A railway car on-line insulation monitoring system and method

Abstract

The application discloses a kind of national railway car online insulation monitoring system and method, related to online insulation monitoring technical field, the system includes: probe signal injection module, for injecting preset broadband active probe signal, synchronous acquisition after response signal set after transmission by the car body ground network;Fingerprint feature construction module is used to construct network fingerprint feature under current insulation state;Spatial distribution map acquisition module is used to obtain the admittance spatial distribution map of whole car by inversion calculation;Insulation abnormal area identification module is used to identify insulation abnormal area.The application solves the technical problems that the insulation state of the car body ground network of the national railway car cannot be monitored online in real time, the positioning accuracy of the abnormal area is low, which leads to a long time-consuming and poor accuracy in troubleshooting, achieves online monitoring of the insulation state of the car body ground network of the national railway car and precise positioning of the abnormal area, and improves the efficiency and accuracy of insulation troubleshooting.

Description

Technical Field

[0001] This invention relates to the field of online insulation monitoring technology, specifically to an online insulation monitoring system and method for national railway vehicles. Background Technology

[0002] During the operation of national railway vehicles, the insulation status of the vehicle's grounding network directly affects the safe and stable operation of the entire vehicle's electrical system. Insulation deterioration or damage can not only cause faults such as leakage and partial discharge, but in severe cases, it can also affect the normal operation of the traction system and auxiliary systems, and even threaten driving safety. Currently, insulation monitoring of the grounding network of national railway vehicles mostly adopts offline detection or fixed-point sampling detection methods. These methods not only cannot achieve real-time monitoring under operating conditions, but also have difficulty covering the entire grounding network link, making it easy to miss highly concealed local insulation defects. At the same time, traditional monitoring methods lack precise spatial positioning capabilities. When an insulation anomaly is detected, maintenance personnel need to check all grounding components of the vehicle one by one, which not only consumes a lot of manpower and time costs, but also prolongs the vehicle's maintenance downtime, failing to meet the actual needs of high-density operation and high-efficiency maintenance of national railway vehicles.

[0003] Existing technologies have technical problems such as the inability to monitor the insulation status of the grounding network of railway vehicles in real time online, low accuracy in locating abnormal areas, resulting in time-consuming and inaccurate troubleshooting. Summary of the Invention

[0004] This application provides an online insulation monitoring system and method for national railway vehicles, which addresses the technical problems in the prior art where the insulation status of the vehicle body grounding network of national railway vehicles cannot be monitored online in real time, the location accuracy of abnormal areas is low, and the fault diagnosis is time-consuming and inaccurate.

[0005] In view of the above problems, this application provides an online insulation monitoring system and method for national railway vehicles.

[0006] The first aspect of this application provides an online insulation monitoring system for national railway vehicles, the system comprising:

[0007] The detection signal injection module injects a preset broadband active detection signal into the vehicle grounding network through multiple signal injection points deployed on the low-voltage side of the vehicle. Simultaneously, it collects the response signal set transmitted through the vehicle grounding network through multiple signal receiving points deployed at different locations on the vehicle. The fingerprint feature construction module processes the response signal set, extracts channel impulse response parameters characterizing the transmission characteristics of the vehicle grounding network, and constructs the network fingerprint feature under the current insulation state. The spatial distribution map acquisition module inputs the network fingerprint feature into a pre-constructed whole-vehicle electromagnetic-circuit joint digital twin model for inversion calculation to obtain the admittance spatial distribution map of the whole vehicle. The insulation anomaly region identification module analyzes the admittance spatial distribution map to identify insulation anomaly regions.

[0008] A second aspect of this application provides a method for online insulation monitoring of national railway vehicles, the method comprising:

[0009] By deploying multiple signal injection points on the low-voltage side of the vehicle, a preset broadband active detection signal is injected into the vehicle grounding network. Simultaneously, by deploying multiple signal receiving points at different locations on the vehicle body, a set of response signals transmitted through the vehicle grounding network is collected synchronously. The set of response signals is processed to extract channel impulse response parameters characterizing the transmission characteristics of the vehicle grounding network, and a network fingerprint feature under the current insulation state is constructed. The network fingerprint feature is input into a pre-constructed vehicle electromagnetic-circuit joint digital twin model for inversion calculation to obtain the admittance spatial distribution map of the entire vehicle. The admittance spatial distribution map is analyzed to identify insulation anomaly areas.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] The detection signal injection module injects preset broadband active detection signals into the vehicle grounding network through multiple signal injection points deployed on the low-voltage side of the vehicle. Simultaneously, it collects response signal sets transmitted through the vehicle grounding network through multiple signal receiving points deployed at different locations on the vehicle body. The fingerprint feature construction module constructs network fingerprint features under the current insulation state. The spatial distribution map acquisition module performs inversion calculations to obtain the admittance spatial distribution map of the entire vehicle. The insulation anomaly area identification module analyzes the admittance spatial distribution map to identify insulation anomaly areas. This achieves the technical effect of online monitoring of the insulation state of the national railway vehicle's grounding network and precise location of anomaly areas, improving the efficiency and accuracy of insulation fault diagnosis. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This application provides a schematic diagram of the structure of an online insulation monitoring system for national railway vehicles.

[0014] Figure 2 This is a flowchart illustrating an online insulation monitoring method for national railway vehicles, provided as an embodiment of this application.

[0015] Figure labeling: Detection signal injection module 10, fingerprint feature construction module 20, spatial distribution map acquisition module 30, insulation abnormal area identification module 40. Detailed Implementation

[0016] This application provides an online insulation monitoring system and method for national railway vehicles, which addresses the technical problems in the prior art where the insulation status of the vehicle body grounding network of national railway vehicles cannot be monitored online in real time, the location accuracy of abnormal areas is low, and the fault diagnosis is time-consuming and inaccurate.

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] Example 1, as Figure 1 As shown, this application provides an online insulation monitoring system for national railway vehicles, the system comprising:

[0019] The detection signal injection module 10 is used to inject a preset broadband active detection signal into the vehicle grounding network through multiple signal injection points deployed on the low-voltage side of the vehicle. At the same time, it synchronously collects the response signal set transmitted through the vehicle grounding network through multiple signal receiving points deployed at different locations on the vehicle body.

[0020] Specifically, the detection signal injection module 10 injects a preset broadband active detection signal into the vehicle grounding network through multiple signal injection points deployed on the low-voltage side of the vehicle, including the vehicle's main integrated grounding busbar and the local grounding busbar of the electrical cabinets housing the traction and auxiliary systems. Simultaneously, it synchronously collects the response signal set transmitted through the vehicle grounding network through multiple signal receiving points deployed at different locations on the vehicle body. These signal receiving points include a first receiving antenna array deployed symmetrically on both sides of the longitudinal centerline inside the vehicle body to sense the asymmetric components of the vehicle's lateral electromagnetic field, and a second receiving antenna array deployed on the low-voltage side within a preset range of the high-voltage insulation components to couple electromagnetic radiation generated by partial discharge or insulation leakage. Furthermore, this module integrates an interference spectrum detection unit and a detection signal adjustment unit. The interference spectrum detection unit can monitor the electromagnetic interference spectrum of the vehicle's operating environment in real time, while the detection signal adjustment unit can dynamically adjust the frequency, time slot, or coding strategy of the broadband active detection signal based on the interference frequency points in the electromagnetic interference spectrum and a preset interference cancellation mechanism, ensuring the effectiveness of the detection signal and the accuracy of the response signal acquisition.

[0021] The fingerprint feature construction module 20 is used to process the response signal set, extract channel impulse response parameters that characterize the transmission characteristics of the vehicle grounding network, and construct the network fingerprint features under the current insulation state.

[0022] Specifically, the fingerprint feature construction module 20 is used to perform full-process fine processing on the response signal set collected by the detection signal injection module to extract channel impulse response parameters characterizing the transmission characteristics of the vehicle grounding network and construct network fingerprint features under the current insulation state: First, the spectrum matrix acquisition unit performs synchronous time-domain sampling and short-time Fourier transform on each response signal in the response signal set to obtain the spectrum matrix of the signal changing over time; then, the response component establishment unit uses the code pattern of the broadband active detection signal as a reference to perform adaptive filtering and coherent processing on the spectrum matrix to establish a multipath response component containing response components of the direct path, the main reflection path, and the scattering path; subsequently, the response sequence acquisition unit performs inverse Fourier transform on the multipath response component to calculate the discrete channel impulse response sequence of each response signal; finally, the network fingerprint feature construction unit extracts the delay spread, coherent bandwidth, and amplitude and phase characteristics at least three frequency points from the discrete channel impulse response sequence, and integrates these key parameters to form a network fingerprint feature that can accurately reflect the current insulation state of the vehicle grounding network.

[0023] The spatial distribution map acquisition module 30 is used to input the network fingerprint features into a pre-constructed vehicle electromagnetic-circuit joint digital twin model for inversion calculation to obtain the admittance spatial distribution map of the whole vehicle.

[0024] Specifically, the spatial distribution map acquisition module 30 is used to extract three-dimensional surface models of the vehicle body metal structure, cable bundles, and electrical equipment housings based on vehicle design drawings. It establishes an initial model by equating the vehicle grounding network to a circuit topology network. Then, it defines the electromagnetic properties of materials in the initial model and parameterizes the insulation state to be solved as admittance values ​​of the mesh attached to the surface of the insulating components. It reproduces the positions of the actual deployed signal injection and receiving points and performs transient co-simulation to construct a joint electromagnetic-circuit digital twin model of the entire vehicle. Finally, it inputs the network fingerprint features obtained by the fingerprint feature construction module into this digital twin model. The model is used for inversion calculation. The initial admittance distribution hypothesis is based on the health insulation state admittance distribution corresponding to the real-time operating conditions of the vehicle. The theoretical network fingerprint feature vector is predicted by full-wave electromagnetic simulation and circuit coupling calculation. The difference measure value between the theoretical network fingerprint feature vector and the actual network fingerprint feature vector is calculated. The gradient of the difference measure value is used as the accompanying excitation source and applied to the signal receiving point of the model to solve the accompanying electromagnetic field distribution in reverse. The gradient vector is obtained by conjugate dot product integration combined with the forward simulation electric field. The admittance distribution hypothesis is iteratively updated according to the gradient vector until the convergence condition is met. Finally, the optimal admittance distribution hypothesis is output as the admittance spatial distribution map of the whole vehicle.

[0025] The insulation abnormality region identification module 40 is used to analyze the admittance spatial distribution map and identify insulation abnormality regions.

[0026] Specifically, the insulation anomaly area identification module 40 is used to deeply analyze the spatial distribution map of the entire vehicle's admittance output by the spatial distribution map acquisition module. It combines the correlation characteristics between the insulation status and admittance distribution of the national railway car to carry out accurate identification: based on the healthy state characteristics of high resistivity and stable electromagnetic wave propagation path when the insulation material is intact, and the deterioration characteristics such as decreased resistivity, changed electromagnetic field distribution, and generation of micro discharge when the insulation is damp, damaged, dirty, or aged, the admittance value of each grid cell in the admittance spatial distribution map is compared with the healthy insulation state admittance distribution benchmark corresponding to the real-time operating conditions of the vehicle, and abnormal grid areas where the admittance parameters deviate from the benchmark range are screened out. By associating the physical location of the vehicle body, the distribution of insulation components, and the electromagnetic field propagation law corresponding to the abnormal grid areas, the specific location of leakage channels, partial discharge, or electromagnetic radiation anomalies caused by insulation deterioration is further located, and finally the insulation anomaly areas throughout the entire vehicle are accurately identified, providing a clear basis for the insulation status monitoring and fault diagnosis of the national railway car.

[0027] In one possible implementation, the fingerprint feature construction module 20 includes:

[0028] The spectrum matrix acquisition unit is used to perform synchronous time-domain sampling and short-time Fourier transform on each response signal in the response signal set to obtain the spectrum matrix of the signal as a function of time.

[0029] The response component establishment unit is used to establish multipath response components by adaptively filtering and coherently processing the spectrum matrix with the code pattern of the broadband active detection signal as a reference.

[0030] The response sequence acquisition unit is used to perform inverse Fourier transform on the multipath response components to calculate the discrete channel impulse response sequence of each response signal.

[0031] The network fingerprint feature construction unit is used to extract the delay spread, coherence bandwidth, and amplitude and phase characteristics at least three frequency points from the channel impulse response sequence to form the network fingerprint feature.

[0032] Specifically, the spectrum matrix acquisition unit performs synchronous time-domain sampling on each response signal in the response signal set acquired by the probe signal injection module. It accurately captures the continuous amplitude changes of each response signal in the time dimension according to the preset sampling frequency to ensure the integrity of the signal's time-domain characteristics. Then, it performs short-time Fourier transform processing on the synchronously sampled time-domain signal, dividing the long time-domain signal into multiple short-time stationary signal segments through a sliding time window. It performs Fourier transform on each signal segment to obtain its frequency distribution characteristics, and finally integrates them to form a two-dimensional spectrum matrix that can simultaneously reflect the signal's time evolution and frequency component changes, laying the foundation for the subsequent separation of multipath response components and extraction of channel impulse response parameters.

[0033] The response component establishment unit uses the code pattern of the broadband active detection signal injected by the detection signal injection module as a precise reference benchmark. For the time-varying spectrum matrix of the signal obtained by the spectrum matrix acquisition unit, it first uses adaptive filtering to dynamically suppress electromagnetic interference, noise and irrelevant signal components in the vehicle operating environment, and adjusts the filtering parameters in real time to adapt to the dynamic changes of the spectrum matrix to ensure the purity of the effective signal. Then, it performs coherent processing on the filtered spectrum matrix, uses the correlation between the reference code pattern and the target signal to strengthen the components of the same source signal, accurately separates and extracts the response components corresponding to the direct path, main reflection path and scattering path of the signal propagation in the vehicle grounding network, and finally establishes a complete multipath response component.

[0034] The response sequence acquisition unit is used to separate and extract the multipath response components, including the direct path, the main reflection path, and the scattering path, from the response component establishment unit. First, frequency domain calibration and phase alignment processing are performed on each path response component to eliminate frequency domain distortion and phase deviation caused by path differences during signal transmission, ensuring the consistency of the multipath signal. Then, an inverse Fourier transform is performed on the calibrated multipath response components to convert the multipath signal representation in the frequency domain back to the time domain signal form. The discrete channel impulse response sequence corresponding to each response signal is calculated. This sequence can accurately reflect the core transmission characteristics of the signal in the vehicle grounding network, such as transmission delay, amplitude attenuation, and path superposition.

[0035] The network fingerprint feature construction unit is used to process the discrete channel impulse response sequence (CIR sequence) output by the response sequence acquisition unit through a multi-step refinement algorithm. The system extracts the delay spread, coherence bandwidth, and amplitude-phase characteristics at at least three frequency points and integrates them into the network fingerprint feature. Specifically, the process is as follows: First, the CIR sequence is normalized using an energy accumulation distribution algorithm. The power value of each sampling point is calculated and accumulated over time. The start and end sampling points are located using 90% of the total energy as a threshold. The time difference between the two points is calculated using the sampling frequency as the delay spread parameter, accurately reflecting the multipath propagation time difference. Then, the CIR sequence is converted into a frequency domain response sequence using a Fast Fourier Transform (FFT). The adjacent frequency Pearson correlation coefficient analysis algorithm is used, with a correlation coefficient threshold of 0.7, to statistically analyze continuously satisfying... The maximum frequency range of the threshold, and the difference between its upper and lower limits, is the coherence bandwidth parameter, which characterizes the channel's transmission consistency for signals of different frequencies. Subsequently, within the working frequency band of the broadband active detection signal, at least three representative characteristic frequency points are selected through an equal-interval frequency screening algorithm. The frequency response accuracy of each frequency point is refined using a frequency domain interpolation algorithm, and the corresponding channel gain amplitude is extracted. The phase unwrapping algorithm is then used to eliminate phase entanglement and obtain accurate phase values, forming a set of amplitude-phase characteristic parameters consisting of amplitude-phase pairs for each frequency point. Finally, parameter normalization and dimension alignment algorithms are used to convert the three types of parameters into a unified dimension feature vector. After dimensionality reduction optimization and removal of redundant components using the principal component analysis (PCA) algorithm, a network fingerprint feature with fixed dimensions that can uniquely characterize the electromagnetic transmission characteristics and current insulation state of the vehicle grounding network is formed.

[0036] In one possible implementation, the fingerprint feature construction module 20 includes:

[0037] The multipath response components include response components for the direct path, the main reflection path, and the scattering path.

[0038] Specifically, the multipath response components are three types of core signal components extracted from the spectrum matrix by the response component establishment unit after adaptive filtering and coherent processing, using the broadband active detection signal code pattern as a reference. These include: the direct path response component, which is the signal component that, after the detection signal originates from the injection point, it is transmitted directly to the receiving point along a pre-set metal conductor in the vehicle grounding network, such as the vehicle body or cable, without any reflection or scattering. This component has the shortest propagation path, the least attenuation, and the most stable phase, and is the core component characterizing the basic transmission characteristics of the channel; and the main reflection path response component, which refers to the detection signal that, during transmission, passes through the vehicle body's metal structure, cable bundle shell, etc. The signal component that reaches the receiving point after specular reflection from the surface of a large, regular conductor has a clear propagation path and strong energy. Its propagation delay and amplitude attenuation can reflect the physical location of the reflecting surface and the characteristics of the conductor. The scattering path response component is the multidirectional propagating signal component formed after non-directional scattering when the detection signal encounters small irregular structures, insulation component interfaces, or local non-uniform areas inside the vehicle body. This component can capture the electromagnetic scattering characteristics of the fine structures inside the vehicle body, and is particularly sensitive to changes in local medium parameters caused by insulation degradation. Together, these three components present the complex propagation process of the detection signal in the vehicle body grounding network.

[0039] In one possible implementation, the spatial distribution map acquisition module 30 includes:

[0040] The network fingerprint feature vector prediction unit is used to perform full-wave electromagnetic simulation and circuit coupling calculation on the initial admittance distribution assumption in the whole vehicle electromagnetic-circuit joint digital twin model, and predict the theoretical network fingerprint feature vector collected under the initial admittance distribution assumption.

[0041] The difference metric calculation unit is used to calculate the difference metric between the theoretical network fingerprint feature vector and the network fingerprint features.

[0042] The optimal admittance distribution hypothesis output unit is used to adjust the admittance distribution hypothesis based on the difference metric value, and iterate with the goal of minimizing the difference metric value until the convergence condition is met, and output the optimal admittance distribution hypothesis as the admittance spatial distribution map.

[0043] Specifically, the network fingerprint feature vector prediction unit adopts a full-wave electromagnetic simulation and circuit-coupled collaborative computing approach. It completes the theoretical network fingerprint feature vector prediction within a pre-constructed vehicle electromagnetic-circuit joint digital twin model. The process is as follows: First, the initial admittance distribution assumption is taken as the healthy insulation state admittance distribution corresponding to the vehicle's real-time operating conditions. This assumption is mapped to the surface mesh of the insulating components in the digital twin model, clarifying the admittance parameters of each mesh element. Then, the full-wave electromagnetic simulation process is initiated, and Maxwell's equations are solved using the finite element method (FEM) to accurately simulate the radiation, propagation, reflection, and scattering of broadband active detection signals in the complex electromagnetic environment composed of the vehicle's metal structure, cable bundles, and insulating components. The process involves acquiring information on signal amplitude attenuation, phase shift, and time delay along each propagation path. Simultaneously, circuit coupling calculation is enabled to interface-adapt the equivalent circuit topology of the vehicle grounding network with the electromagnetic simulation model. The port electrical characteristics of the signal injection point and the receiving point are calculated using the node voltage method or branch current method, achieving a coordinated mapping of physical quantities in the electromagnetic domain and the circuit domain. Based on the time-domain signal of the receiving point obtained from the simulation, synchronous time-domain sampling, short-time Fourier transform, multi-path response component separation, inverse Fourier transform, and core parameter extraction are performed sequentially according to the algorithm flow consistent with the actual network fingerprint feature extraction. Finally, a theoretical network fingerprint feature vector that perfectly matches the feature dimension and parameter type of the actual network fingerprint is generated.

[0044] The difference metric calculation unit employs a multi-dimensional quantization comparison and weighted fusion approach to accurately calculate the difference metric between the theoretical network fingerprint feature vector and the actually extracted network fingerprint features. First, it normalizes the common parameters in the two types of feature vectors, such as delay spread, coherence bandwidth, and amplitude-phase characteristics at each frequency point, mapping parameters of different dimensions to the [0, 1] interval to eliminate the influence of dimensional differences on the calculation results. Then, for each core parameter, a corresponding difference calculation algorithm is used: for scalar parameters such as delay spread and coherence bandwidth, the difference is calculated using absolute error... The single-point difference value is calculated using the method of relative error or the relative error method. For the binary parameters such as amplitude and phase characteristics at each frequency point, the amplitude error and phase error are calculated separately and then weighted and summed to obtain the comprehensive difference value at each frequency point. Then, based on the sensitivity weight of each parameter to the change of insulation state, it is determined through offline sample training, and all single-point difference values ​​are weighted and accumulated to obtain the preliminary comprehensive difference value. Finally, the preliminary comprehensive difference value is mapped to a fixed interval, [0, 1], through normalization processing to form the final difference metric value, which intuitively reflects the degree of fit between the theoretical admittance distribution assumption and the actual insulation state.

[0045] The optimal admittance distribution hypothesis output unit, with minimizing the difference metric as the core optimization objective, achieves precise optimization of the admittance distribution hypothesis through a closed-loop iterative mechanism of gradient solving, hypothesis adjustment, and simulation verification. First, the gradient of the difference metric with respect to the theoretical network fingerprint feature vector is used as the adjoint excitation source and applied to each signal receiving point in the vehicle's electromagnetic-circuit joint digital twin model. The accompanying electromagnetic field distribution throughout the vehicle space is then deduced using the finite element inverse problem solution method. Subsequently, forward full-wave electromagnetic simulation and circuit coupling calculations are performed on the model corresponding to the current admittance distribution hypothesis to obtain the forward simulated electric field of each grid cell. The forward simulated electric field and the accompanying electromagnetic field distribution are then subjected to conjugate dot product integration to accurately calculate the difference metric. The gradient vector of the value relative to the admittance value of each surface grid cell of the insulating component in the model is used. Based on this gradient vector, the admittance parameters of each grid cell are dynamically adjusted using the gradient descent method to generate a new admittance distribution hypothesis. The new admittance distribution hypothesis is input into the digital twin model, and the theoretical network fingerprint feature vector prediction and difference metric calculation process are repeatedly executed to obtain the updated difference metric. The above adjustment and verification steps are continuously iterated until the difference metric drops below the preset threshold or the number of iterations reaches the set upper limit and the convergence condition is met. At this time, the output admittance distribution hypothesis is the optimal admittance distribution hypothesis, which is presented in the form of a spatial grid map to form an admittance spatial distribution map that can accurately reflect the spatial distribution of the insulation state of the entire vehicle.

[0046] In one possible implementation, the optimal admittance distribution assumption output unit includes:

[0047] The accompanying excitation source acquisition subunit is used to take the gradient of the difference metric value with respect to the fingerprint feature vector of the theoretical network as the accompanying excitation source.

[0048] The accompanying electromagnetic field distribution acquisition subunit is used to apply the accompanying excitation source to each signal receiving point position of the vehicle electromagnetic-circuit joint digital twin model, and solve the accompanying electromagnetic field distribution of the entire vehicle space in reverse.

[0049] The gradient vector calculation subunit is used to perform forward simulation of the initial admittance distribution assumption through the vehicle electromagnetic-circuit joint digital twin model, obtain the forward simulation electric field, perform conjugate dot product integration of the forward simulation electric field and the accompanying electromagnetic field distribution, and calculate the gradient vector of the difference metric value relative to the admittance value of each grid cell.

[0050] The admittance distribution hypothesis update sub-unit is used to update the admittance distribution hypothesis according to the gradient vector, and continue to perform full-wave electromagnetic simulation and circuit coupling calculation until the convergence condition is met.

[0051] Specifically, the accompanying excitation source acquisition sub-unit is used to construct the accompanying excitation source based on the nonlinear mapping relationship between the theoretical network fingerprint feature vector and the difference metric value. This is achieved through a precise numerical differentiation strategy. First, the constituent dimensions of the theoretical network fingerprint feature vector are defined, encompassing time delay spread, coherence bandwidth, and amplitude and phase characteristics at at least three frequency points. Each frequency point corresponds to two sub-parameters: amplitude and phase, forming a multi-dimensional feature parameter set. Then, the gradient of the difference metric value is solved using either automatic differentiation or finite difference methods. Taking automatic differentiation as an example, a computational graph of the difference metric value calculation process is constructed, and backpropagation is used to solve for the partial derivative of each feature parameter with respect to the difference metric value. (Measure of difference) / (Latency Spread) (Measure of difference) / (coherent bandwidth) (Measure of difference) / (Amplitude at each frequency point) (Measure of difference) / (Phase at each frequency point); All partial derivatives are integrated into a high-dimensional gradient vector according to the arrangement of characteristic parameters. This gradient vector is the accompanying excitation source. The magnitude of each element directly represents the sensitivity of the corresponding characteristic parameter to the difference measure value, and the element sign reflects the increasing or decreasing trend of the difference measure value when the characteristic parameter changes. This provides a quantitative basis for subsequent back-inference of the accompanying electromagnetic field distribution and precise positioning of the admittance parameter adjustment direction.

[0052] The accompanying electromagnetic field distribution acquisition sub-unit is used to first map and distribute the high-dimensional gradient vector obtained by the accompanying excitation source acquisition sub-unit to the deployment positions of each signal receiving point in the vehicle electromagnetic-circuit joint digital twin model. This ensures that each receiving point is loaded with the accompanying excitation signal of the corresponding dimension, and that the amplitude and phase of the excitation signal are consistent with the numerical characteristics of the corresponding elements in the gradient vector. Subsequently, based on the three-dimensional electromagnetic simulation environment and equivalent circuit topology of the vehicle body metal structure, cable bundle, and insulation components already constructed in the digital twin model, the accompanying equations of Maxwell's equations are solved. Numerical calculation methods such as the finite element method (FEM) or the finite-difference time-domain method (FDTD) are used to reverse-engineer the radiation, propagation, reflection, and coupling process with the circuit topology of the accompanying excitation source after it starts from each receiving point in the complex electromagnetic environment of the vehicle body. During this process, the electric field strength, magnetic field strength, and electromagnetic energy distribution of each grid cell in the entire vehicle space are calculated simultaneously, and finally, accompanying electromagnetic field distribution data that can completely characterize the propagation characteristics of the accompanying excitation source throughout the entire vehicle are generated.

[0053] The gradient vector calculation subunit is used to perform forward simulation and field quantity calculation based on the vehicle's electromagnetic-circuit joint digital twin model. It accurately solves the gradient vector of the difference metric value relative to the admittance value of each grid cell. First, it maps the current initial admittance distribution assumption to the surface grid of the insulating components in the model, clarifying the admittance parameters of each grid cell. Then, it initiates the full-wave electromagnetic simulation and circuit-coupled collaborative calculation process. By solving Maxwell's equations and circuit node voltage equations, it simulates the transmission path, amplitude attenuation, and phase change of the broadband active detection signal in the vehicle's grounding network. Simultaneously, it acquires the forward simulation electric field data corresponding to each grid cell within the entire vehicle space. This data is complete. The electromagnetic propagation characteristics under the current admittance distribution assumption are characterized. Then, the accompanying electromagnetic field distribution data of the sub-unit output is obtained by calling the accompanying electromagnetic field distribution. According to the principle of one-to-one correspondence of spatial positions, the conjugate dot product operation is performed on the forward simulated electric field and the accompanying electromagnetic field of each grid unit to obtain the field product result of each grid unit. Finally, the spatial integration operation is performed on the field product result of all grid units in the whole vehicle. The integration result is classified and organized according to grid units to form a gradient vector of the difference metric value relative to the admittance value of each grid unit on the surface of the insulating component. The magnitude and direction of the gradient vector directly determine the adjustment range and direction of the admittance value of each grid unit.

[0054] The admittance distribution hypothesis is updated in the sub-cell. This is used to calculate the gradient vector of the difference metric value output by the sub-cell relative to the admittance value of each grid cell, based on the gradient vector. The gradient descent optimization algorithm is used to perform iterative updates of the admittance distribution hypothesis. First, the admittance distribution hypothesis G for the current iteration is retrieved. old And the corresponding gradient vector ▽L, according to the preset gradient descent formula G new =G old -α▽L, completing the update of the admittance parameters, where α is the adaptive learning rate. This learning rate can be dynamically adjusted according to the iteration rounds. A larger value is used in the early stages of the iteration to accelerate the convergence speed, and a smaller value is used in the later stages of the iteration to improve the parameter accuracy. At the same time, the change amplitude of the difference metric is adjusted in real time to avoid parameter oscillations. Then, the updated admittance distribution hypothesis G is applied. new Mapping the surface mesh of the insulating component to the joint digital twin model of the vehicle's electromagnetic-circuit system, restarting the full-wave electromagnetic simulation and circuit coupling calculation process, generating a new theoretical network fingerprint feature vector and calculating the corresponding difference metric value; continuously repeating the above closed-loop process of gradient calculation-parameter update-simulation verification until the difference metric value drops below the preset threshold, or the number of iterations reaches the set upper limit, and the iteration is terminated when the convergence condition is met, ensuring that the final output admittance distribution hypothesis can accurately match the insulation state of the actual vehicle grounding network.

[0055] In one possible implementation, the spatial distribution map acquisition module 30 further includes:

[0056] The initial model building unit is used to extract the three-dimensional surface model of the vehicle body metal structure, cable bundle and electrical equipment shell based on the vehicle design drawings. In the three-dimensional surface model, the vehicle body grounding network is equivalent to the circuit topology network to build the initial model.

[0057] The first definition model acquisition unit is used to define the electromagnetic properties of the material in the initial model and parameterize the insulation state to be solved into the admittance value of the mesh attached to the surface of the insulation component, thereby obtaining the first definition model.

[0058] The digital twin model acquisition unit is used to reproduce the actual deployment of signal injection points and signal receiving points in the first defined model, perform transient co-simulation, simulate the full-band electromagnetic response generated by the injection signal excitation at each signal receiving point, and obtain the vehicle electromagnetic-circuit joint digital twin model.

[0059] Specifically, the initial model building unit uses detailed vehicle design drawings, including vehicle body structure construction drawings, cable wiring diagrams, and electrical equipment installation diagrams, as data input benchmarks. Utilizing 3D modeling and reverse reconstruction technology, it accurately extracts 3D surface models of core structures such as the vehicle body metal frame, the routing and laying paths of cable bundles in various functional areas, and the outer contours of electrical equipment housings, ensuring that the model's geometric dimensions and spatial positions are completely consistent with the actual vehicle. Subsequently, based on the actual connection logic and electrical characteristics of the vehicle body grounding network, the grounding conductor is equivalent to a circuit element with specific resistance, inductance, and capacitance parameters, and the conductor connection nodes are equivalent to circuit topology nodes. A one-to-one mapping is established between the electromagnetic physical structure in the 3D surface model and the circuit topology network. By establishing the correlation and constraint relationship between the electromagnetic structure and the circuit topology, an initial model that combines geometric structural features, electromagnetic propagation features, and circuit topology features is finally constructed.

[0060] The first definition model acquisition unit is used to define material properties and parameterize insulation states of the initial model constructed by the initial model building unit, which has both geometric and circuit topology features. First, for different components in the model, such as the metal structure of the vehicle body, the cable insulation layer, the electrical equipment shell, and the insulating support, core electromagnetic property parameters matching their actual materials are assigned, including relative permittivity, conductivity, and permeability, to clarify the dielectric characteristics of each component in the electromagnetic simulation. Then, a structured mesh generation algorithm is used to perform high-precision mesh discretization on the surface of all insulating components in the model, generating uniformly sized insulating surface mesh elements. On this basis, the insulation state parameters to be solved are transformed into admittance values ​​attached to each insulating surface mesh element. The magnitude of the admittance value is used to quantitatively characterize the insulation performance of the corresponding mesh region. At the same time, the correlation mapping relationship between the admittance value and the electrical parameters of the corresponding nodes in the circuit topology network is established, and finally, the first definition model with complete material electromagnetic property definition and insulation state parameterization is formed.

[0061] The digital twin model acquisition unit is used to perform physical scene reproduction and transient co-simulation on the first defined model, thereby constructing a joint digital twin model of the vehicle's electromagnetic-circuit system. First, based on the coordinates, quantity, and installation orientation of the signal injection and receiving points deployed in the actual engineering project, it accurately reproduces these parameters in the first defined model. Simultaneously, it defines the broadband detection signal excitation parameters for the injection points, including signal frequency band, amplitude, phase, and excitation timing, and configures the signal acquisition frequency band and sampling frequency for the receiving points. Then, it initiates the electromagnetic-circuit transient co-simulation process, using the finite-difference time-domain (FDTD) method or the finite element method (FEM) to solve Maxwell's equations, simulating the detection signal's interaction with the vehicle's metal structure, cable bundle, and insulation. This study investigates the radiation, propagation, reflection, scattering, and coupling processes within a complex electromagnetic environment composed of grounding components. Simultaneously, it utilizes the nodal voltage method to solve for the electrical response of the equivalent circuit topology network, enabling real-time interaction and data coupling between physical quantities in the electromagnetic and circuit domains. During this process, electromagnetic response data from each receiving point is simultaneously acquired across the entire frequency band, including signal amplitude attenuation, phase shift, time delay variations, and multipath superposition characteristics. Finally, based on simulation data, the electromagnetic and circuit characteristics of the model are calibrated and optimized to ensure a high degree of consistency between the model's output response characteristics and the actual vehicle grounding network test data. Ultimately, this yields a comprehensive vehicle electromagnetic-circuit joint digital twin model that accurately reflects the vehicle's true state.

[0062] In one possible implementation, the spatial distribution map acquisition module 30 further includes:

[0063] The initial admittance distribution is assumed to be based on the admittance distribution configuration corresponding to the healthy insulation state of the vehicle under real-time operating conditions.

[0064] Specifically, the initial admittance distribution assumption is based on the admittance distribution configuration corresponding to the healthy insulation state under the vehicle's real-time operating conditions. Specifically, it uses the admittance parameters of the insulating components in a healthy state without aging, damage, or moisture under the vehicle's current operating conditions, such as driving speed, load power, and ambient temperature and humidity, as a benchmark. Through a combination of offline testing and simulation calibration, the admittance benchmark values ​​of different regions on the surface of each insulating component in the vehicle's grounding network are obtained. Then, according to the grid division rules of the insulating component surface in the vehicle's electromagnetic-circuit joint digital twin model, the aforementioned admittance benchmark values ​​are mapped one-to-one to the corresponding grid cells, forming an initial admittance distribution matrix that corresponds one-to-one with each grid cell. This initial admittance distribution matrix fully considers the impact of real-time operating conditions on insulation performance. For example, under high temperature and high humidity conditions, the benchmark admittance values ​​of the insulating components are appropriately adjusted to match the performance fluctuations caused by the environment, ensuring that the initial admittance distribution assumption not only conforms to the essential characteristics of the healthy insulation state but also adapts to the vehicle's current actual operating conditions. This provides accurate and realistic initial input parameters for subsequent theoretical network fingerprint feature vector prediction and iterative optimization.

[0065] In one possible implementation, the detection signal injection module 10 further includes:

[0066] The multiple signal injection points include the vehicle's main integrated grounding busbar and the local grounding busbar of the electrical cabinets containing the traction system and auxiliary system.

[0067] The plurality of signal receiving points include a first receiving antenna array deployed at symmetrical positions on both sides of the longitudinal centerline inside the vehicle body, for sensing the asymmetric components of the lateral electromagnetic field of the vehicle body.

[0068] And a second receiving antenna array deployed on the low-voltage side within a preset range of the high-voltage insulation component, for coupling electromagnetic radiation generated by partial discharge or insulation leakage.

[0069] Specifically, the multiple signal injection points are precisely selected based on the grounding topology of the vehicle's electrical system, specifically covering the vehicle's main integrated grounding busbar and the local grounding busbars of the electrical cabinets housing the traction and auxiliary systems. The main integrated grounding busbar, as the core grounding junction node of the entire vehicle's electrical system, is responsible for the collection and distribution of the vehicle's grounding current. Setting it as a signal injection point enables broadband detection signals to excite the entire vehicle's grounding network. The local grounding busbars of the electrical cabinets housing the traction and auxiliary systems correspond to the key subsystems of the vehicle's power supply and auxiliary power supply, respectively. Including these two as signal injection points allows for targeted excitation of the electromagnetic characteristics of the grounding branches of the traction high-voltage circuit and the auxiliary low-voltage circuit, ensuring that the detection signal covers the core electrical grounding link of the entire vehicle. This provides comprehensive and effective excitation source support for the subsequent accurate acquisition and analysis of electromagnetic responses.

[0070] The multiple signal receiving points include a first receiving antenna array, which adopts a symmetrical layout and is precisely deployed at corresponding positions on both sides of the longitudinal centerline inside the vehicle body. The model specifications, installation height, orientation angle, and distance from the longitudinal centerline of the antennas in the array are all strictly consistent, thereby constructing a symmetrical electromagnetic signal acquisition benchmark. The first receiving antenna array is equipped with a broadband electromagnetic induction sensor, which can collect parameters such as electromagnetic field strength, phase, and frequency characteristics in the lateral space of the vehicle body in real time. By performing differential calculations on the data collected by the antennas on both sides, the asymmetric component of the lateral electromagnetic field of the vehicle body can be accurately sensed. This asymmetric component can directly reflect problems such as abnormal grounding circuit on one side of the vehicle body, local deterioration of insulation components, or failure of electromagnetic shielding, providing highly identifiable electromagnetic feature data for subsequent inversion of the insulation state through a digital twin model.

[0071] The second receiving antenna array employs a close-range, precise deployment strategy, strategically positioned within the low-voltage side of high-voltage insulating components, such as high-voltage cable terminals, insulators, and surge arresters, within a pre-defined sensitive range. The distance between the array and the high-voltage insulating components is controlled within a reasonable range that allows for efficient coupling of electromagnetic radiation while avoiding interference from the high-voltage electric field. The array is equipped with a highly sensitive, wideband electromagnetic sensor whose operating frequency band covers the UHF and VHF electromagnetic signal bands generated by partial discharge and insulation leakage. It can capture in real time the weak electromagnetic radiation signals generated by insulation breakdown and charge release in high-voltage insulating components under defective conditions. Simultaneously, the array uses multi-channel synchronous acquisition technology, combined with signal noise reduction algorithms to filter out environmental electromagnetic interference, ensuring accurate identification and extraction of characteristic signals from partial discharge and insulation leakage. This provides crucial local electromagnetic response data for subsequent location of insulation defects and assessment of defect severity using digital twin models.

[0072] In one possible implementation, the detection signal injection module 10 further includes:

[0073] Interference spectrum detection unit is used to monitor the electromagnetic interference spectrum of the vehicle's operating environment in real time.

[0074] The detection signal adjustment unit is used to dynamically adjust the frequency, time slot, or coding strategy of the broadband active detection signal based on the interference frequency points in the electromagnetic interference spectrum and a preset interference cancellation mechanism.

[0075] Specifically, the interference spectrum detection unit relies on multi-channel spectrum monitoring sensors deployed in key electromagnetically sensitive areas of the vehicle, such as around electrical cabinets, near high-voltage insulation components, and within the acquisition range of the antenna array, to collect various electromagnetic interference signals during vehicle operation in real time. These interference signals cover a variety of types, including pulse interference generated by the start-stop of the traction system, power frequency harmonic interference generated by the operation of auxiliary systems, and radio frequency interference radiated by external communication equipment. The unit uses a built-in high-speed signal acquisition module and real-time spectrum analysis algorithms, such as Fast Fourier Transform (FFT) and Short-Time Fourier Transform (STFT), to perform frequency domain conversion and feature extraction on the collected time-domain interference signals. It accurately identifies the frequency distribution, signal strength, bandwidth range, time-domain persistence characteristics, and variation patterns of the interference signals, and finally generates a real-time electromagnetic interference spectrum map containing interference information across the entire frequency band, providing comprehensive and accurate interference data source support for the dynamic adjustment of subsequent detection signals.

[0076] The detection signal adjustment unit receives real-time electromagnetic interference spectrum data output by the interference spectrum detection unit. It extracts strong interference frequency points, interference bandwidth, and interference time-domain distribution patterns from the spectrum. Then, based on a preset interference cancellation mechanism, it dynamically adjusts the transmission parameters of the broadband active detection signal: For continuous strong interference at fixed frequencies, a frequency hopping adjustment strategy is adopted to automatically switch the operating frequency of the detection signal to a clean frequency band outside the interference spectrum, avoiding direct superposition of the signal and interference; for periodic time-domain interference, a time-slot avoidance strategy is adopted, adjusting the transmission time slot of the detection signal by analyzing the time-domain periodic characteristics of the interference signal, ensuring that the signal transmission period is completely staggered from the interference occurrence period; for complex and variable broadband interference, a coding optimization strategy is adopted, modulating the detection signal into an orthogonal coding sequence or a spread spectrum coding sequence. Utilizing the orthogonality of the coding and the anti-interference characteristics of spread spectrum technology, the identification and penetration capabilities of the detection signal in interference environments are improved. Ultimately, this ensures that the adjusted broadband active detection signal can effectively avoid various types of electromagnetic interference, stably excite the vehicle grounding network, and be accurately acquired by the receiving antenna array.

[0077] Example 2, based on the same inventive concept as the online insulation monitoring system for national railway vehicles in the aforementioned examples, such as... Figure 2 As shown, this application provides a method for online insulation monitoring of national railway vehicles. The method and system embodiments in this application are based on the same inventive concept. The method includes:

[0078] Step S100: Inject a preset broadband active detection signal into the vehicle grounding network through multiple signal injection points deployed on the low-voltage side of the vehicle. At the same time, collect the response signal set transmitted through the vehicle grounding network through multiple signal receiving points deployed at different locations on the vehicle.

[0079] Step S200: Process the response signal set, extract the channel impulse response parameters that characterize the transmission characteristics of the vehicle grounding network, and construct the network fingerprint features under the current insulation state.

[0080] Step S300: Input the network fingerprint features into the pre-constructed vehicle electromagnetic-circuit joint digital twin model for inversion calculation to obtain the admittance spatial distribution map of the whole vehicle.

[0081] Step S400: Analyze the admittance spatial distribution map to identify insulation abnormality regions.

[0082] Furthermore, the method also includes:

[0083] Synchronous time-domain sampling and short-time Fourier transform are performed on each response signal in the response signal set to obtain a spectrum matrix of the signal changing over time. Using the code pattern of the broadband active detection signal as a reference, adaptive filtering and coherent processing are performed on the spectrum matrix to establish multipath response components. Inverse Fourier transform is performed on the multipath response components to calculate the discrete channel impulse response sequence of each response signal. From the channel impulse response sequence, delay spread, coherence bandwidth, and amplitude and phase characteristics at least three frequency points are extracted to constitute the network fingerprint feature.

[0084] Furthermore, the method also includes:

[0085] The multipath response components include response components for the direct path, the main reflection path, and the scattering path.

[0086] Furthermore, the method also includes:

[0087] In the vehicle electromagnetic-circuit joint digital twin model, full-wave electromagnetic simulation and circuit coupling calculation are performed on the initial admittance distribution assumption to predict the theoretical network fingerprint feature vector collected under the initial admittance distribution assumption; the difference metric between the theoretical network fingerprint feature vector and the network fingerprint feature is calculated; based on the difference metric, the admittance distribution assumption is adjusted, and the iteration is performed with the goal of minimizing the difference metric until the convergence condition is met, and the optimal admittance distribution assumption is output as the admittance spatial distribution map.

[0088] Furthermore, the method also includes:

[0089] The gradient of the difference metric with respect to the fingerprint feature vector of the theoretical network is used as the adjoint excitation source. The adjoint excitation source is applied to each signal receiving point of the vehicle electromagnetic-circuit joint digital twin model, and the adjoint electromagnetic field distribution of the whole vehicle space is solved in reverse. The initial admittance distribution assumption is simulated in the forward direction using the vehicle electromagnetic-circuit joint digital twin model to obtain the forward simulation electric field. The forward simulation electric field and the adjoint electromagnetic field distribution are multiplied by conjugate to calculate the gradient vector of the difference metric with respect to the admittance value of each grid cell. The admittance distribution assumption is updated according to the gradient vector, and the full-wave electromagnetic simulation and circuit coupling calculation are continued until the convergence condition is met.

[0090] Furthermore, the method also includes:

[0091] Based on vehicle design drawings, a three-dimensional surface model of the vehicle body metal structure, cable bundle, and electrical equipment housing is extracted. In the three-dimensional surface model, the vehicle grounding network is equivalent to a circuit topology network to establish an initial model. The electromagnetic properties of the materials are defined in the initial model, and the insulation state to be solved is parameterized as the admittance value of the mesh attached to the surface of the insulating component to obtain a first defined model. The positions of the signal injection point and the signal receiving point actually deployed are reproduced in the first defined model, and transient co-simulation is performed to simulate the full-band electromagnetic response generated by the injected signal excitation at each signal receiving point position, thus obtaining the whole vehicle electromagnetic-circuit joint digital twin model.

[0092] Furthermore, the method also includes:

[0093] The initial admittance distribution is assumed to be based on the admittance distribution configuration corresponding to the healthy insulation state of the vehicle under real-time operating conditions.

[0094] Furthermore, the method also includes:

[0095] The multiple signal injection points include the vehicle's main integrated grounding busbar and the local grounding busbar of the electrical cabinets containing the traction system and auxiliary system; the multiple signal receiving points include a first receiving antenna array deployed at symmetrical positions on both sides of the longitudinal centerline inside the vehicle body, used to sense the asymmetrical component of the vehicle body's lateral electromagnetic field; and a second receiving antenna array deployed on the low-voltage side within a preset range of the high-voltage insulation components, used to couple electromagnetic radiation generated by partial discharge or insulation leakage.

[0096] Furthermore, the method also includes:

[0097] The electromagnetic interference spectrum of the vehicle's operating environment is monitored in real time; based on the interference frequency points in the electromagnetic interference spectrum, the frequency, time slot, or coding strategy of the broadband active detection signal is dynamically adjusted according to a preset interference cancellation mechanism.

[0098] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0099] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0100] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. An online insulation monitoring system for national railway vehicles, characterized in that, include: The detection signal injection module is used to inject a preset broadband active detection signal into the vehicle grounding network through multiple signal injection points deployed on the low-voltage side of the vehicle. At the same time, it synchronously collects the response signal set transmitted through the vehicle grounding network through multiple signal receiving points deployed at different locations on the vehicle. The fingerprint feature construction module is used to process the response signal set, extract channel impulse response parameters that characterize the transmission characteristics of the vehicle grounding network, and construct the network fingerprint features under the current insulation state. The spatial distribution map acquisition module is used to input the network fingerprint features into a pre-constructed vehicle electromagnetic-circuit joint digital twin model for inversion calculation to obtain the admittance spatial distribution map of the whole vehicle; An insulation anomaly region identification module is used to analyze the admittance spatial distribution map and identify insulation anomaly regions; The fingerprint feature construction module includes: The spectrum matrix acquisition unit is used to perform synchronous time-domain sampling and short-time Fourier transform on each response signal in the response signal set to obtain the spectrum matrix of the signal as a function of time. The response component establishment unit is used to establish multipath response components by adaptively filtering and coherently processing the spectrum matrix with the code pattern of the broadband active detection signal as a reference. The response sequence acquisition unit is used to perform inverse Fourier transform on the multipath response components and calculate the discrete channel impulse response sequence of each response signal. The network fingerprint feature construction unit is used to extract the delay spread, coherence bandwidth, and amplitude and phase characteristics at least three frequency points from the channel impulse response sequence to form the network fingerprint feature; The spatial distribution map acquisition module includes: The network fingerprint feature vector prediction unit is used to perform full-wave electromagnetic simulation and circuit coupling calculation on the initial admittance distribution assumption in the whole vehicle electromagnetic-circuit joint digital twin model, and predict the theoretical network fingerprint feature vector collected under the initial admittance distribution assumption. The difference metric calculation unit is used to calculate the difference metric between the theoretical network fingerprint feature vector and the network fingerprint feature; The optimal admittance distribution hypothesis output unit is used to adjust the admittance distribution hypothesis based on the difference metric value, and iterate with the goal of minimizing the difference metric value until the convergence condition is met, and output the optimal admittance distribution hypothesis as the admittance spatial distribution map.

2. The online insulation monitoring system for national railway vehicles as described in claim 1, characterized in that, The multipath response components include response components for the direct path, the main reflection path, and the scattering path.

3. The online insulation monitoring system for national railway vehicles as described in claim 1, characterized in that, The optimal admittance distribution hypothesis output unit includes: The accompanying excitation source acquisition subunit is used to take the gradient of the difference metric value with respect to the fingerprint feature vector of the theoretical network as the accompanying excitation source; The accompanying electromagnetic field distribution acquisition subunit is used to apply the accompanying excitation source to each signal receiving point position of the whole vehicle electromagnetic-circuit joint digital twin model, and solve the accompanying electromagnetic field distribution of the whole vehicle space in reverse. The gradient vector calculation subunit is used to perform forward simulation of the initial admittance distribution assumption through the vehicle electromagnetic-circuit joint digital twin model, obtain the forward simulation electric field, perform conjugate dot product integration of the forward simulation electric field and the accompanying electromagnetic field distribution, and calculate the gradient vector of the difference metric value relative to the admittance value of each grid unit. The admittance distribution hypothesis update sub-unit is used to update the admittance distribution hypothesis according to the gradient vector, and continue to perform full-wave electromagnetic simulation and circuit coupling calculation until the convergence condition is met.

4. The online insulation monitoring system for national railway vehicles as described in claim 1, characterized in that, The spatial distribution map acquisition module also includes: The initial model building unit is used to extract the three-dimensional surface model of the vehicle body metal structure, cable bundle and electrical equipment housing based on the vehicle design drawings. In the three-dimensional surface model, the vehicle body grounding network is equivalent to the circuit topology network to build the initial model. The first definition model acquisition unit is used to define the electromagnetic properties of the material in the initial model and parameterize the insulation state to be solved into the admittance value of the mesh attached to the surface of the insulation component to obtain the first definition model; The digital twin model acquisition unit is used to reproduce the actual deployment of signal injection points and signal receiving points in the first defined model, perform transient co-simulation, simulate the full-band electromagnetic response generated by the injection signal excitation at each signal receiving point, and obtain the vehicle electromagnetic-circuit joint digital twin model.

5. The online insulation monitoring system for national railway vehicles as described in claim 1, characterized in that, The initial admittance distribution is assumed to be based on the admittance distribution configuration corresponding to the healthy insulation state of the vehicle under real-time operating conditions.

6. The online insulation monitoring system for national railway vehicles as described in claim 1, characterized in that, The multiple signal injection points include the vehicle's main integrated grounding busbar and the local grounding busbar of the electrical cabinets containing the traction system and auxiliary system; The plurality of signal receiving points include a first receiving antenna array deployed at symmetrical positions on both sides of the longitudinal centerline inside the vehicle body, for sensing the asymmetric components of the lateral electromagnetic field of the vehicle body. And a second receiving antenna array deployed on the low-voltage side within a preset range of the high-voltage insulation component, for coupling electromagnetic radiation generated by partial discharge or insulation leakage.

7. The online insulation monitoring system for national railway vehicles as described in claim 1, characterized in that, The detection signal injection module also includes: Interference spectrum detection unit, used to monitor the electromagnetic interference spectrum of the vehicle's operating environment in real time; The detection signal adjustment unit is used to dynamically adjust the frequency, time slot, or coding strategy of the broadband active detection signal based on the interference frequency points in the electromagnetic interference spectrum and a preset interference cancellation mechanism.

8. A method for online insulation monitoring of national railway vehicles, characterized in that, The method is implemented using an online insulation monitoring system for national railway vehicles as described in any one of claims 1-7, and the method includes: By deploying multiple signal injection points on the low-voltage side of the vehicle, a preset broadband active detection signal is injected into the vehicle grounding network. At the same time, by deploying multiple signal receiving points at different locations on the vehicle, the response signal set transmitted through the vehicle grounding network is collected synchronously. The response signal set is processed to extract channel impulse response parameters that characterize the transmission characteristics of the vehicle grounding network, and network fingerprint features under the current insulation state are constructed. The network fingerprint features are input into a pre-constructed vehicle electromagnetic-circuit joint digital twin model for inversion calculation to obtain the admittance spatial distribution map of the entire vehicle; Analyze the spatial distribution map of the admittance to identify areas of insulation abnormality.

Images