A high-precision magnetic levitation electronic bearing displacement and rotating speed sensor design method

Abstract

The application relates to the technical field of magnetic suspension bearing state detection and sensing, and discloses a high-precision magnetic suspension electronic bearing displacement and rotating speed sensor design method, which comprises the following steps: integrating a displacement and rotating speed detection component in a stator base and arranging an insulating shielding partition plate; converting inductive signals and pulse signals into direct-current analog signals and square wave signals respectively through a hardware circuit and inputting the signals into a digital processing unit; calling a neural network module by the digital processing unit to compensate errors of displacement data, adopting a double-algorithm switching mechanism to analyze square wave signals, and utilizing an adaptive multi-stage composite filtering module to carry out filtering output. Through the method, electromagnetic coupling interference between detection components is blocked, nonlinear errors caused by temperature drift are compensated, the measurement limitations of a single analysis algorithm in different rotating speed intervals are overcome, time synchronization alignment of multiple source signals is realized, and the precision of high-end equipment magnetic suspension bearing system state monitoring is improved.

Description

Technical Field

[0001] This invention relates to the field of magnetic levitation bearing condition detection and sensing technology, specifically to a design method for a high-precision magnetic levitation electronic bearing displacement and speed sensor, which is mainly applied to real-time condition monitoring and closed-loop stability control of magnetic levitation bearings in high-end equipment such as high-speed motors, precision machine tools, and aerospace turbines. Background Technology

[0002] The stable operation of magnetically levitated electronic bearings highly depends on the control system's accurate measurement of the rotor's multi-degree-of-freedom displacement and real-time rotational speed. In traditional sensor designs, displacement and rotational speed probes are typically installed independently inside the stator, or lack a robust electromagnetic isolation structure during physical integration. Because the motor operation and high-frequency drive control within the magnetic levitation system generate a strong alternating magnetic field, electromagnetic coupling interference easily occurs between different detection components. Simultaneously, the probe coil material is affected by ambient temperature changes during system operation, exhibiting thermal expansion and permeability drift. These physical disturbances and changes introduce nonlinear measurement errors during the initial signal acquisition stage.

[0003] In the signal analysis and processing stage, existing sensors mostly rely on a single frequency or period measurement algorithm to process the rotational speed square wave signal. During the extremely low-speed start-up phase and when the rotor crosses the critical speed range at high speed, the frequency characteristics and noise distribution of its output signal differ significantly. A single algorithm cannot simultaneously meet the measurement resolution requirements at both ends of the wide speed range, leading to fluctuations in speed measurement accuracy across the entire speed range. Furthermore, traditional multi-channel signal acquisition systems often use software interrupt mechanisms to trigger data sampling and latching. In complex microcontroller systems, the response and scheduling of software instructions inherently incur time overhead and execution jitter, causing a misalignment between the sampling points of the displacement analog signal and the count value of the rotational speed square wave signal on the time axis. Once this time phase delay between multi-source signals is transmitted to the back-end closed-loop control system, it directly affects the dynamic response performance and levitation stability of the magnetic levitation bearing. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a design method for a high-precision magnetic levitation electronic bearing displacement and speed sensor, which solves the problems of electromagnetic interference coupling, full-speed domain measurement accuracy fluctuation, and time phase delay of multi-source signals during the displacement and speed signal detection process in the magnetic levitation bearing system.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] This invention provides a design method for a high-precision magnetic levitation electronic bearing displacement and speed sensor, comprising: integrating a radial displacement detection component, an axial displacement detection component, a speed detection component, and a temperature detection component into a stator base, with an insulating shielding partition between adjacent components; generating a sinusoidal alternating excitation signal through a high-frequency excitation circuit of a hardware circuit system and injecting it into the coils of the aforementioned radial and axial displacement detection components; The sensing signals from the various displacement detection components are converted into DC analog signals by the signal acquisition and conditioning circuit, and the pulse signals from the speed detection components are converted into square wave signals by the hysteresis comparison circuit. The DC analog signals are processed by the analog-to-digital conversion unit and then synchronously input into the digital processing unit along with the temperature data. The square wave signals are input into the high-speed timer module inside the digital processing unit. The digital processing unit calls the error calibration neural network module to perform error compensation on the acquired displacement data, calls the dual-algorithm switching mechanism to parse the square wave signals, and uses the adaptive multi-level composite filter module to filter the output.

[0007] The core of this invention lies in blocking the electromagnetic coupling path between components through spatial integration and an insulating shielding structure; at the digital processing end, a neural network module is introduced to compensate for the nonlinear error of the sensor's sensing signal, and a dual-algorithm switching and adaptive composite filtering mechanism are combined to process the signal characteristics of different speed frequency bands, overcoming the limitations of a single analytical algorithm in measuring the accuracy of displacement and speed signals.

[0008] Preferably, the specific arrangement steps of the radial displacement detection component and the axial displacement detection component include: symmetrically arranging four sets of radial displacement detection components on the inner circumference of the stator base to form two symmetrical differential detection channels, and symmetrically arranging two sets of axial displacement detection components at both ends of the axial direction; independently arranging the speed detection component inside the stator base, facing the outer surface of the end circumference of the rotor; using nanocrystalline ribbon with a single layer thickness of 20μm and a stacking factor of not less than 0.95 to prepare the core of the radial displacement detection component, the axial displacement detection component and the speed detection component, and performing overall insulation encapsulation after forming.

[0009] The differential detection channel formed by the above-described arrangement has the function of suppressing common-mode interference signals. The core uses nanocrystalline ribbon with a specified thickness and stacking factor, utilizing its high magnetic permeability and low high-frequency loss characteristics to reduce eddy current loss and heat generation caused by high-frequency alternating excitation.

[0010] In a preferred embodiment of the present invention, a high-frequency excitation circuit comprising a quartz crystal oscillator, an LC parallel resonant network, and a voltage regulation and current limiting module is constructed. The voltage regulation and current limiting module is used to limit the output amplitude fluctuation of the sinusoidal alternating excitation signal within a set threshold of ±0.5%. A signal acquisition and conditioning circuit comprising a preamplifier, a bandpass filter, a rectifier circuit, and a low-pass filter is constructed. The induced signal is a multi-channel symmetrical differential induced signal. The preamplifier is used to perform linear amplification and differential operation processing on the input multi-channel symmetrical differential induced signals.

[0011] In the aforementioned hardware circuit, the voltage regulation and current limiting module is used to control the consistency of the excitation source amplitude and reduce the measurement reference error caused by the fluctuation of the excitation source; the preamplifier converts the differential sensing signal into a single-ended signal through differential operation processing, thereby improving the signal-to-noise ratio of the subsequent analog-to-digital conversion stage.

[0012] Preferably, the input data dimension of the error calibration neural network module is set to two dimensions, specifically including the digital voltage signal representing the original inductance difference output by the analog-to-digital conversion unit, and the temperature data output by the temperature detection component set by the system; the digital processing unit calls the maximum-minimum-maximum normalization algorithm to uniformly map the two-dimensional input data to the numerical range between negative one and positive one before inputting it into the hidden layer; the hidden layer nodes of the error calibration neural network module are activated using a linear rectification function, and the output layer nodes are processed using a linear activation function to output the true displacement.

[0013] Since the inductance characteristics of the magnetic levitation probe coil exhibit temperature drift, this scheme combines temperature data and voltage signals into a two-dimensional input variable. It utilizes a neural network to construct a temperature-inductance-displacement mapping relationship, directly compensating for errors caused by thermal expansion and permeability temperature drift at the algorithm level.

[0014] In a preferred embodiment of the present invention, the error calibration neural network module performs offline training before deployment, the steps of which include: setting the combination of the digital voltage signal and the temperature data as training samples, and setting the reference displacement data measured by the laser interferometer as training labels; using the mean square error as the loss function, and using the gradient descent optimization algorithm to backpropagate and iteratively update the weight coefficients and bias parameters of each level until the mean square error decreases to 10. -14 Training stops when the parameters reach the order of powers of 1, and the final parameters are stored in internal non-volatile memory.

[0015] The offline training process incorporates laser interferometer data as a benchmark, and solidifies the model parameters through multidimensional fitting. When the sensor runs online, it only needs to call the solidified parameters to perform forward inference, controlling the real-time computing power overhead.

[0016] Preferably, a frequency switching threshold is preset within the digital processing unit, and the value of the frequency switching threshold is jointly determined by the system clock frequency of the digital processing unit and a preset speed detection error requirement; the parsing steps of the dual-algorithm switching mechanism are as follows: When the frequency of the square wave signal is lower than the frequency switching threshold, the time period of adjacent square wave pulses is measured to obtain the signal frequency; when the frequency of the square wave signal is greater than or equal to the frequency switching threshold, the signal frequency is obtained by combining the statistical number of square wave pulses and their corresponding timestamps within a fixed time window; finally, the real-time rotational speed data of the rotor is calculated based on the ratio between the obtained signal frequency and the number of magnetic tooth slots preset at the rotor end.

[0017] The above analysis steps employ statistical methods of period measurement and frequency measurement for the low-frequency and high-frequency bands respectively, and perform branch processing based on the frequency switching threshold to solve the measurement error problem of a single frequency measurement algorithm at both ends of different speed ranges.

[0018] In a preferred embodiment of the present invention, a lower speed limit threshold and an upper speed limit threshold are set inside the adaptive multi-level composite filter module. The lower speed limit threshold is 80% of the first-order critical speed of the rotor, and the upper speed limit threshold is 120% of the first-order critical speed of the rotor. When the real-time speed data is less than or equal to the lower speed limit threshold, the low-speed composite filter channel containing the median filter module and the moving average filter module connected in series is activated. When the real-time speed data is greater than or equal to the upper limit threshold of the speed, the high-speed composite filter channel, which includes a combination of a first-order hysteresis filter module and a Kalman filter module, is activated; when the real-time speed data is in the numerical transition range between the two thresholds, the dynamic weighting coefficient is calculated based on the speed position, and the output values ​​of the low-speed composite filter channel and the high-speed composite filter channel are linearly weighted and fused.

[0019] The vibration characteristics and noise distribution of the rotor differ before and after the critical speed. In the segmented composite filter architecture, the low-speed segment focuses on eliminating pulse interference, the high-speed segment focuses on state prediction and high-frequency noise suppression, and the linear weighted processing in the transition segment ensures the continuity of data output.

[0020] Preferably, a clock synchronization parsing module is provided inside the digital processing unit, and the clock synchronization parsing module has a global hardware timer. The global hardware timer generates a synchronization trigger pulse according to the set system control cycle and directly distributes it to the analog-to-digital conversion unit and the high-speed timer module. When the synchronization trigger pulse arrives, the analog-to-digital conversion unit is forced to start sampling and latching the displacement voltage signal, and the high-speed timer module is simultaneously forced to latch the high-precision timestamp triggered by the latest square wave edge and the current square wave pulse count value in parallel.

[0021] The synchronization pulses generated by the global hardware timer replace the conventional software interrupt triggering method, eliminating the waiting jitter during instruction execution. The forced alignment of displacement data sampling points and speed data calculation points on the time axis provides synchronous status feedback for subsequent control systems.

[0022] In a preferred embodiment of the present invention, a multi-layer shielded cable assembly with an internal twisted-pair structure and an external metal braided shielding mesh for single-end grounding is used to transmit the original detection signals of the above-mentioned detection components; the radial displacement detection component, axial displacement detection component, and speed detection component are fixed on a mounting bracket made of non-magnetic material, and an alloy shielding cover made of permalloy material is encapsulated on the outside of the mounting bracket; a polymer shock-absorbing insulating pad is provided at the mounting interface between the circuit board of the hardware circuit system and the stator base.

[0023] The solution suppresses common-mode and ground loop interference through a twisted-pair cable structure and a single-ended grounded shielding mesh; it utilizes the magnetic permeability characteristics of permalloy to shield the low-frequency alternating magnetic field of the external motor; and a polymer shock-absorbing insulating pad is used to block the transmission of mechanical vibration to the internal circuit board.

[0024] Preferably, the sensor design method further includes a full-range performance calibration step: using a micro-motion stage to drive the calibration target to move in multiple points, and measuring the physical displacement using a reference measuring instrument as calibration reference data; the data acquisition and analysis module synchronously acquires the measured displacement data output by the digital processing unit, records the bidirectional travel measurement data, and uses the least squares method to perform linear fitting on the above multiple sets of data; controlling a standard servo motor connected to the rotor to perform acceleration and deceleration operation, synchronously recording the standard speed data and the measured speed data parsed by the speed detection component, and comparing the time phase difference of the two sets of speed data under transient conditions and the numerical deviation under steady-state conditions.

[0025] The above full-range performance calibration steps verify the static linearity and dynamic response of displacement and speed measurements through static stepping and dynamic acceleration / deceleration tests, ensuring that the measurement system after hardware and software integration meets engineering application specifications.

[0026] This invention provides a design method for a high-precision magnetic levitation electronic bearing displacement and speed sensor. It offers the following advantages: 1. This invention integrates multiple displacement and speed detection components within the stator base. Combined with an insulating shielding partition, a permalloy shield, and a differential detection channel, the hardware structure effectively blocks electromagnetic coupling paths between different detection modules and prevents interference from the alternating magnetic field of the external motor. Furthermore, by incorporating a voltage regulation and current limiting module to control the amplitude of the high-frequency excitation signal, the stability of the original sensing signal is ensured from the physical structure and hardware source, reducing the impact of complex electromagnetic environments on the measurement reference.

[0027] 2. This invention introduces an error calibration neural network module and a dual-algorithm switching mechanism in the data processing stage. Temperature parameters and digital voltage signals are used as joint inputs, and the algorithm model compensates for nonlinear errors caused by thermal expansion and permeability temperature drift of the probe coil material. For different rotor operating conditions, frequency measurement and period measurement calculations are switched according to preset thresholds, and a segmented multi-level composite filtering channel is used to solve the measurement error problem of a single analytical algorithm at extremely low speeds and across critical speed ranges, ensuring the accuracy and continuity of data across the entire speed range.

[0028] 3. This invention utilizes a global hardware timer within the clock synchronization parsing module to generate a synchronization trigger pulse, directly forcing the analog-to-digital conversion unit and the high-speed timer module to perform displacement voltage sampling and square wave pulse latching operations in parallel. This pure hardware synchronization mechanism replaces the traditional software interrupt triggering method, eliminates clock jitter caused by instruction execution waiting, and achieves strict alignment of displacement data and rotational speed data on the time axis, providing synchronous state feedback with no time phase delay for the magnetic levitation control system. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the overall integrated structure layout of an embodiment of the present invention; Figure 2 This is a block diagram of the hardware circuit system architecture of an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the magnetic cutting principle based on rotational speed detection in an embodiment of the present invention; Figure 4 This is a diagram illustrating the dual-algorithm switching mechanism architecture for speed detection in an embodiment of the present invention. Figure 5 This is a schematic diagram of the error calibration neural network module architecture in an embodiment of the present invention; Figure 6 This is a flowchart of the adaptive multi-stage composite filtering algorithm according to an embodiment of the present invention; Figure 7 This is a block diagram of the piecewise linear interpolation and nonlinear error compensation logic in an embodiment of the present invention; Figure 8 This is a block diagram illustrating the dynamic temperature compensation and clock synchronization analysis mechanism in an embodiment of the present invention. Figure 9 This is a schematic cross-sectional view of the anti-interference physical structure according to an embodiment of the present invention; Figure 10 This is a flowchart of the full-range performance calibration method according to an embodiment of the present invention; Figure 11 This is a statistical chart showing the performance test comparison and advantage analysis of embodiments of the present invention; Figure 12 This is a comparison curve of displacement nonlinearity error and rotational speed analytical accuracy in an embodiment of the present invention. Detailed Implementation

[0030] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] See attached document Figure 1 The present invention provides a design method for a high-precision magnetic levitation electronic bearing displacement and speed sensor, which may include: In this embodiment, the sensor adopts a stator-embedded integrated structure. The sensor is entirely embedded and integrated inside the stator base of the magnetic levitation bearing. In a specific embodiment, the sensor includes a radial displacement detection component, an axial displacement detection component, a rotational speed detection component, and a temperature detection component.

[0032] On the inner circumference of the stator base of the magnetic levitation bearing, four sets of radial displacement detection components with basically the same structure and electrical parameters are arranged radially. The four sets of radial displacement detection components are symmetrically distributed in pairs in the X and Y axis directions, forming two symmetrical differential detection channels to detect the bidirectional micro-displacement of the rotor in the radial direction. Two sets of axial displacement detection components are symmetrically arranged at both ends of the axial direction to collect the axial offset of the rotor.

[0033] The speed detection component is independently arranged inside the stator base, directly opposite the outer circumferential surface of the rotor end. Simultaneously, the temperature detection component is also integrated and installed inside the stator base. Insulating shielding partitions are installed between the radial displacement detection component, axial displacement detection component, speed detection component, and temperature detection component to reduce electromagnetic crosstalk between detection signals of different frequency bands through physical isolation layers.

[0034] The aforementioned radial displacement detection component, axial displacement detection component, and speed detection component all include an iron core and a coil. The iron core is prepared by laminating 1K107 nanocrystalline ribbon, with a single layer thickness of 20μm and a lamination factor of not less than 0.95. After lamination, the iron core is integrally insulated and encapsulated. The magnetic levitation bearing stator base has pre-set positioning slots for installing the above detection components. The radial displacement detection component is opposite to the outer cylindrical surface of the rotor, and the axial displacement detection component is parallel and coaxial with the rotor end face. The working air gap length is controlled between 0.3mm and 0.5mm. The coil is wound with 0.08mm diameter enameled copper wire. The radial and axial displacement detection components each have 120 turns, while the speed detection component has 150 turns.

[0035] Displacement detection is primarily based on the principle of self-inductive electromagnetic induction. In actual detection, a high-frequency alternating current is applied to the displacement detection component, establishing an alternating magnetic field between the stator and rotor. According to Ohm's law for magnetic circuits, the coil inductance can be approximately expressed as: ;in, For coil inductance, The permeability of free space, The number of coil turns. The cross-sectional area of ​​the magnetic poles. This is the working air gap length.

[0036] When the rotor is in equilibrium, the working air gap length remains in its initial state. When the rotor is displaced due to external load, the working air gap length changes, causing a change in magnetic reluctance, which in turn leads to a change in coil inductance. By measuring the difference in inductance between two symmetrically arranged coils through a differential detection channel, the common-mode deviation caused by changes in ambient temperature can be offset, thereby extracting the inductance change related to the displacement.

[0037] The integrated structure, along with the hardware circuitry and digital processing unit, completes the overall detection workflow. Upon power-up, the high-frequency excitation circuit within the hardware circuitry generates a 2MHz sinusoidal alternating excitation signal. This excitation signal is injected into the coils of the radial and axial displacement detection components, establishing an alternating magnetic field within the working air gap length.

[0038] When the rotor undergoes radial or axial displacement, the working air gap length of the corresponding detection component changes, causing changes in the magnetic reluctance and inductance of the coils within the radial and axial displacement detection components, thereby outputting an induced signal. Simultaneously, the rotor end structure, as the spindle rotates, cuts the magnetic field established by the speed detection component, generating periodic alternating electromotive force pulses.

[0039] The induced signals output from the radial displacement detection component and the axial displacement detection component are input to the signal acquisition and conditioning circuit within the hardware circuit system. The signal acquisition and conditioning circuit sequentially performs pre-amplification, band-pass filtering, precision rectification, and low-pass filtering on the induced signals, converting the alternating induced signals into DC analog signals. The pulse signal output from the speed detection component is synchronously input to the hysteresis comparator circuit within the signal acquisition and conditioning circuit for shaping, outputting a square wave signal.

[0040] The DC analog signal is converted from analog to digital under a single clock drive and then input to the digital processing unit. The square wave signal is synchronously and directly input to the high-speed timer module in the digital processing unit. The digital processing unit performs nonlinear and temperature error compensation processing on the input data, and then combines filtering algorithms to reduce noise. Finally, it outputs displacement data and speed data to the external closed-loop controller.

[0041] See attached document Figure 2The present invention provides a design method for a high-precision magnetic levitation electronic bearing displacement and speed sensor, which may include: In this embodiment, the hardware circuit system mainly consists of a high-frequency excitation circuit, a displacement signal acquisition and conditioning circuit, and a rotational speed signal conditioning circuit. The hardware circuit system is electrically connected to the digital processing unit to perform the conversion, conditioning, and acquisition of physical quantity signals.

[0042] The high-frequency excitation circuit includes a quartz crystal oscillator and an LC parallel resonant network. The quartz crystal oscillator provides the system's fundamental clock reference pulse. This reference pulse, after being driven, is input to the LC parallel resonant network. The LC parallel resonant network utilizes its resonant characteristics for frequency selection filtering, thereby generating a sinusoidal alternating excitation signal. The formula for calculating the resonant frequency can be expressed as: ; in, The resonant frequency, For the network equivalent inductance, The equivalent capacitance of the network is used. In the specific implementation, by reasonably setting the parameter values ​​of the network equivalent inductance and network equivalent capacitance, the resonant frequency is maintained at around 2MHz. This sinusoidal alternating excitation signal is used as the excitation source for the radial displacement detection component and the axial displacement detection component.

[0043] In some implementations, the high-frequency excitation circuit incorporates a voltage regulator and current limiter module. This module clamps and limits the output voltage and current of the sinusoidal alternating excitation signal. It controls the amplitude fluctuation of the sinusoidal alternating excitation signal within ±0.5%, thereby improving the stability of the magnetic field at the front end of the displacement detection coil.

[0044] The displacement signal acquisition and conditioning circuit adopts a symmetrical differential signal processing architecture. The circuit includes a preamplifier, a bandpass filter, a rectifier circuit, and a low-pass filter. The differential sensing signals output from the radial displacement detection component and the axial displacement detection component are first input to the preamplifier.

[0045] The preamplifier is constructed using an operational amplifier chip. It linearly amplifies weak differential sensing signals. Due to the differential processing architecture, the output voltage of each channel of the preamplifier is proportional to the difference between the input signals at the corresponding differential nodes. This structure can suppress common-mode noise introduced by ambient temperature changes and external electromagnetic interference, thereby improving the signal-to-noise ratio of subsequent signal processing links.

[0046] Mathematical models for the radial and axial displacement detection components are established based on the principle of self-inductive electromagnetic induction. After the system is powered on, an alternating excitation signal is input to the coils inside the radial and axial displacement detection components, forming an alternating magnetic field within the working air gap. The core of electromagnetic induction detection lies in using the change in the air gap magnetic reluctance in the magnetic circuit to reflect the change in the measured displacement. By converting the mechanical displacement into a change in electrical quantity, a foundation is provided for subsequent signal conditioning and analysis.

[0047] When the rotor is in the mechanical center equilibrium position, the working air gap lengths corresponding to the two coils symmetrically arranged inside the radial displacement detection component are equal. This initial air gap length is set to... At this point, the initial inductances of the two symmetrical coils are equal, and the inductance difference output by the differential detection channel is zero. When the rotor is loaded and displaced, it deviates from the mechanical center equilibrium position. The displacement generated by the rotor is set to... The movement of the rotor causes a change in the working air gap length on the symmetrical coil side.

[0048] At this point, the length of the working air gap on one side decreases to The working air gap length on the other symmetrical side increases to Because the core uses high-permeability nanocrystalline ribbon, its internal magnetic reluctance is much smaller than that in the working air gap. Therefore, the effects of core magnetic reluctance and magnetic field edge effects can be reasonably ignored when constructing the model. Under these physical conditions, the real-time inductance of a single-sided coil is inversely proportional to the length of the working air gap. The coil inductance on the side with the reduced air gap can be expressed as: ; The coil inductance corresponding to the side with increased air gap can be expressed as: ; in, The coil inductance on the side with reduced air gap. The coil inductance on the side with increased air gap. The permeability of free space, The number of coil turns. Where is the cross-sectional area of ​​the magnetic pole. The displacement signal acquisition and conditioning circuit extracts the inductance change of the two symmetrical coils through differential operation. The formula for calculating the inductance difference can be expressed as: ; in, This represents the inductance difference. In actual magnetic levitation bearing micro-displacement conditions, the displacement... Typically on the order of micrometers, their values ​​generally range from positive to negative. Within, and the initial air gap length Set at to Between. Due to displacement Much smaller than the initial air gap length In the denominator of the above formula The impact of this term on the overall value is relatively small. To reduce the computational complexity of the digital processing unit, in engineering calculations, [the following can be done]: The term is ignored. Therefore, an approximately simplified formula for the inductance difference is derived: ; Based on the simplified formula, the inductance difference is obtained. With displacement A linear proportional relationship is exhibited. The hardware circuit system extracts the induced signal based on this model and outputs an analog signal representing the change in inductance. This signal is then processed by a digital processing unit to analyze and calculate the real-time displacement data of the rotor. This differential operation structure not only extracts displacement change characteristics but also effectively suppresses common-mode deviation caused by ambient temperature drift. The axial displacement detection component uses the same mathematical model to complete the extraction and analysis of axial displacement.

[0049] See attached document Figure 3 The present invention provides a design method for a high-precision magnetic levitation electronic bearing displacement and speed sensor, which may include: In this embodiment, a mathematical model of the speed detection component is established based on the principle of magnetic induction cutting. The speed detection component is installed on the outer circumference of the rotor end. A permanent magnet is disposed inside the speed detection component to provide a constant bias magnetic field. The circumferential surface of the rotor end is machined with a uniformly distributed magnetic tooth groove structure.

[0050] The magnetically guided toothed groove structure is used to change the equivalent magnetic reluctance within the working air gap at the end of the speed detection component. During actual detection, the bias magnetic field passes through the iron core of the speed detection component, the working air gap, and the magnetically guided toothed groove structure at the rotor end, forming a closed magnetic circuit. As the rotor rotates with the main shaft, the teeth and slots of the magnetically guided toothed groove structure alternately pass over the sensing end face of the speed detection component, causing a periodic change in the magnetic reluctance at the working air gap. This change in the working air gap magnetic reluctance results in alternating fluctuations in the magnetic flux passing through the coil inside the speed detection component.

[0051] According to Faraday's law of electromagnetic induction, a change in magnetic flux induces an alternating electromotive force (EMF) across the coil of the speed detection component. The formula for calculating this alternating EMF can be expressed as: ;in, It is an alternating electromotive force. The number of coil turns within the speed detection component. The magnetic flux passing through the coil, For each revolution of the rotor, the internal coil of the speed detection component outputs a fixed number of alternating electromotive force pulses.

[0052] The number of magnetic guide tooth slots set at the rotor end is set to The frequency of the alternating electromotive force (EMF) pulse is linearly proportional to the real-time rotor speed. The formula for calculating the frequency of the alternating EMF pulse can be expressed as: ;in, The frequency of the alternating electromotive force pulse. This represents the real-time rotational speed of the rotor, with a value of 60 serving as a time unit conversion factor. The alternating electromotive force is conditioned by hardware circuitry and converted into a square wave signal, which is then input to the digital processing unit. The digital processing unit contains a high-speed timer.

[0053] The digital processing unit extracts the frequency parameters of the square wave signal and then calculates the real-time rotational speed of the rotor. The formula for calculating the real-time rotational speed can be expressed as: To address the issue of varying resolution accuracy across a wide frequency range, the digital processing unit employs a dual-algorithm switching mechanism based on a frequency switching threshold. The specific wide-frequency domain resolution algorithm architecture and threshold calculation logic will be detailed in subsequent appendices. Figure 4 Please provide a detailed explanation.

[0054] See attached document Figure 4 The present invention provides a design method for a high-precision magnetic levitation electronic bearing displacement and speed sensor, which may include: In this embodiment, based on the aforementioned magnetic cutting principle at the hardware level, a wide-frequency domain analytical algorithm architecture for the rotational speed signal within the digital processing unit is further disclosed. The alternating electromotive force is converted into a square wave signal after conditioning by the front-end hardware circuit, and then input to the digital processing unit. The digital processing unit contains a high-speed timer module. The digital processing unit calculates the real-time rotational speed data of the rotor by analyzing the frequency parameters of the square wave signal. The rotational speed analytical formula is expressed as: ;in, This represents the real-time rotational speed of the rotor. The frequency of the alternating electromotive force pulse. This refers to the number of magnetic tooth slots.

[0055] To address the issue of varying analytical accuracy across a wide speed range, the digital processing unit incorporates a frequency switching threshold. During online operation, the system utilizes historical signal frequencies acquired in the previous control cycle as the basis for determining the algorithm branch for the current cycle. When the frequency of the current square wave signal is determined to be below the frequency switching threshold, the digital processing unit calculates the signal frequency by measuring the time period of adjacent square wave pulses. When the frequency of the current square wave signal is determined to be greater than or equal to the frequency switching threshold, the digital processing unit combines the statistically counted number of square wave pulses with their corresponding timestamps within a fixed time window to obtain the signal frequency using a frequency measurement method. This dual-algorithm switching mechanism avoids the calculation errors caused by insufficient pulses in low-speed conditions with a single frequency measurement method, and the calculation errors caused by insufficient clock count values ​​in high-speed conditions with a single period measurement method.

[0056] The frequency switching threshold is determined by the system clock frequency of the digital processing unit and the speed detection accuracy. Specifically, the relative quantization error of the period measurement method is directly proportional to the frequency of the measured signal, while the relative quantization error of the frequency measurement method is inversely proportional to the frequency of the measured signal. Based on the error crossover principle of the period and frequency measurement methods, the threshold is set at 80MHz for the system clock frequency of the digital processing unit, and the speed detection error requirement for the system across the entire range is set to be no greater than 0.08%. Based on these parameters, the frequency switching threshold is set to 1kHz. This threshold switching mechanism helps maintain the consistency of sensor detection accuracy in both low-speed and high-speed ranges.

[0057] See attached document Figure 5 The present invention provides a design method for a high-precision magnetic levitation electronic bearing displacement and speed sensor, which may include: In this embodiment, based on the aforementioned mathematical models of the radial displacement detection component and the axial displacement detection component, the inductance difference and the rotor displacement exhibit an approximately linear relationship within a small displacement range. However, in actual operating conditions, as the rotor's displacement from the central equilibrium position increases, the higher-order terms neglected in the aforementioned mathematical model generate truncation errors.

[0058] The magnetic field edge effect present in the working air gap region at the end of the detection component, and the fluctuation of the permeability of the core material under different magnetization intensities, superimpose to cause nonlinear distortion of the inductor air gap characteristics. To reduce the measurement error caused by the above nonlinear distortion, an error calibration neural network module is built and run inside the digital processing unit.

[0059] The system includes a temperature detection component. This component is independently installed inside the stator base of the magnetic levitation bearing. The temperature detection component acquires real-time temperature data of the sensor's surrounding environment.

[0060] The error calibration neural network module employs a fully connected feedforward-backward propagation network structure. The module comprises an input layer, hidden layers, and an output layer. For each differential detection channel, the digital processing unit calls an independent error calibration neural network module for parallel processing. The input data dimension of a single error calibration neural network module is two-dimensional. The specific data items received by the input layer include the digital voltage signal representing the original inductance difference output by the corresponding channel of the analog-to-digital converter and the temperature data output by the temperature detection component.

[0061] Before the input hidden layer, the digital processing unit performs preprocessing logic on the two-dimensional input data received by the input layer. The preprocessing logic uses a max-min normalization algorithm to map the digital voltage signal and temperature data to a numerical range between -1 and +1, thereby unifying the dimensions of data from different dimensions and accelerating network convergence. The specific calculation formula for the normalization algorithm can be expressed as: ;in, These are the normalized data values. The original input data values, This represents the minimum value of the original input data within the set operating condition range. This represents the maximum value of the original input data within the set operating conditions.

[0062] The hidden layer of the error calibration neural network module has ten neurons. These neurons receive data from the input layer and undergo nonlinear activation processing using a linear rectified function. The output layer contains one neuron. This neuron processes the output features of the hidden layer using a linear activation function. The output of the error calibration neural network module corresponds to the actual displacement after nonlinear and temperature compensation. For the model construction and offline training phases of the error calibration neural network module, a laser interferometer is used as the reference measurement device. Under different set operating temperature gradients, the rotor is controlled to move within its full displacement range.

[0063] The testing system synchronously records the digital voltage signal output by the analog-to-digital converter, the temperature data output by the temperature detection component, and the reference displacement data measured by the laser interferometer. In the training step, the digital voltage signal and temperature data are combined as training samples, and the reference displacement data measured by the laser interferometer is used as training labels. The mean square error (MSE) is used as the loss function during training. The formula for calculating the MSE is: ;in, Mean square error, The total number of training samples, For training labels, The predicted displacement output by the error calibration neural network module is used. The offline training process utilizes a gradient descent optimization algorithm, backpropagating based on the gradient of the loss function and iteratively updating the weight coefficients and bias parameters of each layer within the error calibration neural network module. Training stops when the mean squared error decreases to a set threshold. The value of this threshold is determined based on the displacement detection resolution required by the magnetic levitation bearing system. In this embodiment, the target detection resolution is set to 0.3 micrometers. To ensure that the network output error does not affect the overall resolution, the threshold for the mean square error is set to 10. -14 The order of magnitude. At this point, the meter (m) is used as the unified unit for both the actual displacement and the reference displacement data. This threshold is directly related to 0.1 micrometers (i.e., 10...). 7The square of the physical displacement deviation (in meters) corresponds to this (leaving sufficient computational margin for the final resolution of the system), serving as a hard cutoff condition for network model convergence and avoiding underfitting that introduces additional quantization noise. The digital processing unit extracts the weight coefficients and bias parameters after training and stores them in internal non-volatile memory, completing the deployment of the error calibration neural network module. This network structure reduces the Runge phenomenon that occurs in polynomial fitting under complex temperature drift conditions.

[0064] See attached document Figure 6 The present invention provides a design method for a high-precision magnetic levitation electronic bearing displacement and speed sensor, which may include: In this embodiment, the magnetic levitation electronic bearing system is typically subjected to mechanical vibrations and electromagnetic noise interference at different frequency bands during operation. To obtain stable physical state parameters, an adaptive multi-level composite filter module is installed inside the digital processing unit. The adaptive multi-level composite filter module receives the actual displacement output by the error calibration neural network module and the real-time rotational speed data obtained by the speed measurement algorithm.

[0065] Single filtering algorithms often struggle to balance dynamic response speed and noise suppression requirements across a wide speed range. Therefore, an adaptive multi-level composite filtering module dynamically adjusts the filtering strategy based on real-time speed data. To avoid data jumps caused by hard switching of algorithms at a single speed threshold, the adaptive multi-level composite filtering module sets a lower speed threshold and an upper speed threshold, forming a complete numerical transition range between them. The specific values ​​of these thresholds are determined based on the critical speed characteristics of the magnetic levitation rotor; in this embodiment, the lower speed threshold is set to 80% of the rotor's first-order critical speed, and the upper speed threshold is set to 120% of the rotor's first-order critical speed.

[0066] When the real-time rotational speed data is less than or equal to the lower speed threshold, the adaptive multi-level composite filter module activates the low-speed composite filter channel. Under low-speed conditions, the dynamic changes in the displacement signal are relatively smooth, mainly affected by random impulse noise. The low-speed composite filter channel employs a cascaded structure combining a median filter module and a moving average filter module. The median filter module extracts the median value from the data sequence to remove impulse-type glitches. Subsequently, the moving average filter module performs summation and averaging on the data sequence, outputting smooth displacement data.

[0067] When the real-time speed data is greater than or equal to the upper speed threshold, the adaptive multi-stage composite filter module activates the high-speed composite filter channel. Under high-speed conditions, the frequency of dynamic signal changes increases, requiring the filtering algorithm to have low phase delay. The high-speed composite filter channel adopts a combined architecture of a first-order hysteresis filter module and a Kalman filter module. The first-order hysteresis filter module is used to suppress high-frequency white noise. The Kalman filter module recursively estimates the data at the current moment based on the system state equation and measurement equation to maintain the dynamic tracking performance of the data output.

[0068] When the real-time speed data falls between the lower and upper speed thresholds, the system enters a numerical transition range. Within this range, the adaptive multi-level composite filter module calculates dynamic weighting coefficients based on the relative position of the real-time speed data within the range. The formula for calculating the dynamic weighting coefficients is as follows: ;in, For dynamic weighting coefficients, For real-time rotational speed data, This is the lower limit threshold for rotational speed. This is the upper limit threshold for rotational speed.

[0069] After calculating the dynamic weighting coefficients, the adaptive multi-level composite filter module uses these coefficients to perform linear weighted fusion of the output values ​​of the low-speed composite filter channel and the high-speed composite filter channel. The principle of weighted fusion is as follows: as the real-time rotational speed increases from the lower limit to the upper limit within the transition range, the weight ratio of the low-speed composite filter channel decreases linearly, while the weight ratio of the high-speed composite filter channel increases linearly simultaneously. The weighted fusion formula is expressed as: ; in, The final output displacement after weighted fusion. This is the output value of the low-speed composite filter channel. This is the output value of the high-speed composite filter channel. This weighted fusion algorithm maintains the continuity of the filtered data during smooth switching of operating conditions.

[0070] See attached document Figure 7 The present invention provides a design method for a high-precision magnetic levitation electronic bearing displacement and speed sensor, which may include: In another preferred embodiment, to optimize the computing power allocation of the digital processing unit, the system is equipped with a piecewise linear interpolation combined with algebraic operations as a safety backup mode for degraded operation in the event of extreme high load conditions or anomalies (i.e., in extreme abnormal situations such as limited hardware computing power, for example, when the system bottom layer determines through real-time monitoring that the CPU load rate of the digital processing unit exceeds a preset safety threshold, or when the forward inference task of the neural network module times out, the system will automatically trigger a switch, temporarily suspend the neural network module of the main path and enable lightweight compensation logic to ensure that the closed-loop control is uninterrupted, while the neural network module is still the main execution under normal conditions), thereby improving the reliability of the system's all-weather operation.

[0071] Based on the aforementioned physical model, the inductance difference and displacement exhibit an approximately linear proportional relationship near the mechanical center equilibrium position. As the distance between the rotor of the magnetic levitation bearing and the mechanical center equilibrium position increases, the leakage flux within the magnetic circuit and the edge effect of the magnetic field will cause a nonlinear distortion in the relationship between the inductance difference and displacement.

[0072] To reduce nonlinear measurement errors over large displacement ranges, the digital processing unit incorporates a piecewise linear interpolation module and a storage module. The storage module pre-loads a nonlinear calibration data table to provide a discrete mapping between the digital inductance difference and the standard physical displacement value. During the calibration phase, the system completes standard processing, dividing the full displacement range into multiple continuous calibration sub-intervals. To balance approximation accuracy and memory overhead, the sub-interval division density is dynamically set based on the curvature of the inductor air gap characteristic curve; high-density division is used in regions with high curvature and severe nonlinearity, while low-density division is used in approximately linear regions. The extracted set of node coordinates is stored in the storage module.

[0073] To further correct the nonlinear deviations caused by structural dimensional drift due to thermal expansion, a nonlinear error compensation module is also included within the digital processing unit. This module receives displacement measurement data from the piecewise linear interpolation module and performs algebraic compensation calculations in conjunction with the acquired ambient temperature data. The specific formula for thermal error compensation is as follows: ;in, This is the final output of the actual displacement. The displacement measurement data has been processed by piecewise linear interpolation. is the linear thermal expansion coefficient of the stator material of the magnetic levitation bearing. To obtain real-time ambient temperature data, This is the baseline ambient temperature for offline calibration experiments. This interpolation compensation mechanism helps to complete error correction across the entire measurement range without consuming a large amount of hardware computing power.

[0074] See attached document Figure 8The present invention provides a design method for a high-precision magnetic levitation electronic bearing displacement and speed sensor, which may include: In this embodiment, as a backup temperature calibration strategy when the system enters the aforementioned degraded operation mode and enables lightweight compensation logic, the magnetic levitation electronic bearing system experiences fluctuations in the resistivity of the detection coil and the permeability of the core material due to changes in ambient temperature during operation. Specifically, an increase in temperature increases the resistivity of the coil wires and reduces the quality factor, while the magnetic flux density and permeability of the core material also shift nonlinearly with ambient temperature. These fluctuations in physical parameters cause temperature drift in the original electromagnetic signal output by the sensor. To correct the temperature drift at the underlying electrical signal level, a dynamic temperature compensation module is incorporated within the digital processing unit.

[0075] The dynamic temperature compensation module receives real-time temperature data output from the temperature detection component. The module incorporates a polynomial temperature compensation algorithm. This algorithm calculates the temperature compensation coefficient based on the difference between the real-time temperature data and the reference ambient temperature. The formula for calculating the temperature compensation coefficient is as follows: ;in, This is the temperature compensation coefficient. It is the first-order temperature drift coefficient. It is the second-order temperature drift coefficient.

[0076] The first-order and second-order temperature drift coefficients were derived using a least-squares curve fitting algorithm based on data from offline high and low temperature cycling tests. During the calibration phase, the test system recorded the relative deviation of the sensor output signal under different temperature gradients. By constructing a polynomial error matrix and solving the normal equations, the coefficients of the aforementioned second-order polynomial were extracted and stored in the internal memory. The temperature dynamic compensation module uses the calculated temperature compensation coefficients to multiply and correct the digital voltage signal output by the analog-to-digital converter unit, thereby reducing the interference of temperature changes on system parameters. The aforementioned piecewise linear interpolation module, nonlinear error compensation module, and temperature dynamic compensation module together constitute the system's degraded backup logic for extreme operating conditions.

[0077] When the system triggers the degradation standby logic, its signal processing pipeline sequence is strictly defined as follows: First, the temperature dynamic compensation module performs multiplication correction on the original digital voltage signal to eliminate the temperature drift of the underlying electrical parameters; then, the piecewise linear interpolation module receives the corrected digital voltage signal and outputs the initial displacement measurement data according to the mapping relationship; finally, the nonlinear error compensation module performs thermal expansion algebraic compensation calculation on the initial displacement measurement data based on the ambient temperature, thereby outputting the final true displacement.

[0078] Returning to the system's overall main operating logic, in addition to the temperature drift issue, the sensor system simultaneously outputs displacement and rotational speed signals. During the fusion processing of multi-dimensional parameters and system calls, the time phase difference between signals introduces control delay. To ensure the time consistency of multi-source detection data, a clock synchronization and parsing module is incorporated into the digital processing unit.

[0079] The clock synchronization parsing module has an internal global hardware timer. This timer periodically generates synchronization trigger pulses based on the set system control cycle. The system control cycle is determined according to Shannon's sampling theorem and the bandwidth requirements of the closed-loop control of the magnetic levitation bearing system; in this embodiment, to match the 10kHz closed-loop control frequency, the system control cycle is set to 100 microseconds. The clock synchronization parsing module directly distributes the synchronization trigger pulses to the analog-to-digital converter unit and the high-speed timer module.

[0080] When the synchronization trigger pulse arrives, the analog-to-digital converter immediately initiates sampling and latching operations on the displacement voltage signal. Simultaneously, the high-speed timer module latches in parallel the high-precision timestamp triggered by the latest square wave edge and the current square wave pulse count value. Through a hardware-level trigger alignment mechanism, the clock synchronization parsing module reduces the time jitter caused by traditional software polling mechanisms, maintaining the synchronization of displacement and speed data in the time domain, and providing a reliable data source for subsequent high-speed closed-loop control.

[0081] See attached document Figure 9 The present invention provides a design method for a high-precision magnetic levitation electronic bearing displacement and speed sensor, which may include: In this embodiment, the magnetic levitation electronic bearing system contains high-power switching noise and electromagnetic leakage interference. To ensure the stability of displacement and speed parameter detection, the system has an isolation and shielding structure at the hardware physical level. The radial displacement detection component, axial displacement detection component, and speed detection component mentioned in the first part are all fixed above the mounting bracket in the preset positioning slot of the stator base.

[0082] Regarding the installation layout, the mounting bracket maintains a predetermined spatial distance from the power drive electromagnet on the stator base, and the mounting bracket is made of a non-magnetic material. The specific value of this spatial distance is determined based on the spatial attenuation characteristics of the leakage magnetic field of the power drive electromagnet; the system extracts the leakage magnetic flux density distribution through finite element simulation, setting this spatial distance to a safe threshold distance that prevents the leakage magnetic field strength from attenuating to a level that will not cause local saturation of the detection core. The non-magnetic material reduces the conduction path of the leakage magnetic flux generated by the power drive electromagnet to the internal magnetic circuits of each detection component. This physical isolation structure reduces the superposition interference of the main execution magnetic field on the weak magnetic field measured by the sensor.

[0083] At the electrical signal transmission level, the sensor system employs a multi-layered shielded cable assembly. Internally, this assembly uses a twisted-pair structure to transmit weak, raw detection signals. The physical principle of the twisted-pair structure ensures that the two parallel conductors experience equal amplitude and consistent direction of external electromagnetic interference, thus utilizing differential common-mode rejection to cancel spatially coupled common-mode noise. The multi-layered shielded cable assembly is externally wrapped with a metal braided shield. One end of the metal braided shield is connected to the system ground, while the other end remains suspended. This single-ended grounding mechanism prevents potential differences from multiple grounding points, thus avoiding ground loop currents, while simultaneously guiding high-frequency electromagnetic radiation coupled to the shield into the ground.

[0084] The radial displacement detection component, axial displacement detection component, and rotational speed detection component are all externally encapsulated with alloy shielding covers. These alloy shielding covers are made of high-permeability materials such as permalloy, forming a closed magnetic shielding cavity around the detection components. Stray magnetic fields from the external environment are conducted and closed along the surface of the alloy shielding cover, which has low magnetic reluctance, thereby reducing the magnetic flux penetrating to the internal detection coils.

[0085] The magnetically levitated rotor experiences mechanical vibration during high-speed rotation. To ensure the absolute mechanical reference for displacement detection remains intact, high-rigidity hard connections are used between each detection component, the mounting bracket, and the stator base. Simultaneously, to prevent high-frequency vibration energy from being transferred to circuit components and causing microscopic mechanical vibrations such as parasitic capacitance changes or poor contact, a polymer damping and insulating pad is installed at the mounting interface between the signal conditioning circuit board of the hardware circuit system and the stator base. This polymer damping and insulating pad is made of a polymer material (such as silicone or polyurethane) with a high internal loss factor.

[0086] The thickness and dynamic stiffness parameters of the polymer damping insulation layer are selected based on the high-frequency vibration main frequency band of the magnetic levitation rotor to ensure that the inherent resonant frequency of the mechanical structure avoids the effective measurement frequency band of the displacement sensor. The polymer damping insulation layer absorbs and dissipates the high-frequency mechanical vibration energy transmitted by the stator base. This damping design maintains the static stability of the physical positions of various electronic components in the hardware circuit system, reducing the mechanical modulation noise of the detection signal from the hardware source.

[0087] See attached document Figure 10 The present invention provides a design method for a high-precision magnetic levitation electronic bearing displacement and speed sensor, which may include: During the offline phase, this embodiment is equipped with a calibration platform component to perform static and dynamic performance calibrations across the entire measurement range, in order to evaluate the measurement performance of the radial displacement detection component, axial displacement detection component, and rotational speed detection component.

[0088] A micro-stage and a reference measuring instrument are mounted on the base of the calibration platform assembly. A calibration target simulating a rotor material is fixed on the motion axis of the micro-stage. The micro-stage drives the calibration target to perform multi-point stepping movement according to the set displacement step size. The reference measuring instrument uses a dual-frequency laser interferometer to measure the actual physical displacement of the calibration target in real time and output calibration reference data.

[0089] In the displacement performance calibration process, the micro-motion stage drives the calibration target to move sequentially from the set center zero point to the full-range boundaries at both ends. A data acquisition and analysis module is set up within the testing environment. This module synchronously receives calibration reference data output from the reference measuring instrument and measured displacement data output from the digital processing unit. The module records multiple sets of data at each step position point during the forward and reverse strokes, and uses the least squares method to perform linear fitting on these multiple sets of data to obtain the theoretically fitted straight line.

[0090] To quantify static indicators, the data acquisition and analysis module calculates the nonlinear error of the sensor system. The formula for calculating the nonlinear error is as follows: ;in, This is a nonlinear error. This represents the absolute value of the maximum deviation between the measured displacement data and the theoretically fitted straight line over the entire measurement range. This represents the full-scale output range of the sensor. The data acquisition and analysis module uses a similar method to further extract repeatability and hysteresis error indicators.

[0091] The system feeds back the specific measurement residual data of each step position point to the digital processing unit, and updates the calibration data table a second time by superimposing correction values ​​to reduce residual mechanical assembly errors. For the performance calibration of the speed detection component, a standard servo motor is introduced into the test system. The spindle of the standard servo motor and the signal gear plate are rigidly connected coaxially. The control unit of the test system controls the standard servo motor to perform step acceleration and deceleration within the rated speed range.

[0092] The data acquisition and analysis module acquires standard speed data fed back from the internal encoder of the standard servo motor in real time and simultaneously records the measured speed data parsed and output by the speed detection component. The module compares the time phase difference between the two sets of speed data under transient conditions and the numerical deviation under steady-state conditions to verify the coordination and matching between the underlying hardware design and the software processing algorithm. See attached document Figure 11 The present invention provides a design method for a high-precision magnetic levitation electronic bearing displacement and speed sensor, which may include: The system, in conjunction with calibration platform components and a data acquisition and analysis module, performs performance comparison tests between traditional independent sensor groups and the integrated sensor structure of this application under the same test conditions. The test conditions cover the sensor's full-scale displacement range and the range from zero to the rated maximum speed, and a pre-defined wideband electromagnetic interference signal is injected simultaneously. The frequency band and amplitude parameters of this wideband electromagnetic interference signal are set based on the switching frequency and high-order harmonic characteristics of the high-power switching amplifier for the magnetic levitation bearing, to simulate the harsh electromagnetic environment of a real industrial setting.

[0093] To improve displacement detection performance, the data acquisition and analysis module records the output data of the two sets of sensors at synchronized traverse positions. Traditional independent sensors, due to their separate mounting, often suffer from large full-range nonlinear errors due to accumulated structural tolerances and multi-point temperature drift. This solution integrates the radial and axial displacement detection components within the same non-magnetic mounting bracket. The digital processing unit utilizes a multi-stage composite filtering module and a nonlinear error compensation module to perform static and dynamic corrections on the displacement electrical signal. Test data comparison shows that the nonlinear error and hysteresis error of the displacement signal output in this embodiment exhibit a significant downward trend compared to traditional independent sensor sets.

[0094] To assess the speed detection performance, the test system compared the standard speed data output by the standard servo motor with the measured speed data analyzed and output by the speed detection component. Under the rated maximum speed condition, the clock synchronization analysis module within the digital processing unit uses a hardware global timer to uniformly switch the sampling timing. This timing alignment mechanism helps reduce the phase delay caused by traditional software polling. Through calculation and comparison, this embodiment shows lower time phase difference under transient acceleration / deceleration conditions and lower numerical deviation under steady-state conditions compared to the control group. To intuitively quantify the optimization effect of the overall system performance, the data acquisition and analysis module calculates the relative reduction rate of the comprehensive error. The formula for calculating the relative reduction rate of the comprehensive error is expressed as: ; in, The overall error reduction rate This refers to the root mean square error of measurement data from a traditional independent sensor array under specific operating conditions. This represents the root mean square error of the measurement data of the integrated sensor structure in this embodiment under the same operating conditions.

[0095] At the overall system level, the multi-dimensional detection components are integrated into a single layout, reducing the overall axial physical dimension of the magnetic levitation bearing spindle. This reduction in axial dimension improves the rotor's mechanical stiffness and optimizes critical speed characteristics. The physical isolation structure at the hardware level, combined with the composite compensation algorithm at the software level, demonstrates in practice the effectiveness of maintaining data resolution accuracy of the sensor system under complex electromagnetic environments and wide temperature ranges.

[0096] Specific application examples (combined with numerical calculations) This integrated sensor design was applied to a magnetic levitation centrifugal compressor. The equipment has a rated maximum speed of 60,000 RPM, a first-order critical speed of 15,000 RPM, a maximum ambient temperature of 85℃ under full load, and a displacement detection resolution of 0.3 μm.

[0097] The initial working air gap length is set to x0 = 0.4 mm. The displacement detection coil is wound with 0.08 mm diameter enameled copper wire, with N = 120 turns and a magnetic pole cross-sectional area S = 50 mm². The rotor end has Z = 4 magnetic tooth slots, and the speed detection coil has 150 turns. A quartz crystal oscillator is used to build an LC parallel resonant network to generate a sinusoidal alternating excitation signal, and the resonant frequency is tuned to f0. The amplitude fluctuation is controlled within ±0.5% of the MHz. The system clock frequency of the digital processing unit is 80MHz.

[0098] Assuming a transient condition, the rotor generates [something] along the positive X-axis direction. =50μm (0.05mm) displacement. At this time, the air gap on one side decreases to x1=0.35mm, and the air gap on the other side increases to x2=0.45mm.

[0099] After the working air gap length changes, according to the simplified formula Perform the calculation. Substitute the free permeability μ0= H / m, from which the change in inductance can be calculated: μH. The differential inductor signal is then input to a signal conditioning circuit that includes a preamplifier, a bandpass filter with a center frequency of 2MHz, a precision rectifier, and a 10kHz low-pass filter.

[0100] To meet the requirement of a maximum speed of 60,000 RPM, the system is based on an 80MHz clock frequency, and to satisfy the speed error... According to the performance requirements, the frequency switching threshold is set to 1kHz. When the compressor is in the 12,000RPM startup phase, the pulse frequency is calculated as f=(12000×4) / 60=800Hz based on the formula f=(n×Z) / 60.

[0101] Since 800Hz < 1kHz, the system calls the periodicity method for calculation; when the spindle speed increases to 45,000RPM, the pulse frequency f = (45000 × 4) / 60 = 3000Hz ≥ 1kHz, the system switches to the pulse and timestamp joint calculation method within a fixed time window (M / T method), thereby achieving full-range blind-zone-free detection.

[0102] To accommodate filtering requirements at both high and low speeds, the digital processing unit incorporates a multi-level composite filtering algorithm. At lower speeds, the system increases the weights of median filtering and moving average filtering to eliminate random fluctuations. When the speed exceeds 12,000 RPM and enters the numerical transition range, eventually crossing the 18,000 RPM upper limit threshold to enter high-speed operation, the system invokes an adaptive multi-level composite filtering algorithm to automatically reduce the weighting of the low-speed composite filtering channel, thereby minimizing high-speed dynamic response delay.

[0103] Experimental verification and effect comparison analysis A calibration test platform was built, and a dual-frequency laser interferometer was used as an independent benchmark for comparative testing in a wide temperature range of ±0.5mm radial and ±1mm axial range, and -40℃ to 85℃.

[0104] See attached document Figure 12 First, the displacement nonlinearity error and temperature drift characteristics are analyzed. Traditional split-type sensor assemblies, when the temperature exceeds 60℃ and large displacements occur, suffer from severe linearity divergence due to accumulated assembly tolerances and temperature drift, resulting in an error of up to 5.8μm. In this embodiment, after the system calls the error calibration neural network module (or switches to the piecewise linear interpolation calibration algorithm and temperature dynamic compensation module when the system triggers a high-load degraded operation mode), the error curve remains flat across the entire temperature range and the entire measurement range. The measured displacement resolution reaches 0.3μm, and the full-range linearity error remains stable below 0.05%FS.

[0105] Continue to refer to the appendix Figure 12 The accuracy of the speed signal resolution was analyzed. Within the bench acceleration range of 0 to 60,000 RPM, the traditional control group exhibited discrete fluctuations in its error curve at extremely low and high speeds due to detection blind spots and software polling mechanisms, with a maximum speed measurement error of approximately 0.25%. This embodiment utilizes a dual-channel signal synchronous acquisition driven by a co-source 80MHz high-speed clock, coupled with a Schmitt hysteresis comparator and shaping circuit, achieving a stable relative speed error of 0.05% across the entire range (meeting the ≤0.08% requirement), and demonstrating high uniformity within the 0 to 60,000 RPM range.

[0106] Under steady-state conditions of 50,000 RPM, external high-frequency electromagnetic interference and wide-band mechanical vibration were introduced. Test results showed that the internal processing phase delay of the traditional control group due to software polling waiting was as long as 150 μs or more, and the waveform glitches were obvious. In this embodiment, with the physical isolation of a 0.5 mm permalloy high-permeability magnetic shield, the synergy of a polymer shock-absorbing insulating pad (such as silicone material) and a co-source clock synchronization mechanism, the internal operation and synchronization phase delay of the measured signal were reduced to 15 μs (meeting the ≤18 μs requirement). Extremely low data processing overhead was achieved within a 100 μs system control cycle, fully meeting the high-speed closed-loop control requirements of high-precision magnetic levitation bearings.

Claims

1. A design method for a high-precision magnetic levitation electronic bearing displacement and speed sensor, characterized in that, include: The radial displacement detection component, axial displacement detection component, speed detection component and temperature detection component are integrated into the stator base, and an insulating shielding partition is provided between adjacent components. A sinusoidal alternating excitation signal is generated by a high-frequency excitation circuit and injected into the coils of the radial displacement detection component and the axial displacement detection component. The induced signals generated by the radial displacement detection component and the axial displacement detection component are converted into DC analog signals by the signal acquisition and conditioning circuit, and the pulse signal of the speed detection component is converted into a square wave signal by the hysteresis comparison circuit. The DC analog signal, after being processed by the analog-to-digital conversion unit, is synchronously input into the digital processing unit along with the temperature data collected by the temperature detection component. The square wave signal is input into the high-speed timer module of the digital processing unit. The digital processing unit calls the error calibration neural network module to perform error compensation on the displacement data obtained by converting the DC analog signal, calls the dual algorithm switching mechanism to parse the square wave signal, and uses the adaptive multi-level composite filter module to filter the output.

2. The method according to claim 1, characterized in that, The specific arrangement steps for the radial displacement detection component and the axial displacement detection component include: Four sets of radial displacement detection components are symmetrically arranged on the inner circumference of the stator base to form two symmetrical differential detection channels, and two sets of axial displacement detection components are symmetrically arranged at both ends of the axial direction. The speed detection component is independently arranged inside the stator base and is directly opposite the outer circumferential surface of the rotor end of the magnetic levitation bearing. The cores of the radial displacement detection component, the axial displacement detection component, and the rotational speed detection component are prepared by laminating nanocrystalline ribbons with a single layer thickness of 20 μm and a stacking factor of not less than 0.95, and are then encapsulated in an overall insulating manner after forming.

3. The method according to claim 1, characterized in that, The high-frequency excitation circuit, which includes a quartz crystal oscillator, an LC parallel resonant network, and a voltage regulator and current limiter module, is constructed. The voltage regulator and current limiter module is used to limit the output amplitude fluctuation of the sinusoidal alternating excitation signal to within a set threshold of ±0.5%. The signal acquisition and conditioning circuit is constructed, which includes a preamplifier, a bandpass filter, a rectifier circuit, and a low-pass filter. The sensed signal is a multi-channel symmetrical differential sensed signal. The preamplifier is used to perform linear amplification and differential operation processing on the input multi-channel symmetrical differential sensed signal.

4. The method according to claim 1, characterized in that, The input data dimension of the error calibration neural network module is set to two dimensions, specifically including the digital voltage signal representing the original inductance difference output by the analog-to-digital conversion unit, and the temperature data output by the temperature detection component; The digital processing unit calls the maximum-minimum normalization algorithm to uniformly map the two-dimensional input data to the numerical range between negative one and positive one, and then inputs it into the hidden layer of the error calibration neural network module. The hidden layer nodes of the error calibration neural network module are activated using a linear rectification function, and the output layer nodes of the error calibration neural network module are processed using a linear activation function to output the true displacement.

5. The method according to claim 4, characterized in that, The error calibration neural network module undergoes offline training before deployment, including the following steps: The combination of the digital voltage signal and the temperature data is set as the training sample, and the reference displacement data measured by the laser interferometer is set as the training label. The mean squared error is used as the loss function. The gradient descent optimization algorithm is used to backpropagate and iteratively update the weight coefficients and bias parameters of each layer of the error calibration neural network module until the mean squared error drops to the order of 10-14. The training stops then and the final parameters are stored in the non-volatile memory inside the digital processing unit.

6. The method according to claim 1, characterized in that, A preset frequency switching threshold is set inside the digital processing unit. The value of the frequency switching threshold is determined by the system clock frequency of the digital processing unit and the preset speed detection error requirement. The parsing steps of the dual-algorithm switching mechanism are as follows: When the frequency of the square wave signal is lower than the frequency switching threshold, the time period of adjacent square wave pulses is measured to obtain the frequency of the square wave signal. When the frequency of the square wave signal is greater than or equal to the frequency switching threshold, the frequency of the square wave signal is obtained within a fixed time window by combining the statistical number of square wave pulses with their corresponding timestamps. Finally, the real-time rotational speed data of the rotor is calculated based on the ratio between the frequency of the acquired square wave signal and the number of magnetic tooth slots preset at the rotor end of the magnetic levitation bearing.

7. The method according to claim 1, characterized in that, The adaptive multi-level composite filter module sets a lower speed threshold and an upper speed threshold. The lower speed threshold is 80% of the first-order critical speed of the rotor of the magnetic levitation bearing, and the upper speed threshold is 120% of the first-order critical speed of the rotor. When the real-time rotational speed data obtained by analyzing the square wave signal is less than or equal to the lower limit threshold of rotational speed, the low-speed composite filter channel containing the median filter module and the moving average filter module connected in series is activated. When the real-time speed data is greater than or equal to the upper limit threshold of the speed, the high-speed composite filter channel, which combines a first-order hysteresis filter module and a Kalman filter module, is activated. When the real-time speed data is in the numerical transition range between the upper speed limit threshold and the lower speed limit threshold, a dynamic weighting coefficient is calculated based on the relative position of the real-time speed data in the numerical transition range, and the output values ​​of the low-speed composite filter channel and the high-speed composite filter channel are linearly weighted and fused.

8. The method according to claim 1, characterized in that, A clock synchronization parsing module is provided inside the digital processing unit, and the clock synchronization parsing module is equipped with a global hardware timer. The global hardware timer generates synchronous trigger pulses according to the set system control cycle and directly distributes them to the analog-to-digital conversion unit and the high-speed timer module; When the synchronization trigger pulse arrives, the analog-to-digital conversion unit is forced to start sampling and latching the DC analog signal, and the high-speed timer module is simultaneously forced to latch the high-precision timestamp triggered by the latest square wave signal edge and the current square wave pulse count value in parallel.

9. The method according to claim 1, characterized in that, A multi-layer shielded cable assembly with an internal twisted-pair structure and an external metal braided shielding mesh for single-end grounding transmits the original detection signals generated by the radial displacement detection component, the axial displacement detection component, the speed detection component, and the temperature detection component. The radial displacement detection component, the axial displacement detection component, and the rotational speed detection component are fixed on a mounting bracket made of non-magnetic material, and an alloy shield made of permalloy material is encapsulated on the outside of the mounting bracket. A polymer shock-absorbing and insulating pad is provided at the mounting interface between the circuit board containing the hardware circuit system including the high-frequency excitation circuit, the signal acquisition and conditioning circuit, and the hysteresis comparison circuit and the stator base.

10. The method according to claim 1, characterized in that, The sensor design method also includes a full-range performance calibration step: A micro-motion stage is used to drive the calibration target to move in multiple steps, and the physical displacement is measured by a reference measuring instrument as calibration reference data. The data acquisition and analysis module synchronously acquires the measured displacement data output by the digital processing unit, records the measured displacement data of the bidirectional stroke, and uses the least squares method to perform linear fitting on the above multiple sets of data; The rotor of the standard servo motor connected to the magnetic levitation bearing is controlled to perform acceleration and deceleration. The standard speed data output by the standard servo motor and the measured speed data parsed by the digital processing unit are recorded simultaneously. The time phase difference of the two sets of speed data under transient conditions and the numerical deviation under steady-state conditions are compared.

Images