Mechanical and electrical system fault early warning method, device and equipment based on digital twinning, and medium

Abstract

The application relates to the technical field of digital twinning. By providing an electromechanical system fault early warning method, device, equipment and medium based on digital twinning, the method comprises: collecting and compressively encoding high-frequency vibration signals and low-frequency control instructions of an electromechanical system to generate synchronous compressed data; based on photonic crystal compressive sensing technology and FPGA hardware acceleration, time-frequency domain alignment processing is performed on the synchronous compressed data to generate a cross-scale alignment signal; a digital twinning model is constructed based on the cross-scale alignment signal, and multi-scale fusion features are extracted; fault prediction is performed based on the multi-scale fusion features to generate an early warning instruction; and according to the deviation of the actual fault diagnosis result and the early warning instruction, the digital twinning model is dynamically calibrated to obtain a dynamically calibrated digital twinning model, so as to solve the problems of high-frequency signal time sequence misplacement, key feature loss and cross-scale modeling misalignment, thereby improving the synchronization accuracy and feature fidelity of fault early warning.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, and in particular to a method, apparatus, equipment and medium for early warning of electromechanical system faults based on digital twins. Background Technology

[0002] With the continuous improvement of industrial intelligence, electromechanical systems are increasingly widely used in high-end manufacturing, energy equipment, rail transportation, and other fields. Fault early warning, as a core link in ensuring the safe operation of equipment, hinges on accurately extracting the cross-scale characteristics of high-frequency vibration signals and low-frequency control commands.

[0003] However, the relevant technologies have the following problems in cross-scale synchronization of high-frequency signals and low-frequency control: traditional timestamp synchronization protocols (such as IEEE 1588v2) cause timing misalignment due to microsecond-level clock drift, making it impossible to achieve accurate alignment between vibration signals and PLC control commands; existing compression algorithms are forced to downsample due to bandwidth limitations, resulting in the loss of high-frequency fault features. Both of these factors together cause distortion in fault feature extraction. Summary of the Invention

[0004] Therefore, it is necessary to provide methods, devices, equipment and media for fault early warning of electromechanical systems based on digital twins to address the above-mentioned technical problems, so as to solve the problems of high-frequency signal timing misalignment, loss of key features and inaccuracy of cross-scale modeling, thereby improving the synchronization accuracy and feature fidelity of fault early warning.

[0005] In a first aspect, this application provides a method for early warning of faults in electromechanical systems based on digital twins, the method comprising:

[0006] Cross-scale acquisition and compression encoding of high-frequency vibration signals and low-frequency control commands of electromechanical systems are performed to generate synchronous compressed data.

[0007] Based on photonic crystal compressed sensing technology and FPGA hardware acceleration, time-frequency domain alignment processing is performed on synchronous compressed data to generate cross-scale aligned signals.

[0008] A digital twin model is constructed based on cross-scale aligned signals, and multi-scale fusion features are extracted;

[0009] Fault prediction is performed based on multi-scale fusion features to generate early warning instructions;

[0010] Based on the deviation between the actual fault diagnosis results and the early warning instructions, the digital twin model is dynamically calibrated to obtain the dynamically calibrated digital twin model.

[0011] Furthermore, the high-frequency vibration signals and low-frequency control commands of the electromechanical system are acquired and compressed across scales to generate synchronous compressed data, including:

[0012] The following formula is used to generate a sparse coefficient matrix by projecting the high-frequency vibration signal of an electromechanical system onto a sparse basis function using a photonic crystal sensor:

[0013]

[0014]

[0015] in, This represents the raw vibration signal acquired by the photonic crystal sensor. Represents the sparsity coefficient. Denotes basis functions. Indicates the number of basis functions. Indicates the noise term; Represents the observed signal matrix, Represents the basis function matrix, This represents the sparse coefficient matrix to be solved. Represents the noise matrix;

[0016] Entropy encoding is performed on the sparse coefficient matrix to generate sparse compressed data with adjustable compression ratio;

[0017] The low-frequency control commands are timestamped to generate control signals with phase markers.

[0018] The sparse compressed data is jointly encoded with the phase-marked control signal to generate synchronous compressed data.

[0019] Furthermore, a sparse coefficient matrix is ​​generated by projecting the high-frequency vibration signal of the electromechanical system onto a sparse basis function using a photonic crystal sensor, including:

[0020] Based on the spectral distribution characteristics of high-frequency vibration signals, the optimal basis function set is dynamically matched from the photonic crystal basis function library;

[0021] By selectively modulating the photonic crystal wavelength using the optimal basis function set, a nonlinear sparse projection is performed on the high-frequency vibration signal to generate an initial sparse coefficient vector.

[0022] Using the following formula, based on the initial sparse coefficient vector and the transient energy distribution of the signal, the projection parameters are adaptively adjusted to optimize the sparsity of the initial sparse coefficient vector and generate an optimized sparse coefficient vector:

[0023]

[0024]

[0025] in, Indicates the adaptive projection parameters. Indicates the initial projection coefficients. Indicates the attenuation factor. This represents the energy of the k-th time window. Indicates the maximum energy value. This represents optimizing sparse vectors. Indicates the number of coefficients. Indicates the weighting factor. Represents the initial sparsity coefficients. Indicates the corresponding projection parameters;

[0026] The spatiotemporal correlation between the optimized sparse coefficient vector and the multi-channel vibration signal is reconstructed by matrix transformation to generate a sparse coefficient matrix.

[0027] Furthermore, based on photonic crystal compressed sensing technology and FPGA hardware acceleration, time-frequency domain alignment processing is performed on the synchronous compressed data to generate cross-scale aligned signals, including:

[0028] The synchronous compressed data is decoupled in the frequency domain based on the photonic crystal multi-scale decomposition algorithm to separate the high-frequency resonance component and the low-frequency trend component.

[0029] The high-frequency resonant component and the low-frequency control command are time-stamped and phase-aligned using a nanosecond-level synchronous clock signal generated by FPGA hardware.

[0030] The high-frequency components after phase alignment are fused with low-frequency control commands without distortion using a time-frequency domain stitching algorithm to generate a cross-scale aligned signal.

[0031] Furthermore, based on the photonic crystal multi-scale decomposition algorithm, frequency domain decoupling processing is performed on the synchronous compressed data to separate the high-frequency resonance component and the low-frequency trend component, including:

[0032] Based on the frequency domain energy distribution of synchronous compressed data, the high-frequency resonant frequency band and the low-frequency trend frequency band are dynamically selected by the bandgap characteristics of photonic crystals.

[0033] The signal in the high-frequency resonant band is subjected to photonic crystal band-stop filtering to suppress non-fault-related noise components and generate high-frequency resonant components.

[0034] The signal in the low-frequency trend band is processed by photonic crystal bandpass filtering to separate the basic operating status characteristics of the equipment and generate low-frequency trend components.

[0035] Furthermore, based on the deviation between the actual fault diagnosis results and the early warning commands, the digital twin model is dynamically calibrated to obtain a dynamically calibrated digital twin model, including:

[0036] The actual fault diagnosis results are aligned with the early warning instructions in a time sequence to generate an error feedback signal.

[0037] Based on the error feedback signal, the weight parameters of the data-driven model are updated through an incremental learning mechanism to generate the updated data-driven model.

[0038] When the deviation between the simulation results of the digital twin model and the actual state of the equipment continues to exceed a preset threshold, the physical parameter self-calibration mechanism is triggered to generate calibrated digital twin model parameters.

[0039] The updated data-driven model is fused with the calibrated digital twin model parameters to generate a dynamically calibrated digital twin model.

[0040] Furthermore, fault prediction is performed based on multi-scale fusion features to generate early warning instructions, including:

[0041] Time-frequency domain correlation analysis was performed on the multi-scale fusion features to extract a subset of fault-sensitive features;

[0042] Based on multi-dimensional matching between a subset of fault-sensitive features and a historical fault pattern library, a fault pattern confidence distribution is generated.

[0043] Based on the probability entropy value of the failure mode confidence distribution, the warning threshold boundary conditions are dynamically adjusted to generate dynamically adjusted warning threshold boundary conditions.

[0044] Based on the dynamically adjusted warning threshold boundary conditions, warning instructions are generated.

[0045] Secondly, this application also provides a fault early warning device for electromechanical systems based on digital twins, the device comprising:

[0046] The cross-scale acquisition and compression module is used to acquire and compress high-frequency vibration signals and low-frequency control commands of electromechanical systems across scales to generate synchronous compressed data.

[0047] The time-frequency alignment processing module is used to perform time-frequency domain alignment processing on synchronous compressed data based on photonic crystal compressed sensing technology and FPGA hardware acceleration to generate cross-scale aligned signals.

[0048] The model building and feature extraction module is used to build a digital twin model based on cross-scale aligned signals and extract multi-scale fused features;

[0049] The fault prediction and early warning module is used to predict faults based on multi-scale fusion features and generate early warning instructions.

[0050] The model dynamic calibration module is used to dynamically calibrate the digital twin model based on the deviation between the actual fault diagnosis results and the early warning instructions, so as to obtain the dynamically calibrated digital twin model.

[0051] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods in the first aspect of this application.

[0052] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods in the first aspect of this application.

[0053] The technical solution provided in this application includes the following technical effects: By providing a method, device, equipment, and medium for fault early warning of electromechanical systems based on digital twins, the method includes: cross-scale acquisition and compression encoding of high-frequency vibration signals and low-frequency control commands of electromechanical systems to generate synchronous compressed data; time-frequency domain alignment processing of the synchronous compressed data based on photonic crystal compressed sensing technology and FPGA hardware acceleration to generate cross-scale aligned signals; construction of a digital twin model based on the cross-scale aligned signals and extraction of multi-scale fusion features; fault prediction based on the multi-scale fusion features to generate early warning commands; and dynamic calibration of the digital twin model based on the deviation between the actual fault diagnosis results and the early warning commands to obtain a dynamically calibrated digital twin model, thereby solving the problems of high-frequency signal timing misalignment, loss of key features, and inaccurate cross-scale modeling. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying 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.

[0055] Figure 1 This is a flowchart of a fault early warning method for electromechanical systems based on digital twins, according to one embodiment of the present invention.

[0056] Figure 2 The flowchart illustrates how, in one embodiment of the present invention, time-frequency domain alignment processing is performed on synchronous compressed data to generate cross-scale aligned signals based on photonic crystal compressed sensing technology and FPGA hardware acceleration.

[0057] Figure 3 This is a structural diagram of a digital twin-based electromechanical system fault early warning device according to one embodiment of the present invention. Detailed Implementation

[0058] To make the above-mentioned objects, features, and advantages of this application more apparent and understandable, the specific implementation methods of this application will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a full understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0059] like Figure 1 As shown, this application provides a fault early warning method for electromechanical systems based on digital twins, the method comprising:

[0060] S101: Performs cross-scale acquisition and compression encoding of high-frequency vibration signals and low-frequency control commands of electromechanical systems to generate synchronous compressed data.

[0061] Specifically, when collecting high-frequency vibration signals from electromechanical systems, it is necessary to consider the potential impact of surrounding environmental factors such as noise and electromagnetic interference. Therefore, during the acquisition process, the collected signals undergo preprocessing, including filtering and noise reduction, to improve signal quality and usability. The preprocessed signals then require feature extraction to identify key features reflecting the equipment's operating status, such as signal frequency, amplitude, and phase. These features will serve as crucial data for subsequent analysis and fault prediction. Simultaneously, low-frequency control commands are collected, including output signals from monitoring and control systems, such as control commands from programmable logic controllers (PLCs). These commands encompass the equipment's operating parameters and control logic, which are essential for understanding the equipment's operating status and predicting faults. After acquiring both high-frequency vibration signals and low-frequency control commands, the collected data is compressed and encoded to reduce data volume and storage space usage, while also facilitating data transmission and processing. During compression, the integrity of key features is ensured to prevent information loss and ensure accurate fault prediction. The compressed high-frequency vibration signals and low-frequency control commands are then jointly encoded to generate synchronously compressed data. This step ensures that the two are precisely aligned in time, so that subsequent fault warning analysis can accurately reflect the actual operating status of the electromechanical system.

[0062] S102: Based on photonic crystal compressed sensing technology and FPGA hardware acceleration, time-frequency domain alignment processing is performed on synchronous compressed data to generate cross-scale aligned signals.

[0063] Specifically, photonic crystal compressed sensing technology is used to process synchronously compressed data. Photonic crystal compressed sensing is a signal sampling and reconstruction method based on the characteristics of photonic crystals. Its core lies in utilizing the bandgap characteristics of photonic crystals to perform sparse sampling and encoding of signals, thereby achieving effective signal compression. It can reduce the amount of data while preserving key signal features, reducing storage and transmission pressure. Then, FPGA (Field-Programmable Gate Array) hardware acceleration technology is introduced to improve processing speed and efficiency. FPGAs have high parallelism and programmability, enabling complex computational tasks to be converted into hardware circuits for rapid processing. Customized hardware logic accelerates specific algorithms, making them suitable for signal processing tasks with high real-time requirements.

[0064] In this step, the FPGA is used to accelerate computationally intensive operations such as matrix operations and signal transformations in the photonic crystal compressed sensing process, significantly reducing signal processing time. Next, time-frequency domain alignment is performed. The goal of this step is to precisely align the high-frequency vibration signal and low-frequency control commands in both time and frequency dimensions to accurately reflect the actual operating state of the electromechanical system. Specifically, a precise clock signal generated by the FPGA is used to synchronize the high-frequency vibration signal in the time domain, while digital signal processing techniques are used to extract and match frequency domain features of the low-frequency control commands to achieve alignment in both time and frequency. Afterward, a cross-scale aligned signal is generated. After completing the time-frequency domain alignment, the aligned high-frequency vibration signal and low-frequency control commands are fused to form a cross-scale aligned signal. This signal contains detailed information reflecting the transient characteristics of the equipment in the high-frequency vibration signal, as well as information reflecting the equipment's operating logic and trends in the low-frequency control commands, providing an accurate, complete, and cross-scale data foundation for subsequent digital twin model construction and fault early warning.

[0065] S103: Construct a digital twin model based on cross-scale aligned signals and extract multi-scale fusion features.

[0066] Specifically, based on the physical characteristics and functional requirements of the electromechanical system, the overall architecture of the digital twin model is designed, including the model's hierarchical structure and module division, to reflect the operating principles and characteristics of the electromechanical system. A suitable modeling method is selected based on the actual situation of the electromechanical system, such as physical modeling, data-driven modeling, or a combination of both. Physical modeling establishes a mathematical model describing the system's behavior based on the physical laws and principles of the electromechanical system; data-driven modeling utilizes collected cross-scale aligned signal data and employs machine learning and other methods to mine implicit information in the data, establishing an input-output relationship model of the system. Based on prior knowledge, design drawings, or operational experience of the electromechanical system, the parameters of the digital twin model are initialized to provide an initial state for subsequent model calibration and optimization. Key features are extracted from the high-frequency vibration signals and low-frequency control commands in the cross-scale aligned signals. For high-frequency vibration signals, features such as frequency components, amplitude, phase, and kurtosis are extracted; these features reflect the vibration state and potential fault information of the equipment. For low-frequency control commands, features such as amplitude changes, frequency characteristics, and pulse sequence characteristics of the control signals are extracted; these features reflect the operating logic and control state of the equipment.

[0067] The extracted high-frequency vibration signal features and low-frequency control command features are fused to form multi-scale fusion features. Fusion methods include simple feature concatenation and weighted summation, or more complex feature fusion algorithms such as principal component analysis, linear discriminant analysis, and multi-scale convolutional neural networks, to achieve complementarity and correlation between features and extract more representative and discriminative fusion features. The extracted multi-scale fusion features are further optimized to remove redundant information and noise interference, improving feature quality and effectiveness. Feature selection algorithms, such as feature selection based on filters, wrappers, or embedding methods, are used to select the subset of features most relevant to fault diagnosis and early warning; or feature dimensionality reduction techniques, such as singular value decomposition (SVD) and locally linear embedding (LLE), are used to reduce feature dimensionality and improve computational efficiency while retaining key feature information.

[0068] S104: Fault prediction is performed based on multi-scale fusion features, and early warning instructions are generated.

[0069] Specifically, fault-sensitive features, or fault-related features, are selected from multi-scale fusion features. Feature selection algorithms, such as feature selection based on filters, wrappers, or embedding methods, can be used to select the subset of features most relevant to fault diagnosis and early warning. Alternatively, feature dimensionality reduction techniques, such as singular value decomposition (SVD) and locally linear embedding (LLE), can be used to reduce feature dimensionality and improve computational efficiency while preserving key feature information. The extracted fault-sensitive features are then matched and analyzed with a historical fault pattern database. This database includes feature patterns and related information for various known faults. By analyzing the similarity between current features and historical fault patterns, the fault type and severity are determined. The early warning threshold boundary conditions are dynamically adjusted based on the probability entropy value of the fault pattern confidence distribution. The probability entropy value reflects the uncertainty and complexity of the fault pattern. By analyzing changes in this value, the early warning threshold is adjusted appropriately to improve the accuracy and timeliness of early warnings. Based on the dynamically adjusted early warning threshold boundary conditions and the fault prediction results, corresponding early warning instructions are generated. These instructions include information such as fault type, fault location, and fault severity.

[0070] S105: Based on the deviation between the actual fault diagnosis results and the early warning instructions, the digital twin model is dynamically calibrated to obtain the dynamically calibrated digital twin model.

[0071] Specifically, the actual fault diagnosis results and early warning commands are time-series aligned to ensure consistency and comparability in the time dimension, thereby accurately calculating the deviation between them. The deviation calculation includes deviations in fault type judgment, fault location, and fault severity estimation. The calculated deviations are then analyzed in depth to determine their causes and influencing factors, including sensor data acquisition accuracy, model prediction accuracy, and environmental interference. An error feedback signal is generated based on the deviation analysis results. Based on this error feedback signal, the weight parameters of the data-driven model are updated through an incremental learning mechanism. Incremental learning allows the model to gradually adapt to new data and information without retraining the entire model, improving the model's prediction accuracy and adaptability to actual faults.

[0072] When the deviation between the simulation results of the digital twin model and the actual state of the equipment continuously exceeds a preset threshold, a physical parameter self-calibration mechanism is triggered. By analyzing the differences between the actual fault diagnosis results and the model prediction results, the physical parameters in the digital twin model, such as the equipment's stiffness, damping, and mass, are automatically adjusted to make the model's simulation results closer to the actual operating state of the equipment. The updated data-driven model is then fused with the calibrated digital twin model parameters to generate a dynamically calibrated digital twin model. Subsequently, the dynamically calibrated digital twin model is verified and evaluated to ensure that the model can accurately reflect the actual operating state of the electromechanical system and has high accuracy and reliability in fault prediction.

[0073] One embodiment of this application provides a fault early warning method for electromechanical systems based on digital twins, comprising: cross-scale acquisition and compression encoding of high-frequency vibration signals and low-frequency control commands of the electromechanical system to generate synchronous compressed data; time-frequency domain alignment processing of the synchronous compressed data based on photonic crystal compressed sensing technology and FPGA hardware acceleration to generate cross-scale aligned signals; constructing a digital twin model based on the cross-scale aligned signals and extracting multi-scale fusion features; performing fault prediction based on the multi-scale fusion features to generate early warning commands; and dynamically calibrating the digital twin model according to the deviation between the actual fault diagnosis results and the early warning commands to obtain a dynamically calibrated digital twin model, thereby solving the problems of high-frequency signal timing misalignment, key feature loss, and cross-scale modeling inaccuracy, and thus improving the synchronization accuracy and feature fidelity of fault early warning.

[0074] Furthermore, the high-frequency vibration signals and low-frequency control commands of the electromechanical system are acquired and compressed across scales to generate synchronous compressed data, including:

[0075] The following formula is used to generate a sparse coefficient matrix by projecting the high-frequency vibration signal of an electromechanical system onto a sparse basis function using a photonic crystal sensor:

[0076]

[0077]

[0078] in, This represents the raw vibration signal acquired by the photonic crystal sensor. Represents the sparsity coefficient. Denotes basis functions. Indicates the number of basis functions. Indicates the noise term; Represents the observed signal matrix, Represents the basis function matrix, This represents the sparse coefficient matrix to be solved. Represents the noise matrix;

[0079] Entropy encoding is performed on the sparse coefficient matrix to generate sparse compressed data with adjustable compression ratio;

[0080] The low-frequency control commands are timestamped to generate control signals with phase markers.

[0081] The sparse compressed data is jointly encoded with the phase-marked control signal to generate synchronous compressed data.

[0082] Specifically, a photonic crystal sensor is used to acquire high-frequency vibration signals from an electromechanical system. The photonic crystal sensor can accurately sense and convert high-frequency vibration signals into electrical signals for subsequent processing. The acquired high-frequency vibration signals are projected using sparse basis functions to generate a sparse coefficient matrix. Based on the signal's spectral distribution characteristics, an optimal basis function set is dynamically matched from a photonic crystal basis function library. Through photonic crystal wavelength-selective modulation of this basis function set, a nonlinear sparse projection is performed on the high-frequency vibration signals to generate an initial sparse coefficient vector. Then, based on the initial sparse coefficient vector and the signal's transient energy distribution, the projection parameters are adaptively adjusted to optimize the sparsity of the initial sparse coefficient vector, generating an optimized sparse coefficient vector. Subsequently, the spatiotemporal correlation between the optimized sparse coefficient vector and the multi-channel vibration signals is matrix-reconstructed to generate a sparse coefficient matrix. The sparse coefficient matrix is ​​then subjected to entropy encoding. Entropy encoding is a coding method based on information entropy, which can efficiently encode and compress signals according to their probability distribution characteristics. By calculating the entropy of the signal, the probability of each symbol appearing in the signal is determined, and a corresponding code is assigned to each symbol based on the probability. This step can reduce the redundancy of the signal, achieve effective signal compression, and generate sparse compressed data with an adjustable compression ratio.

[0083] Timestamp tags are embedded in low-frequency control commands. Embedding timestamp tags in low-frequency control commands provides accurate time information, enabling precise time synchronization. Simultaneously, phase markers can be added to the control commands to better reflect the phase characteristics of the control signal, generating a phase-marked control signal. The sparse compressed data and the phase-marked control signal are then jointly encoded. Joint encoding is a technique that fuses different types of signals, effectively integrating high-frequency vibration signals and low-frequency control command information to generate synchronized compressed data. During joint encoding, the characteristics of different signals must be considered, employing appropriate encoding methods and parameters to ensure the integrity and consistency of the encoded data while minimizing data volume and transmission bandwidth.

[0084] Furthermore, a sparse coefficient matrix is ​​generated by projecting the high-frequency vibration signal of the electromechanical system onto a sparse basis function using a photonic crystal sensor, including:

[0085] Based on the spectral distribution characteristics of high-frequency vibration signals, the optimal basis function set is dynamically matched from the photonic crystal basis function library;

[0086] By selectively modulating the photonic crystal wavelength using the optimal basis function set, a nonlinear sparse projection is performed on the high-frequency vibration signal to generate an initial sparse coefficient vector.

[0087] Using the following formula, based on the initial sparse coefficient vector and the transient energy distribution of the signal, the projection parameters are adaptively adjusted to optimize the sparsity of the initial sparse coefficient vector and generate an optimized sparse coefficient vector:

[0088]

[0089]

[0090] in, Indicates the adaptive projection parameters. Indicates the initial projection coefficients. Indicates the attenuation factor. This represents the energy of the k-th time window. Indicates the maximum energy value. This represents optimizing sparse vectors. Indicates the number of coefficients. Indicates the weighting factor. Represents the initial sparsity coefficients. Indicates the corresponding projection parameters;

[0091] The spatiotemporal correlation between the optimized sparse coefficient vector and the multi-channel vibration signal is reconstructed by matrix transformation to generate a sparse coefficient matrix.

[0092] Specifically, the spectral distribution characteristics of high-frequency vibration signals are analyzed. Different signals exhibit varying energy distributions across different frequency bands due to their physical origins and transmission characteristics. Based on this, the optimal basis function set best suited for the current high-frequency vibration signal is dynamically matched from a photonic crystal basis function library. This library contains various basis functions with different characteristics, each corresponding to a specific vibration mode or frequency response. Through the matching process, the selected optimal basis function set can retain the key feature information of the signal to the greatest extent in sparse representation, laying the foundation for subsequent processing. Utilizing the wavelength-selective modulation characteristics of the selected optimal basis function set, a nonlinear sparse projection operation is performed on the high-frequency vibration signal. The wavelength selectivity of the photonic crystal allows for selective response and modulation of vibration waves at specific frequencies. The nonlinear sparse projection process converts the original high-frequency vibration signal into a sparse domain representation, obtaining an initial sparse coefficient vector. This vector uses a small number of non-zero coefficients to represent the main features of the signal, retaining key information while reducing data volume. Based on the initial sparse coefficient vector and the transient energy distribution of the signal, the projection parameters are adaptively adjusted to optimize the sparsity of the initial sparse coefficient vector, generating an optimized sparse coefficient vector.

[0093] The transient energy distribution of a signal reflects its energy variation characteristics in both the time and frequency domains. By analyzing this information, the projection parameters are adaptively modified to enhance the sparse representation of the signal. This allows the optimized sparse coefficient vector to more accurately reflect the key features of the original signal, improving the efficiency and accuracy of subsequent processing. Based on the spatiotemporal correlation between the optimized sparse coefficient vector and the multi-channel vibration signal, matrix reconstruction is performed to generate a sparse coefficient matrix. Since the multi-channel vibration signals are correlated in time and space, the matrix reconstruction process arranges and combines the optimized sparse coefficient vector into a matrix form according to certain rules. The sparse coefficient matrix generated by these steps systematically and comprehensively represents the characteristics of high-frequency vibration signals across multiple channels and at different times, providing an efficient and compact data foundation for fault early warning of electromechanical systems based on digital twins.

[0094] like Figure 2 As shown, based on photonic crystal compressed sensing technology and FPGA hardware acceleration, time-frequency domain alignment processing is performed on synchronous compressed data to generate cross-scale aligned signals, including:

[0095] S201: Based on the photonic crystal multi-scale decomposition algorithm, the synchronous compressed data is decoupled in the frequency domain to separate the high-frequency resonance component and the low-frequency trend component.

[0096] S202: The nanosecond-level synchronous clock signal generated by the FPGA hardware is used to perform timestamp phase interpolation and alignment between the high-frequency resonant component and the low-frequency control command;

[0097] S203: Based on the time-frequency domain splicing algorithm, the phase-aligned high-frequency components are fused with the low-frequency control commands without distortion to generate a cross-scale aligned signal.

[0098] Specifically, the photonic crystal multi-scale decomposition algorithm is a method for decomposing signals using the multi-scale characteristics of photonic crystals. Synchronous compressed data contains information such as high-frequency vibration signals and low-frequency control commands from electromechanical systems. This algorithm performs frequency domain decoupling processing on this data, separating signals with different frequency components, thereby separating the high-frequency resonant component and the low-frequency trend component. FPGA hardware possesses powerful parallel processing capabilities and high-precision clock signal generation capabilities. Using the FPGA to generate nanosecond-level synchronous clock signals provides a precise time reference for the high-frequency resonant component and the low-frequency control command. Through timestamp phase interpolation alignment, the two are precisely aligned in the time dimension to ensure they have the same sampling rate and time starting point, preparing for subsequent fusion processing. The time-frequency domain stitching algorithm is an algorithm that fuses signals with different frequency components in the time-frequency domain. After completing the timestamp phase interpolation alignment of the high-frequency resonant component and the low-frequency control command, the time-frequency domain stitching algorithm is used to fuse them without distortion. The algorithm fully considers the time-frequency characteristics of the signal during the fusion process to avoid distortion or error introduced by the fusion operation. It then generates a cross-scale aligned signal that can simultaneously reflect the high-frequency vibration characteristics and low-frequency control state of the electromechanical system.

[0099] Furthermore, based on the photonic crystal multi-scale decomposition algorithm, frequency domain decoupling processing is performed on the synchronous compressed data to separate the high-frequency resonance component and the low-frequency trend component, including:

[0100] Based on the frequency domain energy distribution of synchronous compressed data, the high-frequency resonant frequency band and the low-frequency trend frequency band are dynamically selected by the bandgap characteristics of photonic crystals.

[0101] The signal in the high-frequency resonant band is subjected to photonic crystal band-stop filtering to suppress non-fault-related noise components and generate high-frequency resonant components.

[0102] The signal in the low-frequency trend band is processed by photonic crystal bandpass filtering to separate the basic operating status characteristics of the equipment and generate low-frequency trend components.

[0103] Specifically, frequency domain analysis is performed on the synchronously compressed data to determine its energy distribution across different frequency bands. Different electromechanical systems and operating states will generate different frequency component energy distributions in high-frequency vibration signals and low-frequency control commands. Analyzing these distribution characteristics can provide a basis for subsequent frequency band selection. Based on the analysis results and combined with the bandgap characteristics of photonic crystals, the high-frequency resonant band and the low-frequency trend band are dynamically selected. The bandgap characteristics of photonic crystals refer to their ability to selectively reflect or transmit electromagnetic waves within a specific frequency range. This characteristic can be used to more accurately select the resonant band of the vibration signal to be processed and the trend band of the control command.

[0104] Signals within the selected high-frequency resonant band are subjected to photonic crystal band-stop filtering. Band-stop filtering is a filtering technique that suppresses signals within a specific frequency range. This method can suppress non-fault-related noise components in high-frequency vibration signals and extract fault-related high-frequency resonant components, thereby improving the accuracy of fault detection. Signals within the selected low-frequency trend band are subjected to photonic crystal band-pass filtering. Band-pass filtering, unlike band-stop filtering, allows signals within a specific frequency range to pass while suppressing signals at other frequencies. These methods can separate the low-frequency trend components of the equipment's basic operating state characteristics, enabling a better understanding of the equipment's normal operating mode.

[0105] Furthermore, based on the deviation between the actual fault diagnosis results and the early warning commands, the digital twin model is dynamically calibrated to obtain a dynamically calibrated digital twin model, including:

[0106] The actual fault diagnosis results are aligned with the early warning instructions in a time sequence to generate an error feedback signal.

[0107] Based on the error feedback signal, the weight parameters of the data-driven model are updated through an incremental learning mechanism to generate the updated data-driven model.

[0108] When the deviation between the simulation results of the digital twin model and the actual state of the equipment continues to exceed a preset threshold, the physical parameter self-calibration mechanism is triggered to generate calibrated digital twin model parameters.

[0109] The updated data-driven model is fused with the calibrated digital twin model parameters to generate a dynamically calibrated digital twin model.

[0110] Specifically, the actual fault diagnosis results and early warning commands are time-aligned to generate an error feedback signal. By analyzing the magnitude and direction of the error feedback signal, the difference between the digital twin model's predictions and the actual results is determined. Based on the error feedback signal, the weight parameters of the data-driven model are updated through an incremental learning mechanism. Incremental learning is a machine learning method that can update the model using new data or information without retraining the entire model. In this way, the model can gradually adapt to new data and information, improving prediction accuracy. When the deviation between the simulation results of the digital twin model and the actual state of the equipment continuously exceeds a preset threshold, a physical parameter self-calibration mechanism is triggered.

[0111] By analyzing the discrepancies between actual fault diagnosis results and model predictions, the physical parameters in the digital twin model, such as equipment stiffness, damping, and mass, are automatically adjusted to make the simulation results more closely resemble the actual operating state of the equipment. The updated data-driven model is then fused with the calibrated digital twin model parameters to generate a dynamically calibrated digital twin model. During the fusion process, the consistency of all parts of the model is ensured to achieve accurate simulation and prediction of the equipment's operating state. Simultaneously, the dynamically calibrated digital twin model is validated and evaluated to ensure its accuracy and reliability.

[0112] Furthermore, fault prediction is performed based on multi-scale fusion features to generate early warning instructions, including:

[0113] Time-frequency domain correlation analysis was performed on the multi-scale fusion features to extract a subset of fault-sensitive features;

[0114] Based on multi-dimensional matching between a subset of fault-sensitive features and a historical fault pattern library, a fault pattern confidence distribution is generated.

[0115] Based on the probability entropy value of the failure mode confidence distribution, the warning threshold boundary conditions are dynamically adjusted to generate dynamically adjusted warning threshold boundary conditions.

[0116] Based on the dynamically adjusted warning threshold boundary conditions, warning instructions are generated.

[0117] Specifically, the feature parameters of the multi-scale fusion features are calculated in the time domain, frequency domain, and time-frequency domain to construct an initial high-dimensional feature dataset. Feature selection algorithms, such as kernel principal component analysis (KPCA), are used to reduce the dimensionality of the high-dimensional feature dataset, extracting a subset of sensitive features related to faults. By analyzing the correlation between the aforementioned feature parameters, features of significant importance for fault diagnosis are selected, forming a subset of fault-sensitive features. This subset of fault-sensitive features is then matched with various fault patterns in a historical fault pattern library, and the matching results are evaluated from multiple dimensions (such as feature similarity, fault type correlation, etc.). Based on the matching results, the confidence score for each fault pattern is calculated, forming a fault pattern confidence distribution. This step identifies the historical fault pattern most similar to the current features and assigns a confidence score value to its corresponding fault type, representing the probability of that fault type occurring.

[0118] Based on the probability entropy value of the failure mode confidence distribution, the warning threshold boundary conditions are dynamically adjusted to generate dynamically adjusted warning threshold boundary conditions. Probability entropy is an indicator that measures system uncertainty. By calculating the probability entropy value of the failure mode confidence distribution, the uncertainty of the current failure prediction result is assessed. The warning threshold boundary conditions are dynamically adjusted according to the magnitude of the probability entropy value. When the probability entropy value is high, it indicates that the uncertainty of the failure prediction result is large, and the warning threshold is appropriately relaxed; conversely, when the probability entropy value is low, the warning threshold is tightened to improve the accuracy of the warning. Based on the dynamically adjusted warning threshold boundary conditions, a warning command is generated. The current failure mode confidence distribution is compared with the dynamically adjusted warning threshold boundary conditions. When the confidence of a certain failure mode exceeds the corresponding warning threshold, the warning mechanism is triggered, and a warning command is generated. The warning command contains information such as the failure type, confidence level, and possible causes of the failure, providing a basis for equipment maintenance and fault handling.

[0119] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0120] In one embodiment, such as Figure 3As shown, this application also provides a digital twin-based electromechanical system fault early warning device 300, which includes:

[0121] The cross-scale acquisition and compression module 301 is used to acquire and compress high-frequency vibration signals and low-frequency control commands of electromechanical systems across scales to generate synchronous compressed data.

[0122] The time-frequency alignment processing module 302 is used to perform time-frequency domain alignment processing on synchronous compressed data based on photonic crystal compressed sensing technology and FPGA hardware acceleration to generate cross-scale aligned signals.

[0123] The model building and feature extraction module 303 is used to build a digital twin model based on cross-scale aligned signals and extract multi-scale fusion features;

[0124] The fault prediction and early warning module 304 is used to predict faults based on multi-scale fusion features and generate early warning instructions.

[0125] The model dynamic calibration module 305 is used to dynamically calibrate the digital twin model based on the deviation between the actual fault diagnosis results and the early warning instructions, so as to obtain the dynamically calibrated digital twin model.

[0126] Specifically, the cross-scale acquisition and compression module 301 acquires high-frequency vibration signals and low-frequency control commands from the electromechanical system across scales to capture signal characteristics at different frequencies. The acquired signals are compressed and encoded to generate synchronously compressed data, reducing data volume and improving subsequent processing efficiency. The time-frequency alignment processing module 302 processes the synchronously compressed data using photonic crystal compressed sensing technology and FPGA hardware acceleration. Time-frequency domain alignment is performed to ensure consistency of signals at different frequencies in time and frequency, generating cross-scale aligned signals. The model construction and feature extraction module 303 constructs a digital twin model reflecting the actual operating state of the electromechanical system based on the cross-scale aligned signals. Multi-scale fusion features are extracted from the model, which integrate key information at different scales, providing a more comprehensive basis for fault prediction. The fault prediction and early warning module 304 uses the extracted multi-scale fusion features to perform fault prediction analysis and generates early warning commands based on the prediction results. The model dynamic calibration module 305 dynamically calibrates the digital twin model based on the deviation between the actual fault diagnosis results and the early warning commands. By continuously calibrating the model, we can improve its accuracy and reliability, ensuring that it can truly reflect the operating status of the electromechanical system.

[0127] The cross-scale acquisition and compression module 301 is also used for:

[0128] The following formula is used to generate a sparse coefficient matrix by projecting the high-frequency vibration signal of an electromechanical system onto a sparse basis function using a photonic crystal sensor:

[0129]

[0130]

[0131] in, This represents the raw vibration signal acquired by the photonic crystal sensor. Represents the sparsity coefficient. Denotes basis functions. Indicates the number of basis functions. Indicates the noise term; Represents the observed signal matrix, Represents the basis function matrix, This represents the sparse coefficient matrix to be solved. Represents the noise matrix;

[0132] Entropy encoding is performed on the sparse coefficient matrix to generate sparse compressed data with adjustable compression ratio;

[0133] The low-frequency control commands are timestamped to generate control signals with phase markers.

[0134] The sparse compressed data is jointly encoded with the phase-marked control signal to generate synchronous compressed data.

[0135] The cross-scale acquisition and compression module 301 is also used for:

[0136] Based on the spectral distribution characteristics of high-frequency vibration signals, the optimal basis function set is dynamically matched from the photonic crystal basis function library;

[0137] By selectively modulating the photonic crystal wavelength using the optimal basis function set, a nonlinear sparse projection is performed on the high-frequency vibration signal to generate an initial sparse coefficient vector.

[0138] Using the following formula, based on the initial sparse coefficient vector and the transient energy distribution of the signal, the projection parameters are adaptively adjusted to optimize the sparsity of the initial sparse coefficient vector and generate an optimized sparse coefficient vector:

[0139]

[0140]

[0141] in, Indicates the adaptive projection parameters. Indicates the initial projection coefficients. Indicates the attenuation factor. This represents the energy of the k-th time window. Indicates the maximum energy value. This represents optimizing sparse vectors. Indicates the number of coefficients. Indicates the weighting factor. Represents the initial sparsity coefficients. Indicates the corresponding projection parameters;

[0142] The spatiotemporal correlation between the optimized sparse coefficient vector and the multi-channel vibration signal is reconstructed by matrix transformation to generate a sparse coefficient matrix.

[0143] The time-frequency alignment processing module 302 is also used for:

[0144] The synchronous compressed data is decoupled in the frequency domain based on the photonic crystal multi-scale decomposition algorithm to separate the high-frequency resonance component and the low-frequency trend component.

[0145] The high-frequency resonant component and the low-frequency control command are time-stamped and phase-aligned using a nanosecond-level synchronous clock signal generated by FPGA hardware.

[0146] The high-frequency components after phase alignment are fused with low-frequency control commands without distortion using a time-frequency domain stitching algorithm to generate a cross-scale aligned signal.

[0147] The time-frequency alignment processing module 302 is also used for:

[0148] Based on the frequency domain energy distribution of synchronous compressed data, the high-frequency resonant frequency band and the low-frequency trend frequency band are dynamically selected by the bandgap characteristics of photonic crystals.

[0149] The signal in the high-frequency resonant band is subjected to photonic crystal band-stop filtering to suppress non-fault-related noise components and generate high-frequency resonant components.

[0150] The signal in the low-frequency trend band is processed by photonic crystal bandpass filtering to separate the basic operating status characteristics of the equipment and generate low-frequency trend components.

[0151] The model dynamic calibration module 305 is also used for:

[0152] The actual fault diagnosis results are aligned with the early warning instructions in a time sequence to generate an error feedback signal.

[0153] Based on the error feedback signal, the weight parameters of the data-driven model are updated through an incremental learning mechanism to generate the updated data-driven model.

[0154] When the deviation between the simulation results of the digital twin model and the actual state of the equipment continues to exceed a preset threshold, the physical parameter self-calibration mechanism is triggered to generate calibrated digital twin model parameters.

[0155] The updated data-driven model is fused with the calibrated digital twin model parameters to generate a dynamically calibrated digital twin model.

[0156] The fault prediction and early warning module 304 is also used for:

[0157] Time-frequency domain correlation analysis was performed on the multi-scale fusion features to extract a subset of fault-sensitive features;

[0158] Based on multi-dimensional matching between a subset of fault-sensitive features and a historical fault pattern library, a fault pattern confidence distribution is generated.

[0159] Based on the probability entropy value of the failure mode confidence distribution, the warning threshold boundary conditions are dynamically adjusted to generate dynamically adjusted warning threshold boundary conditions.

[0160] Based on the dynamically adjusted warning threshold boundary conditions, warning instructions are generated.

[0161] In one embodiment, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0162] In one embodiment, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0163] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0164] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A fault early warning method for electromechanical systems based on digital twins, characterized in that, The method includes: Cross-scale acquisition and compression encoding of high-frequency vibration signals and low-frequency control commands of electromechanical systems are performed to generate synchronous compressed data. Based on photonic crystal compressed sensing technology and FPGA hardware acceleration, the synchronous compressed data is subjected to time-frequency domain alignment processing to generate cross-scale aligned signals. A digital twin model is constructed based on the cross-scale aligned signal, and multi-scale fusion features are extracted; Fault prediction is performed based on the multi-scale fusion features, and early warning instructions are generated. Based on the deviation between the actual fault diagnosis results and the warning instructions, the digital twin model is dynamically calibrated to obtain a dynamically calibrated digital twin model.

2. The method for early warning of electromechanical system faults based on digital twins according to claim 1, characterized in that, The process of cross-scale acquisition and compression encoding of high-frequency vibration signals and low-frequency control commands from the electromechanical system to generate synchronous compressed data includes: The following formula is used to generate a sparse coefficient matrix by projecting the high-frequency vibration signal of the electromechanical system onto a sparse basis function using a photonic crystal sensor: in, This represents the raw vibration signal acquired by the photonic crystal sensor. Represents the sparsity coefficient. Denotes basis functions. Indicates the number of basis functions. Indicates the noise term; Represents the observed signal matrix, Represents the basis function matrix, This represents the sparse coefficient matrix to be solved. Represents the noise matrix; The sparse coefficient matrix is ​​subjected to entropy encoding to generate sparse compressed data with adjustable compression ratio. The low-frequency control command is timestamped to generate a control signal with a phase tag. The sparse compressed data is jointly encoded with a phase-marked control signal to generate the synchronous compressed data.

3. The method for early warning of electromechanical system faults based on digital twins according to claim 2, characterized in that, The step of generating a sparse coefficient matrix by projecting a sparse basis function onto the high-frequency vibration signal of the electromechanical system using a photonic crystal sensor includes: Based on the spectral distribution characteristics of the high-frequency vibration signal, the optimal basis function set is dynamically matched from the photonic crystal basis function library; The high-frequency vibration signal is nonlinearly sparsely projected by the photonic crystal wavelength selective modulation of the optimal basis function set to generate an initial sparse coefficient vector. Using the following formula, based on the initial sparse coefficient vector and the transient energy distribution of the signal, the projection parameters are adaptively adjusted to optimize the sparsity of the initial sparse coefficient vector, thereby generating an optimized sparse coefficient vector: in, Indicates the adaptive projection parameters. Indicates the initial projection coefficients. Indicates the attenuation factor. This represents the energy of the k-th time window. Indicates the maximum energy value. This represents optimizing sparse vectors. Indicates the number of coefficients. Indicates the weighting factor. Represents the initial sparsity coefficients. Indicates the corresponding projection parameters; The spatiotemporal correlation between the optimized sparse coefficient vector and the multi-channel vibration signal is reconstructed by matrix transformation to generate the sparse coefficient matrix.

4. The method for early warning of electromechanical system faults based on digital twins according to claim 1, characterized in that, The method based on photonic crystal compressed sensing technology and FPGA hardware acceleration performs time-frequency domain alignment processing on the synchronous compressed data to generate cross-scale aligned signals, including: The synchronous compressed data is decoupled in the frequency domain based on the photonic crystal multi-scale decomposition algorithm to separate the high-frequency resonance component and the low-frequency trend component. The high-frequency resonant component and the low-frequency control command are time-stamped and phase-interpolated and aligned using a nanosecond-level synchronous clock signal generated by FPGA hardware. The cross-scale aligned signal is generated by seamlessly fusing the phase-aligned high-frequency components with the low-frequency control commands using a time-frequency domain splicing algorithm.

5. The method for early warning of electromechanical system faults based on digital twins according to claim 4, characterized in that, The synchronous compressed data is decoupled in the frequency domain using a photonic crystal multi-scale decomposition algorithm to separate high-frequency resonance components and low-frequency trend components, including: Based on the frequency domain energy distribution of the synchronous compressed data, the high-frequency resonant frequency band and the low-frequency trend frequency band are dynamically selected by the bandgap characteristics of the photonic crystal. The signal within the high-frequency resonant band is subjected to photonic crystal band-stop filtering to suppress non-fault-related noise components and generate the high-frequency resonant components. The signal within the low-frequency trend band is subjected to photonic crystal bandpass filtering to separate the basic operating state characteristics of the equipment and generate the low-frequency trend component.

6. The method for early warning of electromechanical system faults based on digital twins according to claim 1, characterized in that, The step of dynamically calibrating the digital twin model based on the deviation between the actual fault diagnosis result and the early warning command to obtain a dynamically calibrated digital twin model includes: The actual fault diagnosis results are time-aligned with the early warning instructions to generate an error feedback signal. Based on the error feedback signal, the weight parameters of the data-driven model are updated through an incremental learning mechanism to generate an updated data-driven model. When the deviation between the simulation results of the digital twin model and the actual state of the device continues to exceed a preset threshold, a physical parameter self-calibration mechanism is triggered to generate calibrated digital twin model parameters. The updated data-driven model is fused with the calibrated digital twin model parameters to generate the dynamically calibrated digital twin model.

7. The method for early warning of electromechanical system faults based on digital twins according to claim 1, characterized in that, The step of predicting faults based on the multi-scale fusion features and generating early warning instructions includes: Time-frequency domain correlation analysis is performed on the multi-scale fusion features to extract a subset of fault-sensitive features; Based on the fault-sensitive feature subset and the historical fault pattern library, a multi-dimensional matching is performed to generate a fault pattern confidence distribution. Based on the probability entropy value of the confidence distribution of the fault mode, the warning threshold boundary conditions are dynamically adjusted to generate dynamically adjusted warning threshold boundary conditions. The warning instruction is generated based on the dynamically adjusted warning threshold boundary conditions.

8. A fault early warning device for electromechanical systems based on digital twins, characterized in that, The device includes: The cross-scale acquisition and compression module is used to acquire and compress high-frequency vibration signals and low-frequency control commands of electromechanical systems across scales to generate synchronous compressed data. The time-frequency alignment processing module is used to perform time-frequency domain alignment processing on the synchronous compressed data based on photonic crystal compressed sensing technology and FPGA hardware acceleration to generate cross-scale alignment signals. The model building and feature extraction module is used to build a digital twin model based on the cross-scale alignment signal and extract multi-scale fusion features; The fault prediction and early warning module is used to predict faults based on the multi-scale fusion features and generate early warning instructions. The model dynamic calibration module is used to dynamically calibrate the digital twin model based on the deviation between the actual fault diagnosis results and the warning instructions, so as to obtain a dynamically calibrated digital twin model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the electromechanical system fault early warning method based on digital twins as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the electromechanical system fault early warning method based on digital twins as described in any one of claims 1 to 7.

Images