A space stepping motor stick suppression method based on multi-physical field coupling

Abstract

The application discloses a kind of based on multi-physical field coupling's aerospace stepping motor card stagnation inhibition method, it is related to motor reliability control and health management technical field, this method includes: the original data of motor operation is collected, the magnetic flux density, thermal induced displacement, permanent magnet demagnetization rate and physical damage index are calculated;While extracting feature vector input data driven model, obtaining data driven damage index.The comprehensive damage index is obtained by weighting fusion of the two, and the control target quantity is calculated.According to the comparison of thermal induced displacement, demagnetization rate and comprehensive damage index with preset threshold, the card stagnation risk level is determined, and the corresponding inhibition instruction is sent to the actuator, and the above steps are repeated after execution to form a closed loop control.The method realizes the leap from passive monitoring and early warning to active source inhibition, directly hits the physical nature of aerospace stepping motor card stagnation, can accurately intervene in early fault, thereby significantly improving the on-orbit operation reliability and mission success rate of motor system.

Description

Technical Field

[0001] This invention relates to the field of motor reliability control and health management technology, and more specifically to a method for suppressing jamming in aerospace stepper motors based on multi-physics coupling. Background Technology

[0002] Aerospace stepper motors are core actuators for spacecraft attitude control, antenna drive, and precision pointing mechanisms, and their long-term reliable operation in orbit is crucial. In the extreme environment of space, such as drastic temperature changes, high vacuum, and radiation, critical motor components, especially bearings and permanent magnets, experience performance degradation due to the strong coupling of multiple physical fields involving electromagnetism, force, and heat. Mechanical deformation caused by uneven thermal expansion and irreversible demagnetization of permanent magnets are the two main modes of motor failure, seriously threatening the safety of space missions.

[0003] Traditional reliability assurance methods mainly focus on two directions: First, improving the motor's environmental tolerance by optimizing material and structural design, such as using radiation-resistant materials, improving bearings, and enhancing heat dissipation. For example, Zhang Dayin, in his paper "An Aerospace Stepper Motor and Its Radiation Hardening Method," reduced the impact of radiation on motor performance by optimizing the motor's internal structure and shielding materials. Second, predicting the motor's remaining service life using various data-driven or model-driven algorithms to enable condition-based maintenance. For instance, the patent with publication number CN121389022B, entitled "A Stepper Motor Life Prediction Method and System Based on Multiphysics Coupling," achieves high-precision assessment of the motor's health status and remaining life by integrating a multiphysics coupling model and a data-driven model.

[0004] However, existing lifespan prediction methods, including the aforementioned advanced methods, ultimately output status information or warning signals. For spacecraft with long lifespans, high reliability, and extremely difficult maintenance, predictive capabilities alone are insufficient; a closed-loop control system capable of automatically triggering and executing targeted mitigation actions based on accurate prediction results must be developed.

[0005] Therefore, there is an urgent need for an innovative method that can inherit and utilize high-precision life prediction information, directly convert it into control commands to suppress motor jamming faults, and form a closed-loop execution, so as to substantially improve the on-orbit survivability and mission success rate of aerospace stepper motors. Summary of the Invention

[0006] To achieve the goal of moving from passive monitoring and early warning to active root cause suppression, and to improve the on-orbit reliability and mission success rate of aerospace stepper motors, this invention adopts the following technical solution:

[0007] This invention provides a method for suppressing jamming in aerospace stepper motors based on multi-physics coupling, comprising the following steps:

[0008] S1. Collect multi-physics field raw data during motor operation, calculate magnetic flux density and thermally induced displacement based on the raw data; calculate permanent magnet demagnetization rate based on the magnetic flux density; calculate physical damage index based on permanent magnet demagnetization rate and thermally induced displacement;

[0009] S2. Extract feature vectors from the original data, input the feature vectors into the trained data-driven model, and obtain the data-driven damage index.

[0010] S3. The physical damage index and the data-driven damage index are weighted and fused to obtain the comprehensive damage index;

[0011] S4. Calculate the control target quantity, which specifically includes the desired bearing preload and the compensated reference current;

[0012] S5. Based on the comparison of the thermally induced displacement, permanent magnet demagnetization rate and comprehensive damage index with the preset threshold, determine the current motor jamming risk level, generate the corresponding suppression command according to the risk level, and send the suppression command and the corresponding control target quantity to the corresponding actuator.

[0013] Furthermore, in step S1, the raw data is acquired using current, temperature, position, and vibration sensors; the raw data includes: ab two-phase current. i a ( k )and i b ( k ), bearing temperature T (k), rotor position mechanical angle θ m (k) and vibration acceleration a v ( k ), where k is the sampling time.

[0014] Furthermore, in step S1,

[0015] The formula for calculating the magnetic flux density B(k) is as follows:

[0016] Two-phase current i a ( k )and i b ( k ), Number of turns N in motor windings, and constant magnetomotive force F generated by the permanent magnet. pm Substituting into the following formula, we obtain the total magnetomotive force F(k):

[0017]

[0018] Where θe (k) is the rotor electrical angle, which is determined by the mechanical angle and the number of pole pairs p of the motor. n The conversion yields:

[0019]

[0020] Based on the total magnetomotive force F(k) and rotor angle θ m (k) corresponds to the air gap permeability Λ(θ) m (k)), the magnetic flux density B(k) is calculated:

[0021]

[0022] Where Ae is the effective cross-sectional area of ​​the air gap magnetic field;

[0023] The demagnetization rate η of the permanent magnet d (k) is calculated using the following formula:

[0024]

[0025] Where, k d denoted as the demagnetization rate coefficient, β as the magnetic field strength coefficient, e as the base of the natural logarithm, and Δt as the sampling time interval;

[0026] The thermally induced displacement δ T (k) is calculated using the following formula:

[0027]

[0028] Where T(k) is the temperature, α is the coefficient of thermal expansion of the material, L is the bearing length, and T0 is the room temperature.

[0029] Furthermore, in step S1, the physical damage index D p ( k The calculation formula is:

[0030]

[0031] in, or dmax To allow the maximum demagnetization rate, d max To allow the maximum safe deformation; α 1, α 2 is the weighting coefficient, and α 1+ α 2 = 1.

[0032] Furthermore, in step S2, the feature vector includes: time-domain statistical features, frequency-domain features, and instantaneous values ​​and statistical features extracted based on the magnetic flux density, thermally induced displacement, and permanent magnet demagnetization rate;

[0033] The time-domain statistical features include the domain mean μ. x and variance σ x 2 The specific formula is as follows:

[0034]

[0035]

[0036] Where N is the length of the data window used to calculate statistical characteristics, and x is the signal with time-domain characteristics;

[0037] The frequency domain features are extracted using the main frequency amplitude value obtained through Fast Fourier Transform. A y The specific formula is as follows:

[0038]

[0039] in, Let y be the sequence of signal values ​​from the (k-N+1)th sampling point to the kth sampling point, where y is a signal with frequency domain characteristics.

[0040] Furthermore, in step S2, the data-driven damage index D d (k) The formula is:

[0041] At least one feature vector X(k) Input the trained data-driven model to obtain the real-time data-driven damage index. D d ( k ):

[0042]

[0043] in, f model This refers to a machine learning model trained on historical data that has the ability to model time series, specifically a Long Short-Term Memory (LSTM) network or a Convolutional Neural Network (CNN).

[0044] Furthermore, in step S3, the specific formula for calculating the comprehensive damage index is as follows:

[0045] Real-time data will drive the damage index D d (k) and the physical damage index D calculated in real time p (k) The final comprehensive damage index D(k) is obtained by weighting and fusing the data using dynamic weighting coefficients:

[0046]

[0047] Weighting coefficient ωp (k) and ω d (k) Adjust dynamically according to the following rules:

[0048]

[0049] in, and These represent the prediction uncertainties of the physical model and the data-driven model at the current moment, respectively. The uncertainty coefficient of the physical model is the variance of its predicted values, obtained through error propagation analysis; the uncertainty coefficient of the data-driven model is the variance of its multiple prediction results, obtained through ensemble learning methods.

[0050] Furthermore, in step S4, the specific formula for the desired bearing preload is:

[0051] According to thermally induced displacement Calculate the amount of bearing preload adjustment required to compensate for this deformation. :

[0052]

[0053] in, K The preload-displacement compensation coefficient is obtained through finite element simulation or experimental calibration of the bearing.

[0054] The expected preload at the next moment is:

[0055]

[0056] Among them, bearing assembly preload The following constraints need to be met:

[0057]

[0058] in, m The coefficient of friction, R As the lever arm, T rated The rated torque of the motor. C The preset safety factor threshold;

[0059] The specific formula for the compensated reference current is as follows:

[0060] Demagnetization rate of permanent magnets or d ( k Substitute the values ​​into the following formula to calculate the reference current of the motor controller current loop after compensation. , :

[0061]

[0062]

[0063] in , For the first k The reference current of the motor controller current loop at the sampling time, and , This is the reference current for the maximum allowable current loop of the motor.

[0064] Furthermore, in step S5, the determination of the motor jamming risk level is specifically as follows:

[0065] If thermally induced displacement d T ( k Exceeding the first-level warning threshold d th If 1, it is determined that there is a risk of mechanical jamming;

[0066] If demagnetization rate or d ( k Exceeding the Level 2 warning threshold or th 2 ， This indicates a risk of electromagnetic jamming.

[0067] If the comprehensive damage index D ( k Exceeding the third-level security threshold D th 3 If so, it is determined that there is a high overall risk.

[0068] Furthermore, in step S5, the suppression command specifically includes: when it is determined that there is a risk of mechanical jamming, generating a first type of suppression command: preload adjustment command;

[0069] When an electromagnetic jamming risk is detected, a second type of suppression command is generated: a current compensation command and a drive mode switching command.

[0070] When a high overall risk is identified, a third type of protection instruction is generated: a derated operation instruction or a safe shutdown sequence instruction.

[0071] The beneficial effects of the above-described technical solutions provided in the embodiments of the present invention include at least the following:

[0072] This invention achieves proactive root-cause suppression of jamming faults in aerospace stepper motors. First, it utilizes multi-physics raw data to calculate magnetic flux density, thermally induced displacement, permanent magnet demagnetization rate, and physical damage index in parallel, and integrates the data-driven damage index to obtain a comprehensive damage index, significantly improving the accuracy and robustness of jamming state assessment. Second, based on the comparison of thermally induced displacement, demagnetization rate, and comprehensive damage index with preset thresholds, it accurately determines the mechanical, electromagnetic, or high-risk level, and dynamically calculates the desired bearing preload and compensated reference current as control targets, ensuring the quantification and feasibility of suppression measures. Finally, it generates targeted suppression commands based on the risk level, such as preload adjustment, current compensation, drive mode switching, or safety protection, and sends them to the corresponding actuators. After execution, the steps are automatically repeated to form a continuous closed loop. This method fundamentally changes the traditional passive early warning or post-event processing mode, enabling proactive intervention in physical roots such as thermal deformation and demagnetization during the fault's incipient stage, effectively preventing jamming and significantly enhancing the on-orbit reliability and mission success assurance capability of aerospace stepper motors in extreme environments. Attached Figure Description

[0073] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0074] Figure 1 This is a flowchart of a method for suppressing jamming in aerospace stepper motors based on multi-physics coupling, provided in an embodiment of the present invention.

[0075] Figure 2 This is a flowchart for calculating thermally induced displacement provided in an embodiment of the present invention.

[0076] Figure 3 This is a flowchart of the calculation process for demagnetizing a permanent magnet provided in an embodiment of the present invention.

[0077] Figure 4 This is a flowchart for calculating the comprehensive damage index provided in an embodiment of the present invention. Detailed Implementation

[0078] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0079] like Figure 1 As shown, this invention discloses a method for suppressing jamming in aerospace stepper motors based on multi-physics coupling, comprising the following steps:

[0080] S1. Collect multi-physics field raw data during motor operation, calculate magnetic flux density and thermally induced displacement based on the raw data; calculate permanent magnet demagnetization rate based on the magnetic flux density; calculate physical damage index based on permanent magnet demagnetization rate and thermally induced displacement; collect raw data such as two-phase current, bearing temperature, rotor mechanical angle and vibration acceleration in real time through current, temperature, position and vibration sensors; calculate air gap magnetic flux density, mechanical displacement caused by thermal expansion and permanent magnet demagnetization rate based on electromagnetic and thermodynamic formulas, and obtain physical damage index by weighted fusion.

[0081] S2. Extract feature vectors from the original data, input at least one feature vector into the trained data-driven model to obtain the data-driven damage index; extract feature vectors such as time-domain mean, variance, frequency-domain main frequency amplitude, and instantaneous values ​​of key physical quantities from the original signal; and output the data-driven damage index using time-series models such as long short-term memory networks or convolutional neural networks.

[0082] S3. The physical damage index and the data-driven damage index are weighted and fused to obtain a comprehensive damage index. The weights are dynamically adjusted according to the prediction uncertainties of the physical model and the data-driven model, and the two are weighted and summed to obtain a more accurate and robust comprehensive damage index.

[0083] S4. Calculate the control target quantity, which specifically includes the desired bearing preload and the compensated reference current; calculate the preload adjustment amount required to compensate for mechanical deformation based on thermally induced displacement, and satisfy the friction torque constraint; compensate the current loop reference current proportionally based on the permanent magnet demagnetization rate to maintain the output torque.

[0084] S5. Based on the comparison of the thermally induced displacement, demagnetization rate, and comprehensive damage index with preset thresholds, determine the current motor jamming risk level, generate corresponding suppression commands based on the risk level, and send the suppression commands and corresponding control target quantities to the corresponding actuators; if the thermally induced displacement exceeds the threshold, mechanical jamming risk is determined; if the demagnetization rate exceeds the threshold, electromagnetic jamming risk is determined; if the comprehensive damage index exceeds the threshold, high comprehensive risk is determined, realizing multi-level risk identification, issuing preload adjustment commands for mechanical risks, issuing current compensation and drive mode switching commands for electromagnetic risks, and issuing derating or safety shutdown commands for high risks.

[0085] This invention proposes a method for suppressing jamming in aerospace stepper motors based on multi-physics coupling. By integrating physical and data-driven damage assessments, it accurately identifies the physical root causes of jamming, namely thermally induced deformation or permanent magnet demagnetization, and automatically generates targeted active suppression commands such as dynamic adjustment of preload, current compensation, and mode switching, forming a closed-loop control. Compared to traditional passive early warning or post-accident handling, this method can intervene at the source of the fault in its early stages, effectively preventing jamming and significantly improving the operational reliability and mission success rate of aerospace stepper motors in extreme environments.

[0086] The following is a detailed description of the aerospace stepper motor jamming suppression method based on multiphysics coupling of the present invention:

[0087] Step S1: Collect raw multiphysics data of the motor during operation. Based on the raw data, calculate the magnetic flux density, thermally induced displacement, permanent magnet demagnetization rate, and physical damage index, specifically including:

[0088] Multiphysics data acquisition:

[0089] Using current, temperature, position, and vibration sensors, the system measures and collects raw multi-physics data of the motor during operation, including two-phase currents a(k) and ib(k), bearing temperature T(k), rotor position mechanical angle θm(k), and vibration acceleration av(k), where k is the sampling time.

[0090] Calculate the magnetic flux density B(k):

[0091] Two-phase current i a ( k )and i b ( k ), Number of turns N in motor windings, and constant magnetomotive force F generated by the permanent magnet. pm Substituting into the following formula, we obtain the total magnetomotive force F(k):

[0092]

[0093] Where θ e (k) is the rotor electrical angle, which is determined by the mechanical angle and the number of pole pairs p of the motor. n The conversion yields:

[0094]

[0095] Based on the total magnetomotive force F(k) and rotor angle θ m (k) corresponds to the air gap permeability Λ(θ) m (k)), the magnetic flux density B(k) is calculated:

[0096]

[0097] Where Ae is the effective cross-sectional area of ​​the air gap magnetic field.

[0098] Calculate thermally induced displacement:

[0099] Substitute the temperature T(K) into the following formula to calculate the thermally induced displacement δ of the mechanical structure. T (k):

[0100]

[0101] Where α is the coefficient of thermal expansion of the material, L is the bearing length, and T0 is the room temperature.

[0102] Calculate the demagnetization rate:

[0103] Substitute the temperature T (K) and magnetic flux density B (K) into the following formula to calculate the demagnetization rate η of the permanent magnet. d (k):

[0104]

[0105] Where, k d denoted as the demagnetization rate coefficient, β as the magnetic field strength coefficient, e as the base of the natural logarithm, and Δt as the sampling time interval;

[0106] Calculate the physical damage index D p (k): Demagnetization rate η of permanent magnet d (k) and the thermally induced displacement δ of the mechanical structure T (k) Substituting into the following formula, we obtain the physical damage index D. p (k) is:

[0107]

[0108] Where, η dmax To allow the maximum demagnetization rate, δ max To allow the maximum safe deformation; α1 and α2 are weighting coefficients, and α1+α2=1.

[0109] Step S2: Extract feature vectors from the original data, and input at least one feature vector into the trained data-driven model to obtain the data-driven impairment index, specifically including:

[0110] Calculation of eigenvectors:

[0111] The dimension of the feature vector X(k) includes a combination of at least one of the following feature parameters:

[0112] 1. Time-domain statistical features extracted from raw signals of current, vibration, and temperature, such as the time-domain mean μ. x With variance σx 2 ;

[0113]

[0114]

[0115] Where N is the length of the data window used to calculate statistical characteristics, and x is a signal with time-domain characteristics, such as current, vibration, temperature, etc.

[0116] 2. Frequency domain features extracted from raw current and vibration signals; such as the main frequency amplitude A extracted by Fast Fourier Transform (FFT). y :

[0117]

[0118] in, Let y be a sequence of signal values ​​from the (k-N+1)th sampling point to the kth sampling point, where y is a signal with frequency domain characteristics, such as current, vibration acceleration, angle, magnetic flux density, etc.

[0119] 3. Magnetic flux density B(k), thermally induced displacement δ T (k), Demagnetization rate η of permanent magnet d The instantaneous value of (k) and its statistical characteristics.

[0120] Data-driven damage index calculation:

[0121] The constructed feature vector X(k) is input into the trained data-driven model to obtain the real-time data-driven damage index D. d (k):

[0122]

[0123] Among them, f model The machine learning model with temporal modeling capabilities trained on historical data is specifically a Long Short-Term Memory Network (LSTM) or a Convolutional Neural Network (CNN).

[0124] Specifically, the architecture of the Long Short-Term Memory (LSTM) network is a single-layer or multi-layer stacked LSTM. For example, in the context of the embodiment described, a feasible configuration is: Input layer [T=100, D] -> LSTM layer (128 units) -> Dropout (0.4) -> LSTM layer (64 units) -> Fully connected layer (32, ReLU) -> Output layer (1). This structure can effectively capture long-term dependencies.

[0125] The architecture of convolutional neural networks (CNNs) is 1D-CNN, which excels at extracting local feature patterns. An exemplary structure is: Input layer [T=100, D] -> 1D convolutional layer (64 kernels, size 5, ReLU) -> Max pooling layer (size 2) -> 1D convolutional layer (128 kernels, size 3, ReLU) -> Global average pooling layer (or flattening layer) -> Fully connected layer (64, ReLU) -> Output layer (1). This model can efficiently identify local distortion features in vibration and current signals.

[0126] Machine learning models can also include hybrid models and attention-based models:

[0127] The hybrid model's architecture is as follows: first, a 1D-CNN layer is used to extract high-level features, and then the feature sequence is input into an LSTM for temporal modeling. For example: Input layer -> 1D-CNN layer (feature extraction) -> remodeling into a sequence -> LSTM layer (temporal learning) -> fully connected output layer. This structure can capture both local features and model long-term dependencies.

[0128] Attention-based models, such as the Transformer encoder, employ a architecture that treats feature sequences as token sequences and computes global dependencies through self-attention, eliminating the need for recursive structures. Its core components include positional encoding, multi-head attention layers, and feedforward network layers. Although relatively new, its powerful sequence modeling capabilities make it a suitable advanced implementation for this invention.

[0129] In practice, one or more of the above models can be selected for training and evaluation based on computing resources, requirements for prediction latency, and characteristics of historical data. The LSTM used in the embodiment is one proven and effective choice.

[0130] The dataset required for model training can be derived from one or more of the following sources to ensure the model's generalization ability and robustness:

[0131] Source 1: Specific test data for the target motor model, including:

[0132] Ground-based accelerated life test data: The motor is subjected to accelerated aging tests in simulated space environments, such as thermal vacuum and radiation, until its performance degrades or it jams, and all sensor data are recorded throughout the process.

[0133] High-fidelity multiphysics simulation data: Using finite element analysis and magnetic-thermal-mechanical coupling simulation software, the operating state of motors under various working conditions and fault modes is simulated to generate synthetic data covering a wide range of scenarios.

[0134] Source 2: Historical on-orbit data.

[0135] Collect historical telemetry data of similar or identical aerospace stepper motors during their on-orbit operation, especially data that includes abnormal or performance degradation phases.

[0136] Source 3: Laboratory standard test data.

[0137] Test data under normal operating conditions and under conditions that introduce known faults, such as insufficient lubrication.

[0138] Source 4: Data-augmented derivative datasets. The original data is subjected to noise addition, scaling, and time-axis distortion to expand the dataset size and improve the model's robustness against interference.

[0139] During training, the construction of the feature vector X(k) is described in step S2. The training labels for the data-driven damage index Dd(k) should be constructed based on physical measurements or expert assessments that are synchronized with the features and can accurately reflect the health of the motor. For example, synchronously measured output torque error, efficiency degradation rate, or a comprehensive health score based on more sophisticated instrument monitoring can be used as supervisory signals.

[0140] The core task of the loss function is regression prediction; therefore, mean squared error (MSE) or mean absolute error (MAE) are standard loss functions. Huber loss can be used in conjunction with it to enhance robustness to outlier labels.

[0141] The optimizer is Adam or AdamW, which are commonly used default choices, and the initial learning rate is usually between 1e-4 and 1e-3.

[0142] In addition to Dropout, regularization methods such as L2 weight decay and batch normalization can also be used to prevent overfitting.

[0143] Key hyperparameters in hyperparameter tuning, such as learning rate, number of network layers, number of units, dropout rate, and batch size, should be determined through cross-validation or grid search or random search on independent validation sets. The parameters given in the examples (e.g., 128 units, Dropout 0.4) can serve as a starting point for optimization.

[0144] Uncertainty estimation for dynamic weighting: To calculate the predictive uncertainty of a data-driven model, in addition to the ensemble learning method already mentioned, the following methods can also be used:

[0145] Monte Carlo Dropout method: Dropout is enabled multiple times during model inference, and the variance of multiple prediction results is used as an estimate of uncertainty.

[0146] Bayesian neural networks: directly model the weight distribution and can naturally output the predicted distribution and its uncertainty.

[0147] By combining the above-mentioned various models and data, the technical effects of the present invention can be verified through systematic comparative experiments:

[0148] On the same test dataset, a baseline model was established using a pure physics model (Dp). Individual models such as LSTM, 1D-CNN, and CNN-LSTM were trained separately. The data-driven models and the physics model Dp were then fused using a dynamic weighting method to obtain multiple fused versions.

[0149] Step S3: Weighted fusion of the physical damage index and the data-driven damage index to obtain the comprehensive damage index, which specifically includes:

[0150] Comprehensive damage index calculation:

[0151] Real-time data will drive the damage index D d (k) and the physical damage index D calculated in real time p (k) The final comprehensive damage index D(k) is obtained by weighting and fusing the data using dynamic weighting coefficients:

[0152]

[0153] Weighting coefficient ω p (k) and ω d (k) Adjust dynamically according to the following rules:

[0154]

[0155] in, and These represent the prediction uncertainties of the physical model and the data-driven model at the current moment, respectively. The uncertainty coefficient of the physical model is the variance of its predicted values, obtained through error propagation analysis; the uncertainty coefficient of the data-driven model is the variance of its multiple prediction results, obtained through methods such as ensemble learning.

[0156] Step S4: Calculate the control target quantities, which specifically include the desired bearing preload and the compensated reference current.

[0157] Calculation of expected bearing assembly preload:

[0158] Based on real-time thermally induced displacement Calculate the amount of bearing preload adjustment required to compensate for this deformation. :

[0159]

[0160] Wherein, K is the preload-displacement compensation coefficient, which is obtained through finite element simulation or experimental calibration of the bearing.

[0161] The expected preload at the next moment is:

[0162]

[0163] Among them, bearing assembly preload The following constraints need to be met:

[0164]

[0165] Where μ is the coefficient of friction, R is the lever arm, and T is the torque. rated C is the rated torque of the motor, and C is the preset safety factor threshold, the value of which needs to be determined based on reliability design requirements and ground test data.

[0166] Calculation of reference current after compensation:

[0167] The demagnetization rate η of the permanent magnet d (k) Substitute into the following formula to calculate the reference current of the motor controller current loop after compensation. , :

[0168]

[0169]

[0170] in, , Let be the reference current of the motor controller current loop at the k-th sampling time, and , This is the reference current for the maximum allowable current loop of the motor.

[0171] Step S5: Based on the comparison of thermally induced displacement, demagnetization rate, and comprehensive damage index with preset thresholds, determine the current motor jamming risk level, generate corresponding suppression commands based on the risk level, and send the suppression commands and corresponding control target quantities to the corresponding actuators. Specifically, this includes:

[0172] The calculated critical state thermally induced displacement δ T (k), Demagnetization rate η of permanent magnet d (k), Comprehensive damage index D(k) and preset threshold δ th 1. η th 2. D th 3. Compare and determine the current motor jamming risk level:

[0173] like Figure 2 As shown, if the thermally induced displacement δ T (k) Exceeds the first-level warning threshold δ thIf 1, it is determined that there is a risk of mechanical jamming. When it is determined that there is a risk of mechanical jamming, a first type of suppression command is generated, including: a preload adjustment command for dynamically adjusting the bearing preload.

[0174] The preload adjustment command is sent to the adjustable preload mechanism to dynamically adjust the bearing preload. To the desired bearing assembly preload .

[0175] like Figure 3 As shown, if the demagnetization rate η d (k) Exceeds the second-level warning threshold η th 2. If an electromagnetic jamming risk is identified, a second type of suppression command is generated, including:

[0176] Current compensation command: Used to adjust the winding current setpoint to compensate for torque loss caused by demagnetization.

[0177] Drive mode switching command: Controls the driver to switch from full step or low microstep mode to high microstep mode.

[0178] The current compensation command and drive mode switching command are sent to the motor driver to adjust the current loop reference current in real time. , Reference current after compensation to the given value , And control the driver to switch from full step or low microstepping mode to high microstepping mode.

[0179] like Figure 4 As shown, if the comprehensive damage index D(k) exceeds the third-level safety threshold D th If the condition is 3, it is determined that there is a high overall risk. When a high overall risk is determined, a third type of protection instruction is generated, including: a derated operation instruction or a safe shutdown sequence instruction.

[0180] The derating operation command or safety shutdown sequence command is sent to the motor driver to execute the corresponding derating or shutdown protection command.

[0181] After the instruction is executed, return to steps S1-S5 to collect data and assess the status for the next sampling cycle, thus forming a closed-loop control.

[0182] The following example illustrates the aerospace stepper motor jamming suppression method based on multiphysics coupling of the present invention:

[0183] Aerospace stepper motors are indispensable in various drive mechanisms of spacecraft. Their long-term reliable operation in orbit faces the risk of failure due to mechanical jamming and electromagnetic demagnetization caused by extreme environments and multi-physics coupling. Traditional fault handling methods are mostly reactive or rely on simple threshold alarms, making precise pre-emptive intervention difficult. The jamming suppression method based on multi-physics coupling proposed in this invention focuses on converting high-precision state prediction information, such as thermally induced displacement and demagnetization rate, into targeted active control actions in real time, thus forming a closed-loop active health assurance system. To clearly demonstrate the implementation process and effects of this method, the following uses a certain type of aerospace stepper motor as an example, and elaborates on the invention in detail with illustrations.

[0184] This embodiment uses a certain type of aerospace stepper motor as the implementation object. The basic parameters of the motor are as follows: number of pole pairs. p n =4, number of winding turns N =100 turns, permanent magnet magnetomotive force F pm =600A, effective cross-sectional area of ​​the air gap Ae =0.01m², bearing thermal expansion coefficient bearing length L =0.015m, demagnetization rate coefficient Magnetic field strength coefficient β =0.1 / T.

[0185] Sampling time interval Δt =0.1s, reference temperature T0=25°C, to sample k Taking time 3600 as an example, sensor data: Phase A current i a =2.8A, phase b current i b =2.5A, bearing temperature T =78.3°C, rotor mechanical angle i m =127°, corresponding air gap permeability radiation-temperature coupling coefficient kr =0.015°C / (T·s), maximum demagnetization rate or dmax =0.2, maximum permissible safety deformation d max =15μm, weighting coefficients for demagnetization rate and deformation. α 1 = 0.4 α 2 = 0.6.

[0186] Preload-displacement compensation coefficient K = 1.5 × 10⁶ N / m, friction coefficient μ = 0.15, lever arm R = 0.1m, rated torque T rated =0.5 The safety factor C = 1.5. The two-phase current reference values ​​given by the motor controller are: I aref (3600) = 2.8A, I bref (3600) = 2.5A. The warning threshold is set as follows: d th 1 = 8μm, or th 2 = 0.15 D th 3 = 0.8. Current time. F P =200N, which satisfies the condition.

[0187]

[0188] In implementation, the deep learning network is implemented using a Long Short-Term Memory (LSTM) network. The network input layer receives the extracted multiphysics feature parameter time-series data, the hidden layer contains 128 LSTM units, and the output layer is a fully connected layer used for regression prediction of damage values. D d Uncertainty in physical model predictions =0.03, uncertainty in data-driven model predictions =0.02,

[0189] 1) Data collection:

[0190] The motor's two-phase current (A and B) is measured and collected using current, temperature, and position sensors. i a and i b Bearing temperature T Rotor position mechanical angle i m ;

[0191] 2) Calculate the magnetic flux density:

[0192] From a mechanical perspective i m Number of pole pairs of the motor p n The rotor electrical angle was calculated. i e :

[0193]

[0194] Utilizing two-phase current i a , i b and the number of turns of the motor windingN With the constant magnetomotive force of permanent magnets F pm The total magnetomotive force was calculated. F :

[0195]

[0196] Based on total magnetomotive force F With rotor mechanical angle i m The corresponding air gap permeability L(θ m ) and the effective cross-sectional area of ​​the air gap magnetic field. Ae The magnetic flux density was calculated. B :

[0197]

[0198] 3) Calculate thermally induced displacement:

[0199] Temperature T Coefficient of thermal expansion α bearing length L Reference temperature T Substitute 0 into the following formula to calculate the thermally induced displacement of the mechanical structure. d T :

[0200]

[0201] 4) Calculate the demagnetization rate:

[0202] Temperature T Magnetic flux density B Demagnetization rate coefficient k d Magnetic field strength coefficient β Substitute into the following formula to calculate the demagnetization rate of the permanent magnet. or d :

[0203]

[0204]

[0205] 5) Calculate the physical damage index:

[0206] Demagnetization rate of permanent magnets or d Thermally induced displacement of mechanical structure d T Maximum allowable safe deformation d max Maximum allowable demagnetization rate ordmax and weighting coefficients α 1, α 2. Substitute into the following formula to obtain the physical damage index. D p :

[0207]

[0208]

[0209] 6) Calculate the eigenvectors:

[0210] Extracting the time-domain mean from the raw signals of current, vibration, and temperature. m x With variance s x 2 ,in, N This is the length of the data window used to calculate statistical properties.

[0211]

[0212] Extracting the main frequency amplitude from the original current and vibration signals using Fast Fourier Transform (FFT) A y :

[0213]

[0214] The average of the original signals of two-phase current, vibration, and temperature will be used. m x With variance s x 2 The dominant frequency amplitude of the two-phase current and vibration signal A y as well as B , d T , or d As feature vectors X The 14 dimensions.

[0215] 7) Calculate the data-driven damage index:

[0216] eigenvectors X Input the trained LSTM model to obtain the real-time data-driven damage index. D d :

[0217]

[0218] 8) Calculate the comprehensive damage index:

[0219] Integrating physical models and data-driven models k Prediction uncertainty at time 3600 and Adjustment rules for the weighting coefficients:

[0220]

[0221] And from oh p +oh d = 1 can be obtained oh p =0.4, oh d =0.6.

[0222] Data-driven damage index D d Physical damage index D p Substitute into the following formula to obtain the final comprehensive damage index. D :

[0223]

[0224] 9) Calculate the desired bearing assembly preload:

[0225] Real-time thermally induced displacement Substitute into the following formula to calculate the bearing preload adjustment required to compensate for the deformation. :

[0226]

[0227] The expected preload at the next moment is:

[0228]

[0229] 10) Calculate the reference current after compensation:

[0230] Demagnetization rate of permanent magnets or d and two-phase current reference values I aref , I bref Substitute the values ​​into the following formula to calculate the reference current of the motor controller current loop after compensation. , :

[0231]

[0232]

[0233] Since the demagnetization rate is extremely small, the compensation amount is negligible, but the calculation process is shown. If the demagnetization is severe, the compensation will be significant.

[0234] 11) Determine the level of risk of congestion:

[0235] thermally induced displacement d T ( k Demagnetization rate of permanent magnets or d ( k Comprehensive damage index D ( k ) and preset threshold d th 1. or th 2 , D th 3 Compare and determine the current motor jamming risk level:

[0236] 1. d T =8.79μm> d th 1=8μm

[0237] 2. or d =2.136×10 5< or th 2 = 0.15

[0238] 3. D =0.39272< D th 3 = 0.8

[0239] because d T > d th 1. The system has determined that there is a risk of mechanical jamming.

[0240] 12) Generate jamming suppression strategy instructions:

[0241] The system detected a risk of mechanical jamming. Based on this, it generated a first-type suppression command: a preload adjustment command, which adjusted the bearing preload to... .

[0242] 13) Execute the jamming suppression command:

[0243] The generated preload adjustment command is sent to the adjustable preload mechanism. One implementation of this mechanism is a bearing sleeve with a built-in shape memory alloy ring. The diameter of the shape memory alloy ring is changed by controlling the heating current of the ring, thereby precisely applying the calculated desired preload. After execution, the system automatically returns to step 1 and begins data acquisition for the next sampling cycle. It uses the new sensor data to evaluate the effects of the preload adjustment, such as changes in temperature and vibration, and then repeats the evaluation and decision-making processes from step 2 to step 11, thereby forming a continuous closed-loop control.

[0244] If demagnetization rate is detected in a certain period or d ( k )> or th 2. The system will determine that there is a risk of electromagnetic jamming and generate and execute a second type of suppression command, that is, simultaneously send a current compensation command and a drive mode switching command to the motor driver. If the comprehensive damage index... D ( k )Exceed D th 3. If the third type of protection instruction is triggered, the limit will be reduced or the system will be shut down safely.

[0245] In summary, this invention proposes a method for suppressing jamming in aerospace stepper motors based on multiphysics coupling, applicable to aerospace stepper motor systems. First, a multiphysics coupling model is used to calculate in real time the key state variables revealing the physical root causes of jamming: thermally induced displacement, demagnetization rate, and comprehensive damage index. Second, intelligent decision-making based on the evaluation results generates specific suppression strategies, calculating specific suppression parameters in real time according to different risk sources. Finally, the execution of the suppression strategy achieves closed-loop feedback, directly sending the generated instructions to the actuator for dynamic adjustment and initiating the next cycle of sensing and evaluation. This method achieves a leap from passive monitoring and early warning to active root cause suppression, directly addressing the physical essence of aerospace stepper motor jamming, enabling precise intervention in the early stages of faults, thereby significantly improving the on-orbit reliability and mission success rate of the motor system.

[0246] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0247] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for suppressing jamming in aerospace stepper motors based on multiphysics coupling, characterized in that, This includes the following steps, which are executed cyclically, to form a closed-loop control: S1. Collect multi-physics field raw data during motor operation, calculate magnetic flux density and thermally induced displacement based on the raw data; calculate permanent magnet demagnetization rate based on the magnetic flux density; calculate physical damage index based on permanent magnet demagnetization rate and thermally induced displacement; The physical damage index D p (k) is calculated using the following formula: Where, η dmax To allow the maximum demagnetization rate, δ max To allow the maximum safe deformation; α1 and α2 are weighting coefficients, and α1 + α2 = 1; S2. Extract feature vectors from the original data, input the feature vectors into the trained data-driven model, and obtain the data-driven damage index. The data-driven damage index D d (k) The formula is: By inputting at least one feature vector X(k) into the trained data-driven model, the real-time data-driven damage index D is obtained. d (k): Among them, f model This refers to a machine learning model trained on historical data that has the ability to model time series, specifically a Long Short-Term Memory (LSTM) network or a Convolutional Neural Network (CNN). S3. The physical damage index and the data-driven damage index are weighted and fused to obtain the comprehensive damage index; The specific formula for calculating the comprehensive damage index is as follows: Real-time data will drive the damage index D d (k) and the physical damage index D calculated in real time p (k) The final comprehensive damage index D(k) is obtained by weighting and fusing the data using dynamic weighting coefficients: Weighting coefficient ω p (k) and ω d (k) Adjust dynamically according to the following rules: in, and These represent the prediction uncertainties of the physical model and the data-driven model at the current moment, respectively. The uncertainty coefficient of the physical model is the variance of its predicted values, obtained through error propagation analysis; the uncertainty coefficient of the data-driven model is the variance of its multiple prediction results, obtained through ensemble learning methods. S4. Calculate the control target quantity, which specifically includes the desired bearing preload and the compensated reference current; The specific formula for the desired bearing preload is as follows: According to thermally induced displacement Calculate the amount of bearing preload adjustment required to compensate for deformation. : Wherein, K is the preload-displacement compensation coefficient, which is obtained through finite element simulation or experimental calibration of the bearing; The expected preload at the next moment is: Bearing assembly preload The following constraints need to be met: Where μ is the coefficient of friction, R is the lever arm, and T is the torque. rated C is the rated torque of the motor, and C is the preset safety factor threshold. The specific formula for the compensated reference current is as follows: The demagnetization rate η of the permanent magnet d (k) Substitute into the following formula to calculate the reference current of the motor controller current loop after compensation. , : in, , Let be the reference current of the motor controller current loop at the k-th sampling time, and , This is the reference current for the maximum allowable current loop of the motor; S5. Based on the comparison of the thermally induced displacement, permanent magnet demagnetization rate and comprehensive damage index with the preset threshold, determine the current motor jamming risk level, generate the corresponding suppression command according to the risk level, and send the suppression command and the corresponding control target quantity to the corresponding actuator.

2. The method as described in claim 1, characterized in that, In step S1, the raw data is acquired using current, temperature, position, and vibration sensors; the raw data includes: the two-phase current i of phases ab. a (k) and i b (k), bearing temperature T(k), rotor position mechanical angle θ m (k) and vibration acceleration a v (k), where k is the sampling time.

3. The method as described in claim 2, characterized in that, In step S1 The formula for calculating the magnetic flux density B(k) is as follows: The two-phase current i a (k) and i b (k) Number of turns in the motor winding N, constant magnetomotive force F generated by the permanent magnet pm Substituting into the following formula, we obtain the total magnetomotive force F(k): Where θ e (k) is the rotor electrical angle, which is determined by the mechanical angle and the number of motor pole pairs p. n The conversion yields: Based on the total magnetomotive force F(k) and rotor angle θ m (k) corresponds to the air gap permeability Λ(θ) m (k)), the magnetic flux density B(k) is calculated: Where Ae is the effective cross-sectional area of ​​the air gap magnetic field; The demagnetization rate η of the permanent magnet d (k) is calculated using the following formula: Where, k d denoted as the demagnetization rate coefficient, β as the magnetic field strength coefficient, e as the base of the natural logarithm, and Δt as the sampling time interval; The thermally induced displacement δ T (k) is calculated using the following formula: Where T(k) is the temperature, α is the coefficient of thermal expansion of the material, L is the bearing length, and T0 is the room temperature.

4. The method as described in claim 1, characterized in that, In step S2, the feature vector includes: time-domain statistical features, frequency-domain features, and instantaneous values ​​and statistical features extracted based on the magnetic flux density, thermally induced displacement, and permanent magnet demagnetization rate; The time-domain statistical features include the domain mean μ. x and variance σ x 2 The specific formula is as follows: Where N is the length of the data window used to calculate statistical characteristics, and x is the signal with time-domain characteristics; The frequency domain features are extracted using the main frequency amplitude value A obtained through Fast Fourier Transform. y The specific formula is as follows: in, Let y be the sequence of signal values ​​from the (k-N+1)th sampling point to the kth sampling point, where y is a signal with frequency domain characteristics.

5. The method as described in claim 1, characterized in that, In step S5, the determination of the motor jamming risk level is specifically as follows: If the thermally induced displacement δ T (k) Exceeds the first-level warning threshold δ th If 1, it is determined that there is a risk of mechanical jamming; If the demagnetization rate η d (k) Exceeds the second-level warning threshold η th 2. If so, it is determined that there is a risk of electromagnetic jamming. If the comprehensive damage index D(k) exceeds the third-level safety threshold D th If the result is 3, it is determined that there is a high overall risk.

6. The method as described in claim 1, characterized in that, In step S5, the suppression command specifically includes: when it is determined that there is a risk of mechanical jamming, a first type of suppression command is generated: preload adjustment command; When an electromagnetic jamming risk is detected, a second type of suppression command is generated: a current compensation command and a drive mode switching command. When a high overall risk is identified, a third type of protection instruction is generated: a derated operation instruction or a safe shutdown sequence instruction.

Images