An information physics and small sample combined driving electromechanical servo state identification method
By combining a cyber-physical fusion model and an adaptive error compensation module with multi-domain physical mechanisms and data augmentation techniques, the state identification problem of electromechanical servo systems in small sample scenarios was solved, achieving high-precision and robust state identification.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SICHUAN UNIV
- Filing Date
- 2026-03-05
- Publication Date
- 2026-06-05
AI Technical Summary
Existing electromechanical servo system state identification technologies suffer from insufficient cyber-physical fusion, difficulty in adapting to small sample learning and robustness under complex working conditions, resulting in insufficient identification accuracy and reliability.
By employing a cyber-physical fusion model that combines multi-domain physical mechanisms and an adaptive error compensation module, and through data augmentation and loss function optimization, a state identification model is constructed to achieve physical consistency and cross-domain feature alignment, thus solving the state identification problem in small sample scenarios.
It improves the identification accuracy and robustness of electromechanical servo systems under complex working conditions, solves the learning bottleneck and domain transfer problem of traditional methods in small sample scenarios, and achieves high-precision state identification.
Smart Images
Figure CN122151650A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of electromechanical control and intelligent diagnostic technology, and more specifically, to an electromechanical servo state identification method driven by cyber-physical and small-sample joint principles. Background Technology
[0002] Electromechanical servo systems, as core execution units in the field of precision control, are widely used in critical scenarios such as CNC machine tools, industrial robots, and aero-engine control. The stability and reliability of their operation directly determine the control accuracy and safety performance of the entire equipment. In practical applications, these systems typically operate under complex conditions of varying loads, varying speeds, and strong interference, making them prone to various failure modes such as gear wear, bearing fatigue, motor demagnetization, and control module drift. These failures pose a serious threat to the system's operational safety and control accuracy. Therefore, accurate state identification of electromechanical servo systems is of great significance.
[0003] However, existing electromechanical servo system state identification technologies have the following significant drawbacks: (1) Insufficient integration of cyber-physical systems: Traditional physical model-based state identification methods rely on precise dynamic equations, which are difficult to adapt to the model mismatch caused by operating condition fluctuations and component aging, resulting in a decrease in identification accuracy; while pure data-driven identification methods are detached from the physical mechanism of the system and rely only on data features for analysis, which are sensitive to data distribution shifts and have insufficient generalization ability under different operating conditions or data distribution change scenarios.
[0004] (2) The bottleneck of small sample learning is prominent: In actual industrial scenarios, it takes a lot of time, manpower and material resources to obtain labeled samples of all fault types and all working conditions, resulting in most fault scenarios being "small sample" scenarios. Traditional deep learning methods have a large demand for labeled samples, and due to the data dependence characteristics, it is difficult to achieve effective state identification in small sample scenarios.
[0005] (3) Significant domain migration problem: Factors such as sudden load changes and speed fluctuations under complex working conditions can cause distortion of monitoring signals, making it difficult for traditional identification models to adapt to changes in signal characteristics across working conditions. This can easily lead to false alarms and missed alarms, affecting the reliability of state identification.
[0006] In summary, existing technologies struggle to balance physical mechanism adaptability, few-shot learning capability, and robustness under complex operating conditions, failing to meet the high-precision and high-reliability state identification requirements of electromechanical servo systems in practical applications. Therefore, there is an urgent need for a technical solution that integrates the advantages of physical mechanisms and few-shot learning, enabling high-precision and fine-grained state identification under complex operating conditions. Summary of the Invention
[0007] To address the aforementioned problems, a first aspect of this invention provides a method for electromechanical servo state identification driven by cyber-physical systems and small sample size analysis, comprising: Establish a cyber-physical fusion model for electromechanical servo systems; Acquire multi-source operating data of the electromechanical servo system; Based on the cyber-physical fusion model, data augmentation processing is performed on small sample fault data in the multi-source operational data to generate a small sample augmentation set. A state identification model containing a feature extractor is constructed, and the state identification model is trained using the multi-source running data and the small sample augmentation set. During the training process, the state identification model is optimized by a joint loss function, which includes at least physical consistency constraint loss and cross-domain feature alignment loss. The state of the electromechanical servo system is identified using the trained state identification model.
[0008] In one optional implementation, the cyber-physical fusion model includes a multi-domain physical mechanism model and an adaptive error compensation module; The multi-domain physical mechanism model includes at least electromagnetic dynamics equations, mechanical dynamics equations, and thermodynamic equations. The adaptive error compensation module is configured to dynamically compensate for the residual between the theoretical output of the multi-domain physical mechanism model and the actual sensor measurement value. The output of the multi-domain physical mechanism model is fused with the compensation output of the adaptive error compensation module to obtain the output of the cyber-physical fusion model.
[0009] In one alternative implementation, the adaptive error compensation module is a neural network with an integrated attention mechanism, which is used to weight and focus on key components in the residual sequence.
[0010] In one optional implementation, the data augmentation processing of the small sample fault data based on the cyber-physical fusion model includes at least one of the following methods: Based on the mapping relationship between fault parameters and signal responses described by the multi-domain physical mechanism model, physical mechanism interpolation is performed between known fault samples of different severity levels to generate interpolated samples of intermediate states; and / or, By using a generative adversarial network, style transfer of data distribution is performed with healthy state samples as the source domain and the small sample fault data as the target domain, thereby generating style transfer samples.
[0011] In one optional implementation, the physical consistency constraint loss term is calculated using the following formula:
[0012] in, Features generated for transfer For physical mechanism mapping function, Theoretical features output by the physical model.
[0013] In one optional implementation, the multi-source operational data includes multi-channel physical signals and multi-channel information signals; The multi-channel physical signal includes at least eight sensor signals from different physical locations of the electromechanical servo system; The multi-channel information signals include at least five control and status signals from a servo driver or programmable logic controller.
[0014] In an optional implementation, the joint loss function further includes a transfer alignment loss term and a classification loss term, the expression of which is:
[0015] in, For the joint loss function, For physical constraint loss, For migration alignment loss, For classification loss.
[0016] A second aspect of this invention provides an electromechanical servo state identification device jointly driven by cyber-physical and small-sample technologies, the device comprising: The cyber-physical fusion modeling module is used to establish a cyber-physical fusion model of an electromechanical servo system; The data acquisition module is used to acquire multi-source operating data of the electromechanical servo system; The small sample augmentation module is used to perform data augmentation processing on small sample fault data in the multi-source operational data based on the cyber-physical fusion model, and generate a small sample augmentation set. A state identification model training module is used to construct a state identification model including a feature extractor, and to train the state identification model using the multi-source running data and the few-sample augmentation set. During training, the state identification model is optimized using a joint loss function, which includes at least a physical consistency constraint loss and a cross-domain feature alignment loss. The state identification and reasoning module is used to identify the state of the electromechanical servo system using the trained state identification model.
[0017] A third aspect of the present invention provides an electronic device, characterized in that it includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement an electromechanical servo state identification method jointly driven by cyber-physical systems and small sample sizes.
[0018] A fourth aspect of the present invention provides a computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and the computer program, when executed by a processor, is a cyber-physical and small-sample jointly driven electromechanical servo state identification method.
[0019] This application has at least the following advantages or beneficial effects: (1) Innovation of the fusion framework: Breaking through the traditional single technical path of "pure physical model" or "pure data driven", we have created a deep fusion framework of "multi-domain physical mechanism model + adaptive error compensation + small sample migration enhancement" to achieve synergistic optimization of physical mechanism and data information, and solve the core problem of insufficient information-physical fusion in the existing technology. (2) Innovation of small sample augmentation strategy: A dual-path small sample augmentation scheme combining physical mechanism interpolation and CycleGAN style transfer is proposed. It not only relies on the physical operation law of the system to ensure the rationality of the augmented sample, but also ensures the consistency of sample data distribution through domain transfer technology. It effectively breaks through the learning bottleneck caused by the scarcity of fault samples in industrial scenarios and avoids the defect of "pseudo-samples" generated by traditional data augmentation. (3) Physical constraints empower innovation: In the process of small sample transfer enhancement, dynamic consistency constraints are introduced. The electromagnetic, mechanical and thermodynamic physical laws of the system are embedded into the model training through the loss function to ensure that the model output is consistent with the essential operating characteristics of the electromechanical servo system. This significantly improves the identification robustness under complex working conditions and solves the problem that traditional transfer models are easily affected by working condition fluctuations and have high false alarm and false alarm rates. (4) Multi-loss joint optimization innovation: Integrating physical constraint loss, cross-domain alignment loss (MMD loss + domain adversarial loss) and Focal Loss classification loss to form a multi-dimensional joint optimization strategy, simultaneously solving the three major technical problems of cross-domain feature alignment, physical consistency guarantee and small sample class imbalance. Compared with a single loss function, it greatly improves the recognition accuracy and generalization ability. (5) Engineering adaptation innovation: While ensuring recognition performance, the model structure is simplified, redundant modules are removed, and a streamlined architecture of "ResNet18 + fully connected classification layer" is adopted, which takes into account both inference efficiency and engineering feasibility. It can meet the real-time monitoring needs of industrial sites and is easy to deploy on industrial-grade hardware, thus reducing the threshold for technology implementation. Attached Figure Description
[0020] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0021] Figure 1 This is a flowchart of an embodiment of the electromechanical servo state identification method jointly driven by cyber-physical and small sample sizes proposed in this application; Figure 2 This is a structural diagram of an electromechanical servo state identification device jointly driven by cyber-physical and small sample sizes, proposed in an embodiment of this application. Figure 3 This is a schematic diagram of an electronic device according to this application. Detailed Implementation
[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0023] Please refer to Figure 1 , Figure 1 This is a flowchart of an embodiment of an electromechanical servo state identification method jointly driven by cyber-physical and small-sample approaches, proposed in this application. Figure 1 As shown, a cyber-physical and small-sample jointly driven electromechanical servo state identification method includes: S100: Establish a cyber-physical fusion model for electromechanical servo systems; Specifically, the cyber-physical fusion model includes a multi-domain physical mechanism model and an adaptive error compensation module. Through the synergistic effect of the two, an accurate description of the operating characteristics of the electromechanical servo system is achieved, as detailed below: S110: A multi-domain physical mechanism model that constructs dynamic equations covering multiple fields such as electromagnetics, mechanics, and thermodynamics, comprehensively and accurately describing the system's operating mechanism. Electromagnetic dynamics equations:
[0024] in, Armature voltage, For armature resistance, For armature current, For armature inductance, The back electromotive force coefficient, This represents the angular velocity of the motor.
[0025] Mechanical dynamics equations:
[0026] in, For electromagnetic torque, For load torque, For rotational inertia, is the damping coefficient.
[0027] Thermodynamic equation:
[0028] in, For heat capacity, For winding temperature, For copper loss, For heat dissipation coefficient, For heat dissipation area, The ambient temperature.
[0029] S120: Adaptive error compensation module, employing a 3-layer BP neural network with integrated attention mechanism, dynamically compensates for the residual between the theoretical output of the multi-domain physical mechanism model and the actual sensor measurement values to correct model bias. Input data: The residual sequence between the theoretical output Xmodel(t) of the multi-domain physical mechanism model and the actual sensor data Xmodel(t). ; Attention mechanism: A spatial attention module is used to focus on key fluctuation components in the residual sequence, with weights calculated as follows: ; Where σ is the Sigmoid activation function, used to enhance the influence of key information on the compensation result; Compensation output: The output of the multi-domain physical mechanism model is fused with the compensation output of the adaptive error compensation module to obtain the final output of the cyber-physical fusion model, i.e. .
[0030] S200: Acquire multi-source operating data of the electromechanical servo system; Specifically, high-quality multi-source operational data is obtained through multi-channel data collection and standardized preprocessing. The specific process is as follows: S210: Multi-source data acquisition, the acquired multi-source operational data includes multi-channel physical signals and multi-channel information signals: Multi-channel physical signals: Sensors are deployed at key physical locations such as the motor end, gearbox input / output end, bearing housing, servo valve, and output shaft of the electromechanical servo system to collect a total of 8 channels of physical signals, including vibration acceleration, motor three-phase current, output torque, bearing temperature, displacement feedback, and servo valve pressure signal. Multi-channel information signals: The PLC of the servo driver acquires a total of 5 channels of information signals, including control commands, target speed, load commands, driver status words, and PLC control signals. Data acquisition parameters: The sampling frequency is set to 25.6kHz, and 15 minutes of data are collected under each operating condition to ensure coverage of the complete system operation cycle and obtain comprehensive operating status information.
[0031] S220: Data preprocessing, performing a series of preprocessing operations on the collected raw multi-source data to generate a standardized sample set D: Signal denoising: A 5-level decomposition is performed using the db6 wavelet basis, and the wavelet coefficients are processed by an improved adaptive threshold function to achieve effective noise suppression; Trend removal: The nonlinear trend term in the signal is removed by using a polynomial fitting method, while retaining the fluctuation component related to the fault. Signal segmentation: The processed signal is segmented using a sliding window of length 5000, with a window overlap rate of 50%. Normalization: Min-Max normalization is used to map the segmented sample data to the [0,1] interval, eliminating differences in data units and improving the stability of model training.
[0032] S300: Based on the cyber-physical fusion model, perform data augmentation processing on the small sample fault data in the multi-source operational data to generate a small sample augmentation set; Specifically, combining the multi-domain physical mechanisms in the cyber-physical fusion model, a dual-path data augmentation strategy is adopted to expand small sample fault data, generating a small sample augmented set that conforms to the system's operating rules and has a consistent data distribution. The specific method is as follows: Physical mechanism interpolation: Based on the fault evolution law described by the multi-domain physical mechanism model and the mapping relationship between fault parameters and signal response, interpolation is performed between known fault samples of different severity to generate interpolated samples of intermediate states. For example, between slight wear and moderate wear fault samples, a transitional sample that conforms to the fault evolution trend is generated based on the physical correlation between wear amount and vibration amplitude, ensuring the physical rationality of the expanded sample.
[0033] Data style transfer: Construct CycleGAN generative adversarial network, using healthy state samples as the source domain and small sample failure data as the target domain, to learn the style transfer rules from healthy samples to different types of failure samples, and generate style transfer samples that are consistent with the distribution of real failure sample data, thus making up for the quantity deficiency of small sample failure data.
[0034] Small sample augmentation set output: The interpolated samples generated by physical mechanism interpolation are fused with the style transfer samples generated by data style transfer to obtain the small sample augmentation set, which can expand the original small sample size by 3 to 5 times, providing sufficient data support for subsequent model training.
[0035] S400: Construct a state identification model containing a feature extractor, and train the state identification model using the multi-source running data and the small sample augmentation set. During the training process, the state identification model is optimized by a joint loss function, which includes at least physical consistency constraint loss and cross-domain feature alignment loss. Specifically, S410: Construction of the state identification model, which includes a feature extractor, a cross-domain alignment module, a physical constraint module, and a classification output layer. The functions of each module are as follows: Feature extractor: A deep residual network (ResNet18) is used as the source domain feature extractor. Input healthy samples and multi-condition samples to extract feature vectors with universality; Cross-domain alignment module: This module achieves cross-domain feature alignment by minimizing the distribution difference between features in the source domain and features in the target domain (small sample fault domain) using the Maximum Mean Difference (MMD) method. The MMD loss formula is as follows:
[0036] in, The number of samples in the source domain. The number of samples in the target domain. Features of the target domain; Physical constraint module: Introduces dynamic consistency constraints to ensure that the samples and model outputs generated during transfer conform to the physical operating laws of the electromechanical servo system. The constraint loss formula is: Where is the feature generated by migration, is the physical mechanism mapping function, and is the theoretical feature output by the physical model; Classification output layer: A fully connected classification layer is added after the ResNet18 feature extractor. The input is a 512-dimensional feature vector. There is one hidden layer (256 neurons, ReLU activation function). The number of neurons in the output layer is the same as the number of state categories. The activation function is Softmax, which is used to implement the state category output.
[0037] S420: The state identification model is trained and optimized using a multi-dimensional joint loss function, the expression of which is:
[0038] Where is the joint loss function, is the physical constraint loss, is the transfer alignment loss, and is the classification loss; Physical constraint loss: the consistency between constraint fusion characteristics and physical analysis characteristics, i.e. ; Migration alignment loss: A weighted sum of MMD loss and domain adversarial loss, i.e. ,in, Losses due to domain confrontation; Classification Loss: Focal Loss is used to address the class imbalance problem in scenarios with few samples, i.e. ,in For category weights, This is the focusing coefficient (value 2). To predict probabilities.
[0039] S430: Model training is implemented based on the TensorFlow framework. The specific process is as follows: Initialization settings: The optimizer uses RMSProp, the initial learning rate is 1e-4, the learning rate decay strategy is ReduceLROnPlateau, the batch size is set to 64, and the number of training iterations is 800; Training cycle: Sample loading: Each iteration loads a mixture of the standardized sample set D and the small sample augmentation set to ensure the diversity of the training data; Forward propagation: Physically fuses the mixed sample input information into the model, cross-domain alignment module, and classification output layer, and calculates the physical constraint loss, transfer alignment loss, classification loss, and total loss. ; Backpropagation: The gradients of each parameter are calculated by automatic differentiation, the total loss is minimized by the optimizer, and all parameters of the model are updated. Model saving: Save the model checkpoint every 100 iterations, record the training loss and validation set performance, and finally save the model with the highest recognition accuracy on the validation set as the optimal state recognition model.
[0040] S500: Using the trained state identification model, the state of the electromechanical servo system is identified.
[0041] Specifically, S510: Model Inference and State Determination, after processing the operating data of the electromechanical servo system to be identified according to the preprocessing standard of S200, it is input into the trained optimal state identification model: The model extracts feature vectors from the data to be identified using the ResNet18 feature extractor, and then adjusts them using the cross-domain alignment module to obtain 512-dimensional standardized feature vectors. The feature vector is input into the classification output layer, which outputs the probability pi (i=1,2,...,C) of each category. The category with the highest probability is the electromechanical servo system state identification result corresponding to the data to be identified.
[0042] S520: Model performance verification and dynamic optimization, test set loading: The model performance is verified using an independent test set. The test set contains samples of different working conditions and unseen fault types, totaling 600 test samples. Dynamic threshold adjustment: Based on the confusion matrix of the test set, an adaptive thresholding algorithm is used to calculate the optimal threshold for each category. The formula is as follows:
[0043] in, The false negative rate for class c is... The false alarm rate for class c; the predicted probability when a sample belongs to class c. If the condition is met, it is determined to be of that category; otherwise, it is re-matched to the next highest probability category to improve the accuracy of identification. Performance evaluation: Calculate indicators such as overall identification accuracy, fault type identification accuracy, severity identification accuracy, and fault location identification accuracy to comprehensively evaluate model performance; Model optimization: If the performance does not meet expectations, adjust the weight coefficients of the joint loss function or the network structure parameters, and repeat the training process of S400 and the verification process of this step until the model performance meets the actual application requirements.
[0044] Please refer to Figure 2 , Figure 2 This is a structural diagram of an electromechanical servo state identification device jointly driven by cyber-physical and small-sample technologies, proposed in an embodiment of this application. Figure 2 As shown, this disclosure also provides an electromechanical servo state identification device jointly driven by cyber-physical systems and few-shot sampling. The device includes: a cyber-physical fusion modeling module 201, a data acquisition module 202, a few-shot enhancement module 203, a state identification model training module 204, and a state identification inference module 205; wherein, Cyber-physical fusion modeling module 201 is used to establish a cyber-physical fusion model of an electromechanical servo system; Data acquisition module 202 is used to acquire multi-source operating data of the electromechanical servo system; The small sample enhancement module 203 is used to perform data enhancement processing on small sample fault data in the multi-source operation data based on the cyber-physical fusion model, and generate a small sample enhancement set. A state identification model training module 204 is used to construct a state identification model including a feature extractor and train the state identification model using the multi-source running data and the few-sample augmentation set. During training, the state identification model is optimized using a joint loss function, which includes at least a physical consistency constraint loss and a cross-domain feature alignment loss. The state identification reasoning module 205 is used to identify the state of the electromechanical servo system using the trained state identification model.
[0045] This disclosure also provides an electronic device, please refer to... Figure 3 , Figure 3 This is a schematic diagram of an electronic device illustrated in an embodiment of this disclosure. For example... Figure 3 As shown, the electronic device 100 includes a memory 110 and a processor 120. The memory 110 and the processor 120 are connected via a bus for communication. The memory 110 stores a computer program that can run on the processor 120 to implement the steps in the electromechanical servo state identification method driven by cyber-physical and small sample joint methods disclosed in this embodiment.
[0046] The disclosed embodiments also provide a computer-readable storage medium that, when the instructions in the computer-readable storage medium are executed by a processor of a computer device, enables the computer device to perform the steps in the cyber-physical and small-sample jointly driven electromechanical servo state identification method of the present disclosure embodiments.
[0047] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, electronic devices, and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0048] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0049] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0050] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0051] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.
[0052] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.
[0053] The foregoing has provided a detailed description of the electromechanical servo state identification method jointly driven by cyber-physical systems and small sample size provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A method for electromechanical servo state identification jointly driven by cyber-physical systems and small sample size analysis, characterized in that, include: Establish a cyber-physical fusion model for electromechanical servo systems; Acquire multi-source operating data of the electromechanical servo system; Based on the cyber-physical fusion model, data augmentation processing is performed on small sample fault data in the multi-source operational data to generate a small sample augmentation set. A state identification model containing a feature extractor is constructed, and the state identification model is trained using the multi-source running data and the small sample augmentation set. During the training process, the state identification model is optimized by a joint loss function, which includes at least physical consistency constraint loss and cross-domain feature alignment loss. The state of the electromechanical servo system is identified using the trained state identification model.
2. The electromechanical servo state identification method jointly driven by cyber-physical systems and small sample size as described in claim 1, characterized in that, The cyber-physical fusion model includes a multi-domain physical mechanism model and an adaptive error compensation module; The multi-domain physical mechanism model includes at least electromagnetic dynamics equations, mechanical dynamics equations, and thermodynamic equations. The adaptive error compensation module is configured to dynamically compensate for the residual between the theoretical output of the multi-domain physical mechanism model and the actual sensor measurement value. The output of the multi-domain physical mechanism model is fused with the compensation output of the adaptive error compensation module to obtain the output of the cyber-physical fusion model.
3. The electromechanical servo state identification method jointly driven by cyber-physical systems and small sample size as described in claim 2, characterized in that, The adaptive error compensation module is a neural network with an integrated attention mechanism, which is used to weight and focus on key components in the residual sequence.
4. The electromechanical servo state identification method jointly driven by cyber-physical systems and small sample size as described in claim 2, characterized in that, The data augmentation processing of the small sample fault data based on the cyber-physical fusion model includes at least one of the following methods: Based on the mapping relationship between fault parameters and signal responses described by the multi-domain physical mechanism model, physical mechanism interpolation is performed between known fault samples of different severity levels to generate interpolated samples of intermediate states; and / or, By using a generative adversarial network, style transfer of data distribution is performed with healthy state samples as the source domain and the small sample fault data as the target domain, thereby generating style transfer samples.
5. The electromechanical servo state identification method jointly driven by cyber-physical systems and small sample size as described in claim 1, characterized in that, The physical consistency constraint loss term is calculated using the following formula: in, Features generated for transfer For physical mechanism mapping function, Theoretical features output by the physical model.
6. The electromechanical servo state identification method jointly driven by cyber-physical systems and small sample size as described in claim 1, characterized in that, The multi-source operational data includes multi-channel physical signals and multi-channel information signals; The multi-channel physical signal includes at least eight sensor signals from different physical locations of the electromechanical servo system; The multi-channel information signals include at least five control and status signals from a servo driver or programmable logic controller.
7. The electromechanical servo state identification method jointly driven by cyber-physical systems and small sample size as described in claim 5, characterized in that, The joint loss function also includes a transfer alignment loss term and a classification loss term, the expression of which is: in, For the joint loss function, For physical constraint loss, For migration alignment loss, For classification loss.
8. The electromechanical servo state identification device jointly driven by cyber-physical systems and small sample size as described in any one of claims 1-7, characterized in that, The device includes: The cyber-physical fusion modeling module is used to establish a cyber-physical fusion model of an electromechanical servo system; The data acquisition module is used to acquire multi-source operating data of the electromechanical servo system; The small sample augmentation module is used to perform data augmentation processing on small sample fault data in the multi-source operational data based on the cyber-physical fusion model, and generate a small sample augmentation set. A state identification model training module is used to construct a state identification model including a feature extractor, and to train the state identification model using the multi-source running data and the few-sample augmentation set. During training, the state identification model is optimized using a joint loss function, which includes at least a physical consistency constraint loss and a cross-domain feature alignment loss. The state identification and reasoning module is used to identify the state of the electromechanical servo system using the trained state identification model.
9. An electronic device, characterized in that, include: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the cyber-physical and small-sample jointly driven electromechanical servo state identification method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the electromechanical servo state identification method jointly driven by cyber-physical and small-sample methods as described in any one of claims 1 to 7.