportional / integral / derivative) learning control method, and the three control parameters can be selected from any torque pulsation extraction filter (regardless of order and filter type). A torque ripple suppression control device for a rotating electrical machine, characterized in that it comprises means for handling Torque pulsation components included in the detected torque value Tdet of the rotating electric machine to be driven are detected in the form of two n-order complex Fourier coefficients TAn and TBn through a torque pulsation extraction filter, and the complex Fourier coefficients TAn and TBn are detected by a learning control system. Control parameter of the rotating electrical machine, which automatically adjusts the control parameter so as to suppress the torque to 0 and suppresses the torque ripple of the rotating electrical machine using the n-th order compensation currents IAn * and IBn * of the torque pulsation obtained by this learning control system In the device
The torque pulsation extraction filter is a high-order low-pass filter configured to detect two high-order complex Fourier coefficients TAn and TBn,
The learning control system is a part for matching only the low-order poles that govern the response of the rotating electrical machine among the n-order compensation currents IAn * and IBn * obtained by the learning control using the high-order complex Fourier coefficients TAn and TBn. Automatic adjustment of the control parameters by a dynamic model matching method,
A torque ripple suppression control device for a rotating electrical machine, comprising means for performing model matching using a learning control response frequency ωC of an arbitrary standard model (normative reference model) as a control parameter . Torque pulsation components included in the detected torque value Tdet of the rotating electric machine to be driven are detected in the form of two n-order complex Fourier coefficients TAn and TBn through a torque pulsation extraction filter, and the complex Fourier coefficients TAn and TBn are detected by a learning control system. Control parameter of the rotating electrical machine, which automatically adjusts the control parameter so as to suppress the torque to 0 and suppresses the torque ripple of the rotating electrical machine using the n-th order compensation currents IAn * and IBn * of the torque pulsation obtained by this learning control system In the method
The torque pulsation extraction filter is a high-order low-pass filter and detects in the form of two high-order complex Fourier coefficients TAn and TBn,
The learning control system is configured by an I-PD (proportional / integral / differential) learning control system, and the three control parameters are associated with an arbitrary torque pulsation extraction filter (regardless of order and filter type). torque ripple suppression control method to be. 4. The torque ripple suppression control method according to claim 3, wherein the learning control system performs calculation of the following expression 1-1 in order to make the three control parameters Kp, Ki, and Kd correspond to the torque pulsation extraction filter .

However, P ^ AN: n-order component real part estimated value of identification result, P ^ BN: n-order component imaginary part estimated value of identification result, ωc: learning control response frequency [rad / s], ωf: Fourier transform response frequency , Torque pulsation components included in the detected torque value Tdet of the rotating electric machine to be driven are detected in the form of two n-order complex Fourier coefficients TAn and TBn through a torque pulsation extraction filter, and the complex Fourier coefficients TAn and TBn are detected by a learning control system. Control parameter of the rotating electrical machine, which automatically adjusts the control parameter so as to suppress the torque to 0 and suppresses the torque ripple of the rotating electrical machine using the n-th order compensation currents IAn * and IBn * of the torque pulsation obtained by this learning control system In the method
The torque pulsation extraction filter is a high-order low-pass filter and detects in the form of two high-order complex Fourier coefficients TAn and TBn,
The learning control system is a part for matching only the low-order poles that govern the response of the rotating electrical machine among the n-order compensation currents IAn * and IBn * obtained by the learning control using the high-order complex Fourier coefficients TAn and TBn. Automatic adjustment of the control parameters by a dynamic model matching method,
A torque ripple suppression control method , wherein model matching is performed using a learning control response frequency ωC of an arbitrary standard system model (reference model as a reference) as a control parameter .

Images