Intelligent start-stop control method and system of magnetic levitation refrigeration compressor

Abstract

The application discloses an intelligent start-stop control method and system for a magnetic suspension refrigeration compressor, which acquires a position perception enhancement source and extracts current response change characteristics by collecting and analyzing current response data of a magnetic bearing coil; estimates a rotor position based on the current response change characteristics, extracts a positioning accuracy driving source from an estimated deviation, and generates a non-impact start-up excitation current instruction; establishes a multi-sensor fusion data network, activates a signal reconstruction configuration through an abnormal response, forms a degraded operation fault-tolerant mode; converts a response preparation gain according to a shutdown instruction signal, evaluates a shutdown state in combination with the signal reconstruction configuration, and generates an emergency shutdown trigger condition; detects a bus capacitor power supply state, releases gradient control potential, determines an active damping control opportunity, and generates a damping force control instruction, so as to finally realize intelligent start-stop control of the magnetic suspension refrigeration compressor, and ensure safe and reliable operation of the system under various working conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of compressor control, in particular to an intelligent start-stop control method and system of a magnetic suspension refrigeration compressor. BACKGROUND

[0002] The magnetic suspension refrigeration compressor adopts electromagnetic bearings to support the rotor, realizes non-contact suspension operation, has the advantages of no friction, no need for lubrication, low vibration and noise, and is widely used in the fields of refrigeration and air conditioning, industrial compression, etc. However, the start-stop process of the magnetic suspension compressor is a key link of system operation, and improper start-stop control can cause serious problems such as rotor collision with standby bearings, vibration overrun, position instability, etc., affecting the service life and operation reliability of the equipment.

[0003] The existing start-stop control method of the magnetic suspension compressor mainly relies on preset control parameters and fixed start-stop programs, and lacks adaptive adjustment capability for real-time running state. In the starting process, the sudden change of excitation current is easy to produce impact and vibration; in the shutdown process, the utilization of bus capacitor energy is not sufficient, and smooth deceleration cannot be realized; at the same time, in the case of sensor drift, signal anomaly and other fault conditions, the system lacks effective fault tolerance mechanism, and it is difficult to ensure safe and reliable start-stop control. Therefore, it is urgent to develop an intelligent start-stop control method which can realize impact-free start and smooth stop. SUMMARY

[0004] The present application discloses an intelligent start-stop control method and system of a magnetic suspension refrigeration compressor, which aims to obtain a position perception enhancement source through fluctuation analysis of current response data, accurately estimate the rotor position and generate an impact-free start excitation current instruction; establish a multi-sensor fusion data network, form a degraded operation fault tolerance mode through signal reconstruction configuration; convert the shutdown instruction signal into a response preparation gain, combine the bus capacitor energy gradient control and active damping control, and realize intelligent start-stop control of the magnetic suspension refrigeration compressor.

[0005] The present application discloses an intelligent start-stop control method and system of a magnetic suspension refrigeration compressor, which aims to obtain a position perception enhancement source through fluctuation analysis of current response data, accurately estimate the rotor position and generate an impact-free start excitation current instruction; establish a multi-sensor fusion data network, form a degraded operation fault tolerance mode through signal reconstruction configuration; convert the shutdown instruction signal into a response preparation gain, combine the bus capacitor energy gradient control and active damping control, and realize intelligent start-stop control of the magnetic suspension refrigeration compressor.

[0006] Collecting current response data of the magnetic bearing coil, performing fluctuation analysis on the current response data to obtain a position perception enhancement source, and obtaining current response change characteristics based on the position perception enhancement source;

[0007] Performing rotor position estimation on the current response change characteristics to extract rotor offset coordinate parameters, extracting a positioning accuracy driving source from the estimation deviation of the rotor offset coordinate parameters, and generating an impact-free start excitation current instruction based on the positioning accuracy driving source processing the current response change characteristics;

[0008] Based on the non-impact starting excitation current instruction, a fusion data network is established by monitoring the multi-sensor operating state, an abnormal response activation signal reconstruction configuration is activated in the fusion data network, and a degraded operation fault-tolerant mode is generated based on the signal reconstruction configuration combined with the current response change characteristics;

[0009] A shutdown instruction signal is received through the degraded operation fault-tolerant mode, the shutdown instruction signal is converted into a response preparation gain, a shutdown state evaluation is performed according to the response preparation gain combined with the signal reconstruction configuration to generate an emergency shutdown trigger condition;

[0010] The bus capacitor power supply state is detected according to the emergency shutdown trigger condition, a gradient control potential is released from the energy change of the bus capacitor power supply state, a proactive damping control opportunity is determined based on the gradient control potential combined with the degraded operation fault-tolerant mode, a damping force control instruction is generated through the proactive damping control opportunity, and intelligent start-stop control of the magnetic suspension refrigeration compressor is realized.

[0011] The second aspect of the present application proposes an intelligent start-stop control system of a magnetic suspension refrigeration compressor, comprising:

[0012] A data acquisition module is used to acquire current response data of a magnetic bearing coil, perform fluctuation analysis on the current response data to obtain a position perception enhancement source, and obtain current response change characteristics based on the position perception enhancement source.

[0013] A position estimation module is used to perform rotor position estimation on the current response change characteristics to extract rotor offset coordinate parameters, extract a positioning accuracy driving source from the estimated deviation of the rotor offset coordinate parameters, and generate a non-impact starting excitation current instruction based on the positioning accuracy driving source processing the current response change characteristics.

[0014] A fault monitoring module is used to monitor the multi-sensor operating state based on the non-impact starting excitation current instruction to establish a fusion data network, activate a signal reconstruction configuration in the fusion data network through an abnormal response, and generate a degraded operation fault-tolerant mode based on the signal reconstruction configuration combined with the current response change characteristics.

[0015] A shutdown control module is used to receive a shutdown instruction signal through the degraded operation fault-tolerant mode, convert the shutdown instruction signal into a response preparation gain, perform a shutdown state evaluation according to the response preparation gain combined with the signal reconstruction configuration to generate an emergency shutdown trigger condition.

[0016] The damping execution module is used for detecting a bus capacitor power supply state according to the emergency stop triggering condition, releasing gradient control potential from energy change of the bus capacitor power supply state, determining an active damping control opportunity based on the gradient control potential in combination with the degraded operation fault-tolerant mode, generating damping force control instructions through the active damping control opportunity, and realizing intelligent start-stop control of the magnetic suspension refrigeration compressor.

[0017] The beneficial effects of the present application are embodied in the following aspects: first, the position sensing enhanced source is constructed by current response fluctuation analysis, indirect measurement of rotor position based on current information is realized, and the measurement limitation of traditional position sensors in high-speed rotating environment is overcome; in this process, the estimation deviation of the rotor position is no longer a pure error, but is extracted as a positioning accuracy driving source, the variation law of which directly guides the generation of the excitation current instruction, so that the control system can adaptively adjust according to the actual deviation characteristics, and the starting impact problem of fixed parameter control is eliminated. Second, a complete fault detection and compensation mechanism is established by combining the multi-sensor fusion data network with the signal reconstruction configuration technology, the identification of the sensor drift mode is not only used for fault diagnosis, but also becomes the basis for triggering cooperative calibration, and a compensation coordinate set is generated through multi-channel cross verification; the degraded operation fault-tolerant mode automatically adjusts the control strategy according to the level of abnormal response, converts the abnormal state information of the system into a decision basis for fault-tolerant control, and ensures that the system can still maintain stable operation when part of the sensors fail. Finally, the torque standing wave formation technology redistributes the oscillation energy of the rotor, injects an opposite torque at a specific phase to form a stable standing wave mode from the originally disordered oscillation, and the standing wave node becomes an ideal shutdown target position; the energy decay process of the bus capacitor is divided into different control regions through gradient analysis, and the residual energy of each region is fully utilized for the corresponding control task, so that the optimization of energy distribution in the shutdown process is realized, and the controllable shutdown time window is extended. BRIEF DESCRIPTION OF DRAWINGS

[0018] The drawings herein show specific examples of the technical solutions described in the present application, and constitute part of the specification together with the specific embodiments, for explaining the technical solutions, principles and effects of the present application.

[0019] Unless specifically stated, the same reference signs in different drawings represent the same or similar technical features, and different reference signs may also be used to represent the same or similar technical features.

[0020] Figure 1 is a flowchart of an intelligent start-stop control method of a magnetic suspension refrigeration compressor according to the present application.

[0021] Figure 2 is a structural block diagram of an intelligent start-stop control system of a magnetic suspension refrigeration compressor according to the present application. DETAILED DESCRIPTION

[0022] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0023] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0024] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0025] The technical solutions of the embodiments of this application are described below.

[0026] like Figure 1 As shown, this embodiment of the invention provides an intelligent start-stop control method for a magnetic levitation refrigeration compressor, including the following steps S110-S150:

[0027] Step S110: Collect current response data of the magnetic bearing coil, perform fluctuation analysis on the current response data to obtain the position sensing enhancement source, and obtain the current response change characteristics based on the position sensing enhancement source.

[0028] Specifically, the current response data of the magnetic bearing coils is collected. High-precision current sensors are deployed in the control loop of the magnetic bearing system to collect the excitation current signals of the radial and axial magnetic bearing coils in real time. The current sensors use Hall effect sensors with a measurement range of 0-50 A, a resolution of 0.001 A, and a response time of less than 1 microsecond, ensuring that rapid changes in current can be captured. The radial magnetic bearing contains four orthogonally distributed electromagnetic coils, each equipped with an independent current sensor, forming X positive, X negative, Y positive, and Y negative four-path control current signals. The axial magnetic bearing contains two sets of upper and lower thrust coils, which monitor the upper and lower thrust currents, respectively. The current data collection uses a synchronous sampling mode, and all sensors achieve strict time synchronization through hardware trigger signals, with a sampling frequency of 20 kHz to ensure that the Nyquist sampling theorem is met. The collection process continuously records the current response of the rotor under different operating conditions, including typical conditions such as the startup phase, steady-state operation, acceleration process, load change, and shutdown process. Ten minutes of current data is collected continuously under each condition to form a complete current response data set containing time series, current amplitude, and phase information.

[0029] In some embodiments, the fluctuation analysis of the current response data obtains a position-aware enhancement source, including: frequency anomaly detection of the current response data to obtain inductance field fluctuation characteristics; identifying a position error compensation current sequence in the inductance field fluctuation characteristics; amplitude calibration of the position error compensation current sequence to form a rotor positioning feature; and constructing a position-aware enhancement source using the rotor positioning feature.

[0030] The frequency anomaly detection is used to obtain the inductance field fluctuation characteristics. The power spectrum of the current signal is estimated by the Welch method, with a segment length of 4096 points and an overlap rate of 50%. The Hanning window is used as the window function. The normal frequency distribution is determined by the steady-state operation data, and the main energy is concentrated in the 0-2 kHz frequency band, with the peak frequency corresponding to the rotational frequency of the rotor and its multiple frequencies. The identification of abnormal frequency components is realized by spectral peak detection. When the power density of a certain frequency point exceeds 3 times the average value of the neighborhood, it is marked as an abnormal frequency peak. The inductance field fluctuation characteristics are characterized by abnormal energy concentration in a specific frequency band, usually in the range of 500-1500 Hz, corresponding to the modulation effect of the interaction between the magnetic field and the rotor. The extraction of the fluctuation characteristics includes the center frequency, bandwidth, energy ratio, and occurrence time of the abnormal frequency. The center frequency is determined by the weighted average method, with the weight being the power density value of each frequency point. The bandwidth is determined by the -3 dB cutoff frequency, reflecting the spectral width of the abnormal frequency component. The energy ratio is calculated by integrating the abnormal frequency band energy and the total energy. An energy ratio of more than 10% indicates significant inductance field fluctuation. The time-domain characteristics are extracted by band-pass filtering, with the filter center frequency set to the detected abnormal frequency and the bandwidth set to 1.5 times the abnormal frequency bandwidth.

[0031] The position error compensation current sequence is identified in the inductance field fluctuation characteristics. The identification of the position error compensation current starts from the 500-1500 Hz abnormal frequency band of the fluctuation characteristics, which corresponds to the characteristic range of the magnetic field modulation effect. The compensation current sequence exhibits a specific energy distribution pattern in this frequency band, and the frequency points with an energy ratio of more than 10% usually contain the compensation signal. The identification process uses the center frequency of the inductance field fluctuation characteristics as the filter design parameter, and sets the bandwidth to the detected abnormal bandwidth value. The radial X-direction compensation current is extracted from the X-component of the fluctuation characteristics, and the modulation component in the 100-1000 Hz range is separated by band-pass filtering. The radial Y-direction compensation current is obtained from the Y-component fluctuation, and its spectral distribution is orthogonal to that of the X-component. The axial compensation current is identified from the Z-direction fluctuation characteristics, and its frequency characteristics are significantly different from those of the radial components, mainly concentrated in the low frequency band. The polarity of the compensation current is determined by the phase information of the fluctuation characteristics, with a positive phase corresponding to a positive compensation force and a negative phase corresponding to a negative compensation force. The continuity of the compensation current sequence is judged by the time-domain envelope of the fluctuation characteristics, and the smoothness of the envelope reflects the stability of the compensation process. The sequence extraction considers the -3 dB bandwidth range of the fluctuation characteristics to ensure that the main compensation energy components are included. The identification result retains the time marker of the fluctuation characteristics, forming a compensation current sequence with accurate time stamps.

[0032] The amplitude calibration of the position error compensation current sequence forms the rotor positioning feature. The position error compensation current sequence is taken as the calibration input, which contains the compensation current time series in X, Y, Z three directions. The amplitude calibration determines the conversion coefficient according to the force-current characteristic and the force-displacement characteristic of the magnetic bearing. The electromagnetic force F generated by the magnetic bearing is F=Ki×i / g, where Ki is the current stiffness coefficient, i is the instantaneous value in the compensation current sequence, and g is the air gap length. The relationship between displacement and current is x=(Ki×i) / (Kx×g), where x is the rotor displacement, and Kx is the displacement stiffness. The compensation current sequence is converted into a displacement sequence point by point through the relationship. The calibration coefficient is determined by the system identification method. The current response is measured under the known displacement excitation, and the conversion coefficient is obtained by least squares fitting. The radial positioning feature includes the X coordinate and Y coordinate time series of the rotor center, with an accuracy of ±1 microns, which is converted from the X and Y direction compensation current sequences respectively. The axial positioning feature includes the axial position Z coordinate sequence of the rotor, with an accuracy of ±2 microns, which is converted from the Z direction compensation current sequence. The rotor attitude feature includes the tilt angles α and β, which are obtained by calculating the difference between the displacement sequences at adjacent time points and the measurement span, with an angle resolution of 0.01 milliradians.

[0033] The rotor positioning feature is used to construct a position perception enhancement source. The construction of the position perception enhancement source starts from the calibrated X, Y, Z coordinate values and tilt angles α, β, and expands to form a complete six-degree-of-freedom motion description of the rotor. The translational degrees of freedom directly use the calibrated X coordinate (±1 micron accuracy), Y coordinate (±1 micron accuracy), and Z coordinate (±2 micron accuracy), which reflect the real-time spatial position of the rotor mass center. The rotational degrees of freedom are obtained by geometric transformation of the calibrated tilt angles α and β. The pitch angle corresponds to the sine component of the α angle, the yaw angle corresponds to the sine component of the β angle, and the roll angle is calculated by the coupling relationship of α and β. The motion trajectory reconstruction connects the discrete X-Y coordinate points in time sequence to form the whirling trajectory of the rotor in the radial plane, and the change of the Z coordinate forms the axial vibration curve. The trajectory feature is extracted from the reconstructed position data. The elliptical fitting parameters of the X-Y plane trajectory reflect the whirling amplitude and direction, and the amplitude statistics of the Z direction reflect the axial stability. The 0.01 milliradian angle resolution obtained by calibration supports accurate attitude change detection, and the angular velocity is obtained by the difference between adjacent time points. The enhancement processing combines these basic positioning parameters with time derivatives to generate a complete state vector containing position, velocity, and acceleration. The fusion process gives higher weight (0.8) to low-frequency position information and lower weight (0.6) to high-frequency attitude information, forming a multi-scale position perception enhancement source.

[0034] The current response variation characteristics are obtained based on the position-aware enhanced source. The multi-dimensional feature vector in the enhanced source is taken as the analysis reference, and the corresponding response law is extracted by comparing with the original current signal. The amplitude variation characteristic is demodulated by the position fluctuation information of the enhanced source. When a radial displacement of 1 micrometer is detected, the corresponding current variation is about 0.2 A. The corresponding relationship is extracted by Hilbert envelope. The phase shift characteristic refers to the phase of the rotation frequency component of the enhanced source, and calculates the deviation of the current signal from the reference phase. The deviation angle reflects the response delay of the control loop. The frequency modulation characteristic starts from the 100-1000 Hz modulation frequency identified by the enhanced source, and tracks the evolution process of these frequency components in the current signal. The time evolution of the variation characteristics is aligned with the time label of the enhanced source, ensuring the synchronization of the feature extraction. The sliding window is set according to the sampling rate of the enhanced source. The feature correlation analysis takes the coupling relationship of the enhanced source as the reference template, and identifies similar coupling modes in the current response. The physical classification is based on the position, speed and acceleration information framework provided by the enhanced source, and the current characteristics are correspondingly assigned to the corresponding categories. The position-related characteristics include displacement amplitude, trajectory shape and center shift parameters, the speed-related characteristics include rotation speed fluctuation, eddy frequency and precession direction parameters, and the acceleration-related characteristics include vibration acceleration, impact response and transient change parameters.

[0035] In step S120, the rotor position estimation is performed on the current response variation characteristics to extract rotor offset coordinate parameters. The positioning accuracy driving source is extracted from the estimation deviation of the rotor offset coordinate parameters. The current response variation characteristics are processed based on the positioning accuracy driving source to generate a non-impact starting excitation current instruction.

[0036] Specifically, the rotor position estimation is extracted from the current response variation characteristics, and the rotor offset coordinate parameters are extracted. Based on the electromagnetic force balance principle, the radial offset distance is calculated according to the amplitude variation parameter in the current variation characteristics. The radial X-direction offset is calculated by the current difference of the X positive and negative coils, Δx=(iX+-iX-)×Kpos, where iX+ is the X positive coil current, iX- is the X negative coil current, and Kpos is the position conversion coefficient (0.5 mm / A). The radial Y-direction offset is determined by the current difference of the Y positive and negative coils, Δy=(iY+-iY-)×Kpos, and the offset positive and negative Y-directions represent the direction of deviation from the center. The axial Z-direction offset is calculated by the current ratio of the upper and lower thrust coils, Δz=(iup-idown) / (iup+idown)×gap, where iup is the upper thrust coil current, idown is the lower thrust coil current, and gap is the axial air gap length (2 mm). The position estimation considers the phase offset characteristics of the current response, and the phase lead indicates that the rotor position lags behind the control command, which needs to be compensated and corrected. The frequency modulation characteristics are used to identify the dynamic offset mode of the rotor, and the modulation below 100 Hz corresponds to slow drift, and the modulation between 100-500 Hz corresponds to rapid vibration. The time resolution of the offset coordinate reaches 50 microseconds, and the discrete estimation points are connected into a continuous position trajectory through interpolation processing. The rotor offset coordinate parameters include three-dimensional position coordinates (Δx, Δy, Δz), offset velocity vector, and offset acceleration, etc. Complete information. The estimated rotor offset coordinate parameters are compared with the expected position to calculate the estimation deviation ε=(ΔX-X0)²+(ΔY-Y0)²+(ΔZ-Z0)², where ΔX, ΔY, and ΔZ are the aforementioned calculated offsets, and X0, Y0, and Z0 are the expected positions (usually 0).

[0037] In some embodiments, the positioning accuracy driving source is extracted from the estimation deviation of the rotor offset coordinate parameters, including: performing sliding window analysis on the estimation deviation of the rotor offset coordinate parameters to obtain local deviation peaks; arranging the local deviation peaks in time sequence and performing gradient analysis to determine the deviation change trend; identifying the deviation turning feature based on the slope mutation point of the deviation change trend; and generating the positioning accuracy driving source according to the time distribution density of the deviation turning feature.

[0038] The estimation deviation of the rotor offset coordinate parameter is analyzed by a sliding window to obtain a local deviation peak value. The calculated estimation deviation value ε is locally analyzed by a sliding window method to identify the peak value points in the deviation sequence. The length of the sliding window is set to 200 sampling points, corresponding to a time span of 10 milliseconds, and the step length is 20 sampling points, ensuring an overlap rate of 85%. The deviation data in the window is identified by an extreme value detection method, and when the deviation value of a point is greater than that of the five points before and after it, it is marked as a local peak value. The threshold for determining the peak value is dynamically adjusted according to the statistical characteristics of the deviation in the window, and the threshold is set to the mean value of the window plus 2 times the standard deviation. The radial deviation peak value reflects the extreme position of the rotor whirl, and the periodicity of the peak value corresponds to the whirl frequency. Non-periodic peak values indicate the presence of random disturbances. The axial deviation peak value corresponds to the extreme point of axial vibration, and continuous peak values indicate the presence of sustained axial instability. The amplitude of the peak value is classified into three levels: first-level peak value (>100 microns), second-level peak value (50-100 microns), and third-level peak value (20-50 microns). Each peak value records its occurrence time, amplitude, duration, and other attribute information such as the coordinate axis it belongs to. The spatial distribution of the local deviation peak value is represented by a three-dimensional scatter plot, and the peak value distribution in the X-Y plane reflects the radial positioning characteristics, and the Z-axis peak value distribution reflects the axial control characteristics.

[0039] The local deviation peak values are arranged in time sequence and gradient analysis is performed to determine the deviation trend. The identified local deviation peak values are arranged in time sequence, and the trend of the peak value sequence is analyzed by gradient calculation. The gradient analysis uses the central difference method to calculate, gi=(Pi+1-Pi-1) / (2Δt), where Pi is the amplitude of the i-th peak value, Pi+1 and Pi-1 are adjacent peak values, Δt is the peak value time interval, and gi is the gradient value. A positive gradient indicates that the deviation peak value is increasing, and the system positioning accuracy is deteriorating. A negative gradient indicates that the deviation peak value is decreasing, and the positioning accuracy is improving. The absolute value of the gradient reflects the rate of change, and when the absolute value of the gradient exceeds 10 microns / s, it indicates rapid change and requires timely adjustment of the control parameters. Continuous multiple positive gradients (more than 5) indicate that the system may enter an unstable state and preventive measures need to be taken. Statistical analysis of the gradient sequence extracts trend features, including gradient mean, gradient variance, and gradient autocorrelation function. The gradient mean reflects the overall trend, and a positive value indicates that the deviation is increasing overall, and a negative value indicates that the deviation is decreasing overall. The gradient variance reflects the degree of change, and the larger the variance, the more unstable the system. The autocorrelation function identifies the periodicity of the gradient sequence, and the time delay corresponding to the correlation peak reflects the characteristic period of the deviation change. The piecewise fitting of the trend uses the least squares method to divide the gradient sequence into rising, stable, and falling segments, and each segment is approximated by a linear function.

[0040] The slope mutation point of the deviation trend is used to identify the turning feature of the deviation. The trend of the deviation is analyzed based on the gradient sequence, and the key feature point where the trend turns is identified by detecting the slope mutation. The second-order difference method is used to detect the slope mutation point, and the mutation position is identified by calculating the rate of change of the gradient. When the rate of change is more than 5 microns per second, it is determined as a mutation point. The mutation point divides the evolution process of the deviation into different stages, and each stage corresponds to different system states or control modes. The rising turning point marks the beginning of the rapid increase of the deviation, which usually corresponds to the introduction of the disturbance or the degradation of the control performance. The falling turning point indicates that the deviation begins to decrease, which corresponds to the effectiveness of the control action or the elimination of the disturbance. The turning points appearing continuously form an oscillation mode, and the oscillation period and amplitude reflect the dynamic characteristics of the system. The extraction of the turning feature includes parameters such as turning angle, turning strength and turning duration. The turning angle is calculated by the difference of the slope before and after the turning, and the larger the angle, the more severe the turning. The turning strength is reflected by the curvature at the turning point, and the larger the curvature, the more acute the turning. The turning duration is the time span from the start of the turning to the completion of the turning, which reflects the response speed of the system. The turning features are classified according to the frequency and impact degree, and the high-frequency turning reflects the rapid adjustment process, and the low-frequency turning reflects the slow drift process.

[0041] The positioning accuracy driving source is generated according to the time distribution density of the deviation turning feature. The time distribution density of the deviation turning feature is calculated by using the kernel density estimation method, and a Gaussian kernel function is placed at each turning point position. The continuous density distribution is obtained by superposition. The high-density area (more than 10 per second) indicates that the turning occurs frequently, which corresponds to the unstable period or key transition period of the system. The low-density area (less than 1 per second) indicates that the system runs smoothly, and the positioning accuracy is relatively stable. The periodic analysis of the density peak value identifies the inherent oscillation mode of the system, and the oscillation with a period in the range of 0.1-1 second is usually related to the control loop. The driving source is constructed by integrating the density distribution, turning strength and trend information, forming a multi-dimensional feature description. The main driving component corresponds to the frequency component with the highest density, and the secondary driving component corresponds to the secondary high-density area. The strength of the driving source is quantified by density integration, and the strength value is normalized to the range of 0-1. The strength value greater than 0.7 is strong driving, the strength value between 0.3 and 0.7 is moderate driving, and the strength value less than 0.3 is weak driving. The generated positioning accuracy driving source contains complete information such as time characteristics, frequency characteristics, strength distribution and phase relationship.

[0042] The current response change feature is processed based on the positioning accuracy driving source to generate a non-impact starting excitation current command. The key to non-impact starting is to eliminate the identified deviation components in the driving source, especially the low-frequency drift and transient impact components. The processing process reversely compensates the deviation amplitude spectrum of the driving source, and when a positioning deviation of 10 microns is detected at a certain frequency, a corresponding compensation component is superimposed in the excitation current. The amplitude of the compensation current is calculated as icomp=-ε×Kcomp×H(f), where ε is the deviation value, Kcomp is the compensation gain (0.02A / μm), and H(f) is the frequency-dependent transfer function. Phase compensation is achieved through a reverse phase process, ensuring that the compensation current and the deviation signal are 180 degrees out of phase, and the deviation is offset. The current command of the starting process uses an S-shaped curve planning, the starting segment has a slope of 0.1A / s, the middle segment gradually increases to 1A / s, and the end segment again decreases to 0.1A / s. The time parameters of the S-shaped curve are determined according to the system response time identified by the driving source, and the total starting time is set to 5-10 times the system time constant. The radial excitation current command includes two parts: the basic suspension current and the dynamic compensation current. The basic current provides static suspension force, and the compensation current eliminates dynamic deviation. The axial excitation current command considers the gravity preload, and the initial current of the upper thrust coil is greater than that of the lower thrust coil, and the difference is equal to the current value corresponding to the rotor gravity. The command generation process monitors the current change rate in real time and limits it within 2A / s to avoid mechanical impact caused by too fast current change.

[0043] Step S130, based on the non-impact starting excitation current command, a fusion data network is established to monitor the operation state of multiple sensors, and an abnormal response activation signal is used to rebuild the configuration of the fusion data network. Based on the signal reconstruction configuration, a degraded operation fault-tolerant mode is generated in combination with the current response change feature.

[0044] Specifically, a fusion data network is established based on the monitoring of the operating state of multiple sensors, including displacement sensors, acceleration sensors, temperature sensors, and current sensors, each type of sensor having multiple measurement points deployed at different locations in the magnetic bearing system. The displacement sensors monitor the actual position change of the rotor after the execution of the excitation current command, and compare the measured displacement with the expected displacement. A deviation of more than 5 microns indicates an abnormality. The acceleration sensors detect the vibration response during the startup process. When the excitation current changes in an S-shaped curve, the acceleration should be kept within 0.1g, and an overrun indicates an impact. The temperature sensors monitor the coil temperature rise. The temperature rise rate of the non-impact start should be less than 2℃ / min, and an excessive temperature rise indicates abnormal loss. The current sensors track the followability of the actual current to the command current, and the tracking error should be less than 2% of the command value. An excessive error indicates a problem in the control loop. The fusion data network is established using a star topology structure, with the excitation current command as the central node and the sensor data as the peripheral nodes. The connection weight between nodes is determined according to the sensitivity of the sensor to the command execution effect. The displacement sensor weight is 0.4, the acceleration sensor weight is 0.3, the temperature sensor weight is 0.2, and the current sensor weight is 0.1. The data fusion uses a weighted average method, and the fusion value F = Σ(wi×si), where wi is the weight of the ith sensor, and si is the normalized value of the sensor measurement. The time synchronization accuracy of the network reaches 100 microseconds, ensuring the alignment of different sensor data on the time axis. When the fusion value F deviates from the normal range or the change rate exceeds the set threshold, an abnormal response signal is generated.

[0045] In some embodiments, the signal reconstruction configuration activated by the abnormal response in the fusion data network includes: utilizing the abnormal response detection sensor in the fusion data network to drift mode; performing multi-channel collaborative calibration through the sensor drift mode to form a collaborative calibration trajectory sequence; implementing spatial reconstruction mapping on the collaborative calibration trajectory sequence to obtain a sensor compensation coordinate set; and forming a signal reconstruction configuration based on the sensor compensation coordinate set.

[0046] Anomaly response detection sensor drift pattern using fusion data network. The detection of sensor drift is achieved by analyzing the long-term trend of the fusion value F, when F is monotonically changing over a certain period of time, it is determined that there is drift. The drift pattern is divided into linear drift and nonlinear drift, the linear drift is manifested as the output value deviates at a constant rate, the change rate of nonlinear drift changes with time. The slope of linear drift is determined by the least square fitting method, the covariance of each time point and the fusion value is calculated to obtain the drift rate. Nonlinear drift uses polynomial fitting, usually using quadratic or cubic polynomial, the coefficients are determined by minimizing the fitting error. Typical drift scenarios include displacement sensors with probe surface contamination leading to slow output increase, showing positive linear drift; temperature sensor due to aging causes the characteristics of thermistor to change, the output presents nonlinear drift characteristics; acceleration sensor due to base loosening produces low frequency vibration superposition, causing periodic drift of output baseline. The identification of drift direction is realized by sign judgment, positive drift indicates that the output value increases, negative drift indicates that the output value decreases. The correlation analysis of multi-sensor drift identifies common mode drift and differential mode drift, common mode drift is usually caused by external factors such as environmental temperature change, all sensors show the same trend; differential mode drift is usually caused by individual sensor aging, the drift direction and rate of different sensors are different. The characteristic parameters of drift pattern include drift start time, drift rate, drift acceleration and predicted failure time.

[0047] Multi-channel collaborative calibration is performed through sensor drift pattern to form a collaborative calibration trajectory sequence. Collaborative calibration uses linear or nonlinear characteristics in drift pattern to establish calibration relationship between adjacent channels, and the channel with smaller drift is used as reference. For linear drift, the slope compensation is used in the calibration process, and the calibration value Ci = Mi - k × (t - t0), where Mi is the measured value, k is the drift slope, t is the current time, and t0 is the drift start time. For nonlinear drift, polynomial compensation is used in the calibration process, and the compensation amount at each time is calculated according to the fitted polynomial function. Multi-channel collaboration is realized through cross comparison, and the calibration value of each channel is compared with the average value of other channels, and the deviation is used to correct the calibration parameter. The generation of collaborative calibration trajectory records the calibration value of each channel in time sequence, forming a two-dimensional trajectory array, with rows corresponding to time points and columns corresponding to different channels. The sampling interval of the trajectory sequence is set to 1 second, and 100 consecutive points are recorded, covering a 100-second calibration process. The smoothing of the trajectory is realized by using the moving average method, and the window length is 5 sampling points, which eliminates the random fluctuations in the calibration process. The calibration effect is evaluated by residual analysis, and the residual less than 10% of the original drift indicates that the calibration is effective.

[0048] The spatial reconstruction mapping is implemented on the collaborative calibration trajectory sequence to obtain a set of sensor compensation coordinates. Spatial reconstruction mapping projects the calibration trajectory in the time domain to the spatial coordinate system, and each trajectory point corresponds to a spatial compensation vector. The mapping relationship is realized through a transformation matrix T, where Xc, Yc, Zc are compensation coordinates, C1-C4 are calibration values of four channels, and T is a 4x3 transformation matrix. The transformation matrix is determined using the geometric arrangement of the sensors. Orthogonal sensors correspond to an orthogonal transformation, and non-orthogonal sensors require an oblique transformation. The physical meaning of the compensation coordinates corresponds to the correction amount of the rotor position. The X and Y direction compensation coordinates correct radial position errors, and the Z direction compensation coordinates correct axial position errors. The construction of the coordinate set collects all compensation coordinates at different times into a point cloud form. The point cloud density reflects the time resolution of the calibration. The determination of the compensation range is based on the envelope analysis of the point cloud. The maximum compensation amount is usually limited to within 5% of the sensor range. The spatial interpolation uses a trilinear interpolation method to generate a continuous compensation field between discrete compensation points. The data structure of the compensation coordinate set includes coordinate values, time labels, effective ranges, and application priorities.

[0049] The signal reconstruction configuration is formed based on the sensor compensation coordinate set. The Xc, Yc, and Zc values in the compensation coordinate set are used as correction parameters and applied to the original output of the corresponding sensor. The reconstruction rule is Srecon = Sraw + Sc, where Srecon is the reconstructed signal, Sraw is the original signal, and Sc is the compensation value extracted from the coordinate set. The selection of the compensation value is based on the current time and sensor position, and the nearest compensation point is found in the set or calculated by interpolation. The configuration scheme includes a main configuration and a backup configuration. The main configuration uses the complete compensation coordinate set, and the backup configuration uses a simplified linear compensation. The configuration parameters include compensation enable flags, compensation intensity coefficients, and compensation range limits. The intensity coefficient is usually set to 0.8-1.0. The signal quality index is evaluated by the improvement of the signal-to-noise ratio before and after compensation. A signal-to-noise ratio improvement of more than 3dB indicates effective reconstruction. The effective conditions for the configuration include sensor drift exceeding the threshold, drift duration exceeding the set value, and other sensor confirmation of abnormality. The storage of the reconstruction configuration uses a structured format, including configuration version, generation time, validity period, and application record management information. The switching of the configuration uses a double buffering method. The new configuration is switched instantly after preparation in the background, avoiding the impact of the reconstruction process on real-time control.

[0050] The degraded operation fault-tolerant mode is generated based on the signal reconstruction configuration combined with the current response change characteristics. The degraded operation mode determines the degradation level according to the type of the signal reconstruction configuration. The parameter adjustment configuration corresponds to the first level degradation (10% performance reduction), the sensor switching configuration corresponds to the second level degradation (30% performance reduction), and the system reconstruction configuration corresponds to the third level degradation (50% performance reduction). The fault-tolerant mode takes the key parameters in the current response change characteristics as supplementary information sources. When the displacement sensor fails, the position is calculated from the current change using the current-displacement conversion relationship. The first level degradation mode maintains the original control bandwidth and only adjusts the control parameters, reducing the proportional gain by 20% and increasing the integral time by 50% to ensure system stability. The second level degradation mode reduces the control bandwidth to 70% of the original value, reducing the dynamic response speed but improving the robustness, and limiting the maximum speed to 80% of the rated value. The third level degradation mode adopts a conservative control strategy, reducing the control bandwidth to 50% of the original value, limiting the speed to 60% of the rated value, and increasing the air gap set value by 20% to provide a larger safety margin. The switching conditions of the fault-tolerant mode include the number of sensor failures, the duration of the failure, and the system performance indicators. Any condition that meets the corresponding degraded mode is triggered. The mode conversion uses a smooth transition method, gradually adjusting the parameters within 5 seconds to avoid transient disturbances caused by mode switching. Each degraded mode retains the ability to degrade to a lower level, ensuring that the system can maintain basic functions in multiple failure conditions.

[0051] In step S140, the degraded operation fault-tolerant mode receives the shutdown instruction signal, converts the shutdown instruction signal into a response preparation gain, and evaluates the shutdown state based on the signal reconstruction configuration to generate an emergency shutdown trigger condition.

[0052] Specifically, the shutdown instruction signal is received by the degraded operation fault-tolerant mode. According to the degradation level of the degraded operation fault-tolerant mode, the receiving strategy of the shutdown instruction is determined, the first level degradation mode keeps the normal instruction receiving channel, the second level degradation enables the redundant receiving channel, and the third level degradation adopts multi-channel voting reception. The shutdown instruction signal contains three types of normal shutdown, fast shutdown and emergency shutdown, each type corresponds to different priority and execution timing. The normal shutdown instruction is received through the main control channel of the fault-tolerant mode, and the signal format is a continuous high level of more than 100 milliseconds, indicating a planned shutdown operation. The fast shutdown instruction is received through the main control and standby channels at the same time, and the signal is a double pulse sequence with a pulse interval of 50 milliseconds, indicating that the shutdown needs to be completed within 30 seconds. The emergency shutdown instruction is received through all available channels, and the signal is a continuous square wave with a frequency of 10 Hz, indicating that the shutdown is executed immediately. The fault-tolerant mode determines the validity of the received instruction, and requires that the signal strength reaches 90% of the threshold in the first level degradation, is reduced to 70% in the second level degradation, and is reduced to 50% in the third level degradation. The time stamp of the instruction reception is accurate to the millisecond level, and records the instruction arrival time, confirmation time and start execution time. The instructions received by multiple channels are aligned through a time window, and the window width is 10 milliseconds, which ensures that the same instruction of different channels is correctly identified.

[0053] In some embodiments, the converting the shutdown instruction signal into a response preparation gain comprises: decomposing the shutdown instruction signal into a dynamic delay sequence according to a time window; performing buffer state analysis on the dynamic delay sequence to generate a buffer time matrix; phase matching the dynamic delay sequence with the buffer time matrix to form a timing buffer diagram; and extracting a preparation enhancement node from the timing buffer diagram to determine the response preparation gain.

[0054] The shutdown instruction signal is decomposed into a dynamic delay sequence by time window. The received shutdown instruction signal is processed by time window segmentation, and the continuous signal is decomposed into sequence segments with different delay characteristics. The setting of the time window is determined according to the characteristics of the instruction signal, and the normal shutdown instruction uses a 100 ms window, the fast shutdown uses a 50 ms window, and the emergency shutdown uses a 20 ms window. The window decomposition uses a sliding mode, and the adjacent windows overlap by 50%, ensuring that the continuity of the signal characteristics is not damaged. The signal segment in each window extracts its start time, peak time and end time to form a time feature triple of the segment. The calculation of the dynamic delay compares the actual arrival time of each signal segment with the ideal arrival time, and the delay value is equal to the actual time minus the ideal time. The delay sequence is arranged in the order of the window to form a time sequence reflecting the delay of the instruction transmission and processing. The delay values in the sequence show dynamic change characteristics, with large initial segment delay (10-20 ms), stable middle segment delay (5-10 ms), and possibly increased end segment delay. Abnormal delay is identified by comparing it with the average delay, and delay points that are more than twice the average value are marked as abnormal points. The length of the dynamic delay sequence is equal to the number of window decomposition segments, and the typical value is 10-50 segments, depending on the duration of the instruction signal. The sequence stores attribute information including delay value, window index, signal intensity and abnormal marker.

[0055] The dynamic delay sequence is analyzed to generate a buffer time matrix. Buffer state analysis checks the maximum buffer time that each delay segment in the dynamic delay sequence can accommodate, and the buffer time is equal to the system response deadline minus the current cumulative delay. The buffer capacity of the ith segment Bi=Tdeadline-∑(Dj), where Tdeadline is the shutdown deadline, Dj is the delay of the jth segment, and the sum is from 1 to i. The construction of the buffer time matrix organizes the buffer capacity of each segment into a two-dimensional structure, with rows corresponding to delay sequence segments and columns corresponding to different buffer strategies. The first column of the matrix is the basic buffer time, which is directly equal to the buffer capacity value calculated for each segment. The second column is the safety buffer time, which reserves a 20% safety margin based on the basic buffer. The third column is the limit buffer time, which considers the maximum processing capacity of the system, usually 1.5 times the basic buffer. The classification of buffer state is based on the adequacy of buffer time, sufficient state (>100 ms), normal state (50-100 ms), tense state (20-50 ms), and critical state (<20 ms). Negative values in the matrix indicate that the segment has exceeded the buffer capacity and needs to be processed immediately or skipped some steps. The buffer time matrix has dimensions N×3, where N is the length of the delay sequence and 3 corresponds to the three buffer strategies. Matrix updating is done in real time, with each segment processed to update the buffer state of the subsequent segment.

[0056] The phase matching of the dynamic delay sequence and the buffer time matrix forms a timing buffer map, which includes converting the dynamic delay sequence into a virtual oscillation signal, making the virtual oscillation signal propagate in the buffer time matrix to generate an interference pattern, extracting the enhanced nodes and the reduced nodes in the interference pattern, and encoding the spatial distribution of the enhanced nodes and the reduced nodes into a timing buffer map.

[0057] The dynamic delay sequence is converted into a virtual oscillation signal. The calculated delay values (the difference between actual time and ideal time) in the dynamic delay sequence are mapped into a form of periodic oscillation signal to capture the dynamic characteristics of the delay. The conversion process maps the delay value Di of the ith segment in the sequence into an amplitude Ai=A0×(Di / Davg), where A0 is a reference amplitude (set to 1) and Davg is the average of all delay values, and this linear mapping preserves the relative size relationship of the delay. The oscillation frequency is determined according to the delay change rate, and segments with a large change rate ΔDi / Δt correspond to high-frequency components (representing rapid changes), where ΔDi is the difference between adjacent delay values and Δt is the time interval, and segments with a small change rate correspond to low-frequency components (representing slow changes). The phase of the virtual oscillation signal is calculated according to the delay cumulative effect, φi=2π×(ΣDj / ΣDmax), where φi is the phase of the ith segment, ΣDj is the cumulative sum of delays from the first segment to the ith segment, and ΣDmax is the cumulative sum of all possible maximum delays, reflecting the cumulative degree of the delay. The envelope of the signal is obtained by connecting the local extreme points, and the upper envelope corresponds to the evolution trend of the delay peak value, and the lower envelope corresponds to the evolution trend of the delay valley value. The modulation characteristics of the oscillation signal reflect the sudden changes of the delay, and a modulation sudden change marker is generated when the change between adjacent delay values exceeds 50%. The complete expression of the signal contains four components: amplitude sequence, change rate sequence, cumulative phase sequence, and envelope function. The time resolution of the virtual signal is consistent with the original sequence, and each delay value corresponds to an oscillation sampling point. The generated virtual oscillation signal serves as another representation of the delay sequence, preserving all timing information.

[0058] The virtual oscillation signal is propagated in the buffer time matrix to generate the interference pattern. The virtual oscillation signal is taken as the input and propagated in the calculation space composed of the buffer time matrix. The propagation process takes each element Bij of the matrix as the propagation coefficient, and the output of the oscillation signal when passing through this position is Oij=Ai×Bij, where Oij is the output signal intensity of the i-th row and j-th column, Ai is the oscillation amplitude of the i-th section, and Bij is the buffer time value of the i-th row and j-th column of the buffer time matrix. The propagation coefficient Bij reflects the buffer capacity at this time. The greater the buffer time, the greater the propagation coefficient, and the more sufficient the signal transmission. The oscillation signal starts from the first row of the matrix and propagates down row by row, generating a new output signal every time it passes a row. When the signal encounters a buffer time change point, the output signal bifurcates: part of it continues to propagate (transmission component), and part of it returns (reflection component). The transmission component Tij=Oij×(Bij / Bij-1)^0.5, and the reflection component Rij=Oij×(1-Bij / Bij-1)^0.5, where Tij is the transmission intensity, Rij is the reflection intensity, and Bij-1 is the buffer time of the same column in the previous row. When Bij>Bij-1, it is mainly transmission, and vice versa, it is mainly reflection. The forward propagating signal and the reflected signal superimpose when they are at the same position to produce interference effect, and the superposition result Iij=Tij+Rij×cos(φi-φj), where Iij is the total interference intensity, φi and φj are the phases of the two signals, and cos is the cosine function. The positive and negative of the interference intensity depends on the phase relationship. When in phase (cos value is positive), it produces enhancement, and when out of phase (cos value is negative), it produces weakening. The interference pattern forms a strong and weak distribution in the matrix space, and the strong interference area corresponds to the period of coordinated system response, and the weak interference area corresponds to the period of response conflict.

[0059] The enhanced nodes and the weakened nodes in the interference pattern are extracted. The identification of the enhanced nodes is realized by threshold judgment. When the interference intensity Iij is greater than the average intensity plus a standard deviation, it is marked as an enhanced node. The judgment standard of the weakened nodes is that the interference intensity is less than the average intensity minus a standard deviation. The system response ability at these positions is weak. The intensity level of the nodes is classified according to the degree of deviation from the mean value. The deviation of more than 2 standard deviations is a strong node, 1-2 standard deviations is a medium node, and within 1 standard deviation is a weak node. The spatial distribution characteristics of the nodes include the node density (the number of nodes / the total number of grids), the average spacing, and the aggregation degree. The connectivity analysis of the enhanced nodes identifies the node group. The continuous enhanced nodes form an enhanced channel, indicating a continuous high response ability period. The distribution pattern of the weakened nodes identifies the weak link of the system. The concentrated appearance of the weakened nodes needs to be focused on and compensated. The time stability of the nodes is evaluated by comparing the node state at adjacent moments. The state remains unchanged for stable nodes, and frequently changes for oscillating nodes. The identification of key nodes is based on the influence range of the nodes. The influence range is quantified by the degree of change of the interference pattern after removing the node. The statistics of the node characteristics include the number, proportion, distribution variance, and other indicators of each type of node.

[0060] The spatial distribution of the enhanced nodes and the weakened nodes is encoded into a time-sequential buffer map. The enhanced nodes are encoded according to the intensity level. The strong enhanced nodes are encoded as 3, the medium enhanced nodes are encoded as 2, and the weak enhanced nodes are encoded as 1. The encoding value reflects the contribution of the node to the system response. The weakened nodes are encoded with negative values. The strong weakened nodes are encoded as -3, the medium weakened nodes are encoded as -2, and the weak weakened nodes are encoded as -1. The negative value represents the weakening effect on the system response. The neutral area (neither enhanced nor weakened) is encoded as 0, indicating that the buffer state at this position is in balance. The spatial position is encoded using a two-dimensional coordinate. Each encoded point is represented as a three-tuple (i, j, Cij), where i is the row number, j is the column number, and Cij is the encoding value of the position (an integer between -3 and +3). The data structure of the time-sequential buffer map uses sparse representation, only recording non-zero encoding points, reducing storage requirements and highlighting key information. The time dimension of the graph is represented by the third coordinate axis, forming a three-dimensional spatiotemporal encoding (i, j, t, Cij), where t is the time index. The encoding density p = Nnonzero / Ntotal reflects the activity level of the system, where p is the encoding density, Nnonzero is the number of non-zero encoding points, and Ntotal is the total number of grid points. A high density indicates that the buffer state changes frequently. The spatial autocorrelation function of the encoding calculates the correlation of adjacent encoding points. Strong correlation indicates that the buffer state has spatial continuity. The integrity of the time-sequential buffer map is guaranteed by interpolation algorithms. The missing encoding points are filled using nearest neighbor interpolation or linear interpolation. The generated time-sequential buffer map contains three parts: the encoding matrix, the statistical characteristics, and the metadata.

[0061] The response preparation gain is determined by extracting the preparation enhancement nodes from the time buffer graph. The preparation enhancement nodes are identified by finding the peak points of the buffer capacity in the time buffer graph, which correspond to the moments when the system response capacity is the strongest. The screening conditions for the nodes include three requirements: the buffer time is greater than 1.5 times the average value, the duration is more than 20 milliseconds, and the state before and after is stable. The strength of the enhancement node is quantified by its prominence in the graph, and the strength value is equal to the ratio of the buffer capacity of the node to the average value around it. The calculation of the response preparation gain integrates the contributions of all enhancement nodes, G = ∑(Si × Wi) / ∑(Wi), where Si is the strength of the i-th node, and Wi is the node weight. The node weight is determined according to its importance in the shutdown process, with an early node weight of 0.3, a medium node weight of 0.5, and a late node weight of 0.2. The gain value is modified considering the uniformity of the node distribution, with a reward factor of 1.1 when the node distribution is uniform, and a penalty factor of 0.9 when the node distribution is concentrated. The final response preparation gain is standardized to the range of 0-2, with 0 indicating complete unpreparedness, 1 indicating normal preparation, and 2 indicating sufficient preparation. The gain value is accompanied by a credibility score based on the number and quality of the enhancement nodes, with more nodes and higher strength resulting in higher credibility.

[0062] The emergency shutdown trigger condition is generated by evaluating the shutdown state according to the response preparation gain combined with the signal reconstruction configuration. The response preparation gain is used as a weight factor to evaluate each parameter in the signal reconstruction configuration. When the gain value is greater than 1.8 and the reconstruction configuration is in the system reconstruction state, the evaluation result tends to immediate shutdown. The evaluation indicators include four aspects: the current rotor speed, the vibration amplitude, the bearing temperature, and the control margin. Each indicator determines its weight in the evaluation according to the gain value. The speed evaluation considers the time requirement from the current speed to zero, with higher gain allowing shorter deceleration time, and a gain of 2.0 requiring shutdown within 10 seconds. The vibration evaluation checks whether the current vibration level allows normal shutdown process, with vibration exceeding 80% of the allowable value requiring adjustment of the shutdown strategy. The temperature evaluation judges whether there is a risk of overheating of the bearing, with temperature exceeding the alarm value triggering a rapid cooling program. The control margin evaluation is determined by the compensation ability of the reconstruction configuration, with the compensation coordinate set covering less than 50% of the required range considered as insufficient margin. The emergency shutdown trigger condition is generated by integrating the evaluation results, with three trigger thresholds set: any two indicators exceeding the limit, the gain value suddenly increasing by more than 0.5, and the reconstruction configuration failing. The trigger condition includes parameters such as trigger type, trigger threshold, delay time, and execution priority. The logical combination of the conditions uses the "or" relationship, and any condition being met starts the emergency shutdown program.

[0063] In step S150, the bus capacitor power supply state is detected according to the emergency stop trigger condition, the gradient control potential is released from the energy change of the bus capacitor power supply state, the active damping control opportunity is determined based on the gradient control potential combined with the degraded operation fault-tolerant mode, the damping force control instruction is generated through the active damping control opportunity, and the intelligent start-stop control of the magnetic suspension refrigeration compressor is realized.

[0064] Specifically, the bus capacitor power supply state is detected according to the emergency stop trigger condition. The detection of the bus capacitor power supply state includes four key parameters of capacitor voltage, discharge current, energy storage capacity and internal resistance change. The capacitor voltage is measured by a high-precision voltage sensor, and the normal working voltage is 540V. When the emergency stop trigger condition is activated, the voltage drop curve is recorded. The discharge current monitors the instantaneous current of the capacitor power supply to the magnetic bearing system. The current is in the range of 10-20A during normal shutdown, and may reach 30-50A during emergency shutdown. The energy storage capacity is obtained by integral calculation, E=0.5xCV2, where C is the capacitance value (typical value 10000μF), V is the current voltage, and E is the energy storage. The internal resistance change reflects the health status of the capacitor, which is calculated by the ratio of voltage drop to current. The increase of internal resistance indicates the aging of the capacitor. The detection frequency is adjusted according to the emergency degree of the trigger condition. The sampling frequency is 1kHz during normal trigger, and is increased to 10kHz during emergency trigger. The grading evaluation of power supply state divides the capacitor state into four levels of sufficient (voltage>500V), normal (450-500V), warning (400-450V) and critical (<400V). The identification of abnormal state includes voltage drop (decrease rate>50V / s), current peak (peak value>80A) and oscillation discharge (voltage fluctuation>10V).

[0065] In some embodiments, the gradient control potential is released from the energy change of the bus capacitor power supply state, including: utilizing the bus capacitor power supply state to monitor the voltage decay gradient mode; performing capacitor discharge optimization control through the voltage decay gradient mode to form an optimized discharge area; implementing hierarchical current regulation on the optimized discharge area to obtain a power current distribution; and forming a gradient control potential based on the power current distribution.

[0066] The voltage decay gradient mode is monitored in the bus capacitor power supply state. By analyzing the voltage variation characteristics in the bus capacitor power supply state, the gradient change mode of voltage decay over time is identified. The voltage decay monitoring starts from the emergency shutdown trigger time, and the voltage value is recorded every millisecond to form a high-density voltage time series. The five-point difference method is used to calculate the decay gradient, which improves the accuracy and noise immunity of the gradient calculation. The gradient mode classification includes three basic types: linear decay, exponential decay, and step decay. Linear decay shows a constant voltage drop rate, usually occurring under constant load conditions, with a decay rate of 20-30V / s. Exponential decay follows the rule V(t)=V0×exp(-t / τ), where V0 is the initial voltage and τ is the time constant, reflecting the natural discharge characteristics of RC circuits. The step decay corresponds to the step switching of the load, with each step load corresponding to a different decay rate, and there is a plateau between steps. The mixed mode identifies the combination of multiple modes in the actual decay process, which may be exponential decay at the beginning, linear at the middle, and step characteristics at the end. The detection of abnormal decay modes includes oscillatory decay (voltage fluctuation), sudden decay (voltage instantaneous drop), and abnormal rise (voltage temporary rise). The mode feature parameter extraction includes average decay rate, decay acceleration, mode transition point, and decay stability.

[0067] The optimization discharge region is formed by performing capacitor discharge optimization control based on the voltage decay gradient mode. According to the identified voltage decay gradient mode, differential discharge control is performed: linear decay mode adopts constant current discharge strategy, maintaining 20A constant current to ensure smooth power supply; exponential decay mode adopts variable current discharge, with high current of 40A in the initial stage to quickly respond, and then gradually reducing to 10A to maintain stability; the step decay mode adjusts the discharge parameters at each voltage step, and increases the current compensation during the step interval. Based on the discharge characteristics of different gradient modes, the optimization region is divided: the high voltage region (>480V) corresponds to the constant current control of the linear mode, the medium voltage region (420-480V) adapts to the variable current regulation of the exponential mode, and the low voltage region (360-420V) adopts the segmented control of the step mode. The setting of the region boundary is dynamically adjusted according to the gradient mode, and the boundary moves forward in the exponential decay mode and uniformly distributes in the linear decay mode. The optimization goals include minimizing shutdown time, ensuring shutdown stability, and reserving safety energy reserve. The discharge path selection prioritizes power supply to critical loads, with magnetic bearing control having the highest priority, followed by auxiliary systems. The region switching uses predictive control, which predicts the voltage trend 5 seconds in advance to smoothly adjust the discharge parameters.

[0068] The current flow distribution is obtained by implementing hierarchical current regulation on the optimized discharge regions. Each optimized discharge region is further subdivided into 3-5 current levels, each corresponding to a specific current range and control objective. The first level (base level) provides the minimum current to maintain suspension, ensuring that the rotor does not contact the backup bearing, with a current range of 5-10 A. The second level (control level) provides the current required for position regulation, enabling precise control of the rotor position, with a current range of 10-20 A. The third level (damping level) provides vibration suppression current, actively damping rotor vibrations, with a current range of 15-25 A. The fourth level (braking level) provides deceleration braking current, rapidly reducing rotor speed, with a current range of 20-35 A. The current flow distribution is obtained through a current sensor array, which monitors the current flow from the bus capacitor to each load branch. The radial magnetic bearing obtains 40-50% of the total current, used to maintain radial stability. The axial magnetic bearing obtains 20-30%, balancing axial forces and gravity. The control system and sensors obtain 10-15%, ensuring normal control functions. The auxiliary system obtains the remaining 10-20%, including cooling and monitoring functions.

[0069] The gradient control potential is formed based on the current flow distribution. The calculation of the gradient control potential is based on the current flow distribution characteristics of each branch: the radial branch carries 40-50% of the total current, corresponding to a controllable control potential of 20 A; the axial branch carries 20-30% of the total current, corresponding to a control potential of 15 A; the cooling branch carries 15-20% of the total current, corresponding to a control potential of 10 A; the auxiliary branch carries 5-10% of the total current, corresponding to a control potential of 5 A. The control potential Pi of each branch is determined according to the flow distribution ratio and current regulation capacity, and the total control potential P = Σ(Pi x ηi), where ηi is the control efficiency coefficient of each branch, the radial branch efficiency is 0.9, the axial branch efficiency is 0.8, and the cooling branch efficiency is 0.7. The time-varying characteristics of the potential decrease as the capacitor voltage decreases, and the potential decay rate is proportional to the voltage decay rate. The potential allocation strategy prioritizes key functions, with 50% of the potential allocated to suspension control, 30% to vibration suppression, and 20% to speed control. The management of potential reserves reserves 20% of the total potential as an emergency reserve for sudden disturbances or control deviations. The determination of the gradient direction is based on the difference between the current state and the target state, determining the priority direction of potential release. The potential evaluation index includes four dimensions: controllable range, response speed, duration, and stability margin.

[0070] The active damping control timing is determined based on the gradient control potential combined with the degradation operation fault-tolerant mode. The determination of the active damping control timing considers two core factors, the energy gradient level and the fault-tolerant mode level. When the gradient control potential P > 30A and the system is in the first fault-tolerant mode, the standard damping control is started; when P is in the range of 15-30A and in the second fault-tolerant mode, the enhanced damping control is started; when P < 15A and in the third fault-tolerant mode, the emergency damping control is started. The potential reserve is also considered in the timing determination, reserving 20% of the control potential for emergency adjustment, and the actual available potential is 80% of the total potential. The fault-tolerant mode level affects the timing selection strategy, and the damping is applied 10 milliseconds before the vibration peak in the first degradation mode, 20 milliseconds in the second degradation, and 30 milliseconds in the third degradation. The setting of the timing window considers the system response delay, and the typical window width is 50 milliseconds, and the optimal intervention point is found within this window. The determination conditions of the optimal intervention point include two requirements, the energy gradient is in the controllable interval and the control margin is sufficient. Multi-objective optimization determines the comprehensive optimal timing, and the objective function includes three items, vibration suppression effect, energy consumption and downtime. The dynamic adjustment of the timing is corrected according to real-time feedback, and if the vibration is not effectively suppressed, the timing is advanced by 5 milliseconds to try again. The duration of the damping control is determined according to the vibration attenuation situation, and the damping is gradually reduced when the vibration decreases to 120% of the target value, and the damping is removed when the vibration decreases to 105%.

[0071] In some embodiments, the generating the damping force control instruction through the active damping control timing comprises: monitoring a moment oscillation mode of the active damping control timing to generate a moment oscillation trajectory; injecting an anti-phase moment at a wave peak position of the moment oscillation trajectory to form a moment standing wave; extracting a deterministic moment component using a stable node of the moment standing wave; and generating the damping force control instruction by superimposing the deterministic moment component in phase.

[0072] The torque oscillation mode of monitoring active damping control opportunity generates torque oscillation trajectory. At the determined active damping control opportunity point, the torque oscillation characteristics of the rotor system are monitored. The monitoring of torque oscillation is calculated by the differential current of the radial magnetic bearing, torque M = (iX+ - iX-) x r, where M is the synthesized control torque, iX+ is the X positive coil current, iX- is the X negative coil current, and r is the force arm, i.e. the rotor radius, usually 0.05-0.08 m. The current difference (iX+ - iX-) generates a radial electromagnetic force, and the product with the force arm r forms the control torque. The identification of the oscillation mode analyzes the variation law of the torque with time, including the oscillation frequency, amplitude and phase characteristics. The primary oscillation frequency is usually equal to the rotor rotation frequency, the amplitude is in the range of 0.5-2.0 N·m, and the phase is related to the rotor angular position. The secondary oscillation components include 2 times frequency, 3 times frequency and other high harmonics, reflecting the influence of rotor imbalance and shape deviation. The generation of the torque oscillation trajectory connects the torque values in chronological order to form a two-dimensional trajectory curve, with the horizontal axis as time and the vertical axis as torque value. The envelope of the trajectory reflects the trend of oscillation energy, with the upper envelope connecting the peak points and the lower envelope connecting the valley points. The classification of the oscillation mode includes stable oscillation (constant amplitude), growing oscillation (increasing amplitude), decaying oscillation (decreasing amplitude) and chaotic oscillation (irregular). The labeling of the trajectory feature points includes peak points, valley points, zero-crossing points and inflection points, which determine the application position of the damping force. The statistical analysis of the oscillation period calculates the average period, period standard deviation and period variation trend.

[0073] In the wave peak position of the torque oscillation trajectory, an opposite phase torque is injected to form a torque standing wave. The wave peak position in the torque oscillation trajectory is identified, and an opposite phase torque is accurately injected at these positions to form a standing wave effect. The accurate positioning of the wave peak position is determined by the first derivative of the trajectory, and the derivative changes from positive to negative at the wave peak. The calculation of the opposite phase torque takes the negative value of the wave peak torque value, Manti = -Mpeak x a, where a is the injection coefficient (0.6-0.8) to avoid excessive compensation. The control of the injection timing takes into account the system response delay, and the injection starts 5-10 milliseconds before the wave peak to ensure that the opposite phase torque reaches its maximum value at the wave peak. The formation mechanism of the torque standing wave is the superposition of the original oscillation and the injected opposite phase torque, which cancels out at certain positions to form nodes and may enhance at other positions to form antinodes. The characteristics of the standing wave mode include fixed node position, stable amplitude of the antinode and regular energy distribution. The node position corresponds to the minimum vibration point, which is the most stable working point of the system and is preferred as the shutdown target position. The antinode position needs to be avoided to prevent amplification of vibration when the system is shut down at these positions.

[0074] The determination of the deterministic moment component by the stable node of the moment standing wave includes: performing oscillation energy analysis on the moment standing wave to obtain oscillation stability data; identifying a moment optimization node and a control optimization node in the oscillation stability data; performing parameter fusion based on the moment optimization node and the control optimization node to generate a moment coordination curve; and performing feature extraction analysis on the moment coordination curve to form the deterministic moment component.

[0075] The oscillation energy analysis is performed on the moment standing wave to obtain oscillation stability data. The energy distribution characteristics in the moment standing wave are analyzed. The standing wave is formed by superimposing an original oscillation moment and an anti-phase damping moment: Mw=Morig+Manti, where Morig is the original oscillation moment, and Manti is the injected anti-phase moment. The standing wave energy is calculated as Ew=∫Mw²dt, and the total energy of the standing wave is obtained by integration. The node position in the standing wave corresponds to the minimum energy, the antinode position corresponds to the maximum energy, and the node spacing reflects the stable period of the standing wave. The energy distribution analysis divides the standing wave space into multiple sections, calculates the energy density of each section, and the section with high density corresponds to the energy concentration area. The stability indicators include the energy fluctuation rate (standard deviation / mean), the energy attenuation rate (energy reduction per unit time), and the energy distribution uniformity. The energy fluctuation rate less than 10% indicates stable oscillation, 10-20% indicates critical stability, and more than 20% indicates instability. The calculation of the instantaneous energy uses Hilbert transform to obtain the envelope, and the square of the envelope is the instantaneous energy density. The energy flow analysis traces the energy transfer path between the standing wave nodes and the wave crests, and the energy flow of the stable standing wave forms a closed loop. The frequency energy spectrum is obtained by power spectral density analysis, which shows the proportion of energy distribution at each frequency component. The main frequency energy proportion more than 70% indicates a single stable oscillation mode, and the multi-frequency energy dispersion indicates the existence of complex coupling. The stability data organization includes three parts: time-domain stability indicators, frequency-domain energy distribution, and spatial energy density diagram.

[0076] The torque optimization nodes and control optimization nodes are identified in the oscillation stability data. The node identification comprehensively utilizes three components of the stability data: identifying stable time periods with torque amplitude changes less than 5% from the time-domain stability indicators, which correspond to potential torque optimization nodes; locating frequency points with main frequency energy concentration from the frequency-domain energy distribution, with frequencies within ±2 Hz around the main frequency corresponding to control optimization nodes; determining energy density minimum positions from the spatial energy density map, which are marked as optimal control intervention points. Cross-validation of the three types of data ensures the accuracy of node identification. Spatial distribution analysis of the nodes shows that torque optimization nodes are mainly located near standing wave nodes, with an interval of about half a wavelength. Control optimization nodes partially overlap with torque optimization nodes, but are more biased towards energy flow convergence points. The temporal characteristics of the nodes show that they appear periodically, with 2-4 torque optimization nodes and 1-2 control optimization nodes appearing in each oscillation period. The classification of node strength divides the nodes into primary nodes (top 20% of stability indicators), secondary nodes (20-50%), and weak nodes (50-80%). Primary nodes are the preferred control targets, secondary nodes are the alternatives, and weak nodes are only used when necessary. Correlation analysis of the nodes identifies the coupling relationship between the nodes, and control of some nodes can affect the stability of other nodes. The effective window of the optimization nodes is defined as the time range in which the node characteristics remain excellent, with a typical window width of 20-50 milliseconds.

[0077] Based on the torque optimization nodes and control optimization nodes, parameter fusion is performed to generate a torque coordination curve. Parameter fusion uses a weighted combination method, with a torque optimization node parameter weight of 0.6 and a control optimization node parameter weight of 0.4. The fusion rule takes the maximum weight at the node overlap and distributes the weight inversely proportional to the distance at the non-overlapping points. The generation of the torque coordination curve connects the fused nodes through spline interpolation, ensuring the second-order continuity of the curve. The shape characteristics of the curve include peak and valley distribution, slope change, and curvature characteristics, with smooth segments corresponding to stable control zones and steep segments corresponding to rapid adjustment zones. The coordination principle prioritizes system stability and pursues torque optimization under the premise of stability, with the stability weight increasing to 0.7 in conflict situations. Dynamic adjustment of the curve is based on real-time feedback correction, with the control optimization node weight increasing if oscillation intensifies, and the torque optimization node weight increasing if response is sluggish. Curve segmentation processing divides the complete curve into a start-up segment, a transition segment, a stable segment, and an end segment, each using different coordination strategies. The boundary conditions of the curve ensure that the torque values at the start and end points match the current state of the system, avoiding jumps.

[0078] The characteristic extraction of the torque coordination curve forms the deterministic torque components. The characteristic extraction uses multi-scale analysis to identify the characteristic patterns of the curve at different time scales. Large-scale characteristics (>100 milliseconds) reflect the overall trend, and the trend line is extracted as the basic torque component. Medium-scale characteristics (10-100 milliseconds) correspond to the main oscillation mode, and the main frequency and its harmonic components are extracted. Small-scale characteristics (<10 milliseconds) contain fast transients, and the high-frequency components that are beneficial to control are selectively extracted. Deterministic determination is achieved through repeatability testing, and characteristics that remain unchanged over multiple cycles have high determinism. The parametric representation of the components includes four parameters: amplitude, frequency, phase, and attenuation coefficient, which fully describe the component characteristics. The identification of dominant components is sorted by energy contribution, and the top several components with a cumulative contribution of 80% are selected. The physical interpretation of the components is related to specific control actions, with low-frequency components corresponding to stiffness compensation, medium-frequency components corresponding to damping provision, and high-frequency components corresponding to disturbance suppression. The independence test of the components ensures that the correlation between each component is less than 0.3, avoiding redundancy and coupling. The final deterministic torque component set contains 5-8 independent components, covering a frequency range from direct current to 5 times the frequency.

[0079] The deterministic torque components are superimposed in phase to generate the damping force control command. Vector synthesis is used, and each component is represented as , where Mk is the amplitude and φk is the phase angle. The superposition formula is Mtotal = Σ(Mk × cos(ωkt + φk)), where ωk is the angular frequency of the kth component and t is the time. The direct current component is directly superimposed to provide constant damping force, the fundamental frequency component is superimposed in synchronization with the rotor phase, and the harmonic component is superimposed according to its own frequency and phase. The embedding of the damping coefficient is achieved by multiplying the superposition result by the damping coefficient Cd, Fd = Cd × Mtotal, and Cd takes a value of 0.1-0.3. The conversion of the control command converts the torque command to a current command, i = Fd / (Km × r), where Km is the torque coefficient and r is the force arm length. The amplitude limiting processing of the command ensures that the current does not exceed the safe range, with the maximum current limited to within 150% of the rated value. Time discretization samples the continuous command at the control period, with a sampling frequency of 10 kHz, ensuring that all important dynamic characteristics are captured. The smoothing processing of the command sequence uses low-pass filtering with a cutoff frequency set to twice the highest control frequency to eliminate high-frequency noise. The final command contains four path control currents and two axial control currents, each with independent amplitude and phase parameters, achieving intelligent start-stop control of the magnetic suspension refrigeration compressor.

[0080] In order to perform the intelligent start-stop control method of the magnetic suspension refrigeration compressor corresponding to the above-mentioned method embodiment to achieve the corresponding functions and technical effects. Referring to Figure 2 , Figure 2A structural block diagram of an intelligent start-stop control system 200 of a magnetic suspension refrigeration compressor provided by an embodiment of the present application is shown. For ease of illustration, only parts related to the present embodiment are shown. The intelligent start-stop control system 200 of the magnetic suspension refrigeration compressor provided by the present embodiment includes:

[0081] A data acquisition module 201 is configured to acquire current response data of a magnetic bearing coil, perform fluctuation analysis on the current response data to obtain a position awareness enhancement source, and acquire a current response change feature based on the position awareness enhancement source.

[0082] A position estimation module 202 is configured to perform rotor position estimation on the current response change feature to extract a rotor offset coordinate parameter, extract a positioning accuracy driving source from an estimated deviation of the rotor offset coordinate parameter, and generate a non-impact start excitation current instruction based on the positioning accuracy driving source and the current response change feature.

[0083] A fault monitoring module 203 is configured to monitor a multi-sensor operating state based on the non-impact start excitation current instruction, establish a fusion data network, activate a signal reconstruction configuration through an abnormal response signal in the fusion data network, and generate a degraded operation fault-tolerant mode based on the signal reconstruction configuration and the current response change feature.

[0084] A shutdown control module 204 is configured to receive a shutdown instruction signal through the degraded operation fault-tolerant mode, convert the shutdown instruction signal into a response preparation gain, perform shutdown state evaluation based on the response preparation gain and the signal reconstruction configuration to generate an emergency shutdown trigger condition.

[0085] A damping execution module 205 is configured to detect a bus capacitor power supply state according to the emergency shutdown trigger condition, release gradient control potential from an energy change of the bus capacitor power supply state, determine an active damping control opportunity based on the gradient control potential and the degraded operation fault-tolerant mode, generate a damping force control instruction through the active damping control opportunity, and implement intelligent start-stop control of the magnetic suspension refrigeration compressor.

[0086] The intelligent start-stop control system 200 of the magnetic suspension refrigeration compressor described above can implement the intelligent start-stop control method of the magnetic suspension refrigeration compressor of the method embodiment described above. The optional items in the method embodiment described above are also applicable to the present embodiment, and will not be described in detail here. The remaining content of the present embodiment can be referred to the content of the method embodiment described above, and will not be described in detail in the present embodiment.

[0087] The purpose of the above embodiments is to exemplarily reproduce and deduce the technical solutions of the present application, and to completely describe the technical solutions, purposes and effects of the present application. The purpose is to make the public understand the disclosure of the present application more thoroughly and comprehensively, and does not limit the protection scope of the present application.

[0088] The above embodiments are also not exhaustive enumeration based on the present application, in addition to which there can be a plurality of other embodiments not listed. Any substitution and improvement made without violating the concept of the present application is within the scope of the present application.

Claims

1. A method of intelligent start-stop control of a magnetic levitation refrigeration compressor, characterized in that, The method comprises the following steps: Collecting current response data of the magnetic bearing coil, performing fluctuation analysis on the current response data to obtain a position sensing enhancement source, and obtaining a current response change feature based on the position sensing enhancement source; Performing rotor position estimation on the current response change feature to extract a rotor offset coordinate parameter, extracting a positioning accuracy driving source from the estimation deviation of the rotor offset coordinate parameter, and processing the current response change feature based on the positioning accuracy driving source to generate a non-impact start excitation current instruction; Based on the non-impact start excitation current instruction, the running state of the multi-sensor is monitored to establish a fusion data network, and the signal reconstruction configuration is activated through the abnormal response of the fusion data network, and the signal reconstruction configuration is combined with the current response change feature to generate a degraded operation fault-tolerant mode; Through the degraded operation fault-tolerant mode, the shutdown instruction signal is converted into a response preparation gain, and the response preparation gain is combined with the signal reconstruction configuration for shutdown state evaluation to generate an emergency shutdown trigger condition; According to the emergency shutdown trigger condition, the bus capacitor power supply state is detected, and the gradient control potential is released from the energy change of the bus capacitor power supply state, including: monitoring the voltage decay gradient mode of the bus capacitor power supply state; and performing capacitor discharge optimization control through the voltage decay gradient mode to form an optimized discharge area; The optimized discharge area is subjected to hierarchical current regulation to obtain a power supply current distribution; based on the power supply current distribution, a gradient control potential is formed, and a proactive damping control opportunity is determined based on the gradient control potential in combination with the degraded operation fault-tolerant mode, a damping force control instruction is generated through the proactive damping control opportunity, and intelligent start-stop control of the magnetic suspension refrigeration compressor is realized.

2. The method of claim 1, wherein, The method comprises the following steps: Performing frequency anomaly detection on the current response data to obtain an inductance field fluctuation feature; Identify the position error compensation current sequence in the inductance field fluctuation feature; Amplitude calibration is performed on the position error compensation current sequence to form a rotor positioning feature; The rotor positioning feature is used to construct a position sensing enhancement source.

3. The method of claim 1, wherein, The method comprises the following steps: Detecting sensor drift mode through the abnormal response of the fusion data network; Performing multi-channel cooperative calibration through the sensor drift mode to form a cooperative calibration trajectory sequence; The sensor compensation coordinate set is obtained by implementing spatial reconstruction mapping on the cooperative calibration trajectory sequence; Forming a signal reconstruction configuration based on the sensor compensation coordinate set.

4. The method of claim 1, wherein, The method comprises the following steps: Performing sliding window analysis on the estimation deviation of the rotor offset coordinate parameter to obtain a local deviation peak value; After arranging the local deviation peak value in time sequence, gradient analysis is performed to determine the deviation change trend; Identify the deviation turning feature based on the slope mutation point of the deviation change trend; Generate a positioning accuracy driving source according to the time distribution density of the deviation turning feature.

5. The method of claim 1, wherein, The method comprises the following steps: The shutdown instruction signal is decomposed by time window to form a dynamic delay sequence; The dynamic delay sequence is subjected to buffer state analysis to generate a buffer time matrix; The dynamic delay sequence is subjected to phase matching with the buffer time matrix to form a timing buffer graph; A response preparation gain is determined from a preparation enhancement node extracted from the timing buffer graph.

6. The method of claim 1, wherein, The damping force control instruction generated by the active damping control opportunity includes: A torque oscillation mode of the active damping control opportunity is monitored to generate a torque oscillation trajectory; An anti-phase torque is injected at a wave crest position of the torque oscillation trajectory to form a torque standing wave; A deterministic torque component is extracted from a stable node of the torque standing wave; The deterministic torque component is superimposed by phase to generate a damping force control instruction.

7. The method of claim 5, wherein, The dynamic delay sequence is subjected to phase matching with the buffer time matrix to form a timing buffer graph, including: The dynamic delay sequence is converted into a virtual oscillation signal; The virtual oscillation signal is allowed to propagate in the buffer time matrix to generate an interference pattern; An enhancement node and an elimination node in the interference pattern are extracted; Spatial distribution of the enhancement node and the elimination node is encoded into a timing buffer graph.

8. The method of claim 6, wherein, The deterministic torque component is extracted from the stable node of the torque standing wave, including: Oscillation energy analysis is performed on the torque standing wave to obtain oscillation stability data; Torque optimization nodes and control optimization nodes are identified in the oscillation stability data; Parameter fusion is performed based on the torque optimization nodes and the control optimization nodes to generate a torque coordination curve; Characteristic extraction analysis is performed on the torque coordination curve to form a deterministic torque component.

9. An intelligent start-stop control system for a magnetic levitation refrigeration compressor, characterized in that, It includes: A data acquisition module is used to acquire current response data of a magnetic bearing coil, perform fluctuation analysis on the current response data to obtain a position sensing enhancement source, and obtain current response change characteristics based on the position sensing enhancement source; A position estimation module is used to perform rotor position estimation on the current response change characteristics to extract rotor offset coordinate parameters, extract positioning accuracy driving sources from the estimation deviation of the rotor offset coordinate parameters, process the current response change characteristics based on the positioning accuracy driving sources to generate a non-impact start excitation current instruction; A fault monitoring module is used to monitor the running state of a plurality of sensors based on the non-impact start excitation current instruction to establish a fusion data network, activate a signal reconstruction configuration through an abnormal response signal in the fusion data network, and generate a degraded operation fault-tolerant mode based on the signal reconstruction configuration combined with the current response change characteristics; A shutdown control module is used to receive a shutdown instruction signal through the degraded operation fault-tolerant mode, convert the shutdown instruction signal into a response preparation gain, and generate an emergency shutdown trigger condition through shutdown state evaluation based on the response preparation gain combined with the signal reconstruction configuration; A damping execution module is used to detect the bus capacitor power supply state according to the emergency shutdown trigger condition, release gradient control potential from the energy change of the bus capacitor power supply state, including: monitoring a voltage decay gradient mode by using the bus capacitor power supply state; and performing capacitor discharge optimization control to form an optimized discharge region through the voltage decay gradient mode. The hierarchical current regulation is implemented on the optimized discharge area to obtain a power current distribution; a gradient control potential is formed based on the power current distribution, an active damping control opportunity is determined based on the gradient control potential in combination with the degraded operation fault-tolerant mode, damping force control instructions are generated through the active damping control opportunity, and intelligent start-stop control of the magnetic suspension refrigeration compressor is realized.

Images