A Method and System for High-Voltage Interlock Fault Detection in New Energy Vehicles Based on a Training Platform

Abstract

This invention discloses a method and system for detecting high-voltage interlock faults in new energy vehicles based on a training platform. It relates to the field of high-voltage electrical system detection and training technology for new energy vehicles. The method includes: constructing a topology graph, selecting key points and electrically central vertices as a set of physical measurement points, generating virtual nodes between these points, and constructing an enhanced detection graph; defining perturbation matrices for vertex and edge faults on the enhanced detection graph, calculating fault sensitivity using the adjoint network method, setting an optimization objective function, solving for it, obtaining the optimal frequency set and corresponding amplitude, and generating a simulated voltage signal. This improves the sensitivity and location accuracy of fault detection, enhances the coverage of key nodes and branches, and, combined with the training platform's HMI operation and student data acquisition functions, enables real-time fault feedback and personalized teaching evaluation, significantly optimizing training effectiveness and learning efficiency.

Description

Technical Field

[0001] This invention relates to the field of testing and training technology for high-voltage electrical systems of new energy vehicles, and in particular to a method and system for detecting high-voltage interlock faults in new energy vehicles based on a training platform. Background Technology

[0002] With the rapid development of the new energy vehicle industry, the safety and reliability of high-voltage electrical systems, as core components of the vehicle, have received high attention.

[0003] High-voltage interlock circuits in new energy vehicles serve as an important protection mechanism for high-voltage systems. They are used to monitor the closure status of high-voltage component interfaces and cables to prevent electrical accidents caused by misoperation. With the development of digital twins, graph theory modeling, and multi-frequency electrical diagnostic technologies, some studies have attempted to construct digital models of high-voltage circuits and use transmission impedance, node admittance matrices, and simulation to achieve fault location.

[0004] However, existing technologies have obvious shortcomings. Traditional loop detection methods lack topology optimization and electrical centrality analysis, resulting in unreasonable selection of measurement points and difficulty in capturing fault signals of key nodes. They can usually only determine whether the loop is open or closed, lack multi-frequency quantization analysis and comprehensive fault sensitivity assessment, and are difficult to accurately locate soft faults. Summary of the Invention

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] In a first aspect, the present invention provides a method for detecting high-voltage interlock faults in new energy vehicles based on a training platform, which includes,

[0007] Offline preprocessing stage: Define the interlocking interface of high-voltage components as vertices and the wiring harness as edges, construct a topology graph, find the key points of the topology graph and calculate the electrical centrality of non-key points, and filter out the set of physical measurement points based on the key points and electrical centrality values;

[0008] On the edges between physical measurement points, candidate virtual nodes are generated. The electrical coupling strength of adjacent candidate virtual nodes is iteratively merged to obtain the effective virtual nodes, which are then combined with the physical measurement points to form an enhanced detection map.

[0009] The perturbation matrices of vertex and edge faults are defined on the enhanced detection map. The fault sensitivity is calculated by the adjoint network method. The goal is to maximize the minimum comprehensive sensitivity among all preset fault modes, while satisfying the frequency range and power constraints. An optimization objective function is set and solved to obtain the optimal frequency set and corresponding amplitude, which are then stored in the embedded controller of the training platform.

[0010] Online testing phase: After connecting the training platform to the safe low-voltage power supply and completing the circuit self-test, the operator selects the fault scenario through the HMI and drives the fault simulation relay group to operate;

[0011] An embedded controller generates analog voltage signals based on an optimal set of frequencies and their corresponding amplitudes.

[0012] The analog voltage signal is injected into the excitation port of the cable system, and the response voltage of the physical measurement point is sampled synchronously. For each optimal frequency point, the optimization problem of the band graph Laplace regularization is solved to obtain the estimated value of the total node voltage. The equivalent series impedance perturbation of each side on the enhanced detection map is calculated, and weighted fusion is performed to obtain the comprehensive impedance perturbation map.

[0013] Peak detection is performed on the real part curve of the integrated impedance disturbance spectrum to obtain candidate fault locations, and frequency domain impedance disturbance vectors are extracted from them. A diagnostic report is generated by matching the vectors with a pre-built fault feature library.

[0014] Based on the diagnostic report, the HMI-driven training platform collects students' test data, generates practical assessment reports, pushes personalized learning content to students, and provides teachers with a visual display of the assessment data.

[0015] As a preferred embodiment of the new energy vehicle high-voltage interlock fault detection method based on a training platform described in this invention, the step of defining the high-voltage component interlock interface as a vertex and the wiring harness as an edge to construct a topology graph, and selecting a set of physical measurement points based on key points and electrical centrality values ​​to construct an enhanced detection graph includes:

[0016] The interlock interface of each high-voltage component is abstracted as a vertex, and the wiring harness connecting the interlock interface is abstracted as an edge. The edge weight is set to the length of the wiring harness, and an HVIL loop topology is constructed.

[0017] Find the key points in the HVIL loop topology graph and add them to the candidate PMP set;

[0018] For non-key vertices, calculate the electrical centrality of the vertex and select the K vertices with the largest electrical centrality values ​​to add to the candidate PMP set;

[0019] For the vertices in the candidate PMP set, they are filtered according to the actual engineering constraints to form the final set of physical measurement points;

[0020] Based on the final set of physical measurement points, under fault-free system conditions, baseline calibration is performed, a baseline feature library is generated, and the baseline series impedance between physical measurement points is calculated.

[0021] Each physical measurement point in the final set of physical measurement points is used as a vertex of the initial weighted undirected graph, the actual connecting wires between the physical measurement points are used as edges of the initial weighted undirected graph, and the baseline series impedance is used as the edge weight of the initial weighted undirected graph.

[0022] On each edge of the initial weighted undirected graph, a series of candidate virtual nodes are generated according to a preset spatial resolution, forming a set of virtual nodes;

[0023] For two adjacent candidate virtual nodes, calculate the electrical coupling strength and preset a coupling threshold. If the electrical coupling strength between the two adjacent candidate virtual nodes is greater than the coupling threshold, take the geometric center point of the two candidate virtual nodes as a new node to represent them.

[0024] The electrical parameters of the new node are determined by the parallel admittance values ​​of the two candidate virtual nodes it represents, and the new node is added to the initial set of effective virtual nodes, while the two candidate virtual nodes it represents are removed from the set of virtual nodes.

[0025] In the remaining set of virtual nodes, the filtering process is repeated until no pair of adjacent virtual nodes can be found with a coupling strength greater than the coupling threshold. Then the iteration stops, and the final set of effective virtual nodes is obtained.

[0026] The final set of physical measurement points is merged with the final set of effective virtual nodes to form an enhanced detection map.

[0027] As a preferred embodiment of the high-voltage interlock fault detection method for new energy vehicles based on a training platform as described in this invention, the method involves defining perturbation matrices for vertex and edge faults on the enhanced detection map, calculating fault sensitivity using the adjoint network method, setting an optimization objective function, solving for the optimal frequency set and corresponding amplitude, and generating an analog voltage signal, including:

[0028] At a given diagnostic angular frequency ω, the physical parameters in the edge set of the enhanced detection diagram are equivalent to a Π-type lumped parameter circuit. The series impedance and parallel admittance to ground of the edge are calculated, and the reciprocal of the series impedance of the edge is taken as the mutual admittance of the edge.

[0029] For each vertex, its ground admittance is defined as the sum of the ground leakage conductance and capacitance of all converging branches at that node.

[0030] On the enhanced detection graph, target fault modes are defined, including vertex faults and edge faults, and a corresponding fault perturbation matrix is ​​generated for each type of fault.

[0031] For each failure mode, modeling is performed by modifying the node admittance matrix of the enhanced detection map, including the baseline node admittance matrices for both off-diagonal and diagonal elements.

[0032] The single-port current injection point is defined as the excitation port, and the observation port is the set of all physical measurement points (PMPs). Under fault-free conditions, the complex voltage change is calculated using the adjoint network method based on the circuit equations and the baseline admittance matrix.

[0033] The overall fault sensitivity is calculated based on the complex voltage change. An optimization objective function is set and solved by the simulated annealing algorithm to obtain the optimal frequency set and corresponding amplitude, which are then stored in the embedded controller of the training platform.

[0034] As a preferred embodiment of the high-voltage interlock fault detection method for new energy vehicles based on a training platform according to the present invention, wherein: after connecting the safe low-voltage power supply of the training platform and completing the circuit self-test, the operator selects a fault scenario through the HMI and drives the fault simulation relay group to operate, including:

[0035] When the safety low-voltage power supply of the training platform is turned on, the control unit sends a DC low-voltage verification signal to the HVIL circuit and detects the end-received signal.

[0036] If the terminal signal is received normally, it enters standby mode; if there is an abnormality, the process is terminated.

[0037] The operator selects a preset fault scenario or a custom fault in the "New Energy Vehicle High Voltage Interlock Fault Database" interface of the HMI to generate layout instructions;

[0038] The control unit receives the layout instructions and drives the fault simulation relay group corresponding to the target position to operate.

[0039] As a preferred embodiment of the high-voltage interlock fault detection method for new energy vehicles based on a training platform as described in this invention, the method involves: injecting an analog voltage signal into the excitation port of the cable system, synchronously sampling the response voltage of the physical measurement points, solving a regularized optimization problem for each optimal frequency point to obtain the estimated voltage value of all nodes, calculating the equivalent series impedance disturbance on each side of the enhanced detection map, performing weighted fusion, and obtaining a comprehensive impedance disturbance spectrum, including:

[0040] After the fault setting is completed, the embedded controller reads the stored optimal frequency set and corresponding amplitude, generates a discrete time series in the digital domain, and sends it to the digital-to-analog converter to generate an analog voltage signal. The analog voltage signal is then injected into the excitation port of the cable system. The system under test generates a voltage response under this excitation. All physical measurement points synchronously sample the response voltage of the system under test under the excitation signal. Windowed FFT is performed on the sampled data of each physical measurement point to extract complex voltage values ​​and construct a multi-frequency measurement vector.

[0041] Based on the baseline feature library, the baseline admittance values ​​of each side at each optimal frequency point are calculated, and the node admittance matrix and node injection current vector at the optimal frequency point are constructed.

[0042] For each optimal frequency point, based on the constructed nodal admittance matrix, injected current vector and measured voltage response, a regularized optimization problem is solved to obtain the voltage estimate of all nodes;

[0043] Based on the voltage estimates of all nodes, the equivalent series impedance disturbance of each edge in the enhanced detection map is calculated using the complex impedance difference method and mapped to its corresponding spatial location to obtain the single-frequency disturbance curve.

[0044] The perturbation curves at each frequency are weighted and fused to obtain a comprehensive impedance perturbation spectrum.

[0045] As a preferred embodiment of the high-voltage interlock fault detection method for new energy vehicles based on a training platform according to the present invention, the method includes: peak detection of the real part curve of the comprehensive impedance disturbance spectrum to obtain candidate fault locations, extracting the frequency domain impedance disturbance vector from it, and generating a diagnostic report by matching it with a pre-built fault feature library, including:

[0046] Plot the curve of the integrated impedance perturbation spectrum on the HMI, and perform peak detection on the curve of the integrated impedance perturbation spectrum to obtain the set of candidate fault locations.

[0047] Extract the frequency domain impedance perturbation vector at each candidate location, match it with a pre-built fault feature library, and generate a diagnostic report, including fault type, fault location, quantization parameters, and a screenshot of the comprehensive impedance perturbation spectrum.

[0048] As a preferred embodiment of the high-voltage interlock fault detection method for new energy vehicles based on a training platform as described in this invention, the method includes: based on diagnostic reports, using an HMI-driven training bench to collect student test data, generating practical evaluation reports, pushing personalized learning content to students, and providing teachers with a visual display of the evaluation data, including:

[0049] According to the fault location in the diagnostic report, the HMI system drives the corresponding LED indicator on the training bench panel to flash and pops up a standard instrument verification task sheet for the trainees, requiring them to use a real multimeter to perform measurements at the designated test holes on the bench.

[0050] The system automatically collects measured resistance values, instrument settings, and operation time through its built-in measurement interface;

[0051] The collected data is intelligently compared with the diagnostic results and standard operating procedures, and a practical skills assessment report is generated for each trainee, including result verification score, operating procedure score and deduction items;

[0052] Based on the practical skills assessment report, the system automatically analyzes the deduction items, matches and pushes personalized learning content to students from the teaching resource library, and aggregates the assessment data of all students for teachers based on the practical skills assessment report, and displays the class skill attainment rate, the list of difficult problems, and the students' performance ranking in a visual way.

[0053] Secondly, this invention provides a high-voltage interlock fault detection system for new energy vehicles based on a training platform, including:

[0054] The self-test setting module is used to connect the training platform's safe low-voltage power supply and complete the circuit self-test. After the operator selects the fault scenario through the HMI, the fault simulation relay group will be driven to operate.

[0055] A generation module is constructed to define the interlocking interfaces of high-voltage components as vertices and the wiring harnesses as edges, construct a topology graph, find the key points of the topology graph and calculate the electrical centrality of non-key points, and filter out the set of physical measurement points based on the key points and electrical centrality values.

[0056] On the edges between physical measurement points, candidate virtual nodes are generated. The electrical coupling strength of adjacent candidate virtual nodes is iteratively merged to obtain the effective virtual nodes, which are then combined with the physical measurement points to form an enhanced detection map.

[0057] A calculation module is set up to define the perturbation matrix of vertex and edge faults on the enhanced detection map. The fault sensitivity is calculated by the adjoint network method. The goal is to maximize the minimum comprehensive sensitivity among all preset fault modes, while satisfying the frequency range and power constraints. An optimization objective function is set, solved, and the optimal frequency set and corresponding amplitude are obtained and stored in the embedded controller of the training platform.

[0058] The acquisition and fusion module is used to inject analog voltage signals into the excitation port of the cable system, synchronously sample the response voltage of the physical measurement points, solve the optimization problem of band graph Laplace regularization for each optimal frequency point, obtain the estimated value of the total node voltage, calculate the equivalent series impedance perturbation of each side on the enhanced detection map, perform weighted fusion, and obtain the comprehensive impedance perturbation map.

[0059] The detection and matching module is used to perform peak detection on the real part curve of the integrated impedance disturbance spectrum, obtain candidate fault locations, extract frequency domain impedance disturbance vectors from them, and generate diagnostic reports by matching them with a pre-built fault feature library.

[0060] The visualization module is used to collect students' test data based on diagnostic reports and HMI-driven training benches, generate practical assessment reports, push personalized learning content to students, and visualize assessment data for teachers.

[0061] The beneficial effects of this invention are as follows: By constructing an HVIL loop topology diagram and combining the selection of key nodes and electrical centrality nodes, this invention forms a set of physical measurement points, further generating an enhanced detection map, thereby achieving comprehensive coverage of signals at key nodes of the loop.

[0062] By injecting multi-frequency excitation signals and calculating using the adjoint network method, a comprehensive impedance disturbance spectrum is established to detect soft faults such as increased contact resistance and decreased insulation with high sensitivity, achieving high-precision fault location. Compared with existing technologies, this invention significantly improves the sensitivity and location accuracy of fault detection, enhances the coverage of key nodes and branches, and, combined with the HMI operation and student data acquisition functions of the training platform, enables real-time fault feedback and personalized teaching evaluation, significantly optimizing training effects and learning efficiency.

[0063] Therefore, this invention solves the problems of difficulty in locating soft faults and lack of practical training feedback in existing HVIL detection methods. Attached Figure Description

[0064] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the 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.

[0065] Figure 1 This is a flowchart of the high-voltage interlock fault detection method for new energy vehicles based on the training platform in Example 1.

[0066] Figure 2 This is a schematic diagram of the high-voltage interlock fault detection system for new energy vehicles based on the training platform in Example 1.

[0067] Figure 3 This is a schematic diagram of the enhanced detection map construction process in Example 1.

[0068] Figure 4 This is a flowchart of the multi-frequency excitation and integrated impedance perturbation spectrum generation process in Example 1. Detailed Implementation

[0069] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0070] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0071] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0072] Example 1, referring to Figures 1 to 4 This is the first embodiment of the present invention, which provides a method for detecting high-voltage interlock faults in new energy vehicles based on a training platform, including the following steps:

[0073] S1. After connecting the safety low-voltage power supply to the training platform and completing the circuit self-test, the operator selects the fault scenario through the HMI and drives the fault simulation relay group to operate.

[0074] Preferably, when the training platform is connected to a safe low-voltage power supply (such as 24V DC), the control unit automatically executes the initialization script to put the relays of all programmable fault injection modules into the "normal circuit" state (i.e., bypass all simulated fault resistors and close the circuit to the normal state).

[0075] The control unit sends a low-voltage DC verification signal (e.g., 5V) to the HVIL circuit and detects the end-received signal;

[0076] If the terminal signal is received normally, the system is determined to have an "intact HVIL basic circuit," the HMI displays "System Ready," and enters standby mode. If an anomaly occurs, an alarm is triggered indicating a "Platform hardware circuit failure," and the process is terminated.

[0077] The operator (teacher or student) selects the testing mode for this training session through the HMI;

[0078] The operator selects a preset fault scenario or a custom fault in the "New Energy Vehicle High Voltage Interlock Fault Database" interface of the HMI to generate layout instructions;

[0079] The control unit receives the layout instructions and drives the fault simulation relay group corresponding to the target location to operate.

[0080] If set to "MCU input contact resistance 250Ω" (used to simulate a fault caused by increased series resistance due to loose terminals or oxidation of contact surfaces), the relay at this node will be controlled to disconnect the normal path and connect a linear resistor element with a resistance of 250Ω.

[0081] It should be noted that this training platform uses a linear resistor to simulate the equivalent series resistance value of a poor contact fault at the diagnostic frequency, in order to simplify the fault injection circuit.

[0082] The nonlinear and time-varying characteristics of real contact failure faults are not within the simulation range of this platform, but this does not affect the teaching objective of students to diagnose faults based on impedance changes.

[0083] It should be understood that the fault detection method based on enhanced detection map and multi-frequency excitation proposed in this invention can be fully extended to nonlinear fault models. Moreover, it only requires that the characteristic parameters of the nonlinear fault be substituted into the digital twin model in the offline preprocessing stage to generate the corresponding fault feature template, thereby realizing the detection and location of real nonlinear faults.

[0084] On the topology diagram of the HMI interface, the target location icon changes to the fault status color (such as red) and displays the fault parameters.

[0085] S2. Define the interlocking interface of the high-voltage component as a vertex and the wiring harness as an edge, construct a topology graph, find the key points of the topology graph and calculate the electrical centrality of the non-key points, and select the set of physical measurement points based on the key points and electrical centrality values.

[0086] On the edges between physical measurement points, candidate virtual nodes are generated. The electrical coupling strength of adjacent candidate virtual nodes is iteratively merged to obtain the effective virtual nodes, which are then combined with the physical measurement points to form an enhanced detection map.

[0087] Preferably, the interlock interface of each high-voltage component (such as the battery pack BMS interface, motor controller MCU interface, on-board charger OBC interface, etc.) is abstracted as a vertex, and the wiring harness connecting the interlock interface is abstracted as an edge. The edge weight is set to the length of the wiring harness to construct the HVIL loop topology.

[0088] Find the key points in the HVIL loop topology graph and add them to the candidate PMP set;

[0089] Key points, specifically, refer to vertices in the HVIL loop topology that increase the number of connected components after the removal of the vertex. These key points correspond to critical connectors or junction boxes that connect multiple branches in the actual loop and are strategic points for monitoring signals flowing through different branches.

[0090] For non-key vertices, calculate the electrical centrality of the vertex and select the K' vertices with the largest electrical centrality values ​​to add to the candidate PMP set. The value of K' depends on the complexity of the actual topology and engineering requirements.

[0091] Electrical centrality is defined as the reciprocal of the sum of the effective resistances from a vertex to all other vertices in a resistive network in which each edge is treated as a unit resistance (i.e., the edge weight is 1, but the actual cable length is only used for the topology and is not included in the resistance value).

[0092] The above method for calculating the effective resistance refers to injecting a 1A current between vertices i and j in a resistor network (resistance of each edge = 1Ω), and the measured voltage difference between the two vertices is the effective resistance.

[0093] For the vertices in the candidate PMP set, all key points and electrical centrality vertices are first forced as mandatory points, and the remaining vertices (non-critical nodes) are screened or supplemented according to actual engineering constraints (such as spatial accessibility and installation cost) to form the final set of physical measurement points.

[0094] It should be noted that electrical critical nodes (joints, vertices with high electrical centrality) should not be excluded due to actual engineering constraints.

[0095] Based on the final set of physical measurement points, under fault-free (or known standard) conditions of the system, baseline calibration is performed to generate a baseline feature library.

[0096] The control unit sequentially injects a set of discrete frequency sinusoidal signals covering the diagnostic frequency band (e.g., 100Hz-100kHz) into the circuit;

[0097] At each frequency, the transfer impedance matrix between all physical measurement points in the final set of physical measurement points is measured synchronously (i.e., the open-circuit voltage generated at all other physical measurement points when current is injected from a certain physical measurement point).

[0098] The transfer impedance matrix at different frequencies, along with the corresponding loop physical parameters (such as cable type, length, and connector model), are stored to form a baseline feature library.

[0099] Each physical measurement point in the final set of physical measurement points is used as a vertex of the initial weighted undirected graph, the actual connecting wires between the physical measurement points are used as edges of the initial weighted undirected graph, and the baseline series impedance is used as the edge weight of the initial weighted undirected graph.

[0100] On each edge of the initial weighted undirected graph, a series of candidate virtual nodes are generated according to a preset spatial resolution (e.g., 0.05 meters), forming a set of virtual nodes;

[0101] For two adjacent candidate virtual nodes, calculate the electrical coupling strength and pre-set a coupling threshold (determined through actual measurement statistics). If the electrical coupling strength between two adjacent candidate virtual nodes is greater than the coupling threshold, it is considered that the two adjacent candidate virtual nodes are electrically too coupled and it is difficult to distinguish their respective influences at the physical measurement point. The geometric center point of these two candidate virtual nodes is taken as a new node to represent them.

[0102] The aforementioned coupling threshold was determined through actual measurement and statistics. Specifically, for a typical HVIL circuit with a total length of approximately 2.5 meters and containing 3 connectors, after generating candidate virtual nodes with a spatial resolution of 0.05 meters, the electrical coupling strength between adjacent candidate nodes was calculated through simulation and ranged from 0.02 to 0.8.

[0103] Through comparative experiments, virtual node merging and subsequent fault location tests were conducted with coupling thresholds set to 0.05, 0.10, 0.15 and 0.20, respectively. It was found that within the range of coupling thresholds from 0.10 to 0.15, the number of virtual nodes can be reduced by about 40% while ensuring that the fault location error is less than 0.1 meters, effectively reducing the online computational complexity.

[0104] Therefore, in this embodiment, the preferred coupling threshold is 0.15. Those skilled in the art should understand that for HVIL loops of different lengths and topologies, a suitable coupling threshold can be selected based on the coupling strength distribution curve through similar simulation or measurement methods to balance positioning accuracy and computational efficiency.

[0105] For two adjacent candidate virtual nodes, the electrical coupling strength is calculated. Specifically, when a unit impedance disturbance is applied at one of the candidate virtual nodes (e.g., increasing the admittance to ground of the node by 1 μS), the voltage change (2-norm of the complex vector) caused at all physical measurement points is compared with the amplitude of the impedance disturbance. This ratio is taken as the coupling strength value. It should be noted that this calculation is performed in the digital twin model after baseline calibration (i.e., the simulation model established based on the enhanced detection map), rather than in the actual physical system.

[0106] The electrical parameters (earth admittance and series impedance) of the new node are determined by the parallel admittance values ​​of the two candidate virtual nodes it represents. The new node is added to the initial set of active virtual nodes, and the two candidate virtual nodes it represents are removed from the set of virtual nodes.

[0107] The electrical parameters of the new node are determined by the parallel admittance values ​​of the two candidate virtual nodes it represents. If the two candidate virtual nodes are located at adjacent positions on the same cable (series relationship), the series impedance of the merged new node is equal to the sum of the series impedances of the two candidate virtual nodes it represents, and the admittance to ground is equal to the sum of the admittances to ground of the two candidate virtual nodes it represents (if the two virtual nodes originally each had their own admittances to ground, then they are summed in parallel). After merging, the new node is located at the geometric center of the two original candidate virtual nodes in space.

[0108] In the remaining set of virtual nodes, the filtering process is repeated until no pair of adjacent virtual nodes can be found with a coupling strength greater than the coupling threshold. Then the iteration stops, and the final set of effective virtual nodes is obtained. (For the remaining nodes in the set of virtual nodes after the iteration stops, i.e., all nodes whose coupling strength with other nodes is lower than the threshold, they are marked as electrically distinguishable independent nodes and added to the final set of effective virtual nodes.)

[0109] The final set of physical measurement points is merged with the final set of effective virtual nodes to form an enhanced detection map;

[0110] Specifically, merging includes merging the vertices in the two sets mentioned above;

[0111] Preserve the edges between all physical measurement points in the enhanced detection map;

[0112] For an effective virtual node, determine its spatial location on the original cable based on its generation source, insert the virtual node into the edge corresponding to the original cable, remove the original edge, and add two new edges to maintain the connectivity and topological consistency of the graph.

[0113] Electrical parameters are assigned to each edge of the enhanced detection map. For edges newly generated due to the insertion of virtual nodes, their resistance per unit length, inductance per unit length, capacitance per unit length to ground, and conductance per unit length to ground are exactly the same as the corresponding parameters of the original cable. Their series impedance and parallel admittance are obtained by multiplying the unit length parameter by the physical length of the new edge, without the need for linear interpolation.

[0114] For unsegmented original edges, their original parameters are directly inherited;

[0115] The aforementioned electrical parameters include the physical length of the edge, resistance per unit length, inductance per unit length, capacitance to ground per unit length, and conductance to ground per unit length.

[0116] S3. Define the perturbation matrix of vertex and edge faults on the enhanced detection map, calculate the fault sensitivity through the adjoint network method, aim to maximize the minimum comprehensive sensitivity among all preset fault modes, and satisfy the frequency range and power constraints. Set the optimization objective function, solve it, obtain the optimal frequency set and corresponding amplitude, and store it in the embedded controller of the training platform.

[0117] Preferably, at a given diagnostic angular frequency ω, the physical parameters in the edge set of the enhanced detection map are equivalent to a Π-type lumped parameter circuit, and the series impedance and parallel admittance to ground of the edges are calculated using the following formula:

[0118] ;

[0119] ;

[0120] in, and These are the series impedance and parallel admittance to ground of edge e at a given diagnostic angular frequency ω. , , , and Let e ​​represent the resistance per unit length, inductance per unit length, physical length, capacitance per unit length to ground, and conductance per unit length to ground for edge e, respectively. The imaginary unit;

[0121] Take the reciprocal of the series impedance of the side as the mutual admittance of the side (i.e., the side admittance).

[0122] For each vertex, its admittance to ground is defined as the sum of the ground leakage conductance and capacitance of all converging branches at that node (or 0 if there is no additional load).

[0123] On the enhanced detection graph, target fault modes are defined, including vertex faults and edge faults, and a corresponding fault perturbation matrix is ​​generated for each type of fault.

[0124] The aforementioned vertex fault refers to an abnormal insulation state of the connection point to the reference ground (vehicle ground), which is quantified as the change in the node's ground admittance.

[0125] A side fault indicates an abnormality in the series impedance characteristics of the guide body path, which is quantified as a change in series impedance. ;

[0126] in, , indicating edge ( The circuit of (j) has poor contact. For example, if the Ω is >10kΩ, the side can be approximated ( The circuit of (j) is open. The complex number represents the edge ( The circuit of (j) is partially short-circuited;

[0127] For each failure mode, modeling is performed by modifying the node admittance matrix of the enhanced detection map, including off-diagonal elements. and diagonal elements The baseline node admittance matrix, formula:

[0128] ;

[0129] ;

[0130] in, For a given diagnostic angular frequency ω, the nth node in the baseline node admittance matrix is... Okay, number Column elements, For the mutual admittance of edge e at a given diagnostic angular frequency ω, To enhance the edge set of the probe graph, For a given diagnostic angular frequency ω, the nth node in the baseline node admittance matrix is... Okay, number Column elements, As vertices The neighbor set of vertex i refers to the set of all vertices directly connected to vertex i in the augmented detection graph. To connect vertices at a given diagnostic angular frequency ω The mutual admittance of j and j To connect vertices The parallel ground admittance of the edge with j, As vertices Earth admittance;

[0131] Vertex faults only affect diagonal elements, yielding the fault perturbation matrix for vertex faults, as shown in the formula:

[0132] ;

[0133] in, Let f be the change in the value of the element in the i-th row and i-th column of the baseline node admittance matrix caused by a fault f at a given diagnostic angular frequency ω. The index of the faulty vertex, from fault f. For angular frequency Below, the change in ground admittance at vertex n;

[0134] For edge faults, the mutual admittance of the relevant edges and the self-admittance of the two endpoints are affected. Let the original edge admittance be . The admittance after the perturbation is ;

[0135] in, For angular frequency below( The admittance value of (j). For angular frequency Below, after the fault ( The admittance value of (j), where j is different from the admittance value of (j,j). The index of the faulty vertex, For edge ( The series impedance of (j) This represents the change in series impedance caused by the fault.

[0136] Define the change in admittance as The admittance disturbance matrix of the edge fault is given by the formula:

[0137] ;

[0138] in, For an edge fault, at a given diagnostic angular frequency ω, the change in the value of the element in the i-th row and i-th column of the baseline node admittance matrix caused by fault f. For vertex index;

[0139] It should be noted that the above method for calculating the change in fault voltage based on the first-order approximation is only applicable to small-signal disturbance scenarios, that is, when the norm of ΔY is much smaller than the norm of Y. For hard faults such as open circuit or short circuit to ground, the linear approximation is no longer effective. In this case, the nonlinear equation should be solved directly. This scheme is mainly for diagnosing soft faults such as increased contact resistance and decreased insulation resistance.

[0140] The single-port current injection point is defined as the excitation port, and the observation port is the set of all physical measurement points (PMPs). Under fault-free (baseline) conditions, the complex voltage change is calculated using the adjoint network method based on the circuit equations and the baseline admittance matrix.

[0141] The above circuit equations and formulas are as follows:

[0142] ;

[0143] in, The baseline admittance matrix describes the frequency domain characteristics of the network in the fault-free state. Given the baseline voltage vector, the complex voltages at all nodes at frequency ω. For the injected current vector, only at the injection point Since the value is non-zero at a given point, this equation is the matrix form of Kirchhoff's Current Law (KCL) in the frequency domain.

[0144] When a disturbance fault occurs, ignoring higher-order minor quantities, calculate the fault state voltage change vector (first-order approximation) using the following formula:

[0145] ;

[0146] in, This is the vector of voltage changes caused by the fault;

[0147] Using the adjoint network method, an adjoint network is constructed in which a unit current source is applied only at the observation point. Based on Tellegen's theorem, the complex voltage change caused by fault f at observation point j is obtained, as shown in the formula:

[0148] ;

[0149] ;

[0150] ;

[0151] in, Let be the adjoint network voltage vector at observation point j at frequency ω. Let be the unit vector of observation point j, indicating that observation point j is 1 if it is 1, and 0 otherwise. Let f be the complex voltage change caused by fault f at observation point j at frequency ω.

[0152] The overall fault sensitivity is calculated based on the complex voltage change, using the following formula:

[0153] ;

[0154] in, The overall fault sensitivity of fault f at frequency ω. Let j be the set of all observation points j;

[0155] Define the objective function for optimization, as shown in the formula:

[0156] ;

[0157] ;

[0158] subject to ;

[0159] ;

[0160] ;

[0161] ;

[0162] in, The overall sensitivity of fault f under multi-frequency excitation. and These are the frequency vector and the amplitude vector, respectively. The frequency vector corresponds to the frequency to be optimized, and the amplitude vector corresponds to the current amplitude of each frequency component. and The encoding uses a discretization method. The fault mode set is a pre-selected representative set of faults, specifically including: vertex faults corresponding to all physical measurement points (insulation to ground drops to 1kΩ, 10kΩ, and 100kΩ) and contact resistance faults on all original edges (non-virtual edges) (series resistance increases to 50Ω, 250Ω, and 1kΩ). To further improve the sensitivity to faults at arbitrary locations, no more than 20% of the original number of virtual nodes (but at least 5) are randomly selected from the final effective virtual node set, and a fault mode with a 250Ω increase in series resistance is added to them, and these are also included. , This is an index representing the total number of frequency components. The total number of frequency components. Let be the power weight of the k-th frequency component. Here, we assume a resistive load, where the power is proportional to the square of the current amplitude. and The lower and upper bounds for operation at each frequency are set based on hardware parameters. For the minimum allowable frequency interval, Total power, based on hardware settings;

[0163] The optimal frequency set and corresponding amplitude are obtained by solving the objective function through simulated annealing algorithm, and then stored in the embedded controller of the training platform along with the baseline feature library and the fault feature library.

[0164] When solving the objective function using the simulated annealing algorithm, the frequency band is evenly divided into 200 candidate frequency points. W frequencies (W is fixed at 3 or 4) are selected from the candidate points and their amplitudes are assigned. The initial value of the amplitude is the average of the total power. Once W is fixed, frequency repetition is not allowed.

[0165] The neighborhood generation method of the simulated annealing algorithm is to randomly replace a frequency point or randomly fine-tune an amplitude (multiply by a random number between 0.8 and 1.2 and then renormalize the power).

[0166] S4. Inject the analog voltage signal into the excitation port of the cable system, synchronously sample the response voltage of the physical measurement point, solve the optimization problem with regularization for each optimal frequency point, obtain the estimated value of the total node voltage, calculate the equivalent series impedance disturbance of each side on the enhanced detection map, perform weighted fusion, and obtain the comprehensive impedance disturbance spectrum.

[0167] Preferably, the embedded controller, based on the optimal frequency set and corresponding amplitude, generates a discrete-time series in the digital domain (such as an FPGA or high-speed DSP), and sends it to a digital-to-analog converter to generate an analog voltage signal.

[0168] An analog voltage signal is injected into the excitation port of the cable system. The system under test generates a voltage response under this excitation. All physical measurement points synchronously sample the response voltage of the system under test under the excitation signal. Windowed FFT (Hanning window) is performed on the sampled data of each physical measurement point to extract complex voltage values ​​and construct a multi-frequency measurement vector.

[0169] The aforementioned cable system refers to the actual physical cable network corresponding to the HVIL loop, including all wire harnesses, connectors, junction boxes, and interlocking interfaces of high-voltage components;

[0170] Based on the baseline feature library, the baseline admittance values ​​of each side at each optimal frequency point are calculated, and the node admittance matrix and node injection current vector at the optimal frequency point are constructed.

[0171] For each optimal frequency point, based on the constructed nodal admittance matrix, injected current vector, and measured voltage response, a regularized optimization problem is solved to obtain the voltage estimate of all nodes (physical measurement points + virtual nodes).

[0172] The above regularization optimization problem has the following formula:

[0173] ;

[0174] in, Let S be the total node voltage estimation vector, and S be the selection matrix (extracting the physical measurement point components from the total node voltage), where rows represent the number of physical measurement points and columns represent the number of total nodes (physical measurement points + virtual nodes). For the _i_th node... There are 1 physical measurement point (whose global number in the global node is 1). ), will the first of S line, number Set the elements of the column to 1, and set the rest of the elements in the row to 0. The regularization parameter is used to balance the weights of the data fitting term (the first term) and the smoothing prior term (the second term). In this embodiment, an empirical value of 10 is used. -4 This value was obtained through simulation verification of a typical HVIL circuit and is used as a fixed parameter during online diagnosis. The Laplacian matrix is ​​constructed based on the topological connectivity of the enhanced probe graph and is used to encode the connectivity relationships between nodes in the enhanced probe graph. This represents the complex voltage vector at the measured physical measurement point. It is the transpose symbol;

[0175] Specifically, the expression for each element in the graph Laplacian matrix is ​​as follows:

[0176] ;

[0177] in, For the first Laplace matrix in the graph Okay, number Column elements, For the index of neighboring nodes, To enhance the detection of nodes in the graph The set of neighboring nodes, For the edge In this embodiment, to simplify the calculation, the weight is taken as... , For the edge The weights, and Similarly, design;

[0178] Based on the voltage estimates of all nodes, the equivalent series impedance disturbance of each edge in the enhanced detection map is calculated using the complex impedance difference method and mapped to its corresponding spatial location to obtain the single-frequency disturbance curve.

[0179] The above mapping means that in the enhanced detection map, each original edge (cable) has a known physical length. When the edge is divided into several segments due to the insertion of virtual nodes, the length of each segment is allocated proportionally to its starting position on the original edge.

[0180] For any virtual node or physical measurement point, its spatial location coordinates are the cumulative cable length traversed from a preset reference point (such as the HVIL start point) along the topology path to that node.

[0181] In the graph, the vertical axis (color or height) represents the equivalent series impedance increment (real and imaginary parts) at that location.

[0182] The perturbation curves at each frequency are weighted and fused (the weighting coefficients are the normalized values ​​of the squared amplitudes of each frequency component) to obtain the comprehensive impedance perturbation spectrum.

[0183] The above-mentioned comprehensive impedance disturbance spectrum is a one-dimensional function of the spatial location of each cable along the HVIL loop, describing the equivalent series impedance increment at each location (the real part is the resistance increment and the imaginary part is the reactance increment). Different colors or heights in the spectrum indicate the degree of impedance anomaly, thereby achieving accurate fault location (e.g., a resistance increment of 250mΩ at 0.3 meters indicates poor contact at that location).

[0184] S5. Peak detection is performed on the real part curve of the integrated impedance disturbance spectrum to obtain candidate fault locations, and frequency domain impedance disturbance vectors are extracted from them. A diagnostic report is generated by matching the vectors with a pre-built fault feature library.

[0185] Preferably, the curve of the integrated impedance disturbance spectrum (the horizontal axis is the loop length and the vertical axis is the impedance increment) is plotted on the HMI, and the peak detection is performed on the curve (real part curve) of the integrated impedance disturbance spectrum (the threshold is set to 3 times the mean of the curve) to obtain the set of candidate fault locations.

[0186] It should be noted that the optimal frequency set used in this embodiment is the globally optimal solution obtained by maximizing the minimum comprehensive sensitivity of all representative fault modes. Although the sensitivity of this frequency set to any specific fault may be lower than that of a dedicated frequency set optimized separately for that fault, this invention believes that the scheme does not have any substantial contradiction and has engineering advantages for the following reasons:

[0187] (1) The actual fault mode is unknown: the system cannot predict the type of fault that is currently occurring during online detection, so it cannot pre-select a dedicated frequency set;

[0188] The globally optimal frequency set guarantees the maximization of the minimum sensitivity to all possible faults, that is, ensuring the reliability of diagnosis in the worst case, which is a robust optimization strategy.

[0189] (2) Feature matching depends on multi-frequency response vectors rather than single-frequency sensitivity: This invention extracts the frequency domain impedance disturbance vector at the candidate fault location (i.e., the vector composed of complex impedance values ​​at multiple optimal frequency points), rather than the scalar value at a single frequency. The distribution of different fault modes in the multi-dimensional feature space has natural distinguishability. Even if the sensitivity of a single frequency is not the highest at the global optimal frequency, the vector pattern composed of multiple frequencies can still uniquely characterize the fault type and degree.

[0190] (3) Avoid multiple measurements: This invention can obtain full-band information by using a single multi-frequency excitation, taking into account both diagnostic speed and accuracy.

[0191] Extract the frequency domain impedance perturbation vector at each candidate location, match it with a pre-built fault feature library (such as minimum Euclidean distance), and generate a diagnostic report, including fault type, fault location, quantization parameters, and a screenshot of the comprehensive impedance perturbation spectrum.

[0192] The aforementioned fault feature library is a pre-constructed structured database. It is a standard feature template library formed by injecting various preset faults into each typical location in the digital twin model of the enhanced detection map, and collecting the frequency domain impedance disturbance vector under each fault state.

[0193] Each template in the fault feature library corresponds to a unique triplet of "fault type - fault severity - fault location";

[0194] During online diagnosis, the measured frequency domain impedance disturbance vector is matched with each template in the feature library for similarity (e.g., minimum Euclidean distance), which can output the fault type (e.g., "increased contact resistance"), quantification parameters (e.g., "250Ω"), and fault location (converted to "how many meters from a certain component interface").

[0195] The aforementioned preset faults include increased contact resistance (50Ω / 250Ω / 1kΩ three levels), open circuit, short circuit to ground, and decreased insulation (1kΩ / 10kΩ / 100kΩ three levels), etc.

[0196] S6. Based on the diagnostic report, the HMI drives the training platform to collect students' test data, generate practical assessment reports, push personalized learning content to students, and provide teachers with a visual display of assessment data.

[0197] Preferably, the HMI system drives the corresponding LED indicator on the training bench panel to flash according to the fault location in the diagnostic report (either the first one or selected by the teacher), and pops up a standard instrument verification task sheet (i.e., a training guide text containing the fault location, expected resistance range, designated test hole number, and measurement precautions) to the student, requiring the student to use a real multimeter to perform measurements at the designated test hole on the bench.

[0198] The system automatically collects measured resistance values, instrument settings, and operation time through its built-in measurement interface;

[0199] The collected data (measured resistance value, instrument setting, operation time) is intelligently compared with the diagnostic results (i.e., the fault type and expected resistance value given by the algorithm) and standard operating procedures (such as correct setting, zeroing of test leads, etc.), and a practical skills assessment report is generated for each trainee, including result verification score, operating procedure score and deduction items.

[0200] The result verification score is calculated based on the relative error between the measured value and the expected value.

[0201] The operating procedure includes checking the correct gear position, whether it has been zeroed, and whether the power has been cut off (determined by using the confirmation button on the HMI or by automatically detecting the zero point offset before measurement using the built-in measurement interface), which is automatically evaluated by the system rules;

[0202] Based on the practical skills assessment report, the system automatically analyzes the deduction items, matches and pushes personalized learning content to students from the teaching resource library, and aggregates the assessment data of all students for teachers based on the practical skills assessment report, and displays the class skill attainment rate, the list of difficult problems, and the students' performance ranking in a visual way;

[0203] Personalized learning content, such as sample documents, prompts students: "Your weak point is 'interpreting the resistance value of a loose connection fault,' and we recommend learning the course 'Measurement of Contact Resistance and Influencing Factors.'"

[0204] This embodiment also provides a high-voltage interlock fault detection system for new energy vehicles based on a training platform, including:

[0205] The self-test setting module is used to connect the training platform's safe low-voltage power supply and complete the circuit self-test. After the operator selects the fault scenario through the HMI, the fault simulation relay group will be driven to operate.

[0206] A generation module is constructed to define the interlocking interfaces of high-voltage components as vertices and the wiring harnesses as edges, construct a topology graph, select the key points and electrical centrality vertices of the topology graph as a set of physical measurement points, generate virtual nodes between the physical measurement points, and construct an enhanced detection graph.

[0207] A calculation module is set up to define the perturbation matrix of vertex and edge faults on the enhanced detection map, calculate the fault sensitivity through the adjoint network method, set the optimization objective function, solve it, obtain the optimal frequency set and corresponding amplitude, and generate a simulated voltage signal.

[0208] The acquisition and fusion module is used to inject analog voltage signals into the excitation port of the cable system, synchronously sample the response voltage of physical measurement points, solve the optimization problem with regularization for each optimal frequency point, obtain the estimated value of the voltage of all nodes, calculate the equivalent series impedance perturbation of each side on the enhanced detection map, perform weighted fusion, and obtain the comprehensive impedance perturbation map.

[0209] The detection and matching module is used to perform peak detection on the real part curve of the integrated impedance disturbance spectrum, obtain candidate fault locations, extract frequency domain impedance disturbance vectors from them, and generate diagnostic reports by matching them with a pre-built fault feature library.

[0210] The visualization module is used to collect students' test data based on diagnostic reports and HMI-driven training benches, generate practical assessment reports, push personalized learning content to students, and visualize assessment data for teachers.

[0211] It should be noted that the training platform of the present invention includes not only a physical test bench (including fault simulation relays, measurement point interfaces, HMI touch screens, etc.), but also a host computer (PC) or high-performance embedded controller that communicates with the test bench.

[0212] All steps in the offline preprocessing stage (constructing the enhanced detection map, calculating the sensitivity of the adjoint network method, optimizing the frequency set by simulated annealing, etc.) are executed in the host computer, utilizing its powerful floating-point operation capabilities and matrix operation library.

[0213] After offline optimization, the obtained optimal frequency set, amplitude, baseline feature library, fault feature library, and other data are stored in the training platform's memory. During online detection, the training platform's controller only needs to read the pre-stored data, generate the corresponding multi-frequency excitation signal, and perform simple FFT and matching operations. Therefore, the technical solution of this invention can be fully implemented on existing new energy vehicle training platforms without requiring unconventional hardware upgrades.

[0214] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for detecting high-voltage interlock faults in new energy vehicles based on a training platform, characterized in that, include: Offline preprocessing stage: Define the interlocking interface of high-voltage components as vertices and the wiring harness as edges, construct a topology graph, find the key points of the topology graph and calculate the electrical centrality of non-key points, and filter out the set of physical measurement points based on the key points and electrical centrality values; On the edges between physical measurement points, candidate virtual nodes are generated. The electrical coupling strength of adjacent candidate virtual nodes is iteratively merged to obtain the effective virtual nodes, which are then combined with the physical measurement points to form an enhanced detection map. The perturbation matrices of vertex and edge faults are defined on the enhanced detection map. The fault sensitivity is calculated by the adjoint network method. The goal is to maximize the minimum comprehensive sensitivity among all preset fault modes, while satisfying the frequency range and power constraints. An optimization objective function is set and solved to obtain the optimal frequency set and corresponding amplitude, which are then stored in the embedded controller of the training platform. Online testing phase: After connecting the training platform to the safe low-voltage power supply and completing the circuit self-test, the operator selects the fault scenario through the HMI and drives the fault simulation relay group to operate; An embedded controller generates analog voltage signals based on an optimal set of frequencies and their corresponding amplitudes. The analog voltage signal is injected into the excitation port of the cable system, and the response voltage of the physical measurement point is sampled synchronously. For each optimal frequency point, the optimization problem of the band graph Laplace regularization is solved to obtain the estimated value of the total node voltage. The equivalent series impedance perturbation of each side on the enhanced detection map is calculated, and weighted fusion is performed to obtain the comprehensive impedance perturbation map. Peak detection is performed on the real part curve of the integrated impedance disturbance spectrum to obtain candidate fault locations, and frequency domain impedance disturbance vectors are extracted from them. A diagnostic report is generated by matching the vectors with a pre-built fault feature library. Based on the diagnostic report, the HMI-driven training platform collects students' test data, generates practical assessment reports, pushes personalized learning content to students, and provides teachers with a visual display of the assessment data.

2. The method for detecting high-voltage interlock faults in new energy vehicles based on a training platform as described in claim 1, characterized in that, The process involves defining the interlocking interfaces of high-voltage components as vertices and the wiring harnesses as edges to construct a topology graph. Based on key points and electrical centrality values, a set of physical measurement points is selected to construct an enhanced detection map, including: The interlock interface of each high-voltage component is abstracted as a vertex, and the wiring harness connecting the interlock interface is abstracted as an edge. The edge weight is set to the length of the wiring harness, and an HVIL loop topology is constructed. Find the key points in the HVIL loop topology graph and add them to the candidate PMP set; For non-key vertices, calculate the electrical centrality of the vertex and select the K vertices with the largest electrical centrality values ​​to add to the candidate PMP set; For the vertices in the candidate PMP set, they are filtered according to the actual engineering constraints to form the final set of physical measurement points; Based on the final set of physical measurement points, under fault-free system conditions, baseline calibration is performed, a baseline feature library is generated, and the baseline series impedance between physical measurement points is calculated. Each physical measurement point in the final set of physical measurement points is used as a vertex of the initial weighted undirected graph, the actual connecting wires between the physical measurement points are used as edges of the initial weighted undirected graph, and the baseline series impedance is used as the edge weight of the initial weighted undirected graph. On each edge of the initial weighted undirected graph, a series of candidate virtual nodes are generated according to a preset spatial resolution, forming a set of virtual nodes; For two adjacent candidate virtual nodes, calculate the electrical coupling strength and preset a coupling threshold. If the electrical coupling strength between the two adjacent candidate virtual nodes is greater than the coupling threshold, take the geometric center point of the two candidate virtual nodes as a new node to represent them. The electrical parameters of the new node are determined by the parallel admittance values ​​of the two candidate virtual nodes it represents, and the new node is added to the initial set of effective virtual nodes, while the two candidate virtual nodes it represents are removed from the set of virtual nodes. In the remaining set of virtual nodes, the filtering process is repeated until no pair of adjacent virtual nodes can be found with a coupling strength greater than the coupling threshold. Then the iteration stops, and the final set of effective virtual nodes is obtained. The final set of physical measurement points is merged with the final set of effective virtual nodes to form an enhanced detection map.

3. The method for detecting high-voltage interlock faults in new energy vehicles based on a training platform as described in claim 2, characterized in that, The process involves defining perturbation matrices for vertex and edge faults on the enhanced detection map, calculating fault sensitivity using the adjoint network method, setting an optimization objective function, solving for the optimal frequency set and corresponding amplitude, and generating an analog voltage signal, including: At a given diagnostic angular frequency ω, the physical parameters in the edge set of the enhanced detection diagram are equivalent to a Π-type lumped parameter circuit. The series impedance and parallel admittance to ground of the edge are calculated, and the reciprocal of the series impedance of the edge is taken as the mutual admittance of the edge. For each vertex, its ground admittance is defined as the sum of the ground leakage conductance and capacitance of all converging branches at that node. On the enhanced detection graph, target fault modes are defined, including vertex faults and edge faults, and a corresponding fault perturbation matrix is ​​generated for each type of fault. For each failure mode, modeling is performed by modifying the node admittance matrix of the enhanced detection map, including the baseline node admittance matrices for both off-diagonal and diagonal elements. The single-port current injection point is defined as the excitation port, and the observation port is the set of all physical measurement points (PMPs). Under fault-free conditions, the complex voltage change is calculated using the adjoint network method based on the circuit equations and the baseline admittance matrix. The overall fault sensitivity is calculated based on the complex voltage change. An optimization objective function is set and solved by the simulated annealing algorithm to obtain the optimal frequency set and corresponding amplitude, which are then stored in the embedded controller of the training platform.

4. The method for detecting high-voltage interlock faults in new energy vehicles based on a training platform as described in claim 3, characterized in that, After the training platform is connected to the safe low-voltage power supply and the circuit self-test is completed, the operator selects a fault scenario through the HMI and drives the fault simulation relay group to operate, including: When the safety low-voltage power supply of the training platform is turned on, the control unit sends a DC low-voltage verification signal to the HVIL circuit and detects the end-received signal. If the terminal signal is received normally, it enters standby mode; if there is an abnormality, the process is terminated. The operator selects a preset fault scenario or a custom fault in the "New Energy Vehicle High Voltage Interlock Fault Database" interface of the HMI to generate layout instructions; The control unit receives the layout instructions and drives the fault simulation relay group corresponding to the target position to operate.

5. The method for detecting high-voltage interlock faults in new energy vehicles based on a training platform as described in claim 4, characterized in that, The process involves injecting an analog voltage signal into the excitation port of the cable system, synchronously sampling the response voltage of physical measurement points, solving a regularized optimization problem for each optimal frequency point to obtain an estimate of the total node voltage, calculating the equivalent series impedance perturbation on each side of the enhanced detection map, performing weighted fusion, and obtaining a comprehensive impedance perturbation spectrum, including: After the fault setting is completed, the embedded controller reads the stored optimal frequency set and corresponding amplitude, generates a discrete time series in the digital domain, and sends it to the digital-to-analog converter to generate an analog voltage signal. The analog voltage signal is then injected into the excitation port of the cable system. The system under test generates a voltage response under this excitation. All physical measurement points synchronously sample the response voltage of the system under test under the excitation signal. Windowed FFT is performed on the sampled data of each physical measurement point to extract complex voltage values ​​and construct a multi-frequency measurement vector. Based on the baseline feature library, the baseline admittance values ​​of each side at each optimal frequency point are calculated, and the node admittance matrix and node injection current vector at the optimal frequency point are constructed. For each optimal frequency point, based on the constructed nodal admittance matrix, injected current vector and measured voltage response, a regularized optimization problem is solved to obtain the voltage estimate of all nodes; Based on the voltage estimates of all nodes, the equivalent series impedance disturbance of each edge in the enhanced detection map is calculated using the complex impedance difference method and mapped to its corresponding spatial location to obtain the single-frequency disturbance curve. The perturbation curves at each frequency are weighted and fused to obtain a comprehensive impedance perturbation spectrum.

6. The method for detecting high-voltage interlock faults in new energy vehicles based on a training platform as described in claim 5, characterized in that, The peak detection of the real part curve of the integrated impedance perturbation spectrum is performed to obtain candidate fault locations, and the frequency domain impedance perturbation vector is extracted from it. This vector is then matched with a pre-built fault feature library to generate a diagnostic report, including: Plot the curve of the integrated impedance perturbation spectrum on the HMI, and perform peak detection on the curve of the integrated impedance perturbation spectrum to obtain the set of candidate fault locations. Extract the frequency domain impedance perturbation vector at each candidate location, match it with a pre-built fault feature library, and generate a diagnostic report, including fault type, fault location, quantization parameters, and a screenshot of the comprehensive impedance perturbation spectrum.

7. The method for detecting high-voltage interlock faults in new energy vehicles based on a training platform as described in claim 6, characterized in that, The system, based on diagnostic reports, uses an HMI-driven training platform to collect student test data, generate practical assessment reports, push personalized learning content to students, and provide teachers with a visual representation of the assessment data, including: According to the fault location in the diagnostic report, the HMI system drives the corresponding LED indicator on the training bench panel to flash and pops up a standard instrument verification task sheet for the trainees, requiring them to use a real multimeter to perform measurements at the designated test holes on the bench. The system automatically collects measured resistance values, instrument settings, and operation time through its built-in measurement interface; The collected data is intelligently compared with the diagnostic results and standard operating procedures, and a practical skills assessment report is generated for each trainee, including result verification score, operating procedure score and deduction items; Based on the practical skills assessment report, the system automatically analyzes the deduction items, matches and pushes personalized learning content to students from the teaching resource library, and aggregates the assessment data of all students for teachers based on the practical skills assessment report, and displays the class skill attainment rate, the list of difficult problems, and the students' performance ranking in a visual way.

8. A high-voltage interlock fault detection system for new energy vehicles based on a training platform, and based on the high-voltage interlock fault detection method for new energy vehicles based on a training platform as described in any one of claims 1 to 7, characterized in that, include: The self-test setting module is used to connect the training platform's safe low-voltage power supply and complete the circuit self-test. After the operator selects the fault scenario through the HMI, the fault simulation relay group will be driven to operate. A generation module is constructed to define the interlocking interfaces of high-voltage components as vertices and the wiring harnesses as edges, construct a topology graph, find the key points of the topology graph and calculate the electrical centrality of non-key points, and filter out the set of physical measurement points based on the key points and electrical centrality values. On the edges between physical measurement points, candidate virtual nodes are generated. The electrical coupling strength of adjacent candidate virtual nodes is iteratively merged to obtain the effective virtual nodes, which are then combined with the physical measurement points to form an enhanced detection map. A calculation module is set up to define the perturbation matrix of vertex and edge faults on the enhanced detection map. The fault sensitivity is calculated by the adjoint network method. The goal is to maximize the minimum comprehensive sensitivity among all preset fault modes, while satisfying the frequency range and power constraints. An optimization objective function is set, solved, and the optimal frequency set and corresponding amplitude are obtained and stored in the embedded controller of the training platform. The acquisition and fusion module is used to inject analog voltage signals into the excitation port of the cable system, synchronously sample the response voltage of the physical measurement points, solve the optimization problem of band graph Laplace regularization for each optimal frequency point, obtain the estimated value of the total node voltage, calculate the equivalent series impedance perturbation of each side on the enhanced detection map, perform weighted fusion, and obtain the comprehensive impedance perturbation map. The detection and matching module is used to perform peak detection on the real part curve of the integrated impedance disturbance spectrum, obtain candidate fault locations, extract frequency domain impedance disturbance vectors from them, and generate diagnostic reports by matching them with a pre-built fault feature library. The visualization module is used to collect students' test data based on diagnostic reports and HMI-driven training benches, generate practical assessment reports, push personalized learning content to students, and visualize assessment data for teachers.

Images