Non-inductive FOC ripple characteristic rotor positioning method and system

Abstract

The invention discloses a non-inductive FOC ripple characteristic rotor positioning method and system, and the method comprises the steps: injecting a multi-direction detection voltage vector with the amplitude increasing gradually into a stator winding at a zero-speed starting stage of a motor, and enabling the multi-direction detection voltage vector to comprise at least three non-collinear space directions; synchronously acquiring three-phase current transient response under the action of each voltage vector, and extracting a current ripple characteristic component matched with the rotor salient pole frequency through band-pass filtering; based on the spatial distribution anisotropy of the current ripple characteristic component, a position sensitive factor set corresponding to the voltage direction is constructed, and each position sensitive factor in the position sensitive factor set represents the distribution weight of ripple energy under excitation in the voltage direction. According to the invention, all-condition robust control is realized through a multi-dimensional cooperative detection mechanism and a dynamic feature compensation architecture.

Description

Technical Field

[0001] The present application relates to the field of motor control technology, and in particular to a method and system for positioning a rotor using a sensorless FOC ripple characteristic. Background Art

[0002] In brushless motor drive solutions, detecting the initial rotor position during the sensorless startup phase is a key challenge.

[0003] The traditional high-frequency signal injection method relies on the inherent salient pole effect of the motor. However, in low-saliency motors or deep magnetic saturation conditions, the current response signal is extremely weak, and conventional unidirectional excitation makes it difficult to extract effective features, resulting in serious lack of zero-speed positioning accuracy. Even more serious is that the actual operation of the motor faces complex dynamic conditions: drastic changes in winding temperature cause resistance parameter drift, sudden load changes cause core saturation characteristic distortion, and permanent magnet demagnetization weakens the magnetic field strength. These factors together cause the current ripple characteristics to produce amplitude attenuation, frequency drift, and phase distortion. Existing ripple detection technology has three limitations:

[0004] First, fixed-direction voltage injection cannot capture the spatial anisotropy of magnetic saturation, and weak ripple features are easily overwhelmed by the nonlinear noise of the inverter. Second, static filter parameters are difficult to adapt to the frequency offset of hot salient poles, and the reliability of feature extraction is sharply reduced. Third, the position solution model does not consider the strong coupling relationship between flux decay and saturation effects, resulting in systematic deviations in angle solution under demagnetization conditions. These shortcomings lead to the risk of positioning failure of traditional methods under extreme conditions. Summary of the Invention

[0005] In order to solve the above problems, an embodiment of the present invention provides a non-sensing FOC ripple characteristic rotor positioning method, the method comprising:

[0006] Obtain visible light reflection images and near-infrared scattering spectrum data of the PCB surface;

[0007] Performing spatial domain filtering on the visible light reflection image to extract interface area pixels, and calculating the coating layer thickness gradient distribution based on the Mie scattering characteristics in the scattering spectrum data, and fusing them to generate an edge gradient distribution map;

[0008] According to the gradient amplitude change rate in the edge gradient distribution map, an adaptive dynamic threshold algorithm is used to segment the effective coverage area of ​​the coating layer, and the curvature extreme point density and boundary fractal dimension parameters at the segmentation boundary are extracted;

[0009] Inputting the curvature extreme point density and boundary fractal dimension parameters into a pre-trained coating curing morphology prediction model, and outputting a coating layer interface bonding strength prediction value and potential peeling risk area coordinates by coupling thermodynamic simulation data;

[0010] The coating surface is scanned using an array of dielectrically coupled sensors, and the impedance phase response of each scanning point at different excitation frequencies is measured. The microscopic porosity distribution is calculated based on the slope of the phase shift versus frequency.

[0011] The interface bonding strength prediction value, potential peeling risk area coordinates and micro-porosity distribution map are temporally and spatially aligned, and a comprehensive evaluation matrix of coating quality is generated through a multi-source data decision fusion algorithm, and a defect location report and quality grade classification results are output.

[0012] Furthermore, the spatial domain filtering processing method includes:

[0013] A non-uniform illumination compensation algorithm is used to perform illumination equalization processing on the visible light reflection image. The algorithm calculates the pixel compensation coefficient by establishing a local illumination distribution model, where the compensation coefficient is dynamically adjusted according to the surface roughness characteristics in the near-infrared band scattering spectrum data.

[0014] Furthermore, the method for generating the edge gradient distribution map includes:

[0015] The gradient phase information of the visible light reflection image is superimposed with the Mie scattering phase delay of the near-infrared scattering spectrum, and the interference of the substrate material texture on the coating boundary identification is eliminated through phase gradient consistency verification.

[0016] Furthermore, the adaptive dynamic threshold algorithm includes:

[0017] A probability density function is established according to the statistical distribution of the gradient amplitude change rate. The variational mode decomposition technique is used to separate the threshold response characteristics of the coating body and the boundary transition zone, and the segmentation threshold interval is dynamically determined.

[0018] Furthermore, the training method of the coating curing morphology prediction model includes:

[0019] A finite element simulation model coupling heat-humidity-force multi-physics fields was constructed. By introducing the time-varying viscoelastic constitutive equation of the coating material, a training set of interfacial stress distribution under different curing conditions was generated. The training set included the mapping relationship between temperature gradient, moisture penetration depth and interfacial peeling stress.

[0020] Furthermore, the analysis method of the impedance phase response includes:

[0021] An equivalent circuit model of the microscopic pores in the coating layer was established, and the phase offset at different frequencies was fitted using the Cole-Cole distribution function to extract the pore connectivity index and equivalent dielectric relaxation time parameters.

[0022] Furthermore, the implementation method of spatiotemporal registration includes:

[0023] A three-dimensional registration coordinate system is established based on the spatial topological structure of the coating surface. The mechanical confidence of the predicted value of the interface bonding strength and the electrical sensitivity of the micro-porosity distribution are weightedly fused to generate a quality evaluation weight matrix with physical interpretability.

[0024] Furthermore, the sensorless FOC ripple characteristic rotor positioning method further includes:

[0025] A multi-frequency eddy current sensing unit is integrated into the array dielectric coupling sensor, and the coating dielectric phase response and substrate eddy current skin effect signals are synchronously collected through frequency division multiplexing technology.

[0026] A frequency-domain blind source separation model is constructed based on the inverse square relationship between the skin depth of the eddy current signal and the frequency, and the cross-coupling noise components caused by metal substrate defects are decoupled from the dielectric phase response.

[0027] The frequency domain characteristics of high-frequency eddy currents, which are sensitive to the pore closure of the coating surface, and low-frequency eddy currents, which represent the continuity of the base metal, are used to reconstruct the microscopic porosity distribution map after removing the base artifacts.

[0028] A non-sensing FOC ripple characteristic rotor positioning method, the system includes:

[0029] Dual-mode acquisition module, which acquires visible light reflection images and near-infrared scattering spectrum data of the PCB surface;

[0030] A gradient map generation module, which performs spatial domain filtering on the visible light reflection image to extract interface area pixels, and calculates the coating thickness gradient distribution based on the Mie scattering characteristics in the scattering spectrum data, and fuses them to generate an edge gradient distribution map;

[0031] A dynamic threshold segmentation module, which uses an adaptive dynamic threshold algorithm to segment the effective coverage area of ​​the coating layer according to the gradient amplitude change rate in the edge gradient distribution map, and extracts the curvature extreme point density and boundary fractal dimension parameters at the segmentation boundary;

[0032] An interface strength prediction module, which inputs the curvature extreme point density and boundary fractal dimension parameters into a pre-trained coating curing morphology prediction model, and outputs a coating layer interface bonding strength prediction value and coordinates of potential peeling risk areas by coupling thermodynamic simulation data;

[0033] The dielectric porosity analysis module scans the coating surface using an array of dielectric coupling sensors, measures the impedance phase response of each scanning point at different excitation frequencies, and calculates the microscopic porosity distribution map based on the slope of the phase shift versus frequency.

[0034] The result output module performs spatiotemporal registration on the interface bonding strength prediction value, the coordinates of the potential peeling risk area and the micro-porosity distribution map, generates a comprehensive evaluation matrix of the coating quality through a multi-source data decision fusion algorithm, and outputs a defect location report and quality grade classification results.

[0035] The technical effects and advantages of the non-sensing FOC ripple characteristic rotor positioning method provided by the present invention are as follows:

[0036] The present invention solves the two core technical problems of insufficient zero-speed positioning accuracy and poor adaptability to operating disturbances during sensorless starting through a multi-dimensional collaborative detection mechanism and a dynamic feature compensation architecture, thus achieving robust control under all operating conditions. The present invention uses a multi-directional rotating voltage excitation strategy to deeply excite the magnetic saturation spatial anisotropy response, and can still extract highly robust rotor orientation characteristics under weak salient pole and deep saturation conditions; constructs a ripple spectrum adaptive tracking mechanism and a flux linkage-saturation coupling compensation model to effectively suppress characteristic distortion caused by parameter drift and demagnetization effects, ensuring the stability of angle solution under extreme operating conditions; pulse excitation and ripple analysis interact in a closed loop in the time-frequency domain, dynamically avoiding single-mode detection blind spots and significantly improving the system's fault tolerance in complex electromagnetic environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flow chart of a method for positioning a rotor with sensorless FOC ripple characteristics according to an embodiment of the present application;

[0038] Figure 2 This is a flow chart of the non-sensing FOC ripple characteristic rotor positioning method according to the second embodiment of the present application;

[0039] Figure 3 This is a connection diagram of the sensorless FOC ripple characteristic rotor positioning system according to the third embodiment of the present application. DETAILED DESCRIPTION

[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0041] Example 1: Please refer to Figure 1 As shown, an embodiment of the present invention provides a non-sensing FOC ripple characteristic rotor positioning method, the method comprising:

[0042] Step S1: During the zero-speed starting phase of the motor, a multi-directional detection voltage vector with increasing amplitude is injected into the stator winding, where the multi-directional detection voltage vector includes at least three non-collinear spatial directions;

[0043] Step S2: synchronously collect the three-phase current transient response under the action of each voltage vector, and extract the current ripple characteristic component matching the rotor salient pole frequency through bandpass filtering;

[0044] Step S3: Based on the spatial distribution anisotropy of the current ripple characteristic component, a set of position sensitive factors corresponding to the voltage direction is constructed, where each position sensitive factor in the set of position sensitive factors represents the distribution weight of the ripple energy under the excitation of the voltage direction;

[0045] Step S4: Input the position sensitive factor set into the pre-trained ripple-position mapping model to output the initial rotor position angle. The ripple-position mapping model establishes a nonlinear relationship between ripple characteristics and position through the motor magnetic saturation effect.

[0046] Step S5: Switch to FOC closed-loop operation according to the initial rotor position angle;

[0047] The extraction of the current ripple characteristic component in step S2 is independent of the back electromotive force signal, and the ripple-position mapping model in step S4 completes the position solution when the motor is stationary.

[0048] During the current ripple characteristic component extraction process in step S2, the collected three-phase current transient response needs to be separated in the frequency domain. Since the current ripple frequency excited by the rotor salient pole effect is directly related to the motor structural parameters, this method uses a dynamically adjustable bandpass filter for feature extraction. The passband center frequency of the bandpass filter needs to match the fundamental frequency of the rotor salient pole harmonics in real time. This frequency is determined by the number of motor pole pairs and the salient pole characteristics. It is a fixed value in the static state but needs to be autonomously identified by the algorithm. In specific implementation, the control system dynamically calibrates the center frequency according to the spectral characteristics of the current injected voltage vector to ensure that the ripple energy is maximized (for example, when the estimated salient pole fundamental frequency is 250Hz, the passband center frequency is automatically locked to 250Hz).

[0049] At the same time, the bandwidth setting of the filter is related to the anisotropy strength of the motor magnetic circuit; according to the principle of electromagnetics, the difference between the quadrature and direct axis inductance (L d −L q ) is larger, the more concentrated the current ripple energy is on the spectrum. Therefore, the filter is designed so that the -3dB bandwidth is proportional to the difference between the quadrature and direct axis inductances. For built-in permanent magnet motors with high saliency (such as L d / L q =0.5), a narrower bandwidth is used to suppress noise. For surface-mount motors with low saliency (Ld≈Lq), the bandwidth is appropriately widened to capture the dispersed ripple energy. This adaptive bandwidth mechanism is compatible with the saliency characteristics of different types of motors while ensuring the signal-to-noise ratio of feature extraction.

[0050] The spatial distribution characteristics of the current ripple characteristic components processed in this way are further converted into position-sensitive factors in step S3. The entire process is completely based on current response and does not rely on back electromotive force signals, which meets the core requirements of static state positioning.

[0051] When constructing the position-sensitive factor set in step S3, it is necessary to quantify the spatial anisotropy of the current ripple characteristic components. For the current ripple signal collected in each voltage direction, two key features, namely the kurtosis coefficient and the envelope root mean square value, are first extracted. The extraction method includes:

[0052] The kurtosis coefficient reflects the sharpness of the ripple signal amplitude distribution. In the direction where the salient pole effect is significant (such as near the direct axis), sharper current pulses will be generated due to magnetic saturation, and the kurtosis value will be higher.

[0053] The RMS value of the envelope characterizes the overall intensity of the ripple energy and is related to the projection component of the voltage vector in the direction of the rotor salient poles.

[0054] Multiplying the two together yields the original sensitivity factor, which incorporates both waveform morphology and energy intensity information. For example, when the injected voltage is aligned with the rotor's direct axis (d-axis), the current ripple exhibits high-frequency, narrow pulses with concentrated energy, and the product significantly increases. However, when the voltage is close to the quadrature axis (q-axis), the ripple distribution is flat and the energy is dispersed, resulting in a sharp decrease in the product.

[0055] In order to unify the dimensions and enhance the directional contrast, the original sensitivity factor is normalized by the exponential function. Mapping to the interval [0,1]: ;

[0056] Where, For the The original sensitivity factor of the direction, 、 are the minimum and maximum values ​​in all directions, As an adjustable gain coefficient, the nonlinear mapping will amplify the difference between the high-sensitivity direction and the low-sensitivity direction, and finally form a set of position-sensitive factors that characterize the ripple energy distribution weights in each voltage direction, providing a highly discriminative input feature for the ripple-position mapping model in step S4.

[0057] Before building the ripple-position mapping model in step S4, offline training is required in the motor pre-calibration phase. The training method includes:

[0058] Full position interval data collection, collection methods include:

[0059] Control the motor rotor to mechanically rotate to N evenly distributed positions (e.g. N = 512, corresponding to a 0.7° resolution);

[0060] At each position θ g Repeat steps S1 to S3 to obtain the position sensitive factor vector F of the point g (including characteristics such as kurtosis coefficient and ripple energy);

[0061] Form the training data set {θ g ,F g}g=1,2,...,N;

[0062] Support vector regression (SVR) modeling is performed using the following methods:

[0063] The position angle θ is used as the regression target and the multidimensional factor vector F is used as the input feature;

[0064] The radial basis kernel function (RBF) is used to construct nonlinear mapping, including: ;

[0065] in, is the number of support vectors, 、 and is the training parameter, is a subscript, indicating the indivual.

[0066] At the same time, hyperparameters are optimized through grid search cross validation.

[0067] The RBF kernel can characterize factor fluctuations caused by magnetic saturation (such as the third harmonic of the position factor caused by the cogging torque). The ϵ-insensitive band ignores small measurement noise, and the support vector only retains key samples to meet the needs of real-time position estimation. The trained SVR model is burned into the controller ROM in the form of a coefficient matrix. During the online operation phase, the position sensitive factor input in step S4 is used to output the rotor position estimate θ^ in real time.

[0068] In the multi-directional detection voltage vector injection phase of step S1, a space vector pulse width modulation (SVPWM) sequence is used to generate precise voltage excitation. In specific implementation, the control system plans at least three non-collinear spatial voltage directions (for example, 0°, 60°, and 120° directions) in the α-β coordinate system. Each target voltage vector is decomposed into the action timing of the three-phase full-bridge circuit using the SVPWM algorithm. Taking the voltage injection in the 60° direction as an example:

[0069] Calculate the duration of the basic voltage vector corresponding to this direction (e.g., vector U1 accounts for 70% and U2 accounts for 30%).

[0070] Generate a PWM drive waveform for the corresponding power tube, where a dead time (e.g., 3μs) is forcibly inserted when switching between adjacent vectors.

[0071] During the dead time, the upper and lower bridge arm drive signals are turned off to completely block the current path and eliminate the risk of direct short circuit of the bridge arm.

[0072] The gradual increase in the amplitude of the voltage vector is achieved by linearly increasing the modulation depth. For example, 5% of the rated voltage is initially injected (to avoid inrush current) and then gradually increased to 20% of the rated voltage (to enhance the ripple signal-to-noise ratio). Each direction lasts for several PWM cycles (such as five cycles) to ensure a stable current response, and the dead time is used to achieve natural current decay when the direction is switched. This injection method ensures the spatial pointing accuracy of the voltage vector while taking into account the safety of the power circuit.

[0073] In step S1, the voltage vector During the injection process, the amplitude increasing strategy uses an exponential rising curve to achieve smooth excitation. Its mathematical expression includes: ;

[0074] Where, is the target peak voltage (e.g. 20% rated voltage), is the critical time constant, is the current time, is a constant; this design overcomes the current step problem caused by linear rise, that is, the initial slope of the exponential curve is large ( =0, d / d = / ) can quickly establish a magnetic field, and the slope gradually slows down in the later stage ( →∞, d / d →0) suppresses overshoot risk.

[0075] Time constant Only adaptive mechanisms are used, including:

[0076] During motor operation, the winding resistance R increases significantly with temperature rise (for example, at 150°C, R increases by 50% compared to the cold state). If the current rises too slowly in the hot state, the ripple feature extraction will be delayed.

[0077] The control system obtains the online identified winding resistance value in real time (shared data through the FOC current loop parameter setting module), press ∝1 / R dynamic adjustment, for example, when the cold resistance R25℃=1Ω, set =50ms, automatically reduced to when hot state R100℃=1.5Ω ≈33ms, which keeps the current transient response speed stable.

[0078] This design ensures that the current ripple energy accumulation rate is consistent under different working conditions, avoiding both cold-state overcurrent shock and hot-state feature extraction delay, laying the foundation for reliable acquisition in step S2.

[0079] In the current ripple feature extraction in step S2, the collected three-phase current transient response is first subjected to a Park transform, converting it from the stationary coordinate system (abc) to the rotor synchronous rotating coordinate system (dq). This transformation eliminates the DC bias of the fundamental current through angle offset compensation, so that the ripple alternating component that was originally submerged by the fundamental current is significantly separated in the q-axis current. For example, when the injection voltage direction coincides with the d-axis, the q-axis current exhibits a high-frequency ripple with an amplitude of approximately 0.5A, while the d-axis current ripple is almost zero.

[0080] For the separated ripple components, the spectral kurtosis is further calculated to quantify the frequency domain energy concentration characteristics, including:

[0081] The short-time Fourier transform (STFT) is used to analyze the ripple signal in the time-frequency two-dimensional plane, and the window function length is set to cover the main harmonic period (for example, a 100μs window length corresponds to a 10kHz analysis bandwidth);

[0082] Calculate the kurtosis coefficient at each frequency point fk , the calculation methods include: ;

[0083] Where, is the time spectrum, It represents the time dimension average. When the spectral kurtosis value is greater than 3 (such as SK=4.2 at 5kHz in the d-axis direction), it indicates that there is a pulsed energy accumulation caused by magnetic saturation in this frequency band. The spectral kurtosis in the q-axis direction is close to 2 (broadband white noise characteristics), reflecting a smooth energy distribution without salient pole effect.

[0084] This method overcomes the information loss problem of traditional bandpass filtering, and the extracted spectral kurtosis feature matrix will serve as the key input for constructing the position-sensitive factor in step S3.

[0085] By real-time monitoring of the confidence index of the position sensitive factor, online calibration of the ripple-position mapping model parameters is triggered when the confidence index is lower than the threshold.

[0086] The generated position sensitive factor vector is obtained in real time. The obtaining method includes:

[0087] The input source is the position sensitive factor vector F g = (m is the feature dimension), here Transpose the sign of the vector;

[0088] Perform time domain stability analysis. The analysis methods include:

[0089] Establish a sliding time window and calculate the coefficient of variation sequence of each characteristic component : ;

[0090] In the formula, is the sliding window width, is the time offset index ( =1,2,3,..., ), is the subscript of the eigenvalue, i.e. eigenvalues.

[0091] Take the weighted norm of the coefficient of variation of each dimension ,Right now , where is the feature weight matrix; th1 is the warning threshold (low risk critical value); th2 is the calibration threshold (high risk critical value).

[0092] Feature weight matrix Based on the feature sensitivity configuration (for example, the ripple energy feature weight is higher than the kurtosis coefficient), as shown in the following table:

[0093] Example 2: Figure 2 As shown, this embodiment further improves the design based on the first embodiment. The difference is that in actual operation, it is found that the detection voltage vector amplitude of the first embodiment is fixed and the direction is limited, resulting in distortion of the ripple characteristics under high temperature and demagnetization conditions, and insufficient spatial resolution of the position sensitivity factor. Based on this, the non-sensing FOC ripple characteristic rotor positioning method further includes:

[0094] The detection voltage amplitude is adjusted in real time according to the winding temperature and the q-axis inductance change rate, and an orthogonal detection direction is added on the basis of the initial three directions; when an abnormal temperature rise is detected or the inductance change rate exceeds the critical value, the basic voltage amplitude is dynamically increased according to the temperature change; an additional compensation voltage is superimposed based on the inductance change rate; and two sets of orthogonal directions are added on the basis of the initial at least three detection directions.

[0095] Exemplary:

[0096] A permanent magnet synchronous motor (PMSM) used to drive a compressor shuts down after running at full load for a long time. At this time:

[0097] The winding temperature is abnormally high: reaching 110°C (much higher than the 25°C during normal temperature calibration).

[0098] According to the preset temperature-voltage mapping table or function, the basic detection voltage amplitude is increased from 0.15V calibrated at room temperature to 0.25V (to compensate for the ripple attenuation and resistance increase caused by high temperature);

[0099] Based on the detected inductance change rate = -18%, the additional compensation voltage is calculated to be ≈ 0.0045V, and the final injection amplitude is 0.2545V.

[0100] A temperature attenuation function is used to suppress the ripple amplitude attenuation effect caused by high temperature; a demagnetization compensation matrix is ​​applied to eliminate the phase distortion caused by weakening magnetic flux; and the corrected position sensitivity factor is input into the ripple-position mapping model to ensure stable rotor angle solution accuracy under all operating conditions.

[0101] Example 3: Figure 3 As shown, based on the same inventive concept as the non-sensing FOC ripple characteristic rotor positioning method in the aforementioned embodiment, the present application provides a non-sensing FOC ripple characteristic rotor positioning system. The system and method embodiments in the present application are based on the same inventive concept. The system includes:

[0102] A gradient injection module injects a multi-directional detection voltage vector with increasing amplitude into the stator winding during the zero-speed startup phase of the motor. The multi-directional detection voltage vector includes at least three non-collinear spatial directions.

[0103] The ripple extraction module synchronously collects the three-phase current transient response under the action of each voltage vector and extracts the current ripple characteristic component that matches the rotor salient pole frequency through bandpass filtering;

[0104] The sensitive configuration module constructs a set of position sensitive factors corresponding to the voltage direction based on the spatial distribution anisotropy of the current ripple characteristic components. Each position sensitive factor in the set of position sensitive factors represents the distribution weight of the ripple energy under the excitation of the voltage direction.

[0105] The angle mapping module inputs the set of position-sensitive factors into a pre-trained ripple-position mapping model and outputs the initial rotor position angle. The ripple-position mapping model establishes a nonlinear relationship between ripple characteristics and position through the motor's magnetic saturation effect.

[0106] The closed-loop switching module switches to FOC closed-loop operation according to the initial rotor position angle.

[0107] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

[0108] The above is only a preferred specific implementation method of the embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes based on the technical solution and concept of the present application within the technical scope disclosed in the present application, and they should be covered by the scope of protection of the present application.

Claims

1. The non-sensing FOC ripple characteristic rotor positioning method is characterized by: Methods include: During the zero-speed starting phase of the motor, a multi-directional detection voltage vector with gradually increasing amplitude is injected into the stator winding, where the multi-directional detection voltage vector includes at least three non-collinear spatial directions; Synchronously collect the three-phase current transient responses under the action of each voltage vector, and extract the current ripple characteristic components that match the rotor salient pole frequency through bandpass filtering; Based on the spatial distribution anisotropy of the current ripple characteristic components, a set of position sensitive factors corresponding to the voltage direction is constructed. Each position sensitive factor in the set of position sensitive factors represents the distribution weight of the ripple energy under the excitation of the voltage direction. The position sensitive factor set is input into the pre-trained ripple-position mapping model to output the initial rotor position angle. The ripple-position mapping model establishes a nonlinear relationship between ripple characteristics and position through the motor's magnetic saturation effect. Switch to FOC closed-loop operation according to the initial rotor position angle.

2. The non-inductive FOC ripple characteristic rotor positioning method according to claim 1 is characterized in that: The passband center frequency of the bandpass filter dynamically tracks the fundamental frequency of the rotor salient pole harmonics, and its -3dB bandwidth is proportional to the difference in the quadrature and direct axis inductances of the motor.

3. The non-inductive FOC ripple characteristic rotor positioning method according to claim 2 is characterized in that: The construction methods of position sensitive factors include: The product of the kurtosis coefficient of the ripple energy under each voltage direction excitation and the root mean square value of the envelope is calculated, and the exponential function is used to normalize it to a weight value in the [0,1] interval.

4. The non-inductive FOC ripple characteristic rotor positioning method according to claim 3 is characterized in that: The training method of the ripple-position mapping model includes: In the motor pre-calibration stage, a set of position sensitive factors in the entire position range is obtained, and a support vector regression mechanism is used to construct a nonlinear mapping function between the position angle and the multidimensional factor vector.

5. The non-sensing FOC ripple characteristic rotor positioning method according to claim 1, characterized in that: The injection of multi-directional detection voltage vectors adopts space vector pulse width modulation sequence, and dead time is inserted between adjacent vectors to suppress the risk of bridge arm shoot-through.

6. The non-inductive FOC ripple characteristic rotor positioning method according to claim 5 is characterized in that: The amplitude of the voltage vector gradually increases using an exponential rising curve, and its time constant is adaptively adjusted according to the thermal winding resistance value of the motor.

7. The non-inductive FOC ripple characteristic rotor positioning method according to claim 1, characterized in that: The extraction methods of current ripple characteristic components include: The Park transform is performed on the three-phase current to separate the ripple alternating component in the rotating coordinate system, and the spectral kurtosis of its short-time Fourier transform is calculated.

8. The non-inductive FOC ripple characteristic rotor positioning method according to claim 4, characterized in that: During the operation of the motor, the confidence index of the position sensitive factor is monitored in real time. When the confidence index is lower than the threshold, the online calibration of the ripple-position mapping model parameters is triggered.

9. The non-inductive FOC ripple characteristic rotor positioning method according to claim 1, characterized in that: The method also includes: The detection voltage amplitude is adjusted in real time according to the winding temperature and the q-axis inductance change rate, and an orthogonal detection direction is added to the initial three directions; The position sensitivity factor distortion is corrected by the temperature attenuation function and the demagnetization compensation matrix, and then output to the ripple-position mapping model.

10. The sensorless FOC ripple characteristic rotor positioning system is characterized by: The system includes: a gradient injection module, wherein the gradient injection module injects a multi-directional detection voltage vector with increasing amplitude into the stator winding during the zero-speed startup phase of the motor, wherein the multi-directional detection voltage vector includes at least three non-collinear spatial directions; a ripple extraction module, which synchronously collects the three-phase current transient responses under the action of each voltage vector and extracts the current ripple characteristic component matching the rotor salient pole frequency through bandpass filtering; A sensitive configuration module, wherein the sensitive configuration module constructs a set of position sensitive factors corresponding to the voltage direction based on the spatial distribution anisotropy of the current ripple characteristic component, wherein each position sensitive factor in the set of position sensitive factors represents the distribution weight of the ripple energy under the excitation of the voltage direction; An angle mapping module, which inputs a set of position sensitive factors into a pre-trained ripple-position mapping model and outputs an initial rotor position angle. The ripple-position mapping model establishes a nonlinear relationship between ripple characteristics and position through the motor's magnetic saturation effect; A closed-loop switching module is configured to switch to FOC closed-loop operation according to the initial rotor position angle.

Images