Linear motor thrust fluctuation real-time compensation control method and system

Abstract

The invention discloses a real-time compensation control method and system for thrust fluctuation of a linear motor, and belongs to the technical field of linear motors, and the method comprises the steps: collecting a hybrid fluctuation source in the operation process of the linear motor, splitting the hybrid fluctuation source, and generating independent geometric nonlinear fluctuation signals; further screening the mixed fluctuation source, screening out a non-geometric nonlinear fluctuation signal, and preprocessing the non-geometric nonlinear fluctuation signal; establishing an independent compensation strategy, and respectively performing targeted compensation on geometric nonlinear fluctuation and non-geometric nonlinear fluctuation; and establishing a feed-forward compensation network, inputting the compensation strategy into the feed-forward compensation network, predicting a fluctuation component, and performing pre-compensation processing in advance. In the implementation process of the technical scheme provided by the invention, the pertinence and effectiveness of a compensation strategy are ensured through precise splitting and preprocessing of the hybrid fluctuation source, and meanwhile, the compensation amount is adjusted in real time by using the prediction function of the feedforward compensation network, the thrust fluctuation is remarkably reduced, and the operation stability of the linear motor is improved.

Description

Technical Field

[0001] This application relates to the field of linear motor technology, specifically to a method and system for real-time compensation and control of thrust fluctuation in linear motors. Background Technology

[0002] A linear motor is a transmission device that converts electrical energy into linear motion mechanical energy without the need for an intermediate conversion structure. It is also called a linear motor, linear motor, etc. It has the advantages of high efficiency and high precision and is widely used in precision machining, automation equipment and other fields.

[0003] In the application of linear motors, thrust fluctuation is a particularly prominent problem, directly affecting the stability of equipment operation and machining accuracy. Thrust fluctuation is mainly caused by the coupling of factors such as cogging force, end effect, current harmonics, and time-varying parameters. It manifests as periodic fluctuations in thrust, leading to increased positioning errors and aggravated vibration and noise. In existing technologies, various methods have been tried to address these problems, such as offline compensation methods, which use offline modeling and parameter identification to pre-set compensation values, and hardware-accelerated compensation methods, such as using high-performance computing units to improve the real-time performance of traditional compensation schemes. For example, DSP-based model predictive control algorithms can be used to identify fluctuation models online and generate compensation quantities, thereby realizing the detection, compensation, and optimization of thrust fluctuation in linear motors.

[0004] However, some problems still exist in the implementation of the above-mentioned existing technologies. For example, offline compensation methods rely on accurate mathematical models of motors, such as cogging force formulas. However, in actual working conditions, motor parameters are easily affected by factors such as temperature and wear, causing compensation errors to accumulate over time. While hardware-accelerated compensation methods improve real-time performance, they require calling complex GPU kernels, leading to increased costs and wasted computing resources.

[0005] Therefore, it is necessary to provide a method and system for real-time compensation and control of thrust fluctuation of linear motors to solve the above problems.

[0006] It should be noted that the information disclosed in this background section is only for understanding the background technology of this application concept, and therefore may include information that does not constitute prior art. Summary of the Invention

[0007] Based on the aforementioned problems in the existing technology, the problem to be solved by this application is to provide a method and system for real-time compensation control of thrust fluctuation of a linear motor, which achieves accurate compensation by analyzing different types of fluctuation signals and formulating corresponding compensation schemes.

[0008] The technical solution adopted by this application to solve its technical problem is: a real-time compensation control method for thrust fluctuation of a linear motor, comprising: The mixed wave sources during the operation of the linear motor are collected and decomposed to generate independent geometric nonlinear wave signals; The mixed wave sources are further screened to identify non-geometric nonlinear wave signals, which are then preprocessed. Establish independent compensation strategies to provide targeted compensation for geometric nonlinear fluctuations and non-geometric nonlinear fluctuations respectively; A feedforward compensation network is established, the compensation strategy is input into the feedforward compensation network, and the fluctuation components are predicted to perform pre-compensation processing in advance.

[0009] In the implementation of the technical solution of this application, the targetedness and effectiveness of the compensation strategy are ensured by accurately splitting and preprocessing the mixed fluctuation source. At the same time, the prediction function of the feedforward compensation network is used to adjust the compensation amount in real time, which significantly reduces thrust fluctuation and improves the operating stability of the linear motor.

[0010] Furthermore, the separation of the hybrid wave source includes the following steps: Geometric nonlinear fluctuations are modeled, and a screening model is established based on geometric feature parameters. This screening model is a feature screening equation used to identify and extract the components of geometric nonlinear fluctuations. The extracted geometric nonlinear wave components were verified by multi-domain analysis, and anomalous wave components were removed. The multi-domain analysis included time domain analysis, frequency domain analysis, and spatial domain analysis. Based on the results of multi-domain analysis, geometric nonlinear fluctuations are labeled, and a corresponding fluctuation characteristic database is established. The labeled geometric nonlinear fluctuation data are re-input into the screening model to verify the separation process, and the fluctuation feature database is updated based on the verification results.

[0011] Furthermore, time-domain analysis uses time-domain waveforms to identify the fluctuation period and amplitude, frequency-domain analysis uses fast Fourier transform to extract the dominant frequency, and spatial-domain analysis calculates the thrust difference between adjacent acquisition devices by collecting thrust fluctuations at fixed intervals along the motion direction of the mover, thereby determining the fluctuation characteristics in the spatial domain.

[0012] Furthermore, the dominant frequency refers to the frequency that plays a dominant role in the geometric nonlinear fluctuation component. This dominant frequency will cause the linear motor to resonate during operation. Once it is locked as the characteristic frequency, it can be used as the reference frequency in subsequent analysis.

[0013] Further, the preprocessing process includes: establishing a flux linkage observation model; comparing observed and actual values ​​based on changes in inductance and resistance during operation to obtain flux linkage deviation; establishing a nonlinear source wave screening equation and filtering out nonlinear terms; using the flux linkage deviation value to weight the nonlinear terms to identify electromagnetic nonlinear wave components; performing wavelet transform multi-scale analysis to decompose transient waves into components of different frequency bands and identify wave characteristics within each frequency band; and judging and marking non-geometric waves based on current harmonic analysis and characteristic thresholds.

[0014] Furthermore, by employing continuous wavelet transform and calculating the cross-correlation wavelet coefficients, the instantaneous coupling fluctuations of thrust and current are extracted, thereby locating the transient impact.

[0015] Furthermore, the feature threshold uses a thrust amplitude threshold and a current harmonic threshold as dual thresholds for judgment, which can identify non-geometric fluctuations while avoiding misjudgments caused by current noise.

[0016] Furthermore, establishing an independent compensation strategy further includes: The two types of wave signals are parameterized. The geometric nonlinear wave is described by a periodic function, and the non-geometric nonlinear wave is described by a nonlinear distortion function. Corresponding compensation models are then established for each. A multimodal fusion classifier is established to process multi-domain data, including time domain, frequency domain, and spatial domain data. Weights are allocated to non-geometric nonlinear fluctuations through a fully connected network, and nonlinear distortions are dynamically adjusted. The compensation strategy output by the multimodal fusion classifier and the fully connected network is used to obtain the compensation scheme.

[0017] Furthermore, the time-domain data uses an LSTM network to capture its time-series correlation data, the frequency-domain data uses a convolutional neural network to extract local frequencies, and the spatial-domain data uses a graph neural network for spatial topology analysis. Finally, the features of the three domains are fused to form a comprehensive discrimination model.

[0018] A real-time compensation control system for thrust fluctuation of a linear motor includes: The hybrid wave source separation module is used to collect hybrid wave sources during the operation of the linear motor, and to separate the hybrid wave sources to generate independent geometric nonlinear wave signals; The filtering module is used to further filter the mixed wave sources, filter out non-geometric nonlinear wave signals, and preprocess them. An independent compensation module is used to establish independent compensation strategies to provide targeted compensation for geometric nonlinear fluctuations and non-geometric nonlinear fluctuations respectively. The feedforward compensation module is used to establish a feedforward compensation network. The compensation strategy is input into the feedforward compensation network, and the fluctuation components are predicted to perform pre-compensation processing in advance.

[0019] The beneficial effects of this application are: The linear motor thrust fluctuation real-time compensation control method and system provided by this application ensures the pertinence and effectiveness of the compensation strategy by accurately decomposing and preprocessing the mixed fluctuation sources. At the same time, by utilizing the prediction function of the feedforward compensation network, the compensation amount is adjusted in real time, which significantly reduces thrust fluctuation and improves the operating stability of the linear motor.

[0020] In addition to the purposes, features, and advantages described above, this application has other purposes, features, and advantages. A further detailed description of this application will be provided below with reference to the figures. Attached Figure Description

[0021] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the overall process of a real-time compensation control method for thrust fluctuation of a linear motor in this application. Figure 2 This is a schematic diagram of the module configuration of a linear motor thrust fluctuation real-time compensation control system according to this application. Detailed Implementation

[0022] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0024] Example 1: As Figure 1 As shown, this application provides a real-time compensation control method for thrust fluctuation of a linear motor. This method is used to detect thrust fluctuation of the linear motor in real time and compensate based on the detection results, thereby effectively suppressing thrust fluctuation and improving the stability of the thrust motor operation. The compensation control method includes the following steps: Step 101: Collect the mixed wave sources during the operation of the linear motor, and decompose the mixed wave sources to generate independent geometric nonlinear wave signals; The operation of a linear motor involves dynamic changes in various physical quantities. These dynamic changes can lead to different thrust fluctuations, which can be broadly categorized into linear and nonlinear sources. For example, periodic fluctuations caused by structural defects, such as the periodic meshing of stator slots with the mover permanent magnet or conductor bars, can cause uneven air gap permeability. When the mover moves, the air gap permeability changes periodically with position, leading to air gap magnetic flux fluctuations and generating periodic electromagnetic forces. There are also some geometric nonlinear sources, such as the non-coincidence of the mover's center of mass and geometric center, which causes changes in air gap length, resulting in reluctance fluctuations and thrust fluctuations at the same frequency as the rotational speed. Nonlinear sources are mainly fluctuations caused by nonlinear electromagnetic characteristics, such as the nonlinear changes in the motor's electromagnetic parameters with current or flux linkage, which generate thrust fluctuations related to the square of the current. In existing technologies, complex filtering algorithms and multivariate control strategies are typically used to suppress these different fluctuation sources. However, due to the high complexity of these algorithms and control strategies, real-time compensation is difficult to achieve, and they consume significant system resources, resulting in unsatisfactory performance in practical applications. Therefore, in this embodiment, the overall fluctuation source is first collected and then broken down, thereby eliminating the need to deploy multiple sensors during the collection process, simplifying the collection process, and reducing the accumulation of errors between collection devices. Specifically, the breakdown of the mixed fluctuation source includes the following steps: Step A: Model the geometric nonlinear fluctuations and establish a screening model based on the geometric feature parameters. This screening model is a feature screening equation used to identify and extract the components of the geometric nonlinear fluctuations. For the extraction of geometric nonlinear fluctuations, establishing a mathematical model is a commonly used method in existing technologies. For example, by using a periodic variation model of air gap magnetic permeability and combining it with the offset of the mover's center of mass, the reluctance fluctuation equation can be derived, thereby extracting the geometric nonlinear fluctuation components. However, this method requires consideration of various complex factors during the derivation process, resulting in limited model accuracy and a large computational load, making it difficult to apply in real time. Especially for linear motors of different models and application scenarios, repeated modeling is required, leading to poor usability. Therefore, in this embodiment, a screening model is established based on geometric feature parameters. This screening model is implemented through feature screening equations. For example, for the periodic meshing of stator slots and mover permanent magnets, feature parameters based on the number of stator slots and the spacing between mover permanent magnets can be set, and the periodic components can be extracted using Fourier transform. In the screening process, since no specific numerical derivation is involved, but rather the feature parameters with periodic repetition are extracted through matching and screening of feature parameters. It should be noted that the selection of the number of stator slots and the spacing of the mover permanent magnets mentioned above is only an example. In practical applications, due to the complexity of linear motors, there will be other parameters, such as mover mass distribution and air gap length change rate, which can all be included as characteristic parameters in the screening model. These will not be elaborated in this embodiment. Specifically, the screening model can use a machine learning feature screening model as a framework, such as a support vector machine or a neural network. In the early stage, it is trained with a large amount of sample data to accurately identify geometric nonlinear fluctuation characteristics and quickly screen out the corresponding components in practical applications. When the geometric parameters of the linear motor are input into the model, the parameters can be automatically matched and the feature screening equation can be called to screen and extract the geometric nonlinear components. Step B: Perform multi-domain analysis on the extracted geometric nonlinear fluctuation components to verify them and remove abnormal fluctuation components. The multi-domain analysis includes time domain analysis, frequency domain analysis and spatial domain analysis. After extracting the geometrically nonlinear wave components, a multi-domain analysis is needed to verify the extracted components, as using only a screening model may misclassify some non-geometrically introduced fluctuations, such as mechanical vibrations and electromagnetic interference, as geometrically nonlinear waves. This multi-domain analysis includes time-domain analysis, frequency-domain analysis, and spatial-domain analysis. Time-domain analysis uses time-domain waveforms to identify the wave period and amplitude. Frequency-domain analysis uses Fast Fourier Transform to extract the dominant frequency, which is then identified as the characteristic frequency. The dominant frequency refers to the frequency that plays a dominant role in the geometrically nonlinear wave components. The dominant frequency, which causes resonance in the linear motor during operation, can be locked as a characteristic frequency and used as a reference frequency in subsequent analysis for more accurate fluctuation identification. Spatial domain analysis, on the other hand, calculates the thrust difference between adjacent acquisition devices by collecting thrust fluctuations at fixed intervals along the motion direction of the mover, thereby determining the fluctuation characteristics in the spatial domain. For example, three sensors are arranged on the mover at fixed intervals and continuously collect thrust data. If the thrust difference between adjacent sensors is greater than a preset threshold, it can be determined that there are significant geometric nonlinear fluctuations in that interval. By performing multi-domain analysis on the geometric nonlinear fluctuation components, we can further verify their source and properties, thereby ensuring the accuracy of the screening results. In this process, some abnormal fluctuation components can be identified and eliminated in time during comparative analysis, thus eliminating the need for additional screening processes in subsequent data analysis and reducing the complexity of data processing. Step C: Based on the results of the multi-domain analysis, the geometric nonlinear fluctuations are labeled, and a corresponding fluctuation characteristic database is established; After performing multi-domain analysis, geometric nonlinear fluctuations can be labeled, and the target fluctuation characteristics can be extracted during the labeling process. This extraction process is called the separation process. The labeling method is to classify and label the fluctuation characteristics based on their performance in the time domain, frequency domain, and spatial domain, combined with preset thresholds. For example, based on the period and amplitude characteristics of the time domain waveform, the dominant frequency in the frequency domain, and the thrust difference between adjacent sensors in the spatial domain, the fluctuations can be divided into three levels: high, medium, and low, each corresponding to different degrees of geometric nonlinear influence, and recorded in the fluctuation characteristic database, thereby providing a basis for subsequent fault diagnosis and fluctuation compensation. Step D: Re-input the labeled geometric nonlinear wave data into the screening model to verify the separation process, and update the wave characteristic database based on the verification results; In addition to initial screening, the screening model can also re-input the labeled geometric nonlinear fluctuation data after screening to verify the accuracy of the separation process. For example, if there are misjudgments or omissions in the labeled data, error feedback will be generated after re-inputting it into the model, that is, the model output does not match the actual fluctuation characteristics. This allows the separation process to be verified, and the fluctuation characteristic database can be updated based on the feedback results after verification. Step 102: Further screen the mixed wave sources to identify non-geometric nonlinear wave signals and preprocess them; In addition to geometrically nonlinear fluctuations, hybrid wave sources also contain other types of wave signals, such as non-geometrically nonlinear fluctuations caused by electromagnetic properties. These fluctuations are mainly caused by electromagnetic induction and current changes, exhibiting non-periodic and nonlinear characteristics, and their amplitude is related to current and frequency changes. Traditional methods based on current harmonics can only suppress fluctuations at specific frequencies and cannot address time-varying parameter problems. Therefore, preprocessing is required after screening. Specifically, this process includes: A flux linkage observation model is established. Based on the changes in inductance and resistance during operation, the observed values ​​are compared with the actual values ​​to obtain the flux linkage deviation. The nonlinearity of electromagnetic parameters is usually due to the decrease in inductance with current caused by core saturation and the increase in resistance with frequency caused by winding eddy currents. Through the flux linkage observation model, the changes in inductance and resistance can be monitored in real time, and the flux linkage deviation can be calculated. This deviation value directly reflects the degree of electromagnetic nonlinearity. For example, when the deviation value exceeds the preset threshold, it can be judged as a significant electromagnetic nonlinear fluctuation. A nonlinear source wave screening equation is established, and the nonlinear terms are screened out. The nonlinear terms are weighted using the flux deviation value, thereby identifying the electromagnetic nonlinear wave components. Non-geometric nonlinear fluctuations are related to the square term and product term of the current in terms of electrical principle. Therefore, by establishing a nonlinear source fluctuation screening equation and accurately extracting these terms, combined with magnetic flux deviation weighting, electromagnetic nonlinear fluctuations can be effectively identified. Furthermore, this equation can be combined with the residual method to extract the fluctuation components that cannot be explained by the linear model, thereby distinguishing between geometric nonlinear (periodic) and non-geometric nonlinear (nonlinear distortion) fluctuations. Wavelet transform multi-scale analysis is performed to decompose transient fluctuations into components of different frequency bands and identify the fluctuation characteristics in each frequency band. Electromagnetic nonlinear fluctuations are often accompanied by transient impacts, such as local saturation caused by sudden current changes. Traditional Fourier transform cannot locate local time-frequency features. Therefore, continuous wavelet transform is used. By calculating the cross-correlation wavelet coefficients, the instantaneous coupling fluctuations of thrust and current can be extracted, such as the thrust spike corresponding to the current rising edge, thereby achieving accurate location of transient impacts. In the process of wavelet transform, transient waves are decomposed into components of different frequency bands. By analyzing the frequency band characteristics, the amplitude and phase changes of the waves in each frequency band are determined, thereby obtaining the frequency domain distribution law of transient impacts. Based on current harmonic analysis and characteristic thresholds, non-geometric fluctuations are identified and marked. Electromagnetic nonlinear fluctuations are significant in higher harmonics of current. By extracting higher harmonic components through harmonic analysis and combining them with characteristic thresholds, it is possible to determine whether there is significant electromagnetic nonlinearity. Furthermore, by combining the thrust amplitude threshold and the current harmonic threshold as a dual threshold for judgment, non-geometric fluctuations can be identified. In addition, it is possible to avoid misjudgments caused by current noise, such as measurement errors or sensor interference. Step 103: Establish independent compensation strategies to compensate for geometric nonlinear fluctuations and non-geometric nonlinear fluctuations respectively; Because geometric and non-geometric nonlinear waves have different causes and characteristics, their mixing can lead to mutual interference between compensation strategies. For example, using geometric compensation to suppress electromagnetic waves can result in compensation lag, while using non-geometric compensation for geometric waves may cause overcompensation. Therefore, separate compensation strategies need to be developed. Specifically: Step 301: Parameterize the two types of wave signals, describe the geometric nonlinear wave with a periodic function, and describe the non-geometric nonlinear wave with a nonlinear distortion function, and establish corresponding compensation models respectively. Among them, the periodic function can be represented by the Fourier series, which is a tool that can decompose a periodic signal into multiple sine and cosine components, and accurately describe the periodic characteristics of geometric nonlinear fluctuations through these components; while the nonlinear distortion function is represented by polynomial approximation, which can capture the complex distortion pattern of non-geometric nonlinear fluctuations, thereby establishing corresponding compensation models respectively. Step 302: Establish a multimodal fusion classifier to process multi-domain data, which includes time domain, frequency domain, and spatial domain data; Traditional classifiers are typically designed as single-modal systems, such as SVMs, which struggle to capture features comprehensively when dealing with mixed multi-domain data. Therefore, a multi-modal fusion classifier is developed. This classifier employs independent methods for processing time-domain, frequency-domain, and spatial-domain data, and then integrates these independent methods into a single classifier. This enables comprehensive feature extraction and accurate classification of multi-domain data. Specifically, an LSTM network is used for time-domain data to capture its time-series correlation data, a convolutional neural network is used for frequency-domain data to extract local frequencies, and a graph neural network is used for spatial topology analysis of spatial-domain data. Finally, the features from the three domains are fused to form a comprehensive discriminative model. Step 303: Weight the non-geometric nonlinear fluctuations using a fully connected network to dynamically adjust the nonlinear distortion; For non-geometric nonlinear data, using a fully connected network for weight allocation can dynamically adjust the compensation intensity according to the fluctuation characteristics, ensuring accurate compensation without overcompensation, thereby effectively improving the stability and response speed of the system. Step 304: Output the compensation strategy from the multimodal fusion classifier and the fully connected network to obtain the compensation scheme; After completing the above steps, the output results of the multimodal fusion classifier and the fully connected network are integrated to form the final compensation scheme. For example, in geometric fluctuations, the Fourier series coefficients are dynamically adjusted to optimize periodic compensation; in non-geometric fluctuations, the polynomial parameters are fine-tuned to accurately correct distortions, ensuring the synergy of the overall compensation strategy. Step 104: Establish a feedforward compensation network, input the compensation strategy into the feedforward compensation network, predict the fluctuation components, and perform pre-compensation processing in advance.

[0025] Feedforward compensation networks not only have predictive capabilities, but can also dynamically adjust compensation strategies based on real-time data. For example, in geometric fluctuations, they can predict periodic trends and adjust Fourier series coefficients in advance; in non-geometric fluctuations, they can predict distortion patterns and dynamically fine-tune polynomial parameters to achieve precise pre-compensation, thereby compensating for thrust fluctuations in linear motors and improving the smoothness and efficiency of motor operation.

[0026] Example 2: Figure 2 As shown, this application also proposes a real-time compensation control system for thrust fluctuation of a linear motor. This system operates the compensation control method described in Embodiment 1. The system includes: The hybrid wave source separation module is used to collect hybrid wave sources during the operation of the linear motor, and to separate the hybrid wave sources to generate independent geometric nonlinear wave signals; The filtering module is used to further filter the mixed wave sources, filter out non-geometric nonlinear wave signals, and preprocess them. An independent compensation module is used to establish independent compensation strategies to provide targeted compensation for geometric nonlinear fluctuations and non-geometric nonlinear fluctuations respectively. The feedforward compensation module is used to establish a feedforward compensation network. The compensation strategy is input into the feedforward compensation network, and the fluctuation components are predicted to perform pre-compensation processing in advance.

[0027] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for real-time compensation and control of thrust fluctuation in a linear motor, characterized in that: include: The mixed wave sources during the operation of the linear motor are collected and decomposed to generate independent geometric nonlinear wave signals; The mixed wave sources are further screened to identify non-geometric nonlinear wave signals, which are then preprocessed. Establish independent compensation strategies to provide targeted compensation for geometric nonlinear fluctuations and non-geometric nonlinear fluctuations respectively; A feedforward compensation network is established, the compensation strategy is input into the feedforward compensation network, and the fluctuation components are predicted to perform pre-compensation processing in advance.

2. The method for real-time compensation and control of thrust fluctuation in a linear motor according to claim 1, characterized in that: Decomposing a hybrid wave source involves the following steps: Geometric nonlinear fluctuations are modeled, and a screening model is established based on geometric feature parameters. This screening model is a feature screening equation used to identify and extract the components of geometric nonlinear fluctuations. The extracted geometric nonlinear wave components were verified by multi-domain analysis, and anomalous wave components were removed. The multi-domain analysis included time domain analysis, frequency domain analysis, and spatial domain analysis. Based on the results of multi-domain analysis, geometric nonlinear fluctuations are labeled, and a corresponding fluctuation characteristic database is established. The labeled geometric nonlinear fluctuation data are re-input into the screening model to verify the separation process, and the fluctuation feature database is updated based on the verification results.

3. The method for real-time compensation and control of thrust fluctuation in a linear motor according to claim 2, characterized in that: Time-domain analysis uses time-domain waveforms to identify the fluctuation period and amplitude. Frequency-domain analysis uses fast Fourier transform to extract the dominant frequency. Spatial-domain analysis calculates the thrust difference between adjacent acquisition devices by collecting thrust fluctuations at fixed intervals along the motion direction of the mover, thereby determining the fluctuation characteristics in the spatial domain.

4. The method for real-time compensation and control of thrust fluctuation in a linear motor according to claim 3, characterized in that: The dominant frequency refers to the frequency that plays a dominant role in the geometric nonlinear fluctuation component. This dominant frequency will cause the linear motor to resonate during operation. Once it is locked as the characteristic frequency, it can be used as the reference frequency in subsequent analysis.

5. The method for real-time compensation and control of thrust fluctuation in a linear motor according to claim 1, characterized in that: The preprocessing process includes: establishing a flux linkage observation model; comparing observed and actual values ​​based on changes in inductance and resistance during operation to obtain flux linkage deviation; establishing a nonlinear source wave screening equation and filtering out nonlinear terms; using flux linkage deviation values ​​to weight the nonlinear terms to identify electromagnetic nonlinear wave components; performing wavelet transform multi-scale analysis to decompose transient waves into components of different frequency bands and identify wave characteristics within each frequency band; and judging and labeling non-geometric waves based on current harmonic analysis and characteristic thresholds.

6. The method for real-time compensation and control of thrust fluctuation in a linear motor according to claim 5, characterized in that: By employing continuous wavelet transform and calculating the cross-correlation wavelet coefficients, the instantaneous coupling fluctuations of thrust and current are extracted, thereby enabling the location of transient impacts.

7. The method for real-time compensation and control of thrust fluctuation in a linear motor according to claim 5, characterized in that: The feature threshold uses a thrust amplitude threshold and a current harmonic threshold as dual thresholds for judgment, which can identify non-geometric fluctuations while avoiding misjudgments caused by current noise.

8. The method for real-time compensation and control of thrust fluctuation in a linear motor according to claim 1, characterized in that: Establishing an independent compensation strategy further includes: The two types of wave signals are parameterized. The geometric nonlinear wave is described by a periodic function, and the non-geometric nonlinear wave is described by a nonlinear distortion function. Corresponding compensation models are then established for each. A multimodal fusion classifier is established to process multi-domain data, including time domain, frequency domain, and spatial domain data. Weights are allocated to non-geometric nonlinear fluctuations through a fully connected network, and nonlinear distortions are dynamically adjusted. The compensation strategy output by the multimodal fusion classifier and the fully connected network is used to obtain the compensation scheme.

9. The method for real-time compensation and control of thrust fluctuation in a linear motor according to claim 8, characterized in that: The time-domain data uses an LSTM network to capture its time-series correlation data, the frequency-domain data uses a convolutional neural network to extract local frequencies, and the spatial domain data uses a graph neural network to analyze the spatial topology. Finally, the features of the three domains are fused to form a comprehensive discrimination model.

10. A real-time compensation control system for thrust fluctuation of a linear motor, used to implement the real-time compensation control method for thrust fluctuation of a linear motor as described in any one of claims 1 to 9, characterized in that: include: The hybrid wave source separation module is used to collect hybrid wave sources during the operation of the linear motor, and to separate the hybrid wave sources to generate independent geometric nonlinear wave signals; The filtering module is used to further filter the mixed wave sources, filter out non-geometric nonlinear wave signals, and preprocess them. An independent compensation module is used to establish independent compensation strategies to provide targeted compensation for geometric nonlinear fluctuations and non-geometric nonlinear fluctuations respectively. The feedforward compensation module is used to establish a feedforward compensation network. The compensation strategy is input into the feedforward compensation network, and the fluctuation components are predicted to perform pre-compensation processing in advance.

Images