Signal reconstruction denoising method for mechanical state evaluation of circuit breaker and storage medium

Abstract

The invention discloses a signal reconstruction denoising method for circuit breaker mechanical state evaluation and a storage medium. In order to solve the problems of multi-source signal aliasing and strong noise interference of the double-break vacuum circuit breaker, a technical route combining mechanical transfer function modeling and a double-flow deep network is innovatively adopted. A complex vibration source is analyzed by constructing a signal coupling model, environmental noise is accurately filtered by using a multi-stage noise reduction strategy, and a multi-dimensional state vector is constructed by fusing time-frequency domain features, so that accurate evaluation of the health state of the equipment is realized. Experiments show that the method is remarkably superior to the prior art in signal denoising, feature extraction precision and diagnosis accuracy, and an efficient solution is provided for intelligent operation and maintenance of power equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of circuit breaker detection, and in particular to a signal reconstruction and denoising method for circuit breaker mechanical state assessment and a storage medium. Background Art

[0002] With the continuous advancement of power technology, vacuum circuit breakers have become core equipment in medium-voltage power systems due to their environmental friendliness, long lifespan, and ease of maintenance. However, technical bottlenecks are particularly prominent in switching capacitor banks. At the 40.5kV voltage level, frequent re-breakdowns lead to overvoltage problems, seriously threatening the stable operation of the power grid. Because the insulation performance of vacuum gaps is physically saturated (i.e., the breakdown voltage does not increase linearly with gap length), optimizing short-gap designs can significantly improve device performance. The double-break structure, by connecting two vacuum interrupters in series, provides an effective solution to the capacitor switching problem.

[0003] Research shows that over 50% of high-voltage circuit breaker failures are caused by mechanical system anomalies. Double-break vacuum circuit breakers are complex electromechanical coupling devices, and their opening and closing processes involve the coordinated action of multiple components. The resulting mechanical vibrations contain rich health status information. Vibration signals can characterize over 20 typical fault characteristics, including trip mechanism jamming, moving part misalignment, core loosening, and pin fracture, providing a non-invasive monitoring method for equipment condition assessment.

[0004] It is worth noting that the double-break structure generates multiple vibration sources during opening and closing. Traditional single-channel sensors are typically mounted on the surface of the operating mechanism, resulting in multi-source aliasing in the acquired signal. This composite signal, which includes the vibration responses from the two interrupter chambers, the transmission linkage, and the operating mechanism, makes decoupling and analysis a key technical challenge for equipment diagnosis. Furthermore, white noise in the complex electromagnetic environment, harmonic interference from communication base stations, and transient impulse noise caused by switching operations further reduce the signal-to-noise ratio, placing higher demands on feature extraction algorithms.

[0005] In summary, vacuum circuit breakers have significant advantages in improving the switching performance of capacitors. However, the health monitoring of their mechanical systems still needs to overcome technical bottlenecks such as multi-source signal separation and feature extraction in strong noise environments to ensure reliable operation of the equipment. Summary of the Invention

[0006] (1) Technical issues to be resolved To solve the above problems, the present invention proposes a signal reconstruction and denoising method and storage medium for circuit breaker mechanical status assessment, aiming to solve the problems in the existing technology that the health monitoring of mechanical systems still needs to break through the problems of multi-source signal separation and feature extraction in strong noise environments.

[0007] (2) Technical solution A signal reconstruction and denoising method for circuit breaker mechanical state assessment according to the present invention comprises: Acquire the original physical signal to build a signal coupling model based on the mechanical transfer function, use cross-correlation analysis to extract the time delay characteristics of the original physical signal, and establish a mapping relationship between the original physical signal and the movement of the mechanical component in combination with the action sequence of the vacuum circuit breaker; Decomposing the original physical signal and reconstructing the decomposed original physical signal based on a two-stream deep network to obtain a first feature signal, and combining the first feature signal with mechanical parameters of the circuit breaker to generate an attention weight matrix dynamic fusion feature; Based on the signal coupling model, a time-varying mixing matrix is constructed and an adjustment factor is generated. The adjustment factor is used to optimize the blind source separation process to obtain a separation matrix. The elements of the separation matrix are modified according to mechanical motion constraints. The separated signal is subjected to multi-level noise reduction processing to obtain a reconstructed signal. A vacuum circuit breaker-specific morphological dictionary and dynamic threshold strategy are used to suppress noise components. Time-frequency domain features are integrated with mechanical state correlation indicators to construct a multidimensional feature vector. A theoretical benchmark signal is generated in conjunction with digital twin simulation. Based on the multi-dimensional difference analysis between the reconstructed signal and the theoretical reference signal, the probability of mechanical state degradation is calculated and the failed components are located using evidence theory fusion.

[0008] In the present invention, the construction of the signal coupling model includes: Acceleration sensors are installed symmetrically on both sides of the circuit breaker operating mechanism box to synchronously collect dual-channel vibration signals; By analyzing the phase difference of double-fracture motion and the difference in vibration propagation path, a multi-source signal coupling equation is established. The signal segments are divided based on the circuit breaker opening and closing cycle. The time delay of the two-channel signals is calculated using the cross-correlation function. The spatiotemporal mapping between the vibration signal and the mechanical action is established based on the angle-displacement relationship of the transmission connecting rod.

[0009] In the present invention, the dual-stream deep network includes: The main network uses a temporal convolution module, whose dilated convolution kernel size is dynamically adjusted according to the circuit breaker opening speed to extract millisecond-level mechanical shock features; The auxiliary network uses a graph convolution module, where nodes are defined as mechanical connectors of the circuit breaker and edge weights are calculated based on the trip spring stiffness and contact pressure parameters. The feature fusion layer encodes the mechanical parameters of the circuit breaker into an attention weight matrix and performs weighted fusion of the high-frequency transient features and low-frequency structural vibration features output by the main and auxiliary networks.

[0010] In the present invention, the process of generating the adjustment factor of the time-varying mixing matrix includes: Obtain real-time monitoring data through the arc extinguishing chamber pressure sensor and the operating mechanism current transformer; Based on finite element simulation, a mapping table of air pressure and vibration attenuation coefficient and a correlation model between current waveform characteristics and electromagnetic force were established; The pressure change rate and the current rising edge steepness are used as the row vector correction factors of the mixing matrix, and a three-dimensional lookup table is constructed to dynamically adjust the separation matrix elements.

[0011] In the present invention, the method for embedding mechanical motion constraint conditions includes: The transmission link angle-displacement differential relationship is converted into an orthogonal constraint condition for a separation matrix; The continuity requirement of contact motion acceleration is added into the independent component analysis objective function using the Lagrange multiplier method. The iterative projection algorithm is used to ensure that the power spectrum density of the separated source signal is consistent with the kinematic simulation results of the circuit breaker.

[0012] In the present invention, the multi-stage noise reduction process includes: Construct a morphological dictionary specifically for circuit breakers: collect typical fault vibration waveforms of contact collision and spring release, and train the atomic basis function library using the K-SVD algorithm; Determine the threshold baseline according to the second-order derivative extreme point of the signal energy accumulation curve in the current operation cycle; Calculate adjacent The Jensen-Shannon divergence of the operating cycle signal is calculated, and the adaptive smoothing filter is started when the divergence exceeds the set threshold; The resonant frequency band under the current current interruption condition is calculated based on the dielectric recovery strength model, and the zero-phase shift IIR notch filter is activated when the resonant frequency band exceeds a predetermined threshold.

[0013] In the present invention, the feature fusion method includes: Calculate the grey correlation between each time-frequency feature and the trip speed and overtravel, and construct a three-dimensional matrix of feature dimension-mechanical parameter-weight value; Bayesian reasoning is used to analyze the joint probability distribution of the statistical characteristics of the vibration signal and the residual of the motion trajectory of the operating mechanism to eliminate outlier features caused by sensor anomalies. A graph neural network is used to perform node embedding learning on the feature-fault knowledge graph constructed from type test data, and output the fault mode confidence vector.

[0014] In the present invention, the digital twin simulation and evaluation method includes: The denoised vibration signal is input into the multi-body dynamics simulation model to generate a theoretical reference waveform that takes into account the wear state of the mechanism; A dynamic time warping algorithm is used to align the mechanical event time series of the measured and simulated waveforms, and the morphological matching degree is calculated; extract The wavelet packet energy distribution of the frequency band is quantified by the Wasserstein distance; A fault tree model is constructed, and the reconstruction error, feature deviation, and simulation difference are used as evidence layer nodes. The failure probability contribution of each mechanical component is calculated through the back propagation algorithm.

[0015] In the present invention, the original physical signal includes one or more combinations of six types: current signal, acoustic emission signal, temperature signal, magnetic field signal, pressure signal, and vibration signal.

[0016] Another computer-readable storage medium of the present invention stores a computer program thereon, which, when executed by a processor, implements the signal reconstruction and denoising method for circuit breaker mechanical state assessment described in any one of the above technical solutions.

[0017] (3) Beneficial effects Compared with the prior art, the present invention has the following beneficial effects: This invention achieves efficient reconstruction and denoising of circuit breaker mechanical vibration signals through the innovative combination of mechanical transfer function modeling and a dual-stream deep neural network. Its core technical advantages are reflected in three aspects: First, the constructed signal coupling model accurately analyzes multi-source aliasing signals, resolving the diagnostic challenges caused by the complex vibration sources in dual-break structures.

[0018] The present invention adopts a multi-level noise reduction system, combined with a morphological dictionary and a dynamic threshold strategy, which can adaptively suppress complex noises such as white noise and pulse interference, greatly improving the signal-to-noise ratio. By fusing time-frequency domain features with mechanical state correlation indicators, a multi-dimensional evaluation vector is formed, which greatly improves the accuracy of fault diagnosis compared with traditional methods and significantly enhances the reliability of power grid operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 Schematic diagram of the logical framework of the signal reconstruction and denoising method; Figure 2 Schematic diagram of the original vibration signal waveform containing noise; Figure 3 Schematic diagram of the vibration signal waveform after noise reduction; Figure 4 Schematic diagram of the reconstructed vibration signal waveform. DETAILED DESCRIPTION

[0021] Example 1

[0022] like Figure 1-Figure 4 The signal reconstruction and denoising method for circuit breaker mechanical condition assessment shown includes the following steps: S100. Acquire the original physical signal to construct a signal coupling model based on the mechanical transfer function, use cross-correlation analysis to extract the time delay characteristics of the original physical signal, and establish a mapping relationship between the original physical signal and the movement of the mechanical component in combination with the action sequence of the vacuum circuit breaker.

[0023] Furthermore, in the signal acquisition phase, the acceleration sensor layout strategy was optimized to address the transient characteristics of the vacuum circuit breaker's opening and closing movements and the symmetry of its mechanical structure. Sensors were installed symmetrically on both sides of the circuit breaker's operating mechanism housing to simultaneously collect dual-channel vibration signals. In this invention, the original physical signals include one or more combinations of six types: current, acoustic emission, temperature, magnetic field, pressure, and vibration.

[0024] Compared to the transient impact characteristics of vibration signals, current signals contain the dynamic response of the electromagnetic actuator coil and the energy characteristics of the arc burning process. When processing, the time-frequency analysis must enhance the ability to capture steep rising edges (coil excitation) and high-frequency oscillations (arc reignition). An improved Prony algorithm is used to decompose the transient components of the current, and the current-mechanical motion coupling equation is constructed in conjunction with the equivalent circuit model of the circuit breaker's electromagnetic mechanism. Unlike vibration signals, the dynamic arc resistance model must be introduced as a priori constraint to distinguish electromagnetic interference from the actual operating current waveform during the blind source separation stage.

[0025] Acoustic signals have a wider frequency domain distribution than vibration signals and are significantly affected by propagation path attenuation. Processing requires adding typical partial discharge waveform basis functions to the morphological decomposition dictionary and employing a waveguide propagation model to compensate for phase distortion caused by sensor location. A key difference from vibration signal processing lies in the establishment of an acoustic-vibration signal transfer function matrix, the use of array signal processing techniques to separate mechanical impact sound from electromagnetic noise, and the introduction of time difference of arrival (TDOA) analysis of acoustic emission signals to locate the source of the anomaly.

[0026] To address the slowly varying characteristics and nonlinear cumulative effects of temperature signals, time-frequency analysis must be replaced with a time series decomposition module. Variational modal decomposition combined with a contact resistance-temperature rise physical model is used to extract abnormal temperature rise components. This differs from vibration signal processing in that a heat conduction network model must be constructed, and a reference temperature field distribution pattern generated through finite element thermal simulation. Cross-modal correlation analysis is then performed between the temperature gradient characteristics and the mechanical wear characteristics of the vibration signal.

[0027] Magnetic field signal processing requires enhanced spatial distribution feature extraction capabilities. A three-dimensional magnetic field tensor decomposition layer must be embedded in the deep learning model. This layer, combined with the circuit breaker's magnetic circuit design parameters, constructs a mapping between magnetic flux density and mechanical displacement. This differs from vibration signals in that a magnetic dipole array model must be used for source localization. Magnetic hysteresis loop characteristics are introduced as physical constraints for blind source separation to eliminate the influence of core remanence on signal reconstruction.

[0028] To address the step response characteristics of pressure signals, gas dynamics equations are used to modify the shock wave propagation model during transient waveform matching. Shock wave management theory is then used to construct a time-delay compensation mechanism between the pressure pulse and mechanical action. This differs significantly from vibration signal processing in that a gas-solid coupling simulation model is established, and the arc chamber pressure change rate is used as a vibration signal envelope correction factor to achieve combined noise reduction of both the pressure and vibration signals.

[0029] By analyzing the phase difference of double-fracture motion and the difference in vibration propagation path, a multi-source signal coupling model based on mechanical transfer function is constructed, and its cross-correlation function is: ,in They correspond to two signals collected by two different sensors. Indicates the time delay variable between signals. , combined with the transmission link angle-displacement relationship ,in is the connecting rod radius, Indicates the real-time rotation angle of the transmission connecting rod Cross-correlation analysis is used to extract the time delay characteristics of the two-channel signals. Combined with the circuit breaker action sequence, the mapping relationship between the vibration signal and the movement of the mechanical components is established, and the equation for time-space mapping is established: ,in It represents the propagation speed of the vibration wave in the circuit breaker mechanism and provides physical constraints for subsequent decomposition.

[0030] S200. Decompose the original physical signal and reconstruct the decomposed original physical signal based on a dual-stream deep network to obtain a first feature signal, and combine the first feature signal with the mechanical parameters of the circuit breaker to generate an attention weight matrix dynamic fusion feature.

[0031] In the decomposition stage, the original vibration signal is first subjected to adaptive noise compensation time-frequency analysis, and the window function parameters are dynamically adjusted according to the circuit breaker opening and closing cycle characteristics. The window length is set based on the real-time monitoring of the opening time and closing time interval to ensure that each mechanical action event (such as contact separation and spring release) is completely included in an analysis window to avoid spectrum aliasing caused by event truncation. Specifically, the window length According to the current operation cycle according to Calculate, where is the sampling frequency. The window parameter calculates the local signal-to-noise ratio (SNR) through a sliding window. When the SNR is lower than the threshold, a Gaussian window is used to suppress sidelobe leakage. In the high SNR area, a rectangular window is switched to preserve transient details. The window function expression is: in Adaptive adjustment is made based on the transient width of the signal. At the same time, the circuit breaker kinematic model is introduced to calculate the moving contact speed through the transmission mechanism connecting rod angle sensor data. , integrating it to obtain the theoretical vibration envelope ,in is the natural frequency of the mechanism. Using this envelope signal as the reference noise, an adaptive filtering algorithm is used to separate the inertial impact component of the operating mechanism from the original signal. The compensation formula is: in To compensate for the intensity coefficient, NLMS stands for Normalized Least Mean Square Adaptive Filter.

[0032] The reconstruction phase uses a two-stream deep network. The main network consists of five layers of temporal convolution modules (TCN). The dilation factor of each dilated convolution layer is and opening speed Dynamic association, press adjustment, The baseline speed is 0.000. The convolution kernel weights are initialized by embedding the spectral characteristics of historical circuit breaker fault data to enhance sensitivity to wear and shock waveforms. The auxiliary network constructs a mechanical connection graph, whose nodes include core components such as the contact system, trip spring, and transmission connecting rod. Edge weights are calculated based on the mechanical relationships between components. For example, the edge weight for the connection between the contact and the connecting rod is 0.000, where 0.000 is the moment arm length. The graph convolution layer captures the characteristics of the mechanical vibration transmission path through third-order neighborhood aggregation.

[0033] The feature fusion layer designs a hybrid attention mechanism to encode parameters such as the trip spring stiffness and contact pressure into a 128-dimensional vector, and performs multi-head attention calculations with the feature matrix output by TCN and GCN. Specifically, the query matrix The key-value matrix is generated by the mechanical parameter vector through the fully connected layer From the network features, the attention weights are calculated as: The final fusion feature dynamically allocates the weights of temporal transient and spatial structural features through the gating mechanism, and the gating value ,in is the sigmoid function, is a trainable parameter.

[0034] To address the problem of mechanical property drift, the online incremental learning module monitors the energy distribution of the vibration signal envelope in real time , when the energy center frequency shift is detected When the proportion of low-frequency energy increases suddenly, the model is fine-tuned. The Elastic Weight Consolidation (EWC) algorithm is used to constrain important parameters. Update range: in are the diagonal elements of the Fisher information matrix, is the original parameter, Control the forgetting rate. Simultaneously, update the convolution kernel function of the feature extractor, extract the principal component direction of the new data through sliding window PCA, and adjust the kernel function basis vector to maintain the ability to capture aging features.

[0035] S300: constructing a time-varying mixing matrix based on the signal coupling model and generating adjustment factors, optimizing a blind source separation process using the adjustment factors to obtain a separation matrix, and correcting elements of the separation matrix using mechanical motion constraints.

[0036] When constructing the time-varying hybrid matrix, we first establish a multi-physics field coupling hybrid model based on the real-time data of the arc chamber pressure sensor and the operating mechanism current transformer. Mapped to vibration propagation attenuation coefficient through preset finite element simulation database , and its relationship is calibrated through experiments as ,in is the baseline attenuation rate, is the pressure sensitivity factor, is the reference pressure. Current waveform characteristics (such as rising edge time With peak ) Predicting electromagnetic interference intensity through pre-trained LSTM network This is combined with the electromagnetic coupling coefficient of the vibration signal propagation path to generate a current correction term. Real-time updates of the hybrid matrix are achieved through a three-dimensional lookup table. The dimensions include pressure bins (0.1 Pa intervals), current intensity bins (10 A intervals), and a time window (5 ms sliding). Each cell stores the corresponding attenuation coefficient and coupling coefficient, and a bilinear interpolation algorithm is used to dynamically calculate the matrix elements at the current moment.

[0037] The embedding of mechanical constraints is achieved by transforming the objective function of independent component analysis (ICA). Contact displacement The differential relationship As a hard constraint, is the transmission ratio, which is converted into the orthogonality condition of the separated signal ( , Corresponding to the mechanical correlation source). The motion continuity requirement is integrated into the objective function through the Lagrange multiplier method to construct the extended cost function: , in is the separation matrix, is the reference signal vector generated by the kinematic equations, , The optimization process uses the improved FastICA algorithm, which applies a projection operation to the separation matrix after each iteration to force its column vector to satisfy ,in is the motion constraint covariance matrix, derived from multibody dynamics simulation data. The power spectral density of the separated signal must be verified for physical consistency. If the correlation coefficient between the frequency band energy distribution and the simulation result is less than 0.8, the mixing matrix backtracking correction mechanism is triggered, reloading the historical optimal parameters and adjusting the lookup table interpolation weights.

[0038] In order to improve real-time performance, the separation matrix is updated using incremental singular value decomposition (Incremental SVD). When a new data window is received, only the previous The principle component directions are calculated, and sliding window cumulative statistics of mechanical motion characteristics (such as average contact acceleration and connecting rod angle variance) are introduced as auxiliary decision factors for mixing matrix correction. When the opening time deviation exceeds 2ms or the arc chamber pressure fluctuation exceeds the limit, the system automatically switches to robust separation mode and re-estimates the mixing matrix using weighted least squares method to reduce the risk of false separation under abnormal conditions.

[0039] S400. Perform multi-level noise reduction processing on the separated signal to obtain a reconstructed signal. Use a vacuum circuit breaker-specific morphological dictionary and dynamic threshold strategy to suppress noise components. Fuse time-frequency domain features with mechanical state correlation indicators to construct a multidimensional feature vector. Combined with digital twin simulation, generate a theoretical benchmark signal.

[0040] During the noise reduction phase, a morphological dictionary specifically for vacuum circuit breakers was first constructed. Vibration signal samples under typical fault conditions, including contact welding, spring fatigue, and connecting rod jamming, were collected in the laboratory and on-site. Waveform segments corresponding to key mechanical events were extracted as training data. The K-SVD algorithm was used for sparse coding learning, with the optimization objective being to minimize the reconstruction error and the sparsity constraint: in is a sample matrix containing waveforms such as impact-type contact collision and damped oscillation-type spring release. A library of atomic basis functions, where the length of each atom matches the duration of a mechanical event (typically 2-5ms), is the sparsity threshold. After training, the dictionary contains three types of atoms: short, narrowband pulses for matching contact collision transients, exponentially decaying sine waves for simulating spring resonance, and frequency-modulated waveforms for characterizing non-stationary friction vibrations. The parameters of each type of atom, such as center frequency and decay rate, are calibrated based on the fault mechanism.

[0041] For the separated signal, a two-level dynamic threshold noise reduction is implemented. The primary threshold is based on the signal energy accumulation curve within the operation cycle. Setting, where is the typical time window for opening or closing action. The threshold baseline is calculated as: in and are the mean and standard deviation of the energy curve, Dynamically adjust according to the noise statistics (usually 2-3). The secondary threshold is calculated by analyzing the adjacent The mutual information entropy difference of the signals in the operation cycles ,when When the value exceeds 3 times the standard deviation of the historical mean, the adaptive smoothing filter is activated, and its window length is according to Adjustment, is the dominant frequency of the current signal.

[0042] Establishing dielectric recovery strength for high-frequency arc noise and breaking current The association model: ,in Determined by the arc extinguishing chamber structural parameters. Based on this, the resonant frequency is calculated in real time , design a zero-phase-shift IIR notch filter with the transfer function: ,in , The notch bandwidth is controlled (typically 0.85-0.95). The filter is cascaded 3-5 times to enhance stopband attenuation, while group delay compensation is used to minimize waveform distortion. The final reconstructed signal is composed of the sparse representation components of the morphological dictionary, the effective components after threshold filtering, and the filter residual, ensuring that the mechanical impact characteristics are fully preserved. During the feature fusion stage, time domain peak values, kurtosis, and frequency band energy ratios are extracted and compared point by point with the ideal waveform generated by the digital twin simulation. Frequency bands where the difference exceeds the threshold trigger dictionary atom optimization iterations, forming a closed-loop noise reduction mechanism.

[0043] S500 , based on the multi-dimensional difference analysis between the reconstructed signal and the theoretical reference signal, the probability of mechanical state degradation is calculated using evidence theory fusion and the failed component is located.

[0044] The feature fusion module incorporates a mechanical state correlation assessment mechanism to calculate the grey correlation between each feature parameter and key mechanical performance indicators of the circuit breaker (such as tripping speed and overtravel), and construct a three-dimensional weight allocation matrix. A spatiotemporal feature cross-validation algorithm was developed to jointly analyze the time-domain statistical characteristics of the vibration signal with the fitting residuals of the circuit breaker operating mechanism's motion trajectory. Bayesian inference was used to eliminate spurious features caused by loose sensor installation. A feature-fault mapping knowledge graph was established using circuit breaker type test data, and a graph neural network was employed to mine implicit correlation patterns between multidimensional features.

[0045] In the final evaluation phase, a hybrid digital twin framework was employed to input real-time vibration signals into a high-precision circuit breaker multibody dynamics simulation model to generate a theoretical vibration reference waveform. A two-dimensional similarity metric was defined: a dynamic time warping algorithm was used in the time domain to calculate the morphological similarity between the measured signal and the theoretical waveform, and the Wasserstein distance was used in the frequency domain to measure energy distribution differences. A fault probability propagation tree model was developed, combining three types of evidence: vibration signal reconstruction error, feature deviation, and simulation comparison results. DS evidence theory was used to fuse multi-source information to calculate the probability of mechanical state degradation. Fault tree reverse reasoning was then used to locate the specific failed component.

[0046] Example 2

[0047] An embodiment of the present invention provides a computer-readable storage medium.

[0048] The computer-readable storage medium provided in the embodiment of the present invention stores a computer program, and when the computer program is executed by a processor, the steps of any one of the above-mentioned signal reconstruction and denoising methods for evaluating the mechanical state of a circuit breaker can be implemented.

[0049] The computer-readable storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc., which can store program codes.

[0050] For an introduction to the computer-readable storage medium provided in an embodiment of the present invention, please refer to the above method embodiment, and the present invention will not elaborate on it here.

[0051] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0052] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0053] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0054] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the concept and scope of the present invention. Any modifications and improvements made to the technical solution of the present invention by a person of ordinary skill in the art without departing from the design concept of the present invention shall fall within the scope of protection of the present invention. The technical content for which protection is sought in the present invention is fully set forth in the claims.

Claims

1. A signal reconstruction and denoising method for circuit breaker mechanical state assessment, characterized in that: include: Acquire the original physical signal to build a signal coupling model based on the mechanical transfer function, use cross-correlation analysis to extract the time delay characteristics of the original physical signal, and establish a mapping relationship between the original physical signal and the movement of the mechanical component in combination with the action sequence of the vacuum circuit breaker; Decomposing the original physical signal and reconstructing the decomposed original physical signal based on a two-stream deep network to obtain a first feature signal, and combining the first feature signal with mechanical parameters of the circuit breaker to generate an attention weight matrix dynamic fusion feature; Based on the signal coupling model, a time-varying mixing matrix is constructed and an adjustment factor is generated. The adjustment factor is used to optimize the blind source separation process to obtain a separation matrix. The elements of the separation matrix are modified according to mechanical motion constraints. The separated signal is subjected to multi-level noise reduction processing to obtain a reconstructed signal. A vacuum circuit breaker-specific morphological dictionary and dynamic threshold strategy are used to suppress noise components. Time-frequency domain features are integrated with mechanical state correlation indicators to construct a multidimensional feature vector. A theoretical benchmark signal is generated in conjunction with digital twin simulation. Based on the multi-dimensional difference analysis between the reconstructed signal and the theoretical reference signal, the probability of mechanical state degradation is calculated and the failed components are located using evidence theory fusion.

2. The signal reconstruction and denoising method for circuit breaker mechanical state assessment according to claim 1, characterized in that: The constructing of the signal coupling model comprises: Acceleration sensors are installed symmetrically on both sides of the circuit breaker operating mechanism box to synchronously collect dual-channel vibration signals; By analyzing the phase difference of double-fracture motion and the difference in vibration propagation path, a multi-source signal coupling equation is established. The signal segments are divided based on the circuit breaker opening and closing cycle. The time delay of the two-channel signals is calculated using the cross-correlation function. The spatiotemporal mapping between the vibration signal and the mechanical action is established based on the angle-displacement relationship of the transmission connecting rod.

3. The signal reconstruction and denoising method for circuit breaker mechanical state assessment according to claim 2, characterized in that: The dual-stream deep network includes: The main network uses a temporal convolution module, whose dilated convolution kernel size is dynamically adjusted according to the circuit breaker opening speed to extract millisecond-level mechanical shock features; The auxiliary network uses a graph convolution module, where nodes are defined as mechanical connectors of the circuit breaker and edge weights are calculated based on the trip spring stiffness and contact pressure parameters. The feature fusion layer encodes the mechanical parameters of the circuit breaker into an attention weight matrix and performs weighted fusion of the high-frequency transient features and low-frequency structural vibration features output by the main and auxiliary networks.

4. The signal reconstruction and denoising method for circuit breaker mechanical state assessment according to claim 3, characterized in that: The process of generating the adjustment factor of the time-varying mixing matrix includes: Obtain real-time monitoring data through the arc extinguishing chamber pressure sensor and the operating mechanism current transformer; Based on finite element simulation, a mapping table of air pressure and vibration attenuation coefficient and a correlation model between current waveform characteristics and electromagnetic force were established; The pressure change rate and the current rising edge steepness are used as the row vector correction factors of the mixing matrix, and a three-dimensional lookup table is constructed to dynamically adjust the separation matrix elements.

5. The signal reconstruction and denoising method for circuit breaker mechanical state assessment according to claim 4, characterized in that: The mechanical motion constraint condition embedding method comprises: The transmission link angle-displacement differential relationship is converted into an orthogonal constraint condition for a separation matrix; The continuity requirement of contact motion acceleration is added into the independent component analysis objective function using the Lagrange multiplier method. The iterative projection algorithm is used to ensure that the power spectrum density of the separated source signal is consistent with the kinematic simulation results of the circuit breaker.

6. The signal reconstruction and denoising method for circuit breaker mechanical state assessment according to claim 5, characterized in that: The multi-stage noise reduction process includes: Construct a morphological dictionary specifically for circuit breakers: collect typical fault vibration waveforms of contact collision and spring release, and train the atomic basis function library using the K-SVD algorithm; Determine the threshold baseline according to the second-order derivative extreme point of the signal energy accumulation curve in the current operation cycle; Calculate adjacent The Jensen-Shannon divergence of the operating cycle signal is calculated, and the adaptive smoothing filter is started when the divergence exceeds the set threshold; The resonant frequency band under the current current interruption condition is calculated based on the dielectric recovery strength model, and the zero-phase shift IIR notch filter is activated when the resonant frequency band exceeds a predetermined threshold.

7. The signal reconstruction and denoising method for circuit breaker mechanical state assessment according to claim 6, characterized in that: The feature fusion method includes: Calculate the grey correlation between each time-frequency feature and the trip speed and overtravel, and construct a three-dimensional matrix of feature dimension-mechanical parameter-weight value; Bayesian reasoning is used to analyze the joint probability distribution of the statistical characteristics of the vibration signal and the residual of the motion trajectory of the operating mechanism to eliminate outlier features caused by sensor anomalies. A graph neural network is used to perform node embedding learning on the feature-fault knowledge graph constructed from type test data, and output the fault mode confidence vector.

8. The signal reconstruction and denoising method for circuit breaker mechanical state assessment according to claim 7, characterized in that: The digital twin simulation and evaluation method includes: The denoised vibration signal is input into the multi-body dynamics simulation model to generate a theoretical reference waveform that takes into account the wear state of the mechanism; A dynamic time warping algorithm is used to align the mechanical event time series of the measured and simulated waveforms, and the morphological matching degree is calculated; extract The wavelet packet energy distribution of the frequency band is quantified by the Wasserstein distance; A fault tree model is constructed, and the reconstruction error, feature deviation, and simulation difference are used as evidence layer nodes. The failure probability contribution of each mechanical component is calculated through the back propagation algorithm.

9. The signal reconstruction and denoising method for circuit breaker mechanical state assessment according to claim 7, characterized in that: The original physical signal includes one or more combinations of six types: current signal, acoustic emission signal, temperature signal, magnetic field signal, pressure signal, and vibration signal.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the signal reconstruction and denoising method for circuit breaker mechanical state assessment according to any one of claims 1 to 9 is implemented.

Images