Magnetic bearing gap on-line measuring device and compensation method

Abstract

The present application belongs to the technical field of magnetic suspension motor, and discloses a magnetic suspension bearing gap online measurement device and a compensation method. The method comprises the following steps: deploying a quantum and eddy current sensor array on the inner wall of a magnetic suspension bearing stator, and correspondingly arranging a surface acoustic wave sensor inside the stator. A scanning signal is injected into the eddy current array, real-time complex impedance is collected, and a differential signal with a baseline is calculated. Real-time characteristic fingerprints are obtained by analysis, and phase changes are determined by comparing a phase change fingerprint library. When the phase changes, a quantum sensor is used to calculate a reference deviation, generate a local space correction field, and obtain a correction amount of each eddy current sensor by interpolation. The surface acoustic wave sensor collects signals to analyze stress and temperature gradient, establishes a feedforward compensation model to calculate a gap compensation amount, and obtains a corrected gap value in combination with the correction amount. The present application realizes the real-time, accuracy and stability of gap data under dynamic operation, avoids rotor instability, equipment vibration intensification and bearing wear caused by inaccurate gap monitoring, and ensures long-term reliable operation of the magnetic suspension bearing.

Description

Technical Field

[0001] This invention relates to the field of magnetic levitation motor technology, and more specifically, to an online measurement device and compensation method for magnetic levitation bearing clearance. Background Technology

[0002] Magnetic levitation bearings are widely used in high-end equipment fields due to their advantages of non-contact operation, low friction, and high speed. Their stable operation is highly dependent on the real-time and accurate monitoring of the gap between the rotor and stator. However, in actual operation, the bearing material is easily affected by factors such as temperature fluctuations and load changes, which can cause microscopic phase changes. These phase changes can directly alter the electromagnetic properties of the material, leading to a sudden change in the measurement benchmark of traditional gap sensing systems.

[0003] Chinese patent application CN115507800A discloses a method, apparatus, terminal device, and storage medium for detecting the clearance of a magnetic levitation bearing. The method includes: controlling the rotor of the magnetic levitation bearing to roll on the inner ring of the magnetic levitation bearing; acquiring radial displacement information between the rotor and the inner ring when the rotor rolls on the inner ring; and determining the clearance uniformity information of the magnetic levitation bearing based on the radial displacement information.

[0004] While the above methods can meet the needs of most scenarios, research and practical application of these methods and existing technologies have revealed at least the following shortcomings:

[0005] The above method is an offline monitoring method, which cannot monitor the gap change in real time under the dynamic state of normal operation of the magnetic levitation bearing. In particular, it is difficult to detect and warn of sudden changes in the gap sensing reference caused by factors such as microscopic phase change of materials during operation, which may lead to increased equipment vibration, rotor instability or even bearing wear.

[0006] In view of this, the present invention proposes an online measurement device and compensation method for the clearance of magnetic levitation bearings to solve the above problems. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: an online measurement device and compensation method for the clearance of magnetic levitation bearings, comprising:

[0008] A quantum sensor array is deployed on the inner wall of the stator of the magnetic levitation bearing to measure the Rabi oscillation frequency and calculate the absolute gap value; planar eddy current coils are arranged alternately between the quantum sensor arrays to form an eddy current sensor array.

[0009] A scanning signal is injected into the eddy current sensor array to acquire real-time complex impedance signals. The difference signal between the real-time complex impedance signal and the corresponding frequency point in the baseline database is calculated to obtain the complex impedance differential signal.

[0010] The complex impedance differential signal is analyzed to obtain real-time feature fingerprints; the real-time feature fingerprints are compared with the phase transition fingerprint database to obtain the phase transition determination results.

[0011] When the phase transition determination result indicates that a phase transition has occurred, the absolute gap value of the quantum sensor at the phase transition location and the original gap measurement value of the eddy current sensor are collected. The deviation between the absolute gap value and the original gap measurement value is calculated to obtain the reference deviation. A local spatial correction field is generated with the phase transition location as the center. The correction amount corresponding to each eddy current sensor is calculated through an interpolation algorithm.

[0012] By collecting the propagation time and resonant frequency of sound waves using a surface acoustic wave sensor, the stress and temperature gradients are analyzed and obtained. A feedforward compensation model is constructed based on a semi-empirical hybrid model. The gap compensation amount is calculated based on the stress and temperature gradients, and the correction gap value is calculated by combining the correction amount.

[0013] Furthermore, methods for obtaining the correction gap value include:

[0014] Acquire the acoustic wave propagation time and resonant frequency of each surface acoustic wave sensor, and perform Kalman filtering and moving average filtering to obtain the filtered acoustic wave propagation time and the filtered resonant frequency.

[0015] The stress is calculated based on the propagation time of the sound wave under stress-free conditions and the propagation time of the sound wave after filtering; the single-point temperature is calculated based on the preset temperature, the resonant frequency under stress-free conditions and the resonant frequency after filtering; and the temperature gradient is calculated based on the temperature difference and spacing between adjacent sensors.

[0016] If the stress is greater than the stress threshold or the temperature gradient is greater than the gradient threshold, it is determined that a phase transition has occurred and proceeds to the next step; otherwise, it is determined that no phase transition has occurred and the output gap compensation amount is 0.

[0017] Based on stress and temperature gradient, a semi-empirical hybrid model is used to build a feedforward compensation model for the gap compensation amount. The feedforward compensation model is solved using the least squares method to obtain the model coefficients. The model coefficients are then substituted into the feedforward compensation model, and the gap compensation amount is calculated by combining real-time stress and real-time temperature gradient.

[0018] Calculate the sum of the correction amount and the gap compensation amount to obtain the correction gap value.

[0019] Furthermore, methods for generating local spatial correction fields include:

[0020] A circular region to be calibrated is defined with the location of the phase transition as the center and the radius of the phase transition influence as the radius. A calibration field is generated only for the eddy current sensor within the circular region to be calibrated.

[0021] Obtain the physical coordinates of all eddy current sensors within the circular area to be calibrated, forming a set of sensors to be calibrated; extract the quantum sensor data adjacent to the phase transition location from the set of sensors to be calibrated.

[0022] An exponential decay model was constructed, and the theoretical reference deviation at coordinates (x, y) was calculated based on the reference deviation, straight-line distance, and phase transition influence radius at the location of the phase transition.

[0023] The exponential decay model is calibrated to obtain the calibrated exponential decay model.

[0024] Centered on the location where the phase transition occurs, a square grid covering the circular area to be corrected is generated according to the grid side length. For each grid node, a preliminary correction value is calculated based on the calibrated exponential decay model. Based on the anchor point weighted correction method, the preliminary correction value of each grid node is adjusted to obtain the optimized grid node correction value. The optimized grid node correction value is smoothed by Gaussian filtering to obtain the smoothed correction value. The local spatial correction field is obtained.

[0025] Furthermore, methods for obtaining the radius of influence of the phase transition include:

[0026] The location of the phase transition is extracted, and the radius of influence of the phase transition is calculated using a semi-empirical formula, taking into account the area of ​​the phase transition and the magnetic permeability and electrical conductivity of the bearing material.

[0027] The quantum sensor data includes the quantum sensor coordinates, the measured absolute gap value, and the original gap value and reference deviation of the eddy current sensor corresponding to the phase transition location.

[0028] Methods for obtaining a calibrated exponential decay model include:

[0029] The reference deviation corresponding to the quantum sensor is used as the interpolation anchor point. The exponential decay model is calibrated using the interpolation anchor point: the interpolation anchor point is substituted into the exponential decay model, the theoretical reference deviation of the interpolation anchor point is calculated, and then the residual between the theoretical reference deviation and the reference deviation of the interpolation anchor point is calculated; then the mean of the residuals is calculated. If the mean of the residuals is greater than the residual threshold, the decay coefficient is adjusted according to the residual step size; if the reference deviation of the interpolation anchor point is less than the theoretical reference deviation of the interpolation anchor point, the decay coefficient is increased according to the residual step size; if the reference deviation of the interpolation anchor point is greater than the theoretical reference deviation of the interpolation anchor point, the decay coefficient is decreased according to the residual step size; until the mean of the residuals is not greater than the residual threshold, the decay function at this time is obtained and substituted into the exponential decay model to obtain the calibrated exponential decay model.

[0030] Furthermore, methods for obtaining optimized mesh node correction values ​​include:

[0031] Calculate the grid distance from each grid node to each interpolation anchor point; then calculate the grid weight of each grid node to each interpolation anchor point based on the grid distance; calculate the anchor point correction amount based on the mean of the grid weights and the residual; finally, calculate the sum of the anchor point correction amount and the initial correction value to obtain the optimized grid node correction value.

[0032] Furthermore, the method for obtaining the correction amount corresponding to each eddy current sensor includes:

[0033] The Gaussian radial basis function is selected as the interpolation kernel function to construct the interpolation function. Based on the interpolation condition that the interpolation function value at the interpolation anchor point is equal to the actual reference deviation, a system of linear equations is established. The Cholesky decomposition method is used to solve the system of linear equations to obtain the interpolation coefficients.

[0034] Iterate through each eddy current sensor in the set of sensors to be calibrated Obtain the corresponding coordinates, substitute them into the interpolation function, and calculate the eddy current sensor value. The initial calibration amount; if the eddy current sensor If the absolute value of the initial correction is greater than the correction threshold, then the eddy current sensor is selected. The initial correction amount is the correction amount threshold;

[0035] A Q×Q neighborhood mean filter is used to smooth the initial calibration value of each eddy current sensor in the sensor set to be calibrated. Taking the current eddy current sensor as the center, the mean of the initial calibration values ​​of the surrounding Q×Q-1 neighboring eddy current sensors is calculated. The eddy current sensor is obtained by weighting the initial calibration value of the current eddy current sensor and the mean of the initial calibration values. The correction amount.

[0036] Furthermore, methods for obtaining real-time feature fingerprints include:

[0037] Calculate the amplitude of the complex impedance differential signal across the entire frequency band to obtain an amplitude sequence. Detect whether the rate of change of amplitude at a frequency point in the amplitude sequence exceeds the amplitude jump threshold. If so, record the corresponding frequency point and amplitude jump value; otherwise, record the corresponding frequency point and amplitude jump value as 0 to obtain amplitude abrupt change characteristics.

[0038] Calculate the phase of the complex impedance differential signal across the entire frequency band to obtain a phase sequence. Detect whether the phase change rate at a frequency point in the phase sequence exceeds the phase jump threshold. If it does, record the corresponding frequency point and phase jump value; otherwise, record the corresponding frequency point and phase jump value as 0 to obtain the phase jump characteristics.

[0039] Local maximum values ​​in the amplitude sequence are extracted as characteristic resonant frequencies. The offset of the characteristic resonant frequency relative to the corresponding resonant frequency in the baseline impedance spectrum is calculated to obtain the characteristic frequency offset feature.

[0040] According to the pre-division rules, the entire frequency band is divided into three sub-bands: low, medium, and high. The integral energy of the amplitude of the complex impedance differential signal in each sub-band is calculated, and the ratio of the integral energy of each sub-band to the total energy is calculated to obtain the frequency band energy variation characteristics.

[0041] The amplitude abrupt change feature, phase abrupt change feature, characteristic frequency offset feature, and frequency band energy change feature are concatenated to obtain a feature vector. Each parameter in the feature vector is normalized to obtain a real-time feature fingerprint.

[0042] Furthermore, methods for obtaining phase transition determination results include:

[0043] The difference between the real-time feature fingerprint and the phase transition fingerprint in the phase transition fingerprint database is calculated using weighted Euclidean distance to obtain the similarity distance. The similarity distance is then converted into a similarity score by combining the maximum distance.

[0044] Traverse all phase transition fingerprints in the phase transition fingerprint database, calculate the similarity between the real-time feature fingerprint and each phase transition fingerprint, and obtain the maximum similarity as the matching degree;

[0045] If the matching degree exceeds the matching threshold, it is determined that a microscopic phase transition has occurred in the sensitive area of ​​the current eddy current sensor, and a phase transition determination signal is output. Simultaneously, the physical coordinates corresponding to the current eddy current sensor are recorded as the location where the phase transition occurred; otherwise, it is determined that no phase transition has occurred, and a phase transition determination signal is output. .

[0046] Furthermore, methods for obtaining complex impedance differential signals include:

[0047] For each frequency point of data from each eddy current sensor, outliers are removed by applying the three-times-standard-deviation principle to obtain valid data points. The arithmetic mean of the valid data points is then calculated to obtain the real-time average of the complex impedance.

[0048] The mean values ​​of the real-time resistance component and the mean values ​​of the real-time reactance component are filtered to obtain the filtered real-time complex impedance.

[0049] For each eddy current sensor traverse all its frequency points Retrieve the corresponding eddy current sensor from the baseline database at the same frequency. The baseline complex impedance;

[0050] For each eddy current sensor Each frequency point The differential components of the filtered real-time complex impedance and the baseline complex impedance are calculated respectively to obtain the complex impedance differential signal.

[0051] Furthermore, methods for obtaining the absolute gap value include:

[0052] Turn off the bias magnetic field of the Helmholtz coil; apply a frequency-sweeping microwave signal of a preset frequency band to the microwave antenna in a preset step size, and at the same time collect the fluorescence intensity signal of the quantum sensor to obtain the resonance spectrum of fluorescence intensity with respect to microwave frequency;

[0053] Identify the resonance peaks in the resonance spectrum corresponding to the transitions between the ground state ms=0 and ms=±1 energy levels of the quantum sensor, obtain the corresponding resonance frequencies, and calculate the background magnetic field according to the quantum sensor energy level splitting formula.

[0054] The background magnetic field is fed back to the Helmholtz coil control system, and a reverse compensation magnetic field of the same magnitude is applied.

[0055] A circular metal sheet of the same material as the bearing rotor and with the same diameter as the rotor is used, and the distance between the circular metal sheet and the surface of the stator sensor is controlled to be a standard gap.

[0056] For each standard gap, a microwave signal source is activated to output a microwave field with a fixed power. The frequency of the microwave field is adjusted to make the quantum sensor generate Rabi oscillations and collect the decay signal of fluorescence intensity over time.

[0057] The attenuation signal is sinusoidally fitted and the Levenberg-Marquardt algorithm is used to iteratively solve the problem to obtain the Rabi oscillation frequency. The correspondence between the Rabi oscillation frequency and the standard gap is established, and the calibration formula is obtained through fitting. The initial absolute gap value is calculated based on the real-time Rabi oscillation frequency and the calibration formula. The initial absolute gap values ​​of all quantum sensors are weighted and fused to obtain the absolute gap value.

[0058] An online measurement device for magnetic levitation bearing clearance, comprising the following steps:

[0059] Sensing deployment module: A quantum sensor array is deployed on the inner wall of the magnetic levitation bearing stator to measure the Rabi oscillation frequency and calculate the absolute gap value; planar eddy current coils are arranged alternately between the quantum sensor arrays to form an eddy current sensor array; surface acoustic wave sensors are deployed inside the magnetic levitation bearing stator at positions corresponding to the quantum sensor array and the eddy current sensor array.

[0060] The acquisition and positioning module injects a scanning signal into the eddy current sensor array, acquires real-time complex impedance signals, calculates the difference signal between the real-time complex impedance signal and the corresponding frequency point in the baseline database, and obtains the complex impedance differential signal.

[0061] Phase transition determination module: Analyzes the complex impedance differential signal to obtain real-time feature fingerprints; compares the real-time feature fingerprints with the phase transition fingerprint database to obtain the phase transition determination result;

[0062] Online calibration module: When the phase transition determination result indicates that a phase transition has occurred, the absolute gap value of the quantum sensor at the phase transition location and the original gap measurement value of the eddy current sensor are collected. The deviation between the absolute gap value and the original gap measurement value is calculated to obtain the reference deviation. A local spatial correction field is generated with the phase transition location as the center. The correction amount corresponding to each eddy current sensor is calculated through an interpolation algorithm.

[0063] Feedforward compensation module: Acquires sound wave propagation time and resonant frequency through surface acoustic wave sensor, analyzes and obtains stress and temperature gradient, constructs feedforward compensation model based on semi-empirical hybrid model, calculates gap compensation amount according to stress and temperature gradient, and calculates correction gap value by combining correction amount.

[0064] The technical effects and advantages of the online measurement device and compensation method for magnetic levitation bearing clearance of the present invention are as follows:

[0065] This invention constructs a comprehensive gap monitoring system by collaboratively deploying quantum sensor arrays, eddy current sensor arrays, and surface acoustic wave (SAW) sensor arrays on the inner wall of the stator of a magnetic levitation bearing. This system integrates absolute reference monitoring, dynamic signal acquisition, phase transition feature identification, deviation correction, and feedforward prevention. It effectively addresses the shortcomings of existing technologies, such as the difficulty in accurately monitoring gaps during dynamic operation and the inability to promptly identify sudden changes in the sensing reference caused by microscopic phase transitions in materials. The quantum sensors provide an absolute gap reference unaffected by phase transitions. Combined with the complex impedance differential signal and real-time feature fingerprint from the eddy current sensors, phase transitions can be accurately determined and located. A local spatial correction field provides differentiated correction values ​​for eddy current sensors at different locations, correcting deviations caused by existing reference changes. The SAW sensors monitor stress and temperature gradients, and a semi-empirical feedforward compensation model calculates compensation values ​​in advance to prevent potential phase transition risks. Ultimately, this achieves real-time, accurate, and stable gap data under dynamic operation, avoiding rotor instability, increased equipment vibration, and bearing wear caused by inaccurate gap monitoring, thus ensuring the long-term reliable operation of the magnetic levitation bearing. Attached Figure Description

[0066] Figure 1 This is a schematic diagram of the online measurement and compensation method for magnetic levitation bearing clearance according to the present invention;

[0067] Figure 2 This is a schematic diagram of the method for calculating the correction amount corresponding to each eddy current sensor using an interpolation algorithm according to the present invention.

[0068] Figure 3 This is a schematic flowchart of the method for obtaining the correction gap value according to the present invention;

[0069] Figure 4 This is a schematic diagram of the structure of the online measurement device for the clearance of magnetic levitation bearings according to the present invention. Detailed Implementation

[0070] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] Example 1:

[0072] Please see Figure 1 As shown, this embodiment provides a method for online measurement and compensation of magnetic levitation bearing clearance, including:

[0073] A quantum sensor array is deployed on the inner wall of the stator of the magnetic levitation bearing to measure the Rabi oscillation frequency and calculate the absolute gap value; planar eddy current coils are arranged alternately between the quantum sensor arrays to form an eddy current sensor array.

[0074] Methods for obtaining absolute gap values ​​include:

[0075] Turn off the bias magnetic field of the Helmholtz coil; apply a frequency sweep microwave signal of a preset band to the microwave antenna in a preset step size, such as applying a frequency sweep microwave signal of 2.7GHz-3.0GHz in a step size of 1MHz, and at the same time collect the fluorescence intensity signal of the quantum sensor of the NV color center to obtain the resonance spectrum of fluorescence intensity with respect to microwave frequency.

[0076] Identify the resonance peaks in the resonance spectrum corresponding to the transitions between the ground state energy levels ms=0 and ms=±1 of the NV color center, obtain the corresponding resonance frequencies, and calculate the background magnetic field using the NV color center energy level splitting formula; such as the background magnetic field. ,in, It is the resonant frequency; The electron gyromagnetic ratio is approximately 28.024 GHz / T;

[0077] The background magnetic field is fed back to the Helmholtz coil control system, and a reverse compensation magnetic field of the same magnitude is applied to counteract the influence of the background magnetic field, ensuring that the magnetic field of the sensor in subsequent measurements is contributed only by the bias magnetic field and the gap magnetic field.

[0078] A circular metal sheet of the same material as the bearing rotor and with the same diameter as the rotor is used, and the distance between the circular metal sheet and the surface of the stator sensor is controlled to be a standard gap.

[0079] For each standard gap, a microwave signal source is activated to output a microwave field with a fixed power. The frequency of the microwave field is adjusted to induce Rabi oscillations in the NV color center, and the decay signal of fluorescence intensity over time is collected.

[0080] The attenuation signal is sinusoidally fitted and the Levenberg-Marquardt algorithm is used to iteratively solve the problem to obtain the Rabi oscillation frequency. The correspondence between the Rabi oscillation frequency and the standard gap is established, and the calibration formula is obtained through fitting. The initial absolute gap value is calculated based on the real-time Rabi oscillation frequency and the calibration formula. The initial absolute gap values ​​of all quantum sensors are weighted and fused to obtain the absolute gap value.

[0081] The methods for iteratively solving the problem using the Levenberg-Marquardt algorithm include:

[0082] Construct a sinusoidal fitting formula for the attenuated signal, such as... Time decay signal ,in, The amplitude; This is the Rabi oscillation frequency; For phase; The decay time constant; For mathematical constants; integrate the parameters to be solved into a parameter vector. , It is the transpose of the vector;

[0083] Based on physical characteristics and prior knowledge, reasonable initial values ​​are given, such as taking the maximum value of the measured attenuated signal as the initial amplitude; performing a Fast Fourier Transform on the measured attenuated signal and taking the main peak frequency of the spectrum as the initial oscillation frequency; taking the phase corresponding to the measured attenuated signal at t=0 as the initial phase. , The measured attenuation signal at time t=0; The initial amplitude is set; the decay time constant is set to an empirical value based on the material properties of the NV color center.

[0084] With the first Taking the next iteration as an example:

[0085] Calculate the difference between the measured attenuation signal and the fitted attenuation signal to obtain the first... The residual vector of the nth iteration is used to calculate the nth iteration. The current sum of squared residuals in the next iteration;

[0086] Calculate the first The Jacobian matrix is ​​obtained by taking the partial derivatives of the residual vector with respect to each parameter to be solved in each iteration.

[0087] Construct and solve the regularized least squares equation with damping terms to obtain the parameter update step size; such as ,in, For the Hessian matrix approximation of the Gauss-Newton method, For the first Jacobian matrix of the next iteration It is the transpose of the vector; For damping term, For the first The damping factor for the next iteration can be obtained through a natural heuristic optimization algorithm. To construct a diagonal matrix by taking the diagonal elements of the matrix; For the first The parameter update step size vector for the next iteration; For gradient terms, For the first The residual vector of the next iteration;

[0088] Calculate the updated candidate parameters based on the parameter update step size, and calculate the corresponding residual sum of squares;

[0089] If the sum of squared residuals is less than the current sum of squared residuals, then accept the update and decrease the damping factor; otherwise, reject the update and increase the damping factor.

[0090] After each iteration, check if the convergence condition is met. If it is, terminate the iteration and output the optimal parameters; otherwise, continue iterating. The convergence condition is determined to be met when any of the following conditions are met: the change in the sum of squared residuals is lower than the first threshold; the norm of the parameter update is lower than the second threshold; or the norm of the residuals is lower than the third threshold.

[0091] If the convergence condition is not met even after the maximum number of iterations is reached, the iteration is terminated and the current parameters are output. At the same time, a message is displayed indicating that the maximum number of iterations has been reached and the initial parameters or hyperparameter settings need to be checked.

[0092] Verify the accuracy of the output parameters: Calculate the root mean square error of the fitting result. If the root mean square error is not higher than the accuracy threshold, the verification is passed and the fitting is deemed qualified. Otherwise, adjust the initial parameters and restart the iteration.

[0093] A linear or polynomial model is used to fit the relationship between the Rabi oscillation frequency and the standard gap. The least squares method is then used to fit the relationship parameters, and the calibration formula is obtained by substituting the relationship parameters into the relationship.

[0094] The above steps involve shutting off the bias magnetic field of the Helmholtz coil, applying a swept-frequency microwave signal to acquire the fluorescence resonance spectrum and calculate the background magnetic field, and then applying a reverse compensation magnetic field to cancel interference, ensuring that subsequent measurements only reflect the bias magnetic field and the gap magnetic field, thus laying a precise magnetic field foundation for gap measurement. Next, a standard gap is constructed using a metal sheet with the same material and diameter as the rotor. The Rabi oscillation of the NV color center is excited by a microwave field, and after acquiring the fluorescence decay signal, the Levenberg-Marquardt algorithm is used to iteratively solve the sinusoidal fitting parameters. Then, the least squares method is used to fit the linear or polynomial relationship between the Rabi oscillation frequency and the standard gap to obtain the calibration formula. Finally, the initial absolute gap value is calculated based on the real-time Rabi oscillation frequency and the calibration formula, and weighted fusion is performed to obtain a reliable absolute gap value. In the above process, the quantum sensor provides an absolute gap reference that is not affected by the microscopic phase transition of the material. The precise Rabi oscillation frequency extraction and calibration formula ensure the real-time performance and accuracy of gap monitoring under dynamic operation, thereby enabling timely detection of abrupt changes in the gap sensing reference caused by the microscopic phase transition of the material, and avoiding consequences such as increased equipment vibration, rotor instability, and bearing wear caused by the abrupt change in the reference.

[0095] A scanning signal is injected into the eddy current sensor array to acquire real-time complex impedance signals. The difference signal between the real-time complex impedance signal and the corresponding frequency point in the baseline database is calculated to obtain the complex impedance differential signal.

[0096] Methods for obtaining complex impedance differential signals include:

[0097] For each frequency point of data from each eddy current sensor, outliers are removed using a three-standard-deviation principle to obtain valid data points. The arithmetic mean of these valid data points is then calculated to obtain the real-time complex impedance mean. ,in, Eddy current sensor At frequency The real-time average of the resistance components; Eddy current sensor At frequency The average real-time reactance component, Eddy current sensor At frequency The real-time average of the complex impedance;

[0098] The mean values ​​of the real-time resistance component and the mean values ​​of the real-time reactance component are filtered to obtain the filtered real-time complex impedance.

[0099] For each eddy current sensor traverse all its frequency points Retrieve the corresponding eddy current sensor from the baseline database at the same frequency. The baseline complex impedance; such as the baseline complex impedance ,in, Eddy current sensor At frequency The baseline resistance component; Eddy current sensor At frequency The baseline reactance component, Eddy current sensor At frequency The baseline complex impedance;

[0100] For each eddy current sensor Each frequency point The differential components between the filtered real-time complex impedance and the baseline complex impedance are calculated to obtain the complex impedance differential signal. (The complex impedance differential signal is then described.) ,in, Eddy current sensor At frequency The resistance differential component is the difference between the resistance component of the filtered real-time complex impedance and the baseline complex impedance. Eddy current sensor At frequency The reactance differential component is the difference between the reactance component of the filtered real-time complex impedance and the baseline complex impedance. Eddy current sensor At frequency The complex impedance differential signal.

[0101] The above steps, for each frequency point of each eddy current sensor, employ a three-standard-deviation principle to eliminate outliers and calculate the arithmetic mean of valid data points to obtain the real-time complex impedance mean. Then, the resistance and reactance components are filtered separately to ensure the reliability and stability of the real-time complex impedance signal under dynamic operation, laying a high-quality data foundation for subsequent differential calculations. Subsequently, according to the principle of one-to-one correspondence between sensors and frequency points, baseline complex impedances under the same conditions are retrieved from the baseline database. By calculating the resistance and reactance differential components of the filtered real-time complex impedance and the baseline complex impedance, a complex impedance differential signal is obtained. The complex impedance differential signal can accurately characterize the difference between the real-time gap change and the baseline state. When a microscopic phase change in the material causes a sudden change in the gap sensing reference, it directly alters the electromagnetic coupling characteristics between the bearing and the sensor, which is reflected in the changes in the resistance and reactance components of the complex impedance. This change can be captured in real time through the complex impedance differential signal, breaking the limitation of existing technologies being insensitive to reference changes in dynamic monitoring. This provides crucial raw sensing data support for timely early warning and prevention of consequences such as increased equipment vibration, rotor instability, and bearing wear.

[0102] The complex impedance differential signal is analyzed to obtain real-time feature fingerprints; the real-time feature fingerprints are compared with the phase transition fingerprint database to obtain the phase transition determination results.

[0103] Methods for obtaining real-time feature fingerprints include:

[0104] Calculate the amplitude of the complex impedance differential signal across the entire frequency band to obtain an amplitude sequence. Check if the rate of change of amplitude at each frequency point in the amplitude sequence exceeds the amplitude jump threshold. If so, record the corresponding frequency point and amplitude jump value; otherwise, record the corresponding frequency point and amplitude jump value as 0. (The last sentence appears to be incomplete and possibly refers to a different calculation method.) Amplitude change rate ,in, Eddy current sensor At frequency The amplitude of the complex impedance differential signal; amplitude jump value ;

[0105] Calculate the phase of the complex impedance differential signal across the entire frequency band to obtain a phase sequence. Check if the phase change rate at each frequency point in the phase sequence exceeds the phase jump threshold. If it does, record the corresponding frequency point and phase jump value; otherwise, record the corresponding frequency point and phase jump value as 0. (The last sentence appears to be incomplete and possibly refers to the phase of the complex impedance differential signal.) Phase change rate ,in, Eddy current sensor At frequency The phase and amplitude jump values ​​of the complex impedance differential signal. ;

[0106] Local maxima in the amplitude sequence are extracted as characteristic resonant frequencies. The offset of these characteristic resonant frequencies relative to the corresponding resonant frequencies in the baseline impedance spectrum is calculated to obtain the characteristic frequency offset feature. ,in, The characteristic resonant frequency; This corresponds to the resonant frequency in the baseline impedance spectrum;

[0107] The entire frequency band is divided into three sub-bands—low, mid, and high—according to pre-defined rules. For example, [0.1MHz, 5MHz] is divided into the low-frequency band, (5MHz, 20MHz) into the mid-frequency band, and (20MHz, 50MHz) into the high-frequency band. The integral energy of the complex impedance differential signal amplitude within each sub-band is calculated, and the ratio of the integral energy of each sub-band to the total energy is calculated to obtain the frequency band energy variation characteristics. ,in, In frequency point The amplitude of the complex impedance differential signal; For the first The lower limit frequency of each sub-band; For the first The upper limit frequency of each sub-band The values ​​are 1, 2, and 3. This corresponds to the low-frequency band. The time corresponds to the mid-frequency band. The time corresponds to the high frequency band; Frequency integral variable; frequency band energy proportion ,in, It is the sum of the integrated energies of the low, medium, and high frequency bands, i.e., the total energy;

[0108] The amplitude abrupt change feature, phase abrupt change feature, characteristic frequency offset feature, and frequency band energy change feature are concatenated to obtain a feature vector. Each parameter in the feature vector is normalized to obtain a real-time feature fingerprint.

[0109] The above steps, by calculating the amplitude and phase jumps of the complex impedance differential signal, can accurately capture the instantaneous signal of electromagnetic characteristic abrupt changes caused by microscopic phase transitions in materials, directly reflecting the abnormal fluctuations of the gap sensing reference. Extracting characteristic resonant frequencies and their offsets relative to the baseline can characterize the overall trend of impedance spectrum changes, corresponding to the systematic changes in the electromagnetic properties of materials caused by phase transitions. Dividing the entire frequency band into low, medium, and high sub-bands and calculating the integral energy ratio of each band can quantify the differences in energy distribution in different frequency ranges, reflecting the influence of phase transitions at different depths on electromagnetic coupling. Finally, the real-time feature fingerprint obtained by splicing and normalizing these features fully integrates abrupt, trend, and distributional characteristics, and can map the abrupt changes in the gap sensing reference caused by microscopic phase transitions in materials in real time during dynamic operation. This overcomes the limitations of existing technologies that are insensitive to reference changes and have a single monitoring dimension, providing accurate feature basis for timely early warning of risks such as increased equipment vibration, rotor instability, and bearing wear.

[0110] Methods for obtaining phase transition determination results include:

[0111] The difference between the real-time feature fingerprint and the phase transition fingerprint in the phase transition fingerprint database is calculated using weighted Euclidean distance to obtain the similarity distance. The similarity distance is then converted into a similarity score by combining the maximum distance. The maximum distance is the maximum distance between the feature vector and the corresponding phase transition fingerprint when there is no phase transition, which is obtained through statistical analysis of experimental data.

[0112] Traverse all phase transition fingerprints in the phase transition fingerprint database, calculate the similarity between the real-time feature fingerprint and each phase transition fingerprint, and obtain the maximum similarity as the matching degree;

[0113] If the matching degree exceeds the matching threshold, it is determined that a microscopic phase transition has occurred in the sensitive area of ​​the current eddy current sensor, and a phase transition determination signal is output. Simultaneously, the physical coordinates corresponding to the current eddy current sensor are recorded as the location where the phase transition occurred; otherwise, it is determined that no phase transition has occurred, and a phase transition determination signal is output. .

[0114] When the phase transition determination result indicates that a phase transition has occurred, the absolute gap value of the quantum sensor at the location where the phase transition occurred, as well as the original gap measurement value of the eddy current sensor, are collected. The deviation between the absolute gap value and the original gap measurement value is calculated to obtain the reference deviation.

[0115] The above steps rely on the absolute measurement characteristics of quantum sensors to directly quantify the reference mutation of eddy current sensors into calculable reference deviations. This not only clarifies the specific magnitude of the reference mutation but also provides a crucial quantitative basis for subsequent correction of eddy current sensor measurements and restoration of accurate gap monitoring. Without this reference deviation, it would be impossible to specifically correct the eddy current measurement errors caused by the reference mutation, leading to problems such as increased equipment vibration and rotor instability caused by erroneous gap data. Therefore, this step is the core link connecting phase transition determination and subsequent precise compensation, effectively filling the gap in the quantification of reference mutations in existing technologies and providing necessary deviation data support to avoid consequences such as bearing wear.

[0116] A local spatial correction field is generated centered on the location where the phase transition occurs; the correction amount corresponding to each eddy current sensor is calculated by interpolation algorithm; if it is determined that no phase transition has occurred, the real-time complex impedance signal is re-acquired for analysis.

[0117] Methods for generating local spatial correction fields include:

[0118] The location of the phase transition is extracted, and the phase transition area, along with the magnetic permeability and conductivity of the bearing material, are used to calculate the phase transition influence radius using a semi-empirical formula. The phase transition area can be obtained through experimental fitting, such as by obtaining a relationship between the phase transition area ratio and the target electromagnetic characteristics, including amplitude jumps, characteristic frequency shifts, and mid-frequency energy proportions. This relationship is then used to calculate the influence radius in conjunction with real-time target electromagnetic characteristics. If the phase transition influence radius is less than half the distance between adjacent quantum sensors, then half the distance between adjacent quantum sensors is taken as the phase transition influence radius. ,in, The correction coefficient is obtained through preliminary experimental calibration, such as conducting multiple phase transition simulation experiments for different bearing materials, measuring the actual influence range, and then fitting the result. The area representing the phase transition; The magnetic permeability of the bearing material; The electrical conductivity of the bearing material;

[0119] A circular region to be calibrated is defined with the location of the phase transition as the center and the radius of the phase transition influence as the radius. A calibration field is generated only for the eddy current sensor within the circular region to be calibrated.

[0120] Obtain the physical coordinates of all eddy current sensors within the circular area to be calibrated, forming a set of sensors to be calibrated; extract the quantum sensor data adjacent to the phase transition location from the set of sensors to be calibrated; including the quantum sensor coordinates, the measured absolute gap value, and the original gap value and reference deviation of the eddy current sensor corresponding to the phase transition location.

[0121] An exponential decay model is constructed, and the theoretical baseline deviation at coordinates (x, y) is calculated based on the baseline deviation, straight-line distance, and phase transition influence radius at the location of the phase transition. For example, the theoretical baseline deviation at coordinates (x, y)... ,in, This represents the reference deviation of the location where the phase transition occurs; The attenuation coefficient is determined by the material properties and can be calibrated experimentally, such as the attenuation coefficient of bearing steel. Attenuation coefficient of ceramic materials The larger the attenuation coefficient, the faster the deviation decreases with distance; The linear distance from the coordinates (x, y) to the location where the phase transition occurs; It is a mathematical constant;

[0122] The reference deviation corresponding to the quantum sensor is used as the interpolation anchor point. The exponential decay model is calibrated using the interpolation anchor point: the interpolation anchor point is substituted into the exponential decay model, the theoretical reference deviation of the interpolation anchor point is calculated, and then the residual between the theoretical reference deviation and the reference deviation of the interpolation anchor point is calculated; then the mean of the residual is calculated. If the mean of the residual is greater than the residual threshold, the decay coefficient is adjusted according to the residual step size; if the reference deviation of the interpolation anchor point is less than the theoretical reference deviation of the interpolation anchor point, the decay coefficient is increased according to the residual step size; if the reference deviation of the interpolation anchor point is greater than the theoretical reference deviation of the interpolation anchor point, the decay coefficient is decreased according to the residual step size; until the mean of the residual is not greater than the residual threshold, the decay function at this time is obtained and substituted into the exponential decay model to obtain the calibrated exponential decay model.

[0123] A square grid covering the circular area to be corrected is generated with the phase transition location as the center and the grid side length as the grid side length.

[0124] For each grid node, a preliminary correction value is calculated based on the calibrated exponential decay model;

[0125] Based on the anchor point weighted correction method, the initial correction value of each grid node is adjusted to obtain the optimized grid node correction value; the optimized grid node correction value is smoothed by Gaussian filtering to obtain the smoothed correction value; and then the local spatial correction field is obtained.

[0126] Methods for obtaining optimized mesh node correction values ​​include:

[0127] Calculate the grid distance from each grid node to each interpolation anchor point; then calculate the grid weight of each grid node to each interpolation anchor point based on the grid distance; calculate the anchor point correction amount based on the mean of the grid weights and the residual; finally, calculate the sum of the anchor point correction amount and the initial correction value to obtain the optimized grid node correction value. For example, the grid node... to interpolation anchor point Grid weights ,in, For grid nodes to interpolation anchor point Grid distance; To obtain a very small number whose denominator is not zero; anchor point correction. ,in, This represents the number of interpolation anchor points; The average of the grid weights; interpolation anchor point The reference deviation; interpolation anchor point The theoretical benchmark deviation.

[0128] The above steps first combine the phase transition area, the magnetic permeability and electrical conductivity of the bearing material, and calculate the phase transition influence radius using a semi-empirical formula and modified boundary conditions to accurately delineate the area to be corrected, avoiding excessive correction range or omissions. Then, using the absolute gap value of the quantum sensor and the corresponding eddy current sensor reference deviation as absolute anchor points, an exponential decay model is built, and the attenuation coefficient is adjusted iteratively through the anchor point residual to complete the calibration, ensuring that the model conforms to the spatial decay physical law of the reference deviation under the influence of material phase transition. Subsequently, the area to be corrected is covered by mesh division, and a smooth correction field is obtained by combining anchor point weighted correction and Gaussian filtering. This provides a precise correction amount for all eddy current sensors in the area that matches the degree of phase transition influence. This fills the gap in existing technology where, after detecting a sudden change in the phase transition reference, it is impossible to perform differentiated and physically compliant correction for the degree of deviation of eddy current sensors at different positions under dynamic operating conditions. This allows eddy current sensors that were originally affected by the reference change to restore accurate gap measurement, thereby continuously acquiring real gap data during the normal operation of the magnetic levitation bearing, avoiding consequences such as increased equipment vibration, rotor instability, and bearing wear caused by gap measurement errors due to missing correction.

[0129] Reference Figure 2 Methods for obtaining the correction value corresponding to each eddy current sensor include:

[0130] The Gaussian radial basis function is selected as the interpolation kernel function to construct the interpolation function. Based on the interpolation condition that the interpolation function value at the interpolation anchor point equals the actual reference deviation, a system of linear equations is established. The Cholesky decomposition method is used to solve the system of linear equations to obtain the interpolation coefficients. For example, the Gaussian radial basis function... ,in, The shape parameters are determined through cross-validation to minimize interpolation error; The distance from the interpolation point to the interpolation anchor point; the interpolation function. ,in, This is the theoretical value of the calibrated exponential decay model; These are the interpolation coefficients to be determined; Radial basis functions can be used as interpolation basis functions; interpolation anchor point The coordinates;

[0131] Iterate through each eddy current sensor in the set of sensors to be calibrated Obtain the corresponding coordinates, substitute them into the interpolation function, and calculate the eddy current sensor value. The initial calibration amount; if the eddy current sensor If the absolute value of the initial correction is greater than the correction threshold, then the eddy current sensor is selected. The initial correction value is the correction threshold; the correction threshold is generally taken as 10% of the absolute gap value measured by the quantum sensor.

[0132] A Q×Q neighborhood mean filter is used to smooth the initial correction values ​​of all units to be calibrated. Taking the current eddy current sensor as the center, the average initial correction value of the surrounding Q×Q-1 neighboring eddy current sensors is calculated. The eddy current sensor is then calculated by weighting the initial correction value of the current eddy current sensor with its average initial correction value. The correction amount.

[0133] The above steps select a Gaussian radial basis function to construct the interpolation function and combine it with the Cholesky decomposition method to solve the interpolation coefficients. This allows the absolute reference deviation anchor point provided by the quantum sensor to be accurately mapped to each eddy current sensor within the region to be calibrated. This overcomes the limitation of existing technologies that cannot provide differentiated calibration for eddy current sensors at different locations based on the spatial differences caused by phase transitions. By setting a calibration threshold, measurement distortion caused by calibration values ​​exceeding the physically reasonable range can be avoided, ensuring the reliability of the calibrated gap data. Furthermore, Q×Q neighborhood mean filtering is used to smooth the initial calibration value, eliminating abrupt changes in local calibration values ​​and ensuring the spatial continuity of calibration values ​​for adjacent sensors, thus preventing increased equipment vibration due to calibration fluctuations. Overall, this method provides each eddy current sensor affected by reference abrupt changes with accurate and stable calibration values ​​that conform to physical laws. This allows eddy current sensors that were previously affected by phase transitions to regain accurate gap monitoring capabilities, thereby continuously outputting true gap data during normal dynamic operation of the magnetic levitation bearing. This effectively avoids consequences such as rotor instability and bearing wear caused by gap measurement errors.

[0134] The propagation time and resonant frequency of sound waves are collected by surface acoustic wave sensors, and the stress and temperature gradients are analyzed. A feedforward compensation model is constructed based on a semi-empirical hybrid model. The gap compensation amount is calculated based on the stress and temperature gradients, and the correction gap value is calculated by combining the correction amount. The surface acoustic wave sensors are deployed inside the stator of the magnetic levitation bearing at positions corresponding to the quantum sensor array and the eddy current sensor array.

[0135] Reference Figure 3 Methods for obtaining the correction gap value include:

[0136] The propagation time and resonant frequency of the sound wave are collected for each surface acoustic wave sensor, and Kalman filtering and moving average filtering are performed to obtain the filtered propagation time and resonant frequency; where the resonant frequency is the frequency point corresponding to the maximum amplitude in the amplitude sequence of the output signal of the surface acoustic wave sensor.

[0137] Stress is obtained by calculating the sound wave propagation time under stress-free conditions and the sound wave propagation time after filtering; such as stress ,in, The stress calibration coefficient is obtained through material tensile testing. The propagation time of the filtered sound wave; The propagation time of sound waves under stress-free conditions;

[0138] The single-point temperature is calculated based on the preset temperature, the resonant frequency under stress-free conditions, and the filtered resonant frequency; the temperature gradient is calculated based on the temperature difference and spacing between adjacent sensors; such as a surface acoustic wave sensor. temperature ,in, The temperature calibration coefficient was obtained through experimental fitting. This is the resonant frequency after filtering; The resonant frequency is at a preset temperature, typically 25°C, under stress-free conditions; temperature gradient. ,in, For surface acoustic wave sensors Temperature; For surface acoustic wave sensors With surface acoustic wave sensor The spacing;

[0139] If the stress is greater than the stress threshold or the temperature gradient is greater than the gradient threshold, it is determined that a phase transition has occurred and proceeds to the next step; otherwise, it is determined that no phase transition has occurred and the output gap compensation amount is 0.

[0140] Based on stress and temperature gradients, a semi-empirical hybrid model is used to construct a feedforward compensation model for the gap compensation amount. The model coefficients are obtained by solving the feedforward compensation model using the least squares method. These coefficients are then substituted into the feedforward compensation model, and the gap compensation amount is calculated by combining real-time stress and real-time gradients. (See feedforward compensation model). ,in, This is the gap compensation amount; and The linear influence coefficient of stress and temperature gradient can be obtained by combining experimental data with the least squares method. For stress; For temperature gradient; The stress-temperature coupling influence coefficient can be obtained by combining experimental data with the least squares method. The basic compensation amount; if the gap compensation amount Then take ,like Then take In other cases, take directly As the gap compensation amount; among which... This is the upper limit of the gap compensation amount, obtained based on historical data statistics;

[0141] Calculate the sum of the correction amount and the gap compensation amount to obtain the correction gap value.

[0142] The above steps involve deploying a surface acoustic wave (SAW) sensor at the stator position corresponding to the quantum and eddy current sensors, collecting the acoustic wave propagation time and resonant frequency, and filtering the data to ensure the accuracy of monitoring the core causes of phase transition. Then, stress and temperature gradients are calculated using calibration formulas, and threshold judgments are used to identify potential phase transition risks in advance, overcoming the shortcomings of existing technologies in providing insufficient early warning of phase transition causes. A feedforward compensation model is constructed based on a semi-empirical hybrid model, and the coefficients are solved using the least squares method, combined with real-time causes to calculate the gap compensation amount. This achieves early compensation for benchmark mutations caused by potential phase transitions, avoiding the lag of existing technologies that only passively correct after a phase transition occurs. Finally, the correction amount from the eddy current sensor is added to the feedforward compensation amount to obtain the corrected gap value, correcting existing measurement deviations and avoiding errors that may be caused by future causes. This ensures that the magnetic levitation bearing can always obtain accurate gap data during dynamic operation, effectively avoiding consequences such as increased equipment vibration, rotor instability, and bearing wear caused by inaccurate gap monitoring.

[0143] Example 2:

[0144] Please see Figure 4 As shown, this embodiment provides an online measurement device for the clearance of magnetic levitation bearings, including:

[0145] Sensing deployment module: A quantum sensor array is deployed on the inner wall of the magnetic levitation bearing stator to measure the Rabi oscillation frequency and calculate the absolute gap value; planar eddy current coils are arranged alternately between the quantum sensor arrays to form an eddy current sensor array; surface acoustic wave sensors are deployed inside the magnetic levitation bearing stator at positions corresponding to the quantum sensor array and the eddy current sensor array.

[0146] The acquisition and positioning module injects a scanning signal into the eddy current sensor array, acquires real-time complex impedance signals, calculates the difference signal between the real-time complex impedance signal and the corresponding frequency point in the baseline database, and obtains the complex impedance differential signal.

[0147] Phase transition determination module: Analyzes the complex impedance differential signal to obtain real-time feature fingerprints; compares the real-time feature fingerprints with the phase transition fingerprint database to obtain the phase transition determination result;

[0148] Online calibration module: When the phase transition determination result indicates that a phase transition has occurred, the absolute gap value of the quantum sensor at the phase transition location and the original gap measurement value of the eddy current sensor are collected. The deviation between the absolute gap value and the original gap measurement value is calculated to obtain the reference deviation. A local spatial correction field is generated with the phase transition location as the center. The correction amount corresponding to each eddy current sensor is calculated through an interpolation algorithm.

[0149] Feedforward compensation module: Acquires sound wave propagation time and resonant frequency through surface acoustic wave sensor, analyzes and obtains stress and temperature gradient, constructs feedforward compensation model based on semi-empirical hybrid model, calculates gap compensation amount according to stress and temperature gradient, and calculates correction gap value by combining correction amount.

[0150] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0151] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for online measurement and compensation of clearance in magnetic levitation bearings, characterized in that, include: A quantum sensor array is deployed on the inner wall of the stator of the magnetic levitation bearing to measure the Rabi oscillation frequency and calculate the absolute gap value; planar eddy current coils are arranged alternately between the quantum sensor arrays to form an eddy current sensor array. A scanning signal is injected into the eddy current sensor array to acquire real-time complex impedance signals. The difference signal between the real-time complex impedance signal and the corresponding frequency point in the baseline database is calculated to obtain the complex impedance differential signal. The complex impedance differential signal is analyzed to obtain real-time feature fingerprints; the real-time feature fingerprints are compared with the phase transition fingerprint database to obtain the phase transition determination results. When the phase transition determination result indicates that a phase transition has occurred, the absolute gap value of the quantum sensor at the phase transition location and the original gap measurement value of the eddy current sensor are collected. The deviation between the absolute gap value and the original gap measurement value is calculated to obtain the reference deviation. A local spatial correction field is generated with the phase transition location as the center. The correction amount corresponding to each eddy current sensor is calculated through an interpolation algorithm. By collecting the propagation time and resonant frequency of sound waves using a surface acoustic wave sensor, the stress and temperature gradients are analyzed and obtained. A feedforward compensation model is constructed based on a semi-empirical hybrid model. The gap compensation amount is calculated based on the stress and temperature gradients, and the correction gap value is calculated by combining the correction amount.

2. The online measurement and compensation method for magnetic levitation bearing clearance according to claim 1, characterized in that, Methods for obtaining the correction gap value include: Acquire the acoustic wave propagation time and resonant frequency of each surface acoustic wave sensor, and perform Kalman filtering and moving average filtering to obtain the filtered acoustic wave propagation time and the filtered resonant frequency. The stress is calculated based on the propagation time of the sound wave under stress-free conditions and the propagation time of the sound wave after filtering; the single-point temperature is calculated based on the preset temperature, the resonant frequency under stress-free conditions and the resonant frequency after filtering; and the temperature gradient is calculated based on the temperature difference and spacing between adjacent sensors. If the stress is greater than the stress threshold or the temperature gradient is greater than the gradient threshold, it is determined that a phase transition has occurred and proceeds to the next step; otherwise, it is determined that no phase transition has occurred and the output gap compensation amount is 0. Based on stress and temperature gradient, a semi-empirical hybrid model is used to build a feedforward compensation model for the gap compensation amount. The feedforward compensation model is solved using the least squares method to obtain the model coefficients. The model coefficients are then substituted into the feedforward compensation model, and the gap compensation amount is calculated by combining real-time stress and real-time temperature gradient. Calculate the sum of the correction amount and the gap compensation amount to obtain the correction gap value.

3. The method for online measurement and compensation of clearance in magnetic levitation bearings according to claim 1, characterized in that, Methods for generating local spatial correction fields include: A circular region to be calibrated is defined with the location of the phase transition as the center and the radius of the phase transition influence as the radius. A calibration field is generated only for the eddy current sensor within the circular region to be calibrated. Obtain the physical coordinates of all eddy current sensors within the circular area to be calibrated, forming a set of sensors to be calibrated; extract the quantum sensor data adjacent to the phase transition location from the set of sensors to be calibrated. An exponential decay model was constructed, and the theoretical reference deviation at coordinates (x, y) was calculated based on the reference deviation, straight-line distance, and phase transition influence radius at the location of the phase transition. The exponential decay model is calibrated to obtain the calibrated exponential decay model. Centered on the location where the phase transition occurs, a square grid covering the circular area to be corrected is generated according to the grid side length. For each grid node, a preliminary correction value is calculated based on the calibrated exponential decay model. Based on the anchor point weighted correction method, the preliminary correction value of each grid node is adjusted to obtain the optimized grid node correction value. The optimized grid node correction value is smoothed by Gaussian filtering to obtain the smoothed correction value. The local spatial correction field is obtained.

4. The online measurement and compensation method for magnetic levitation bearing clearance according to claim 3, characterized in that, Methods for obtaining the radius of influence of phase transition include: The location of the phase transition is extracted, and the radius of influence of the phase transition is calculated using a semi-empirical formula, taking into account the area of ​​the phase transition and the magnetic permeability and electrical conductivity of the bearing material. The quantum sensor data includes the quantum sensor coordinates, the measured absolute gap value, and the original gap value and reference deviation of the eddy current sensor corresponding to the phase transition location. Methods for obtaining a calibrated exponential decay model include: The reference deviation corresponding to the quantum sensor is used as the interpolation anchor point. The exponential decay model is calibrated using the interpolation anchor point: the interpolation anchor point is substituted into the exponential decay model, the theoretical reference deviation of the interpolation anchor point is calculated, and then the residual between the theoretical reference deviation and the reference deviation of the interpolation anchor point is calculated; then the mean of the residuals is calculated. If the mean of the residuals is greater than the residual threshold, the decay coefficient is adjusted according to the residual step size; if the reference deviation of the interpolation anchor point is less than the theoretical reference deviation of the interpolation anchor point, the decay coefficient is increased according to the residual step size; if the reference deviation of the interpolation anchor point is greater than the theoretical reference deviation of the interpolation anchor point, the decay coefficient is decreased according to the residual step size; until the mean of the residuals is not greater than the residual threshold, the decay function at this time is obtained and substituted into the exponential decay model to obtain the calibrated exponential decay model.

5. The online measurement and compensation method for magnetic levitation bearing clearance according to claim 3, characterized in that, Methods for obtaining optimized mesh node correction values ​​include: Calculate the grid distance from each grid node to each interpolation anchor point; then calculate the grid weight of each grid node to each interpolation anchor point based on the grid distance; calculate the anchor point correction amount based on the mean of the grid weights and the residual; finally, calculate the sum of the anchor point correction amount and the initial correction value to obtain the optimized grid node correction value.

6. The method for online measurement and compensation of magnetic levitation bearing clearance according to claim 5, characterized in that, Methods for obtaining the correction value corresponding to each eddy current sensor include: The Gaussian radial basis function is selected as the interpolation kernel function to construct the interpolation function. Based on the interpolation condition that the interpolation function value at the interpolation anchor point is equal to the actual reference deviation, a system of linear equations is established. The Cholesky decomposition method is used to solve the system of linear equations to obtain the interpolation coefficients. Iterate through each eddy current sensor in the set of sensors to be calibrated Obtain the corresponding coordinates, substitute them into the interpolation function, and calculate the eddy current sensor value. The initial calibration amount; if the eddy current sensor If the absolute value of the initial correction is greater than the correction threshold, then the eddy current sensor is selected. The initial correction amount is the correction amount threshold; A Q×Q neighborhood mean filter is used to smooth the initial calibration value of each eddy current sensor in the sensor set to be calibrated. Taking the current eddy current sensor as the center, the mean of the initial calibration values ​​of the surrounding Q×Q-1 neighboring eddy current sensors is calculated. The eddy current sensor is obtained by weighting the initial calibration value of the current eddy current sensor and the mean of the initial calibration values. The correction amount.

7. The method for online measurement and compensation of clearance in magnetic levitation bearings according to claim 1, characterized in that, Methods for obtaining real-time feature fingerprints include: Calculate the amplitude of the complex impedance differential signal across the entire frequency band to obtain an amplitude sequence. Detect whether the rate of change of amplitude at a frequency point in the amplitude sequence exceeds the amplitude jump threshold. If so, record the corresponding frequency point and amplitude jump value; otherwise, record the corresponding frequency point and amplitude jump value as 0 to obtain amplitude abrupt change characteristics. Calculate the phase of the complex impedance differential signal across the entire frequency band to obtain a phase sequence. Detect whether the phase change rate at a frequency point in the phase sequence exceeds the phase jump threshold. If it does, record the corresponding frequency point and phase jump value; otherwise, record the corresponding frequency point and phase jump value as 0 to obtain the phase jump characteristics. Local maximum values ​​in the amplitude sequence are extracted as characteristic resonant frequencies. The offset of the characteristic resonant frequency relative to the corresponding resonant frequency in the baseline impedance spectrum is calculated to obtain the characteristic frequency offset feature. According to the pre-division rules, the entire frequency band is divided into three sub-bands: low, medium, and high. The integral energy of the amplitude of the complex impedance differential signal in each sub-band is calculated, and the ratio of the integral energy of each sub-band to the total energy is calculated to obtain the frequency band energy variation characteristics. The amplitude abrupt change feature, phase abrupt change feature, characteristic frequency offset feature, and frequency band energy change feature are concatenated to obtain a feature vector. Each parameter in the feature vector is normalized to obtain a real-time feature fingerprint.

8. The online measurement and compensation method for magnetic levitation bearing clearance according to claim 1, characterized in that, Methods for obtaining phase transition determination results include: The difference between the real-time feature fingerprint and the phase transition fingerprint in the phase transition fingerprint database is calculated using weighted Euclidean distance to obtain the similarity distance. The similarity distance is then converted into a similarity score by combining the maximum distance. Traverse all phase transition fingerprints in the phase transition fingerprint database, calculate the similarity between the real-time feature fingerprint and each phase transition fingerprint, and obtain the maximum similarity as the matching degree; If the matching degree exceeds the matching threshold, it is determined that a microscopic phase transition has occurred in the sensitive area of ​​the current eddy current sensor, and a phase transition determination signal is output. Simultaneously, the physical coordinates corresponding to the current eddy current sensor are recorded as the location where the phase transition occurred; otherwise, it is determined that no phase transition has occurred, and a phase transition determination signal is output. .

9. The method for online measurement and compensation of magnetic levitation bearing clearance according to claim 1, characterized in that, Methods for obtaining complex impedance differential signals include: For each frequency point of data from each eddy current sensor, outliers are removed by applying the three-times-standard-deviation principle to obtain valid data points. The arithmetic mean of the valid data points is then calculated to obtain the real-time average of the complex impedance. The mean values ​​of the real-time resistance component and the mean values ​​of the real-time reactance component are filtered to obtain the filtered real-time complex impedance. For each eddy current sensor traverse all its frequency points Retrieve the corresponding eddy current sensor from the baseline database at the same frequency. The baseline complex impedance; For each eddy current sensor Each frequency point The differential components of the filtered real-time complex impedance and the baseline complex impedance are calculated respectively to obtain the complex impedance differential signal.

10. The method for online measurement and compensation of clearance in magnetic levitation bearings according to claim 1, characterized in that, Methods for obtaining absolute gap values ​​include: Turn off the bias magnetic field of the Helmholtz coil; apply a frequency-sweeping microwave signal of a preset frequency band to the microwave antenna in a preset step size, and at the same time collect the fluorescence intensity signal of the quantum sensor to obtain the resonance spectrum of fluorescence intensity with respect to microwave frequency; Identify the resonance peaks in the resonance spectrum corresponding to the transitions between the ground state ms=0 and ms=±1 energy levels of the quantum sensor, obtain the corresponding resonance frequencies, and calculate the background magnetic field according to the quantum sensor energy level splitting formula. The background magnetic field is fed back to the Helmholtz coil control system, and a reverse compensation magnetic field of the same magnitude is applied. A circular metal sheet of the same material as the bearing rotor and with the same diameter as the rotor is used, and the distance between the circular metal sheet and the surface of the stator sensor is controlled to be a standard gap. For each standard gap, a microwave signal source is activated to output a microwave field with a fixed power. The frequency of the microwave field is adjusted to make the quantum sensor generate Rabi oscillations and collect the decay signal of fluorescence intensity over time. The attenuation signal is sinusoidally fitted and the Levenberg-Marquardt algorithm is used to iteratively solve the problem to obtain the Rabi oscillation frequency. The correspondence between the Rabi oscillation frequency and the standard gap is established, and the calibration formula is obtained through fitting. The initial absolute gap value is calculated based on the real-time Rabi oscillation frequency and the calibration formula. The initial absolute gap values ​​of all quantum sensors are weighted and fused to obtain the absolute gap value.

11. An online measurement device for magnetic levitation bearing clearance, implementing the online measurement and compensation method for magnetic levitation bearing clearance as described in any one of claims 1-10, characterized in that, include: Sensing deployment module: A quantum sensor array is deployed on the inner wall of the magnetic levitation bearing stator to measure the Rabi oscillation frequency and calculate the absolute gap value; planar eddy current coils are arranged alternately between the quantum sensor arrays to form an eddy current sensor array; surface acoustic wave sensors are deployed inside the magnetic levitation bearing stator at positions corresponding to the quantum sensor array and the eddy current sensor array. The acquisition and positioning module injects a scanning signal into the eddy current sensor array, acquires real-time complex impedance signals, calculates the difference signal between the real-time complex impedance signal and the corresponding frequency point in the baseline database, and obtains the complex impedance differential signal. Phase transition determination module: Analyzes the complex impedance differential signal to obtain real-time feature fingerprints; compares the real-time feature fingerprints with the phase transition fingerprint database to obtain the phase transition determination result; Online calibration module: When the phase transition determination result indicates that a phase transition has occurred, the absolute gap value of the quantum sensor at the phase transition location and the original gap measurement value of the eddy current sensor are collected. The deviation between the absolute gap value and the original gap measurement value is calculated to obtain the reference deviation. A local spatial correction field is generated with the phase transition location as the center. The correction amount corresponding to each eddy current sensor is calculated through an interpolation algorithm. Feedforward compensation module: Acquires sound wave propagation time and resonant frequency through surface acoustic wave sensor, analyzes and obtains stress and temperature gradient, constructs feedforward compensation model based on semi-empirical hybrid model, calculates gap compensation amount according to stress and temperature gradient, and calculates correction gap value by combining correction amount.

Images