Motor permanent magnet health state online evaluation and residual life prediction method and system

Abstract

The invention discloses a motor permanent magnet health state online evaluation and residual life prediction method and system, and belongs to the technical field of motor state monitoring and fault prediction. The method comprises the following steps: state monitoring and signal acquisition working condition judgment; a back electromotive force signal and a d-axis injection optimization high-frequency signal are collected; performing FFT analysis, calculating THD, and identifying d-axis inductance Ld on line; fTHD and delta Ld are calculated; carrying out normalization on the characteristic quantity FTHD and the delta Ld, and carrying out cross validation; under the balance of the three-phase voltage, the health index HImagnet of the permanent magnet is calculated; and fitting the descending trend of the HImagnet, and predicting the remaining useful life RUL. According to the method, the evaluation precision and robustness can be remarkably improved by fusing the two strongly correlated and complementary characteristic quantities of the back electromotive force harmonic distortion FTHD and the d-axis inductance variation delta Ld, real-time early warning of the demagnetization fault of the permanent magnet is realized, and trend prediction of the residual life is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of motor state monitoring and fault prediction, more specifically to a motor permanent magnet health state online evaluation and residual life prediction method and system. BACKGROUND

[0002] Permanent magnet synchronous motor (PMSM) is the core power component of electric vehicles, and the health state of its permanent magnet directly affects the performance, efficiency and safety of the motor. Permanent magnets are prone to irreversible demagnetization under harsh conditions such as high temperature, overcurrent and vibration, resulting in reduced motor performance and efficiency, and even causing serious faults. Therefore, it is crucial to accurately evaluate and predict the life of the permanent magnet health state in real time.

[0003] Chinese invention patent with publication number CN120541787A discloses a permanent magnet motor demagnetization fault diagnosis method based on deep learning. The patent uses "adaptive diagnosis strategy + U-Net prediction", although it introduces a practical strategy of dynamically adjusting the monitoring frequency, but its feature extraction still relies on additional vibration sensors, increasing the system hardware cost and installation complexity. At the same time, its U-Net model and competitive auction mechanism system are complex, and are essentially data-driven black box models, lacking sufficient integration with the physical mechanism of the motor. The deployment and operation of complex neural networks on the vehicle MCU are extremely challenging, with high power consumption and cost.

[0004] Chinese invention patent with publication number CN118686780A discloses a method for estimating the service life of an oil pump based on feedback signals. The patent calculates the deviation rate of multiple parameters (current, speed, temperature) relative to the initial value, sums the weighted values to obtain a comprehensive failure rate, and estimates the life loss according to the failure rate range. This method is too simple and linear, and fails to capture the nonlinear degradation process of permanent magnet demagnetization. Moreover, the weight setting relies on prior knowledge, lacks adaptability and scientificity, and has limited prediction accuracy. SUMMARY

[0005] The present application aims to overcome the shortcomings of the prior art and provide a motor permanent magnet health state online evaluation and residual life prediction method.

[0006] The present application adopts the following technical solutions: The motor permanent magnet health state online evaluation and residual life prediction method comprises the following steps: Step one, state monitoring and signal acquisition working condition judgment, including back electromotive force signal acquisition working condition and d-axis inductance identification working condition; Step two, signal acquisition and reconstruction: S21, under the condition of back electromotive force signal acquisition, the back electromotive force signal is collected, including collecting three-phase terminal voltage and collecting rotor position information; S22, under the condition of d-axis inductance identification, a high-frequency sinusoidal voltage signal or a spread spectrum signal is superimposed on the d-axis voltage command by the control algorithm; Step three, signal alignment and feature extraction: S31, the collected three-phase terminal voltage signal is aligned and coordinate-transformed according to the synchronously collected rotor position information, and is converted from the stationary three-phase coordinate system to the coordinate system synchronous with the rotor to obtain the back electromotive force waveform; S32, the obtained back electromotive force waveform is digitally filtered to extract a pure signal for THD calculation; Step four, waveform transformation and F_THD calculation: the aligned back electromotive force waveform is subjected to fast Fourier transform to obtain its frequency spectrum; the total harmonic distortion rate F_THD of the back electromotive force waveform is calculated; Step five, online parameter identification and feature extraction: S51, the d-axis inductance parameter Ld of the motor is identified in real time online when the motor is in the d-axis inductance identification condition; S52, a long-term historical record of the d-axis inductance parameter Ld is established to track its change trend and calculate its offset ΔLd relative to the initial health value as a characteristic quantity representing the change in the saturation degree of the magnetic circuit; Step six, feature normalization and cross-validation: S61, the characteristic quantities F_THD and ΔLd are normalized to eliminate the dimensional influence to obtain a normalized feature vector: [F_THD_norm, ΔLd_norm]; S62, cross-validation logic is introduced: when F_THD_norm abnormally rises, the system checks whether the collected three-phase voltage instantaneous values are balanced, if the three-phase voltage imbalance degree exceeds the limit, it is determined that there is a winding fault; if balanced, it enters step seven; Step seven, feature fusion and health assessment: the normalized feature vector passing through cross-validation is input into a pre-set weight function to output a comprehensive permanent magnet health index HI_magnet; Step eight, life prediction: the descending trend of HI_magnet is fitted to predict the time length or mileage required for the value to drop to a pre-set failure threshold, i.e. the remaining useful life RUL.

[0007] The above step one specifically includes: S11, the driving torque command is zero and the motor speed is higher than a pre-set threshold, i.e. the no-load coasting state; or the driving torque command is lower than a first threshold and the motor speed fluctuation is lower than a second threshold, i.e. the light-load constant-speed running state, then the back electromotive force signal acquisition condition is entered; S12, the driving torque command is lower than a third threshold and the motor speed is stable, i.e. the steady-state light-load running state, then the d-axis inductance identification condition is entered.

[0008] In a preferred embodiment, the present application preferably uses a simplified model reference adaptive or recursive least square method for d-axis inductance identification.

[0009] In a preferred embodiment, the specific implementation of S21 in step two is as follows: under the working condition of back electromotive force signal collection, the controller keeps the inverter PWM output, but uses the minimum duty cycle mode or blanking strategy, uses the phase voltage sensor built-in the motor controller or collects the DC bus voltage and inverter switch state through software reconstruction to collect three-phase terminal voltage; at the same time, the high-precision resolver is used to synchronously collect the rotor position information.

[0010] In a preferred embodiment, step four uses a sliding window FFT instead of a full-size FFT to calculate F_THD.

[0011] In a preferred embodiment, the normalized boundary value (*_min, *_max) of step six is determined by a bench accelerated aging test.

[0012] In a preferred embodiment, the weight function of step seven has the expression: HI_magnet = 1 - (w1 * F_THD_norm + w2 *ΔLd_norm) where w1 and w2 are weight coefficients, and w1 + w2 = 1, and their values are determined by fitting the data of the accelerated demagnetization aging test; the output range of HI_magnet is [0, 1], 1 represents health, and 0 represents complete failure.

[0013] The weight coefficients w1 and w2 are determined by accelerated demagnetization aging tests on not less than two samples of the same type on a bench, specifically as follows: collect F_THD and ΔLd data under different demagnetization degrees, and record the real performance attenuation degree to normalize the real health index HI_actual; use a least squares fitting algorithm to solve the weight values w1 and w2 that minimize the error of the formula HI_actual≈ 1 - (w1 * F_THD_norm + w2 * ΔLd_norm).

[0014] In a preferred embodiment, step eight uses an exponential smoothing method, linear regression or weighted moving average method to fit the downward trend of HI_magnet.

[0015] In a preferred embodiment, step eight also includes early warning, specifically: when the real-time monitored HI_magnet is lower than the health threshold, or the predicted RUL is lower than the safety value, send hierarchical early warning information to the vehicle instrument panel or cloud monitoring platform through the controller area network (CAN) bus.

[0016] The application further provides a motor control system comprising a processor and a memory, wherein the memory stores a computer program, and the processor implements the motor permanent magnet health state online evaluation and residual life prediction method when executing the computer program.

[0017] From the above description of the application, compared with the prior art, the application has the following advantages: 1. The application calculates the health index (HI_magnet) by fusing the back electromotive force harmonic distortion (F_THD) and the d-axis inductance change (ΔLd), two strongly correlated and complementary characteristic quantities, and predicts the residual life (RUL) accordingly. The application significantly improves the evaluation accuracy and robustness, realizes real-time early warning of permanent magnet demagnetization failure, and realizes trend prediction of the residual life.

[0018] 2. Inductive online monitoring: fully utilizing existing vehicle sensor and controller resources, without increasing any hardware cost, realizing real vehicle end embedded real-time evaluation.

[0019] 3. Optimized signal acquisition strategy: under no-load coasting / light-load constant-speed working conditions, the back electromotive force is acquired through low-interference PWM and filtering technology; under steady-state light-load working conditions, the d-axis inductance is identified by injecting high-frequency signals with optimized parameters, which minimizes the influence on system performance while ensuring the quality of the characteristic.

[0020] 4. Robustness design based on cross-validation: introducing three-phase voltage balance check, cross-validating the back electromotive force characteristics, effectively distinguishing permanent magnet demagnetization from winding failure, and avoiding misjudgment.

[0021] 5. Weight self-adaptation based on bench data: scientifically determining the characteristic weight through reliability test data, making the health index more consistent with the actual physical degradation process, and avoiding the limitations of artificial experience setting. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 The figure is a flowchart of the application.

[0023] Figure 2 The figure is a flowchart of residual life prediction and early warning release of the application. DETAILED DESCRIPTION

[0024] The specific embodiments of the application will be described below with reference to the accompanying drawings. In order to fully understand the application, many details are described below, but the application can be implemented without these details for those skilled in the art. For well-known components, methods and processes, the following will not be described in detail.

[0025] The motor permanent magnet health state online evaluation and residual life prediction method refers to Figure 1 , comprising the following steps: Step one, state monitoring and signal acquisition condition judgment The motor controller monitors the driving torque command and motor speed in real time. When any of the following conditions is met, the corresponding signal acquisition process is triggered.

[0026] 1. Back-EMF signal acquisition condition: driving torque command is zero and motor speed is higher than a preset threshold (e.g. 500 rpm), i.e. the state of no-load coasting; or driving torque command is lower than a first threshold (e.g. 10% rated torque) and motor speed fluctuation is lower than a second threshold (e.g. ±2%), i.e. the state of light load constant speed operation.

[0027] 2. d-axis inductance identification condition: driving torque command is lower than a third threshold (e.g. 20% rated torque) and motor speed is stable, i.e. the state of steady light load operation.

[0028] Step two, signal acquisition and reconstruction 1. Collect back-EMF signal: in the back-EMF signal acquisition condition, the controller keeps the inverter PWM output, but uses the minimum duty cycle mode (e.g. 1%) or a specific blanking strategy to reduce switching noise interference. Use the built-in phase voltage sensor of the motor controller (or through software reconstruction by collecting DC bus voltage and inverter switch state) to collect three-phase terminal voltage (Ua, Ub, Uc).

[0029] 2. d-axis injection inductance identification signal: in the d-axis inductance identification condition, the control algorithm superimposes a high-frequency sinusoidal voltage signal or spread spectrum signal with optimized amplitude (e.g. 5V) and frequency (e.g. 500Hz) on the d-axis voltage command for online identification. The selection of this condition and the optimization of signal parameters aims to minimize the impact on motor NVH and operating efficiency.

[0030] Step three, signal alignment and feature extraction Align and coordinate transform the collected three-phase terminal voltage signal according to the synchronous collected rotor position information, convert it from the stationary three-phase coordinate system (a-b-c) to the coordinate system rotating with the rotor, and get the back-EMF waveform. Perform digital filtering (such as band-pass filter) on the obtained back-EMF waveform to further suppress switching noise and background interference, and extract the pure signal for THD calculation.

[0031] Step four, waveform transformation and F_THD calculation 1. Perform fast Fourier transform (FFT) on the aligned back-EMF waveform to get its frequency spectrum.

[0032] 2. Calculate the total harmonic distortion rate (F_THD) of the back-EMF waveform.

[0033] Step five, online parameter identification and feature extraction (ΔLd) When the motor is in the d-axis inductance identification condition, the d-axis inductance parameter Ld of the motor is identified in real time online by using the high-frequency signal injected by the control algorithm itself or through a model reference adaptive system (MRAS).

[0034] A long-term historical record of the d-axis inductance parameter Ld is established, its change trend is tracked, and its offset ΔLd relative to the initial healthy value is calculated as a characteristic quantity representing the change in the saturation degree of the magnetic circuit.

[0035] Step six, feature normalization and cross-validation The characteristic quantities F_THD and ΔLd are normalized to eliminate the dimensional influence, and the normalized feature vector [F_THD_norm, ΔLd_norm] is obtained.

[0036] The normalized boundary values (*_min, *_max) are determined by the bench accelerated aging test.

[0037] Before health assessment, cross-validation logic is introduced. When F_THD_norm abnormally rises, the system checks whether the three-phase voltage instantaneous values are balanced (for example, whether the difference between the effective values of the three-phase voltages exceeds a preset threshold, such as 5%). If the three-phase voltage imbalance exceeds the limit, it is determined that the winding is likely to fail, and a winding failure warning is triggered, instead of immediately performing permanent magnet health assessment, thereby avoiding misjudgment caused by local failure of the winding.

[0038] Step seven, feature fusion and health assessment The normalized feature vector passed through cross-validation is input into a pre-set weight function, and the comprehensive permanent magnet health index HI_magnet is output.

[0039] The expression of the weight function is: HI_magnet = 1 - (w1 * F_THD_norm + w2 * ΔLd_norm) Where w1 and w2 are weight coefficients, and w1 + w2 = 1; their values are determined by fitting the data of the bench accelerated aging test, reflecting the importance of different features.

[0040] The output range of HI_magnet is [0, 1], 1 represents health, and 0 represents complete failure.

[0041] Step eight: bench data and weight determination The weight coefficients w1 and w2 are determined by performing accelerated demagnetization aging tests on a certain number (e.g., no less than 2) of same model sample pieces on a test bench. F_THD and ALd data at different demagnetization levels are collected, and the real performance attenuation degree (e.g., back electromotive force amplitude drop rate) is recorded to normalize the real health index HI_actual. The least square method or other fitting algorithms are used to solve the weight values w1 and w2 that minimize the error of the formula HI_actual ≈ 1 - (w1 * F_THD_norm + w2 * ALd_norm).

[0042] Step nine: life prediction and early warning The change of HI_magnet over time or running mileage is continuously recorded to form a health index time series.

[0043] The time series prediction algorithm (such as exponential smoothing method, ARIMA model, etc.) is used to fit the downward trend of HI_magnet, and the time length or mileage required for the value to drop to the preset failure threshold, i.e., the remaining useful life (RUL) is predicted.

[0044] When the real-time monitored HI_magnet is lower than the health threshold, or the predicted RUL is lower than the safety value, the controller area network (CAN) bus sends graded early warning information to the vehicle instrument panel or the cloud monitoring platform.

[0045] The application also provides a motor control system, which comprises a processor and a memory, the memory stores a computer program, and the processor implements the motor permanent magnet health state online evaluation and remaining life prediction method when executing the computer program.

[0046] The following is a specific implementation case of the application.

[0047] I. Preset parameters Back electromotive force signal collection working condition judgment threshold: torque = 0 and speed > 600 r / min, or torque < 5% rated torque and speed fluctuation < ± 2%.

[0048] D-axis inductance identification working condition judgment threshold: torque < 10% rated torque and speed is stable.

[0049] Back electromotive force THD calculation range: 2nd to 50th harmonic.

[0050] Inductance parameter identification method: high frequency signal injection, parameters: d-axis, 500 Hz, 5V amplitude, using spread spectrum modulation.

[0051] Health index failure threshold HI_fail: 0.3.

[0052] Health warning threshold HI_warn: 0.6.

[0053] Weighting factor (determined by bench test fitting): w1=0.6, w2=0.4.

[0054] Three-phase voltage unbalance threshold: 5%.

[0055] II. Health assessment process 1. Real-time data acquisition Counter electromotive force acquisition period: when the vehicle enters the counter electromotive force signal acquisition condition, the inverter adopts a low interference PWM mode, and synchronously acquires three-phase voltage and rotor position.

[0056] d-axis inductance identification period: when the vehicle enters the d-axis inductance identification condition, the control algorithm (injects high-frequency signal according to optimized parameters), extracts the response component from the phase current through the filter, and calculates the current value of Ld in real time.

[0057] 2. Feature processing (1) Calculate total harmonic distortion rate F_THD after coordinate system transformation and FFT: : RMS of the fundamental of counter electromotive force.

[0058] : RMS of the nth harmonic.

[0059] N: the highest harmonic number to be considered.

[0060] Take N=50, the calculation result F_THD_current=10.15%.

[0061] (2) Calculate the offset percentage of current Ld_current and new vehicle state initial Ld_initial to get ΔLd: Ld_current: the d-axis inductance value obtained by real-time online identification; the identification result is Ld_current =0.55 mH.

[0062] Ld_initial: the initial d-axis inductance reference value; when the vehicle is powered on for the first time and the system judges that the motor is in a healthy state (normal temperature, no fault code), measure Ld multiple times and take the average value, store it in the MCU control board EEPROM; Ld_initial = 0.5 mH.

[0063] The calculation result ΔLd =10%.

[0064] (3) Normalization of F_THD and ΔLd: F_THD_current: the current calculated total harmonic distortion raw feature value; ΔLd_current: the current calculated Ld offset percentage raw feature value from the initial Ld of the new vehicle state.

[0065] *_min and *_max: normalization boundary values, which are determined by the bench accelerated aging test, representing the reasonable range of the feature from "perfect health" to "severe failure".

[0066] Determined by bench test: F_THD range: about 3% when healthy, about 20% when failed. Therefore, set F_THD_min=3, F_THD_max=20.

[0067] ΔLd range: about 0% when healthy, about 25% when failed. Therefore, set ΔLd_min=0, ΔLd_max=25.

[0068] Then the normalized result is: F_THD_norm=(10.15-3) / (20-3)=0.42; ΔLd_norm=(10-0) / (25-0)=10 / 25=0.40.

[0069] After calculating F_THD_norm, the system checks the three-phase voltage unbalance degree in this collection period. Calculate the effective values of three-phase voltage Ua_rms, Ub_rms, Uc_rms, if Then determine that the voltage is unbalanced, this F_THD feature is invalid, trigger the "suspected winding fault, please check" warning, and do not update HI_magnet temporarily.

[0070] 3. Health index calculation Call the weight function to calculate the current health index: HI_magnet=1-(0.6*F_THD_norm+0.4*ΔLd_norm); calculate HI_magnet=1-(0.6*0.42+0.4*0.4)=0.59.

[0071] 4. Result judgment The calculated HI_magnet=0.59 has fallen below the health warning threshold 0.6, and the system sends a prompt message to the instrument through the CAN bus: "motor permanent magnet performance moderate attenuation, suggest attention".

[0072] III. Life prediction process (refer to Figure 2 ) 1. Historical data record and management The system opens a storage area in the MCU control board EEPROM to continuously record the calculated health index HI_magnet and its corresponding cumulative running mileage Mileage. To save storage space, the "circular queue" or "equal interval mileage record" strategy can be used.

[0073] Historical record table 2. Trend prediction algorithm (exponential smoothing method) The exponential smoothing method is used for trend fitting and prediction, and the calculation formula is: St: the smoothed prediction value of the current period (mileage point t); St-1: the smoothed prediction value of the previous period (mileage point t-1); Xt: the actual observed value (i.e. HI_magnet) of the current period (mileage point t); α: smoothing coefficient, value range (0, 1).

[0074] The last four data points are demonstrated as follows, the smoothing coefficient α=0.3, and the initial smoothed value St−1 is set to the first actual observed value. Mileage (km) Actual HI Xt Smoothed predicted HI St 45 0.73 0.73 Initial value 50 0.65 0.706 55 0.61 0.638 60 0.59 0.604

[0075] The smoothed prediction value St=0.604 of the current mileage point (6 million kilometers) is obtained.

[0076] 3. Residual useful life (RUL) calculation After obtaining the trend line, the mileage required for the health index to drop to the preset failure threshold HI_fail is predicted.

[0077] Define the failure threshold: HI_fail=0.3 (when the health index is lower than this value, consider the permanent magnet function failure).

[0078] Calculate the trend slope: use the last two smoothed prediction values to calculate the current health index decline slope a : The calculation result is a =-0.0068 (decrease rate per 10,000 kilometers).

[0079] Calculate the residual life (RUL): To make the prediction more stable, HI_current takes the current trend value St instead of the instantaneous observation value Xt (which contains various random noises and short-term disturbances).

[0080] The calculated RUL = 44.7 million km.

[0081] 4. Early warning release The system compares the predicted RUL value with the preset safety threshold and releases the corresponding level of early warning information through the CAN bus.

[0082] Predicted RUL > 10 million km: normal state, no need for early warning.

[0083] 5 million km < predicted RUL ≤ 10 million km: level 1 early warning (prompt). Prompt "Suggest checking the motor system when the next maintenance is performed" through the instrument panel.

[0084] 1 million km < predicted RUL ≤ 5 million km: level 2 early warning (warning). Warning "Motor permanent magnet performance degradation, suggest planning maintenance" through the instrument panel.

[0085] Predicted RUL ≤ 1 million km: level 3 early warning (serious). Immediately pop up the warning "Motor system needs immediate repair!" through the instrument panel and limit the motor peak power output.

[0086] According to the above calculation, the current health index trend indicates that the permanent magnet will fail after about 44.7 million km, and the state is normal, no need for early warning.

[0087] The above is only a specific embodiment of the present application, but the design concept of the present application is not limited thereto, and any non-substantial modification of the present application using this concept shall be deemed as an act of infringing the protection scope of the present application.

Claims

1. A method for on-line evaluation of health state and prediction of residual life of permanent magnet of electric machine, characterized in that, Comprise the following steps: Step one, state monitoring and signal acquisition working condition judgment, including back electromotive force signal acquisition working condition and d-axis inductance identification working condition; Step two, signal acquisition and reconstruction: S21, in the back electromotive force signal acquisition working condition, collect back electromotive force signal, including collecting three-phase terminal voltage and collecting rotor position information;S22, in the d-axis inductance identification working condition, control algorithm superimposes a high-frequency sinusoidal voltage signal or spread spectrum signal on the d-axis voltage instruction; Step three, signal alignment and feature extraction: S31, the collected three-phase terminal voltage signal is aligned and coordinate transformed according to the synchronous collected rotor position information, which is converted from the stationary three-phase coordinate system to the coordinate system synchronous with the rotor, to obtain the back electromotive force waveform;S32, the obtained back electromotive force waveform is digitally filtered to extract the pure signal for THD calculation; Step four, waveform transformation and F_THD calculation: the aligned back electromotive force waveform is subjected to fast Fourier transform to obtain its frequency spectrum;Calculate the total harmonic distortion rate F_THD of the back electromotive force waveform; Step five, online parameter identification and feature extraction: S51, extract the d-axis inductance parameter Ld of the motor in the d-axis inductance identification working condition;S52, establish a long-term historical record of the d-axis inductance parameter Ld, track its change trend, calculate its offset ΔLd relative to the initial health value, as a characteristic quantity representing the change of magnetic circuit saturation degree; Step six, feature normalization and cross-validation: S61, normalize the characteristic quantities F_THD and ΔLd to eliminate the influence of dimension, and obtain the normalized feature vector: [F_THD_norm, ΔLd_norm];S62, introduce cross-validation logic: when F_THD_norm abnormally rises, the system checks whether the collected three-phase voltage instantaneous value is balanced, if the three-phase voltage imbalance degree is out of limit, it is judged as winding fault;If balanced, go to step seven; Step seven, feature fusion and health assessment: input the normalized feature vector through cross-validation into the pre-set weight function, and output the comprehensive permanent magnet health index HI_magnet; Step eight, life prediction: fit the descending trend of HI_magnet, predict the time length or mileage required for its value to drop to the pre-set failure threshold, that is, the remaining useful life RUL.

2. The method of claim 1, wherein the method further comprises: The step one specifically comprises: S11, the drive torque instruction is zero and the motor speed is higher than a pre-set threshold, that is, the no-load coasting state;Or the drive torque instruction is lower than the first threshold and the motor speed fluctuation is lower than the second threshold, that is, the light load constant speed running state, then enter the back electromotive force signal acquisition working condition;S12, the drive torque instruction is lower than the third threshold and the motor speed is stable, that is, the steady-state light load running state, then enter the d-axis inductance identification working condition.

3. The method of claim 1, wherein the method further comprises: Simplified model reference adaptive or recursive least squares method is used for d-axis inductance identification.

4. The method of claim 1, wherein the method further comprises: The specific implementation of S21 in step two is as follows: in the condition of back electromotive force signal acquisition, the controller keeps the PWM output of the inverter, but adopts the minimum duty cycle mode or the blanking strategy, and acquires the three-phase terminal voltage by using the phase voltage sensor built in the motor controller or by acquiring the DC bus voltage and inverter switch state through software reconstruction; at the same time, the rotor position information is synchronously acquired by using a high-precision rotary transformer.

5. The method of claim 1, wherein: In step four, a sliding window FFT is used instead of a full-size FFT to calculate F_THD.

6. The method of claim 1, wherein the method further comprises: determining the health state of the permanent magnet of the electric machine based on the magnetic field distribution. The normalized boundary value (*_min, *_max) in step six is determined by a bench accelerated aging test.

7. The method of claim 1, wherein: The weight function in step seven has the expression: HI_magnet = 1 - (w1 * F_THD_norm + w2 *ΔLd_norm) where w1 and w2 are weight coefficients, and w1 + w2 = 1, and their values are determined by fitting the data of the accelerated demagnetization aging test; the output range of HI_magnet is [0, 1], 1 represents health, and 0 represents complete failure.

8. The method of claim 7, wherein the method further comprises: determining the health state of the permanent magnet of the electric machine based on the magnetic field strength and the magnetic field direction. The weight coefficients w1 and w2 are determined by performing an accelerated demagnetization aging test on not less than two same type samples on a bench, and the specific implementation is as follows: the F_THD and ΔLd data under different demagnetization degrees are acquired, and the real performance attenuation degree is recorded to be normalized to the real health index HI_actual; the least square method or other fitting algorithm is used to solve the weight values w1 and w2 that make the error of the formula HI_actual ≈ 1 - (w1* F_THD_norm + w2 * ΔLd_norm) minimum.

9. The method of claim 1, wherein: In step eight, the exponential smoothing method, linear regression or weighted moving average method is used to fit the downward trend of HI_magnet.

10. The method of claim 1, wherein: In step eight, the method further includes early warning, specifically: when the real-time monitored HI_magnet is lower than the health threshold, or the predicted RUL is lower than the safety value, the controller local area network bus sends hierarchical early warning information to the vehicle instrument panel or the cloud monitoring platform.

11. An electric machine control system comprising a processor and a memory, characterized by: The memory stores a computer program, and the processor implements the motor permanent magnet health state online evaluation and remaining useful life prediction method according to any one of claims 1-10 when executing the computer program.

Images