Self-adaptive method for active feedforward compensation of cogging torque of super-multipole permanent magnet synchronous motor

Abstract

The present application relates to the technical field of permanent magnet synchronous motor control, and particularly relates to an adaptive method for actively feeding forward and compensating cogging torque of super-multipole permanent magnet synchronous motor, which is as follows: a cogging torque feedforward compensation model is established, and the amplitude-phase nonlinear parameters are converted into linear weight coefficient form through triangular orthogonal decomposition; secondly, the AC component of the q-axis voltage output by the current PI regulator is extracted as a compensation residual error signal, the weight coefficient is updated on-line and iteratively using the normalized least mean square algorithm, and adaptive tracking of the cogging torque change caused by temperature drift and mechanical wear is realized; finally, combined with a dual-redundant position feedback system, a double-channel independent weight coefficient maintenance, cross-checking and seamless fault switching mechanism is established. The present application can reduce torque ripple under low-speed working conditions, improve the low-speed stability and control accuracy of super-multipole permanent magnet synchronous motor, and meet the functional safety requirements of aviation-grade electric drive systems.

Description

Technical Field

[0001] This invention relates to the field of permanent magnet synchronous motor control technology, and in particular to an adaptive method for active feedforward compensation of cogging torque in a multi-pole permanent magnet synchronous motor. Background Technology

[0002] Permanent magnet synchronous motors (PMSMs), with their core advantages of high efficiency and high power density, have been widely used in high-precision control fields such as electric aircraft, electric vehicles, and high-end servo motors. However, their inherent cogging torque problem has become a key bottleneck restricting their low-speed stability and control accuracy. Cogging torque is the periodic torque pulsation generated by the interaction between the magnetic field of the permanent magnet and the cogging teeth of the stator core. Its characteristics are independent of the stator current. Even when the motor is in an unexcited state, the rotor will still generate significant positioning torque fluctuations. This problem can lead to speed fluctuations and vibrations when the motor is running at low speeds. Especially in the propeller / rotor systems of electric aircraft, the low-frequency torque pulsations caused by this problem can directly reduce hovering stability and attitude control accuracy, and in severe cases, induce fuselage shaking and noise.

[0003] The amplitude and oscillation order of cogging torque are determined by the pole-slot matching relationship of the motor: when the least common multiple of the pole-slot matching is large, the fundamental period is short and the amplitude is relatively small; when it approaches a certain ratio, the amplitude can reach several percent or even higher of the rated torque. For ultra-multiple pole permanent magnet synchronous motors such as 192 poles and 180 slots, its cogging torque spectrum not only includes the fundamental component, but the amplitudes of higher harmonics such as the 3rd and 5th can also reach more than 20% of the fundamental component, and traditional methods that only compensate for the fundamental component cannot effectively suppress them.

[0004] Traditional suppression techniques focus on optimizing the motor's structure, including pole offset, skewed slots / pole configuration, auxiliary slots, and optimized pole arc coefficients. These methods are unaffected by operating conditions and are effective once the design is finalized. However, they have significant limitations: first, they may sacrifice some electromagnetic torque output or increase manufacturing complexity; second, they cannot improve motors already in production; and third, they can only optimize for specific operating points, and their effectiveness deteriorates significantly under light loads, low speeds, or weak magnetic fields.

[0005] With the development of power electronics and digital control technology, control-level suppression methods have become a research hotspot. Mainstream methods include torque ripple feedforward compensation, repetitive control, iterative learning control, and disturbance observer-based compensation methods. Among these, feedforward compensation is the most widely used due to its simple structure and rapid response. However, existing control technologies still suffer from three major shortcomings: First, harmonic modeling is severely inadequate, typically only considering the fundamental or second harmonic, failing to cover the considerable amplitude of third and fifth harmonics in multi-pole motors, and potentially introducing new fluctuations due to phase errors. Second, the compensation coefficients are factory-calibrated fixed values, unable to adapt to changes in cogging torque amplitude and phase drift caused by temperature variations, bearing wear, and magnetic circuit saturation, resulting in a significant decrease in compensation effectiveness after long-term operation. Third, the integration with dual-redundant position sensing systems in safety-critical applications such as electric aircraft is low, failing to utilize redundant information to ensure the reliability of compensation functions and diagnose faults, thus limiting the improvement of system functional safety levels.

[0006] Therefore, this invention proposes an adaptive method for active feedforward compensation of cogging torque in ultra-multipole permanent magnet synchronous motors to solve the above problems. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention proposes an adaptive method for active feedforward compensation of cogging torque in ultra-multipole permanent magnet synchronous motors. This invention aims to solve the technical problems in existing methods for suppressing cogging torque in permanent magnet synchronous motors, such as the compensation model not fully considering higher harmonics, the inability of the compensation coefficient to adapt online, and the lack of deep integration with redundant position sensors.

[0008] The technical solution of this invention to solve the technical problem is an adaptive method for active feedforward compensation of cogging torque in a multi-pole permanent magnet synchronous motor, comprising the following steps: S1. Establish a feedforward compensation model containing odd harmonics of cogging torque in the traditional FOC current loop, orthogonally decompose the model into a linear weighted coefficient form, and then apply the real-time collected motor rotor electrical angle... Calculate the q-axis current feedforward compensation value using the current weighting coefficients. ; S2, feedforward compensation value of q-axis current The original torque current command output by the speed loop Superimposed to form the final q-axis current command And it is fed into the current PI regulator; S3, Extract the q-axis voltage command output from the current PI regulator. The AC component is used as the error signal for cogging torque compensation residual. ; S4. Determine the adaptive update enable condition. If it is met, then use the error signal based on the adaptive algorithm. and rotor electrical angle The linear weight coefficients of the feedforward compensation model are updated iteratively online. S5. Smooth the updated weight coefficients, use the smoothed weight coefficients as the current weight coefficients for the next control cycle, and periodically store them in the controller's non-volatile memory. S6. During power-on initialization, the stored weight coefficients are loaded as the initial weight coefficients. If loading fails, the preset initial weight coefficients are used.

[0009] The establishment and orthogonal decomposition process of the cogging torque feedforward compensation model in step S1 is as follows: Based on cogging torque With rotor electrical angle The periodic relationship, cogging torque The Fourier series expansion is as follows: , in, An odd number indicates the harmonic order; The first Fourier series representing the cogging torque The amplitude of the second harmonic torque component; Indicates the first The initial phase corresponding to the subharmonic torque component; According to the electromagnetic torque of the motor With q-axis current linear relationship , Representing the motor torque constant, the harmonic model of the feedforward compensation current is derived. Due to the magnetic field-oriented control framework, the motor electromagnetic torque... With q-axis current Approximately satisfies a linear relationship Therefore, by adjusting the q-axis current, a compensating electromagnetic torque with the same amplitude but opposite phase to the cogging torque can be generated, thereby achieving active suppression of the cogging torque. The required compensating current can thus be expressed as... Furthermore, the corresponding harmonic compensation model is obtained, as follows: , in, Indicates the cogging torque corresponding to the first The amplitude of the q-axis compensation current for the second harmonic component. ; Indicates the first The initial phase of the secondary compensation current harmonic. ; For ultra-multi-pole permanent magnet synchronous motors, select The compensation model is obtained as follows: , Then, using trigonometric identities, it is orthogonally decomposed into a linear weight coefficient form: , make , ,get: , in, and They represent Adaptive weighting coefficients corresponding to the subharmonic sinusoidal basis function and cosine basis function; The amplitude and phase of each harmonic are recovered from the weighting coefficients as follows: , .

[0010] The preset initial weight coefficients are obtained through offline initial calibration: Drive the motor to a preset constant low speed, so that the current loop is in a zero torque command state, measure the output shaft torque fluctuation, or calculate the cogging torque waveform in reverse through the speed loop PI output. Fourier analysis is performed on the acquired waveforms to extract the amplitude and phase of the 1, 3, and 5 times electrical angular frequency components, which are then converted into initial weighting coefficients and stored in the controller's non-volatile memory.

[0011] The error signal mentioned in step S3 The extraction method is as follows: q-axis voltage command The error signal is obtained by filtering out the DC component using a high-pass filter with a cutoff frequency lower than the current electrical angular frequency. This error signal directly reflects the uncompensated residual of the cogging torque.

[0012] The adaptive algorithm described in step S4 is the normalized least mean square algorithm, and its weight coefficient update formula is: , , in, This indicates the odd harmonic order of the cogging torque compensation. =1, 3, 5, =1 indicates the first fundamental harmonic. =3 indicates the 3rd harmonic. =5 indicates the 5th harmonic; express The weighting coefficients of the sinusoidal terms corresponding to the second harmonic order. express The weighting coefficients of the cosine terms corresponding to the second harmonic order; μ represents the adaptive step size.

[0013] The smoothing process described in step S5 uses a first-order low-pass filter, and the calculation formula is as follows: , in, Represents the filter coefficients. This represents the updated weight coefficients. These are the weight coefficients after smoothing from the previous period. This represents the weighting coefficients after smoothing for this period.

[0014] The method is fully embedded in the pulse width modulation (PWM) interrupt service routine of the motor controller, and is executed in the following timing sequence within each current loop cycle: (1) Read the rotor electrical angle The three-phase current is sampled and the Clarke and Parker transforms are performed to obtain... , Three-phase current , , The three-phase stationary coordinates are transformed into two-phase stationary coordinates using the Clark transformation, and two-phase stationary orthogonal currents are output. and Then , and After Parker transformation, the output d-axis current is obtained. and q-axis current The d-axis corresponds to the direction of the rotor flux linkage; (2) Extract the AC component of the q-axis voltage command from the previous cycle after high-pass filtering as the error signal e(k); (3) Determine the adaptive update enable condition; if it is satisfied, update the weight coefficients. (4) Smooth the updated weight coefficients using a first-order low-pass filter; (5) Calculate the q-axis current feedforward compensation value for the current cycle based on the smoothed weighting coefficients. ; (6) The final q-axis current command is obtained by superimposing the original torque current command output by the speed loop; (7) Perform PI regulation of d-axis and q-axis currents, and output d-axis voltage. and q-axis voltage It also performs the Park inverse transform and space vector pulse width modulation.

[0015] When the motor is equipped with a dual-redundant position feedback system with a main position sensor and a backup position sensor, a set of weight coefficients for a feedforward compensation model is maintained independently for each of the two channels, and the compensation current is calculated and adaptively updated using their respective electrical angles. The system defaults to using the compensation current of the main sensor channel for control, while the backup channel runs synchronously in the background. The compensation current values ​​of the two channels are compared in real time. If the absolute value of the error exceeds the preset threshold and continues for a preset time, a certain channel is determined to be abnormal. At this time, the channel with the smaller effective value of the error signal is selected as the reliable source and the fault is reported. When a hard fault occurs in the main sensor, the compensation channel switches to the backup channel weight coefficient within a single control cycle.

[0016] The adaptive update enable condition in step S4 is that the following conditions must be met simultaneously: The motor speed is less than 20% of the rated speed, the speed fluctuation amplitude is less than 1%, the torque current command change rate is less than the set threshold, and the system has no faults. When the detected motor temperature change exceeds 5°C compared to the last adaptive update, the adaptive update is forcibly started for at least 10 seconds.

[0017] In step S5, checksums are recorded synchronously when storing weight coefficients periodically; In step S6, when loading the stored weight coefficients, the checksum is first verified. If the verification fails, the process reverts to the initial weight coefficients set at the factory.

[0018] The effects described in the invention are merely those of the embodiments, and not all the effects of the invention. The above technical solutions have the following advantages or beneficial effects: This invention discloses an adaptive method for active feedforward compensation of cogging torque in ultra-multipole permanent magnet synchronous motors (ULSMs). Addressing the problems of existing technologies that only compensate for the fundamental frequency, have fixed compensation coefficients, and lack redundant sensing fault tolerance, this invention constructs a cogging torque feedforward compensation model incorporating the fundamental, third, and fifth harmonics. The amplitude-phase parameters are transformed into linear weight coefficients through orthogonal decomposition. Furthermore, an error signal is constructed based on the AC residual in the current PI regulator output, and the weight coefficients are updated online using a normalized least mean square algorithm, achieving adaptive tracking of compensation parameters against temperature drift, mechanical wear, and changes in operating conditions. Simultaneously, a dual-channel redundant structure combining a main position sensor and a sensorless observer is established to create a compensation current cross-validation and fault switching mechanism, thereby improving the cogging torque suppression effect and system reliability. This invention effectively reduces torque ripple and speed fluctuations under low-speed conditions, improving the stability, control accuracy, and safety of the ULSM drive system. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0020] Figure 1This is a schematic diagram of the method flow of the present invention.

[0021] Figure 2 This is a schematic diagram of the system structure of the present invention under the FOC control architecture.

[0022] Figure 3 This is a schematic diagram illustrating the orthogonal decomposition principle of cogging torque multi-harmonics in this invention.

[0023] Figure 4 This is a timing diagram of the execution of the method of the present invention in the PWM interrupt service routine.

[0024] Figure 5 This is a schematic diagram of the dual-channel redundancy compensation system of the present invention.

[0025] Figure 6 Comparison of q-axis current ripple waveforms under different compensation schemes.

[0026] Figure 7 A schematic diagram of the convergence curve of the adaptive weight coefficients of this invention. Detailed Implementation

[0027] To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific implementation methods and in conjunction with the accompanying drawings.

[0028] Example 1 like Figure 1 As shown, an adaptive method for active feedforward compensation of cogging torque in a multi-pole permanent magnet synchronous motor includes the following steps: S1. Establish a feedforward compensation model containing odd harmonics of cogging torque in the traditional FOC current loop, orthogonally decompose the model into a linear weighted coefficient form, and then apply the real-time collected motor rotor electrical angle... Calculate the q-axis current feedforward compensation value using the current weighting coefficients. ; S2, feedforward compensation value of q-axis current The original torque current command output by the speed loop Superimposed to form the final q-axis current command And it is fed into the current PI regulator; S3, Extract the q-axis voltage command output from the current PI regulator. The AC component is used as the error signal for cogging torque compensation residual. ; S4. Determine the adaptive update enable condition. If it is met, then use the error signal based on the adaptive algorithm. and rotor electrical angle The linear weight coefficients of the feedforward compensation model are updated iteratively online. S5. Smooth the updated weight coefficients, use the smoothed weight coefficients as the current weight coefficients for the next control cycle, and periodically store them in the controller's non-volatile memory. S6. During power-on initialization, the stored weight coefficients are loaded as the initial weight coefficients. If loading fails, the preset initial weight coefficients are used.

[0029] In a specific implementation, the adaptive method for active feedforward compensation of cogging torque in ultra-multipole permanent magnet synchronous motors of the present invention is applied to a traditional field-oriented control (FOC) system, such as... Figure 2 As shown, a cogging torque feedforward compensation module is added between the speed loop and current loop of the traditional FOC control architecture. This module receives the electrical angle signal of the motor rotor. The q-axis current feedforward compensation value is calculated based on the real-time updated compensation weight coefficients. This compensation value is then compared with the original torque current command output by the speed loop regulator. Superimposed to form the final q-axis current command The current is fed into the d-axis and q-axis current regulators.

[0030] This invention is applicable to multi-pole permanent magnet synchronous motors, and is especially suitable for drive control systems in electric aviation, electric vehicles, and high-end servo applications that have extremely high requirements for low-speed stability.

[0031] In a specific implementation, during the motor's factory commissioning phase, the cogging torque is initially calibrated through offline bench testing to determine the initial weighting coefficients of the compensation model. The specific steps are as follows: (1) Connect the motor under test to the drive device coaxially. The drive device adopts the speed control mode and drives the motor under test to a preset constant low speed (for example, 5% of the rated speed). (2) Ensure the current loop of the motor under test is in a zero torque command state. =0), at this time the torque fluctuation of the motor output shaft is entirely caused by the cogging torque; (3) The torque waveform is continuously acquired at a preset sampling rate (e.g., 10 times the PWM frequency) for a time not less than a preset number of mechanical cycles; (4) Perform Fourier analysis on the acquired torque waveform and extract the amplitudes of the 1, 3, and 5 times electrical angular frequency components. , , and phase , , ; (5) Based on the torque constant of the motor Calculate the initial compensation current amplitude for each harmonic: , , ; (6) Calculate the initial compensation phase for each harmonic: , , , (7) Convert the amplitude and phase into initial weighting coefficients: , , , ; , ; (8) Store the above 6 initial weight coefficients into the controller's non-volatile memory and record the checksum.

[0032] For situations where torque cannot be directly measured, the cogging torque waveform can be calculated by reverse calculation using the output of the speed loop regulator. The specific method is as follows: run the motor in speed closed-loop mode and give a constant low-speed command. At this time, the output current of the speed loop regulator will be completely used to cancel the cogging torque. Fourier analysis of this current waveform will yield the harmonic components of the cogging torque.

[0033] In a specific implementation, the cogging torque is a periodic torque generated by the interaction between the magnetic field of the permanent magnet and the cogging teeth of the stator core. Its magnitude depends only on the rotor position and is independent of the stator current. It is the rotor electrical angle The periodic function of has the following Fourier series expansion: , in, An odd number indicates the harmonic order; The first Fourier series representing the cogging torque The amplitude of the second harmonic torque component; Indicates the first The initial phase corresponding to the subharmonic torque component.

[0034] Under the FOC control framework, the electromagnetic torque of the motor With q-axis current Satisfies a linear relationship: , in, is the torque constant of the motor.

[0035] To completely suppress cogging torque, an electromagnetic torque component with the same amplitude but opposite phase to the cogging torque needs to be generated, i.e.: , The harmonic model of the feedforward compensation current is thus derived: , in, For the first The amplitude of the compensation current for the subharmonic. For the first Phase of the compensation current for subharmonics.

[0036] For ultra-multipole permanent magnet synchronous motors, actual measurements show that the energy of the cogging torque is mainly concentrated in the fundamental (1st), 3rd, and 5th harmonics, while the amplitude of higher harmonics is less than 5% of the fundamental wave and can be ignored. Therefore, as Figure 3 As shown, the preferred compensation model of the present invention is: Equation (1) To facilitate adaptive updates, trigonometric identities are used. Equation (1) is orthogonally decomposed and transformed into a linear weight coefficient form: (Equation 2) in, and They represent Adaptive weighting coefficients corresponding to the subharmonic sine basis function and cosine basis function.

[0037] The original amplitude and phase of each harmonic can be uniquely recovered from the weighting coefficients: , .

[0038] The above orthogonal decomposition transforms the nonlinear tuning problem of amplitude-phase into a problem of updating linear weight coefficients, which can provide an ideal platform for the implementation of adaptive algorithms.

[0039] In a specific implementation, when the cogging torque is not fully compensated, the residual periodic torque pulsation will force the q-axis current regulator to output an AC component of the same frequency in an attempt to suppress the disturbance. Therefore, this invention selects the q-axis current regulator output voltage command. The AC component is used as the error signal It directly reflects the compensation residual of the cogging torque.

[0040] The specific method for extracting the error signal is as follows: The voltage command output by the q-axis current regulator... By using a high-pass filter with a cutoff frequency lower than the current electrical angular frequency, the DC component and low-frequency load torque component are filtered out, resulting in an AC component containing only cogging torque pulsations. .

[0041] In a specific implementation, the weight coefficients are updated online adaptively based on the normalized LMS algorithm; To address the problem that fixed compensation coefficients cannot track motor parameter drift, this invention derives an online adaptive update law for the compensation weight coefficients based on the instantaneous gradient descent principle. (1) Instantaneous gradient descent method and LMS algorithm For the weight coefficients in the online optimization equation (2), the instantaneous objective function for the k-th control cycle is defined as: , The goal is to adjust the weight coefficients to make the objective function... minimize.

[0042] Find any weight coefficients for the objective function Partial derivatives: , Due to feedforward compensation current When directly superimposed on the q-axis current command, its changes will be linearly reflected in the current error and regulator output. Near steady state, it can be approximated as follows: , in, Represents the weight coefficients The reference input for multiplication (i.e.) or item) This is a positive constant related to the system gain and can be incorporated into the step size factor.

[0043] Updating the weight coefficients along the negative gradient direction yields the basic update law of the Least Mean Square (LMS) algorithm: , Where μ is the adaptive step size.

[0044] (2) Normalized LMS algorithm The convergence speed and stability of the basic LMS algorithm are significantly affected by the power of the reference input signal. To improve convergence stability, this invention employs a normalized LMS algorithm, which normalizes the step size using the instantaneous power of the reference input vector.

[0045] The reference input vector is: , Since the selected 1st, 3rd, and 5th harmonic components are orthogonal to each other, the square of the magnitude of the reference input vector satisfies: , Therefore, the formula for updating the normalized weight coefficients is: , (Equation 3a) (Equation 3b) in, This indicates the odd harmonic order of the cogging torque compensation. =1, 3, 5, =1 indicates the first fundamental harmonic. =3 indicates the 3rd harmonic. =5 indicates the 5th harmonic; express The weighting coefficients of the sinusoidal terms corresponding to the second harmonic order. express The weighting coefficients of the cosine term corresponding to the second harmonic order.

[0046] The adaptive step size μ is determined based on the system's convergence speed and steady-state error requirements, striking a balance between the two. An excessively large step size may cause system oscillations, while an excessively small step size will result in a slow convergence speed.

[0047] In a specific implementation, the smoothing and storage management of the weighting coefficients are as follows: (1) Smoothing of weight coefficients To suppress high-frequency noise during the adaptive update process, the updated weight coefficients are smoothed using a first-order low-pass filter: , Where α is the filter coefficient, which is determined based on the system noise level; Update the weight coefficients obtained from equation (3); These are the weighting coefficients after smoothing from the previous period.

[0048] The smoothed weighting coefficients are used to calculate the feedforward compensation current for the next control cycle.

[0049] (2) Storage and loading of weight coefficients To ensure consistent compensation accuracy throughout the entire lifecycle, this invention periodically stores the smoothed weighting coefficients into the controller's non-volatile memory. The specific steps are as follows: 1) Write the currently smoothed 6 weight coefficients into the designated area of ​​the non-volatile memory according to the preset period; 2) Simultaneously calculate and store the checksums of these 6 weight coefficients; 3) During power-on initialization, the weighting coefficients and checksums stored in the non-volatile memory are read first; 4) Recalculate the checksum of the read weight coefficients and compare it with the stored checksum; 5) If the checksum matches, load the set of weight coefficients as the initial weight coefficients; 6) If the checksum is inconsistent, the system will revert to the initial weighting coefficients set at the factory.

[0050] The aforementioned memory function allows the compensation parameters to be inherited after the motor ages, avoiding repeated calibration.

[0051] In specific implementation methods, such as Figure 4 As shown, the method of the present invention is completely embedded in the PWM interrupt service routine of the motor controller, and is executed in the following timing sequence in each current loop cycle: Read the rotor position signal and calculate the rotor electrical angle. Simultaneously sample the three-phase current. , , ; Perform a Clark transformation on the sampled three-phase currents to obtain the currents in the α-β coordinate system. and Then perform the Park transformation to obtain the current in the dq coordinate system. , ; Extract the output voltage of the q-axis current regulator from the previous cycle. The AC component after high-pass filtering is used as the error signal for the current period. ; Determine the adaptive update enable condition. If it is met, update the 6 weight coefficients according to equations (3a) and (3b); otherwise, freeze the weight coefficient update. The updated weight coefficients are smoothed using a first-order low-pass filter. Based on equation (2) and the current smoothed weight coefficients, calculate the q-axis current feedforward compensation value for the current cycle. ; Will The original torque current command output by the speed loop regulator Superimposed, the final q-axis current command is obtained. d-axis current command Keep it as the default value; Execute d-axis and q-axis current adjustment, output voltage command , ;right , Perform the inverse Park transformation to obtain the voltage command in the α-β coordinate system. , ; The space vector pulse width modulation (SVPWM) is executed to generate six PWM drive signals to drive the inverter.

[0052] This execution timing ensures the phase accuracy of the compensation signal, and the computational overhead of the adaptive algorithm only increases by a small amount of multiplication and addition operations, which puts very little burden on the main control chip.

[0053] In specific implementations, in safety-critical applications, motors are typically equipped with a dual-redundant position feedback system consisting of a primary position sensor and a backup position sensor. This invention maintains an independent set of compensation weighting coefficients for each of the two channels, achieving synchronous redundancy between the compensation function and the position signal.

[0054] (1) Dual-channel parallel operation mechanism Main sensor channel: The compensation current is calculated and the weighting coefficients are updated adaptively using the electrical angle output by the main position sensor, and the compensation current directly participates in the control. Backup channel: The electrical angle output by the backup position sensor is used to perform compensation current calculation and adaptive update of weight coefficients in the background. The compensation current is not connected to the current loop.

[0055] The adaptive update processes of the two channels are completely independent and do not interfere with each other.

[0056] (2) Cross-validation and fault diagnosis like Figure 5 As shown, to ensure the reliability of the dual-channel compensation function, cross-validation logic is set up. The resolver channel and the inductively coupled channel calculate the compensation current vector based on their respective electrical angles. and Error signals are extracted from the AC component of the output voltage command of the q-axis current PI regulator corresponding to each channel. and ,in To compensate for the residuals in the resolver channel. This is to compensate for the residual in the non-sensory channel.

[0057] The controller compares the compensation current vectors of the two channels in real time. If the current exceeds a set threshold (for example, 5% of the rated current of the q-axis) and persists for more than 10 electrical cycles, an abnormal deviation is determined to exist between the two channels. Further comparison is then performed. and The effective value of the error is selected, and the channel with the smaller effective value is chosen as the reliable compensation source, and fault information is reported. When the main sensor experiences a hard fault such as loss of signal (LOS), signal degradation (DOS), or loss of tracking (LOT), the system seamlessly switches to the compensation coefficient and compensation output corresponding to the backup channel within a single control cycle, so as to achieve uninterrupted suppression of cogging torque disturbance.

[0058] (3) Seamless fault switching When the main sensor experiences a hard fault such as signal loss or abnormal amplitude, the system seamlessly switches the compensation channel to the weight coefficient of the backup channel within a single control cycle, ensuring uninterrupted cogging torque suppression. No significant torque surge occurs during the switching process, guaranteeing system stability.

[0059] In this specific implementation, to ensure the stability and convergence of the adaptive algorithm, adaptive update enabling conditions are set. Weight coefficient updates are only initiated when all conditions are simultaneously met; otherwise, updates are frozen. The specific enabling conditions are as follows: Speed ​​condition: The motor speed is less than 20% of the rated speed. Because cogging torque has a significant impact at low speeds, its impact is masked by the load torque at high speeds, and no compensation is required; Steady-state condition: Speed ​​fluctuation amplitude is less than 1%. During dynamic processes, the current loop regulator output contains a large number of dynamic components, which will interfere with the extraction of error signals; Torque condition: The rate of change of torque current command is below a set threshold. This avoids adaptive updates during sudden torque changes. Fault condition: The system has no faults.

[0060] In addition, when the temperature of the motor is detected to change by more than 5°C relative to the temperature at the time of the last adaptive update, the adaptive update is forcibly started for no less than 10 seconds to quickly track the changes in the magnetic properties of the magnet with temperature.

[0061] Example 2 To verify the beneficial effects of the present invention, the following comparative experiments were conducted: Experimental subject: 192-pole, 180-slot ultra-multi-pole permanent magnet synchronous motor, rated torque 100N m, rated speed 150 rpm, torque constant =2.5N m / A; Experimental conditions: no load, speed 38 rpm, corresponding electrical frequency 48 Hz; Comparative Example 1: Cogging torque compensation; Comparative Example 2: Fundamental cogging torque feedforward compensation only; Example: The fundamental wave + 3rd + 5th harmonic combined adaptive compensation of the present invention.

[0062] Experimental results are as follows Figure 6 As shown, the peak-to-peak value of the q-axis current ripple is 18.2A without compensation; after compensation of the fundamental frequency only, the peak-to-peak value is reduced to 8.5A, a reduction of about 53%; after using the compensation method of the present invention, the peak-to-peak value is further reduced to 3.6A, a reduction of about 80%.

[0063] The adaptive convergence process of the weight coefficients is as follows Figure 7As shown, it is divided into three parts, including the initial stage (iteration from 0), the fast convergence region (reaching near the steady-state value in about 1.5 seconds), and the steady-state region (the weights tend to stabilize with slight fluctuations). The weight coefficients converge to the steady-state value in about 1.25 seconds. The steady-state value corresponds to a fundamental sin component amplitude of 12.5, indicating that the method of the present invention has a fast convergence speed, small steady-state error, and can quickly track the changes in cogging torque.

[0064] Although the specific embodiments of the invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the invention. Based on the technical solutions of the invention, various modifications or variations that can be made by those skilled in the art without creative effort are still within the scope of protection of the invention.

Claims

1. An adaptive method for active feedforward compensation of cogging torque in a multi-pole permanent magnet synchronous motor, characterized in that, Includes the following steps: S1. Establish a feedforward compensation model containing odd harmonics of cogging torque in the traditional FOC current loop. Orthogonally decompose this model into a linear weighted coefficient form, and then calculate the model based on the real-time collected motor rotor electrical angle. Calculate the q-axis current feedforward compensation value using the current weighting coefficients. ; The establishment and orthogonal decomposition process of the cogging torque feedforward compensation model in step S1 is as follows: Based on cogging torque With rotor electrical angle The periodic relationship, cogging torque The Fourier series expansion is as follows: ， in, An odd number indicates the harmonic order; The first Fourier series representing the cogging torque The amplitude of the second harmonic torque component; Indicates the first The initial phase corresponding to the subharmonic torque component; According to the electromagnetic torque of the motor With q-axis current linear relationship , The harmonic model of the feedforward compensation current is derived by representing the motor torque constant as follows: ， in, Indicates the cogging torque corresponding to the first The amplitude of the q-axis compensation current for the second harmonic component. ; Indicates the first The initial phase of the secondary compensation current harmonic. ; For ultra-multi-pole permanent magnet synchronous motors, select The compensation model is obtained as follows: ， Then, using trigonometric identities, it is orthogonally decomposed into a linear weight coefficient form: ， make , ,get: ， in, and They represent Adaptive weighting coefficients corresponding to the subharmonic sinusoidal basis function and cosine basis function; The amplitude and phase of each harmonic are recovered from the weighting coefficients as follows: ， ； S2, feedforward compensation value of q-axis current The original torque current command output by the speed loop Superimposed to form the final q-axis current command And it is fed into the current PI regulator; S3, Extract the q-axis voltage command output from the current PI regulator. The AC component is used as the error signal for cogging torque compensation residual. ; S4. Determine the adaptive update enable condition. If it is met, then use the error signal based on the adaptive algorithm. and rotor electrical angle The linear weight coefficients of the feedforward compensation model are updated iteratively online. S5. Smooth the updated weight coefficients, use the smoothed weight coefficients as the current weight coefficients for the next control cycle, and periodically store them in the controller's non-volatile memory. S6. During power-on initialization, the stored weight coefficients are loaded as the initial weight coefficients. If loading fails, the preset initial weight coefficients are used.

2. The adaptive method for active feedforward compensation of cogging torque in a multi-pole permanent magnet synchronous motor according to claim 1, characterized in that, The preset initial weight coefficients are obtained through offline initial calibration: Drive the motor to a preset constant low speed, so that the current loop is in a zero torque command state, measure the output shaft torque fluctuation, or calculate the cogging torque waveform in reverse through the speed loop PI output. Fourier analysis is performed on the acquired waveforms to extract the amplitude and phase of the 1, 3, and 5 times electrical angular frequency components, which are then converted into initial weighting coefficients and stored in the controller's non-volatile memory.

3. The adaptive method for active feedforward compensation of cogging torque in a multi-pole permanent magnet synchronous motor according to claim 1, characterized in that, The error signal mentioned in step S3 The extraction method is as follows: q-axis voltage command The error signal is obtained by filtering out the DC component using a high-pass filter with a cutoff frequency lower than the current electrical angular frequency. This error signal directly reflects the uncompensated residual of the cogging torque.

4. The adaptive method for active feedforward compensation of cogging torque in a multi-pole permanent magnet synchronous motor according to claim 1, characterized in that, The adaptive algorithm described in step S4 is the normalized least mean square algorithm, and its weight coefficient update formula is: ， ， in, This indicates the odd harmonic order of the cogging torque compensation. =1, 3, 5, =1 indicates the first fundamental harmonic. =3 indicates the 3rd harmonic. =5 indicates the 5th harmonic; express The weighting coefficients of the sinusoidal terms corresponding to the second harmonic order. express The weighting coefficients of the cosine terms corresponding to the second harmonic order; μ represents the adaptive step size.

5. The adaptive method for active feedforward compensation of cogging torque in a multi-pole permanent magnet synchronous motor according to claim 1, characterized in that, The smoothing process described in step S5 uses a first-order low-pass filter, and the calculation formula is as follows: ， in, Represents the filter coefficients. This represents the updated weight coefficients. These are the weight coefficients after smoothing from the previous period. This represents the weighting coefficients after smoothing for this period.

6. The adaptive method for active feedforward compensation of cogging torque in a multi-pole permanent magnet synchronous motor according to claim 1, characterized in that, The method is fully embedded in the pulse width modulation (PWM) interrupt service routine of the motor controller, and is executed in the following timing sequence within each current loop cycle: (1) Read the rotor electrical angle The three-phase current is sampled and the Clarke and Parker transforms are performed to obtain... , Three-phase current , , The three-phase stationary coordinates are transformed into two-phase stationary coordinates using the Clark transformation, and the output is a two-phase stationary orthogonal current. and Then , and After Parker transformation, the output d-axis current is... and q-axis current The d-axis corresponds to the direction of the rotor flux linkage; (2) Extract the AC component of the q-axis voltage command from the previous cycle after high-pass filtering as the error signal. ; (3) Determine the adaptive update enable condition; if it is satisfied, update the weight coefficients. (4) Smooth the updated weight coefficients using a first-order low-pass filter; (5) Calculate the q-axis current feedforward compensation value for the current cycle based on the smoothed weighting coefficients. ; (6) The final q-axis current command is obtained by superimposing the original torque current command output by the speed loop; (7) Perform PI regulation of d-axis and q-axis currents, and output d-axis voltage. and q-axis voltage It also performs the Park inverse transform and space vector pulse width modulation.

7. The adaptive method for active feedforward compensation of cogging torque in a multi-pole permanent magnet synchronous motor according to claim 1, characterized in that: When the motor is equipped with a dual-redundant position feedback system with a main position sensor and a backup position sensor, a set of weight coefficients for a feedforward compensation model is maintained independently for each of the two channels, and the compensation current is calculated and adaptively updated using their respective electrical angles. The system defaults to using the compensation current of the main sensor channel for control, while the backup channel runs synchronously in the background. The compensation current values ​​of the two channels are compared in real time. If the absolute value of the error exceeds the preset threshold and continues for a preset time, a certain channel is determined to be abnormal. At this time, the channel with the smaller effective value of the error signal is selected as the reliable source and the fault is reported. When a hard fault occurs in the main sensor, the compensation channel switches to the backup channel weight coefficient within a single control cycle.

8. The adaptive method for active feedforward compensation of cogging torque in a multi-pole permanent magnet synchronous motor according to claim 1, characterized in that, The adaptive update enable condition in step S4 is that the following conditions must be met simultaneously: The motor speed is less than 20% of the rated speed, the speed fluctuation amplitude is less than 1%, the torque current command change rate is less than the set threshold, and the system has no faults. When the detected motor temperature change exceeds 5°C compared to the last adaptive update, the adaptive update is forcibly started for at least 10 seconds.

9. The adaptive method for active feedforward compensation of cogging torque in a multi-pole permanent magnet synchronous motor according to claim 1, characterized in that: In step S5, checksums are recorded synchronously when storing weight coefficients periodically; In step S6, when loading the stored weight coefficients, the checksum is first verified. If the verification fails, the process reverts to the initial weight coefficients set at the factory.

Images