Active vibration suppression method for random frequency mixing of three-phase surface-mounted permanent magnet motor

Abstract

The invention discloses an active vibration suppression method for random frequency mixing of a three-phase surface-mounted permanent magnet motor, and the method comprises the steps: firstly carrying out the LMS and peak estimation of a real domain signal collected by a vibration acceleration sensor in a segmented manner with the step length of 1Hz, and finally obtaining a series of peak radial vibration frequencies x1, x2, x3... xk; establishing a current injection model under a rotating coordinate system, and for each peak radial vibration frequency xk, calculating a positive-sequence symmetric current of xk + f frequency and injecting the positive-sequence symmetric current into the model; combining and injecting all obtained frequency current injection models into the rear end of a current loop PI regulator, and obtaining the vibration of each peak frequency; multi-target vibration deep learning is carried out on the vibration of each peak frequency to obtain an optimal frequency mixing current, and the motor frequency is adjusted to realize active vibration suppression; according to the comprehensive vibration suppression algorithm based on vibration feedback, the coupling problem of vibration of each frequency generated by multi-frequency current is solved, the vibration suppression effect is improved, the stability and the anti-interference capability of the system are enhanced, and the method is easy to implement.

Description

Technical Field

[0001] The invention relates to the technical field of synchronous motor vibration control, and in particular to an active vibration suppression method for random frequency mixing of a three-phase surface-mounted permanent magnet motor. Background Art

[0002] The greatest military value of military unmanned underwater vehicles is their stealth. Their sound waves mainly come from the propulsion motor. The vibration is transmitted to the shell through the base, and the sound waves are transmitted underwater, which can be easily captured by sonar. The three-phase permanent magnet propulsion motor of the underwater vehicle installed horizontally will transmit vibrations through the axial propulsion motor when the propeller is loaded. The low-frequency vibration has large energy and a long transmission distance. It has strong penetration performance and cannot be completely absorbed by the shock absorber. However, it can transmit noise to the underwater through the base and shell. The vibration frequency components of the load are complex and not fixed. In order to achieve the stealth of the low-frequency vibration of the motor, it is necessary to design an active control method to suppress this complex vibration.

[0003] The harmonic current injection method is an active vibration suppression control method. The current in the winding causes the armature magnetic field to interact with the permanent magnet magnetic field, causing radial vibration of the stator. The harmonic current can cause harmonic radial vibration. If the harmonic radial vibration is opposite to the harmonic in the original radial vibration of the motor in phase but has the same amplitude, this harmonic vibration can be suppressed.

[0004] The frequency, amplitude and phase of harmonic current injection determine the frequency, amplitude and phase of the electromagnetic vibration it generates, and a harmonic current of one frequency can suppress a harmonic radial vibration of one frequency. By injecting multiple harmonic currents that can suppress all low-frequency harmonic radial vibrations of the motor into the motor windings in the form of mixed current, the low-frequency and multi-frequency radial vibrations of the motor can be suppressed.

[0005] The existing harmonic injection method only suppresses the harmonic radial vibration, without considering the suppression of vibration of any frequency, and even less considering the influence of phase on vibration suppression, resulting in a weak vibration suppression effect and failure to achieve comprehensive suppression control of multi-frequency vibrations.

[0006] Therefore, the prior art still needs to be improved and enhanced. Summary of the invention

[0007] In view of the above-mentioned deficiencies in the prior art, the purpose of the present invention is to provide an active vibration suppression method for random frequency mixing of a three-phase surface-mounted permanent magnet motor, aiming to solve the problem that the harmonic injection method in the prior art only suppresses the double-frequency radial vibration, does not consider the suppression of vibration at any frequency, and does not consider the influence of the phase on the vibration suppression, resulting in the problem that the vibration suppression effect is not obvious.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions: An active vibration suppression method for random frequency mixing of a three-phase surface-mounted permanent magnet motor, characterized by comprising: Step 1: Start the motor, collect the real-domain signal through the vibration acceleration sensor, and perform LMS and peak estimation on the real-domain signal collected by the vibration acceleration sensor in segments with a step size of 1 Hz. Finally, a series of peak radial vibration frequencies are obtained. x 1 , x 2 , x 3 …, x k ; Step 2: Establish a current injection model in a rotating coordinate system, and for each peak radial vibration frequency x k , calculate x k + f The positive sequence symmetrical current of the frequency is injected into the model; Step 3: Combine all the obtained current injection models of each frequency and inject them into the back end of the current loop PI regulator to obtain the vibration of each peak frequency; Step 4: Perform multi-objective vibration deep learning on the vibration of each peak frequency to obtain the optimal mixing current, adjust the motor frequency, and achieve active vibration suppression.

[0009] Furthermore, the step 1 includes: First, the local frequency spectrum is established from low to high frequency with a step size of 1Hz, and then the peak frequency of the frequency spectrum is estimated. The estimated value is the peak frequency. The adaptive LMS filter is used to extract the amplitude of each frequency signal and establish the local frequency spectrum. After multiple peak sampling, the mean is the frequency corresponding to the peak, and the maximum likelihood estimation method is used to estimate , the estimation formula is as follows In the formula, x 1 , x 2 , x 3 … x k is the sampling point, i.e. the frequency corresponding to a series of peak radial vibrations, y 1 , y 2 , y 3 … y k are the peak vibration values, is the maximum likelihood estimate. Among them, the frequency corresponding to the maximum value of the frequency spectrum is the characteristic vibration frequency; after the characteristic vibration frequency is obtained, the amplitude and phase of each frequency vibration are obtained through the adaptive LMS filter.

[0010] Furthermore, the current injection model in the rotating coordinate system in step 2 is as follows: In the formula, i d and i q They are the dq axis current, x is the required vibration frequency, ω =2π f is the electrical angular frequency, is the current amplitude is the initial phase of the current.

[0011] Furthermore, the multi-objective vibration deep learning of the vibrations of each peak frequency includes: Establish a multi-objective vibration optimization model: In the formula is the objective function vector, i.e., the radial vibration amplitude vector of each harmonic; for x 1 arrive x k Frequency radial vibration, They are injected x k+1 The amplitude and initial phase of the frequency current, For the i constraints.

[0012] Although the various objectives in this multi-objective optimization model are not in the same domain, they are internally coupled to a certain extent, and the adaptive optimization method is used to achieve decoupling.

[0013] Furthermore, the multi-objective vibration deep learning of the vibrations of each peak frequency also includes: A deep learning algorithm is used to find the minimum value of all single objective functions at the same time; k The specific steps of a single objective function are as follows: Assume: decay rate ρ , the initial learning rate η , initial parameters θ ; Assume: Gradient cumulant ν Initialized to 0, the initial constant δ is 10 -7 ; The loop body is as follows: Simultaneous collection m Samples ,in , For sample The true value of calculate m The average gradient of samples: ; The square of the cumulative historical gradient: ; Calculate the update amount: (element-wise operations) Parameter update: The loop body ends; Loop end condition: , , is the fundamental current amplitude; In the above loop, m is the number of samples, yes m samples, For the k The frequency i Sample t -1 moment vibration amplitude, are the corresponding current amplitude and phase, For the k The frequency i The peak value of the vibration amplitude of the samples, For the t The average gradient at time, is the gradient symbol, is the gradient function, is the square of the cumulative historical gradient, is the deep learning coefficient with memory.

[0014] The method of finding the minimum value of multiple single objective functions simultaneously can achieve adaptive decoupling.

[0015] The technical solution adopted by the present invention has the following beneficial effects: 1) Improved vibration suppression sensitivity, because x + f The phase sequence of positive-sequence current injection is consistent with the motor fundamental phase sequence, while the phase sequence of negative-sequence current injection is opposite to the motor fundamental phase sequence. Therefore, the positive-sequence current injection of the same amplitude has a smaller distortion rate on the total current and a higher vibration suppression sensitivity than the negative-sequence current injection.

[0016] 2) The extraction efficiency of peak vibration frequency is improved. By combining LMS and peak frequency estimation, the algorithm is simple and the extraction speed is fast.

[0017] 3) Closed-loop vibration suppression control based on vibration signal feedback is realized, the coupled vibration generated by currents of various frequencies is reduced, and comprehensive vibration suppression of radial vibrations of various frequencies is achieved, with obvious vibration suppression effects.

[0018] In general, compared with the traditional method of only suppressing the frequency-doubled radial vibration, this method takes into account more comprehensive considerations. The vibration of the actual propulsion motor is very complex, so the present invention is highly practical. The comprehensive vibration suppression algorithm based on vibration feedback solves the coupling problem of vibrations of various frequencies generated by multi-frequency currents, improves the vibration suppression effect, enhances the stability and anti-interference ability of the system, and is easy to implement. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The schematic diagram is a system principle diagram of an active vibration suppression method for a three-phase surface-mounted permanent magnet motor with random frequency mixing; Figure 2 It is a flow chart of an active vibration suppression method for random frequency mixing of a three-phase surface-mounted permanent magnet motor; Figure 3 is the experimental waveform of this method. Figure 3 (a) is the original current waveform of the motor. Figure 3 (b) is the current waveform after comprehensive vibration suppression. Figure 3 (c) is the original radial vibration waveform of the motor. Figure 3 (d) is the radial vibration waveform after comprehensive vibration suppression. Figure 3 (e) is the original radial vibration spectrum of the motor. Figure 3 (f) is the radial vibration spectrum after comprehensive vibration suppression.

[0020] 1. PI regulator, 2. iPark transformation, 3. iClark transformation, 4. AC current regulator, 5. SiC Mosfet, 6. Permanent magnet motor, 7. Encoder, 8. Vibration acceleration sensor, 9. LMS and peak frequency estimation module, 10. Multi-objective deep learning algorithm module, 11. x+f frequency current injection model module. DETAILED DESCRIPTION

[0021] In order to more clearly show the purpose, technical scheme and effect of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not limit the present invention. In addition, in each embodiment of the present invention described below, the technical features involved can be combined with each other as long as they do not conflict with each other.

[0022] The present invention provides an active vibration suppression method for random frequency mixing of a three-phase surface-mounted permanent magnet motor. Combining the magnetomotive force formula and Maxwell stress formula of the motor, the following is derived:x The frequency of radial vibration can be x + f The frequency positive sequence symmetrical current is suppressed, where f is the electrical frequency of the motor.

[0023] In order to three-phase x + f The positive frequency sequence symmetrical current is injected into the three-phase winding. The present invention provides a current injection model in a rotating coordinate system. x + f Based on the frequency positive sequence symmetrical current model, the injection model in the synchronous rotating coordinate system is obtained by combining Park transformation, and is injected into the three-phase winding after the current regulator of the current loop.

[0024] 1) x + f Frequency positive sequence symmetrical current injection scheme To inject x + f Positive sequence symmetrical current of frequency, x + f The positive sequence symmetrical current model of the frequency is as follows: In the formula, A phase, B phase, C phase x + f Frequency current; for x + f frequency current amplitude; for x + f The initial phase difference between the frequency current and the fundamental current; It is the electrical angle between the d-axis of the synchronously rotating coordinate system and the A-axis of the stationary coordinate system.

[0025] In order to inject model (1), the ABC axis model (1) is converted into the dq axis model, that is, Park transformation is performed on (1), and the dq axis current model is obtained as follows: On the basis of three-phase sinusoidal current, (2) is injected into the dq axis respectively, and the total current on the dq axis is iPark transformed and converted into ABC three-phase current. The motor drive adopts a current regulator, and the winding is injected x + f Positive sequence symmetrical current with frequency.

[0026] 2) Vibration signal separation solution Incremental Encoder Error Calculation For the DSP28335 controller, When the eQEP unit works in the orthogonal counting mode, the position counter is provided with the counting clock QCLK by detecting the edge of the QEPA and QEPB signals. Therefore, the clock frequency generated by the eQEP logic is 4 times the input clock frequency. For example, when a 2500-line encoder rotates one circle, 10,000 counting pulses can be generated.

[0027] For an 8-pole surface-mounted three-phase permanent magnet propulsion motor, it generally runs at a low speed (less than 500r / min). The low-speed accuracy of T measurement is better than that of the M method and the M / T method. Therefore, this paper chooses the T method to measure speed, and the calculated absolute value of the frequency error is less than 0.4Hz.

[0028] In order to realize closed-loop vibration suppression control with vibration signal as feedback, the present invention proposes a peak frequency estimation method to extract the frequency, amplitude and phase of peak vibration. Because the frequency amplitude extraction algorithm of vibration signal is limited by window function, it is normally distributed in a certain area, and its peak value is the actual frequency, so the frequency estimation method can extract the frequency of the peak part more accurately.

[0029] Peak frequency estimation In order to reduce the influence of other characteristic frequencies and background noise on the detection of harmonic signals, we first establish a local frequency spectrum from low to high frequency with a step size of 1Hz, and then estimate the peak frequency of the frequency spectrum. The estimated value is the peak frequency. Assuming that the frequency measured by the T method is, the measurement error is, and the frequency range of the established local frequency spectrum is , define the spectral line interval according to the required frequency accuracy (for example, if the frequency accuracy is 0.1Hz, the spectral line interval is 0.1Hz), use the adaptive LMS filter to extract the amplitude of each frequency signal, and establish the local frequency spectrum. Theoretically, if the sampling is fixed step size, the local frequency spectrum of the main wave is a sampling function. However, due to the variable step size sampling of LMS and the superposition of other harmonic vibration signals, its graph is slightly deviated. Therefore, it is impossible to use the fitting method to find the maximum value of the local frequency spectrum. It can only be obtained through multiple peak samplings. The mean is the frequency corresponding to the peak value. It can be estimated using the maximum likelihood estimation method. This method is simple, fast, and highly accurate. It is suitable for online estimation. The estimation formula is as follows The frequency corresponding to the maximum value of the frequency spectrum is the characteristic vibration frequency. After the characteristic vibration frequency is obtained, the amplitude and phase of each peak frequency vibration are obtained through the adaptive LMS filter.

[0030] In order to achieve comprehensive suppression of radial vibrations of low frequencies, the present invention proposes a multi-objective deep learning control algorithm, which can reduce the coupled vibrations generated by currents of various frequencies. The algorithm is simple and can make the vibrations of various frequencies quickly reach the lowest point.

[0031] 3) Multi-objective deep learning control scheme Combining the motor's magnetomotive force formula and Maxwell's stress formula, we can get x + f Frequency positive sequence current x Frequency radial vibration plays a major control role. To suppress the radial vibrations of the motor, it is necessary to mix the characteristic frequency currents for joint control. Due to many external factors, the vibration amplitude and phase of each frequency deviate randomly. Therefore, it is necessary to adjust the amplitude and phase of the corresponding current in real time to control the vibration amplitude of the corresponding frequency to the lowest point.

[0032] To achieve comprehensive suppression of radial vibrations at all low frequencies, the designed control algorithm needs to achieve: 1. Through adjustment x + f The amplitude and phase of the frequency current are used to suppress x Frequency of radial vibration.

[0033] 2. Due to x + f The frequency current will also affect x Radial vibrations other than the frequency will be affected, and coupled vibrations will occur between currents of different frequencies. Although the amplitude of the coupled vibration is very small, it still exists and will be superimposed as the number of injected frequency currents increases. A multi-objective optimization control is needed to achieve comprehensive suppression of radial vibrations of different frequencies.

[0034] In order to minimize the amplitude of each low-frequency radial vibration, the present invention proposes a multi-objective optimization model: In the formula is the objective function vector, i.e., the radial vibration amplitude vector of each harmonic; for x 1 arrive x k Frequency radial vibration, They are injected x k+1 The amplitude and initial phase of the frequency current, For the i constraints.

[0035] Although the various objectives in this multi-objective optimization model are not in the same domain, there is a certain degree of internal coupling, which can be decoupled by using adaptive optimization methods.

[0036] This paper uses a deep learning algorithm to find the minimum value of all single objective functions at the same time. This algorithm can automatically adjust the learning rate, converge quickly, and the algorithm is stable.k The specific steps of a single objective function are as follows: Assume: decay rate ρ , the initial learning rate η , initial parameters θ Assume: Gradient cumulant ν Initialized to 0, the initial constant δ is 10 -7 The loop body is as follows: The loop body is as follows: Simultaneous collection m Samples ,in , For sample The true value of calculate m The average gradient of samples: ; The square of the cumulative historical gradient: ; Calculate the update amount: (element-wise operations) Parameter update: The loop body ends; Loop end condition: , , is the fundamental current amplitude; In the above loop, m is the number of samples, yes m samples, For the k The frequency i Sample t -1 moment vibration amplitude, are the corresponding current amplitude and phase, For the k The frequency i The peak value of the vibration amplitude of the samples, For the t The average gradient at time, is the gradient symbol, is the gradient function, is the square of the cumulative historical gradient, is the deep learning coefficient with memory.

[0037] The method of finding the minimum value of multiple single objective functions at the same time can achieve adaptive decoupling, resulting in better suppression effect.

[0038] In the embodiment of the present invention, the principle block diagram of the system adopted by the present invention is as follows Figure 1 It includes PI regulator, iPark transformation, iClark transformation, AC current regulator, SiC Mosfet, permanent magnet motor, encoder, vibration acceleration sensor, LMS and peak frequency estimation module, multi-objective deep learning algorithm module, x + f Frequency current injection model module; Specifically, when in use, the encoder and vibration acceleration sensor are used to obtain the real-time mechanical angle, point angular frequency and radial vibration acceleration value respectively, and the obtained measurement values ​​are estimated by LMS and peak frequency to obtain the vibration amplitude peak point and the corresponding frequency value. The vibration amplitude and phase corresponding to the frequency of each vibration amplitude peak point are used as the input of the multi-objective deep learning algorithm to obtain the various x i The amplitude and phase of the frequency current, when passing through x i + f Frequency current injection model, get the harmonic current injected into the dq axis corresponding i dk and i qk The value is transformed by iPark and iClark, and then tracked in real time by the AC current regulator. After a period of time, the vibration amplitude corresponding to each vibration peak frequency is reduced to the minimum.

[0039] In this embodiment, Figure 2 This is a flow chart of an active vibration suppression method for a three-phase surface-mounted permanent magnet motor with random frequency mixing. In this method, the real-domain signal collected by the vibration acceleration sensor is first segmented with a step size of 1Hz for LMS and peak estimation, and finally a series of peak radial vibration frequencies are obtained. x 1 , x 2 , x 3 … x k ; Establish a current injection model in a rotating coordinate system, and for each peak radial vibration frequency x k , calculate x k + fThe positive sequence symmetrical current of each frequency is injected into the model; all the obtained current injection models of each frequency are combined and injected into the back end of the current loop PI regulator to obtain the vibration of each peak frequency; then multi-objective vibration deep learning is performed on the vibration of each peak frequency to obtain the optimal mixing current, adjust the motor frequency, and realize active vibration suppression; the comprehensive vibration suppression algorithm based on vibration feedback of the present invention solves the coupling problem of vibrations of each frequency generated by multi-frequency current, improves the vibration suppression effect, enhances the stability and anti-interference ability of the system, and is easy to implement.

[0040] In this embodiment, Figure 3 is the experimental waveform of this method. Figure 3 (a) is the original current waveform of the motor. Figure 3 (b) is the current waveform after comprehensive vibration suppression. Figure 3 (c) is the original radial vibration waveform of the motor. Figure 3 (d) is the radial vibration waveform after comprehensive vibration suppression. Figure 3 (e) is the original radial vibration spectrum of the motor. Figure 3 (f) is the radial vibration spectrum after comprehensive vibration suppression; from the radial vibration diagram before and after vibration suppression, it can be seen that except for the double frequency radial vibration within 200Hz, other low-frequency vibrations are significantly reduced, and the maximum harmonic amplitude is reduced by 95.3%. From the vibration waveform diagram before and after vibration suppression, it can be seen that the total vibration amplitude is reduced a lot. From the current waveform diagram before and after vibration suppression, it can be seen that the reduction of vibration does not mean that the current is an ideal sinusoidal current, because even if it is an ideal sinusoidal current, due to the structural characteristics of the motor, vibrations such as pole slot bearings will still exist. Therefore, if you want to reduce these vibrations, the current is not necessarily an ideal sinusoidal current. The measured current distortion rate after vibration suppression is 0.97%, so the distortion rate is not large and will not affect the performance of the motor.

[0041] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the schemes disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed in this disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the claims.

Claims

1. An active vibration suppression method for random frequency mixing of a three-phase surface-mounted permanent magnet motor, characterized in that: include: Step 1: Start the motor, collect the real-domain signal through the vibration acceleration sensor, and perform LMS and peak estimation on the real-domain signal collected by the vibration acceleration sensor in segments with a step size of 1 Hz. Finally, a series of peak radial vibration frequencies are obtained. x 1, x 2, x 3… x k ; Step 2: Establish a current injection model in a rotating coordinate system, and for each peak radial vibration frequency x k , calculate x k + f The positive sequence symmetrical current of the frequency is injected into the model; Step 3: Combine all the obtained current injection models of each frequency and inject them into the back end of the current loop PI regulator to obtain the vibration of each peak frequency; Step 4: Perform multi-objective vibration deep learning on the vibration of each peak frequency to obtain the optimal mixing current, adjust the motor frequency, and achieve active vibration suppression.

2. The active vibration suppression method for random frequency mixing of a three-phase surface-mounted permanent magnet motor according to claim 1 is characterized in that: The step 1 includes: First, the local frequency spectrum is established from low to high frequency with a step size of 1Hz, and then the peak frequency of the frequency spectrum is estimated. The estimated value is the peak frequency. The adaptive LMS filter is used to extract the amplitude of each frequency signal and establish the local frequency spectrum. After multiple peak sampling, the mean is the frequency corresponding to the peak, and the maximum likelihood estimation method is used to estimate , the estimation formula is as follows In the formula, x 1, x 2, x 3… x k is the sampling point, i.e. the frequency corresponding to a series of peak radial vibrations, y 1, y 2, y 3… y k are the peak vibration values, is the maximum likelihood estimate; the frequency corresponding to the maximum value of the frequency spectrum is the characteristic vibration frequency; after the characteristic vibration frequency is obtained, the amplitude and phase of each frequency vibration are obtained through an adaptive LMS filter.

3. The active vibration suppression method of random frequency mixing of a three-phase surface-mounted permanent magnet motor according to claim 1 is characterized in that: The current injection model in the rotating coordinate system in step 2 is as follows: In the formula, i d and i q They are the dq axis current, x is the required vibration frequency, ω =2π f is the electrical angular frequency, is the current amplitude, is the initial phase of the current.

4. The active vibration suppression method of random frequency mixing of a three-phase surface-mounted permanent magnet motor according to claim 1 is characterized in that: The multi-objective vibration deep learning for the vibrations of each peak frequency includes: Establish a multi-objective vibration optimization model: In the formula is the objective function vector, i.e., the radial vibration amplitude vector of each harmonic; for x 1 to x k Frequency radial vibration, They are injected x k+1 The amplitude and initial phase of the frequency current, For the i constraints.

5. The active vibration suppression method of random frequency mixing of a three-phase surface-mounted permanent magnet motor according to claim 1 is characterized in that: The multi-objective vibration deep learning of the vibrations at each peak frequency also includes: A deep learning algorithm is used to find the minimum value of all single objective functions at the same time; k The specific steps of a single objective function are as follows: Assume: decay rate ρ , the initial learning rate η , initial parameters θ ; Assume: Gradient cumulant ν Initialized to 0, the initial constant δ is 10 -7 ; The loop body is as follows: Simultaneous collection m Samples ,in , For sample The true value of calculate m The average gradient of samples: ; The square of the cumulative historical gradient: ; Calculate the update amount: (element-wise operations) Parameter update: The loop body ends; Loop end condition: , , is the fundamental current amplitude; In the above loop, m is the number of samples, yes m samples, For the k The frequency i Sample t -1 moment vibration amplitude, are the corresponding current amplitude and phase, For the k The frequency i The peak value of the vibration amplitude of the samples, For the t The average gradient at time, is the gradient symbol, is the gradient function, is the square of the cumulative historical gradient, is the deep learning coefficient with memory.

Images