Method and system for testing split type safety link structure of high-voltage line mirror bracket

Abstract

The invention relates to the technical field of high-voltage line mirror bracket safety testing, and discloses a high-voltage line mirror bracket split type safety link structure testing method and system. According to the method, plugging and bending tests are carried out on a high-voltage line mirror bracket split type link structure through a mechanical arm, and interface pressure and structure deformation image data are collected in real time. And analyzing interface wear and structural fatigue damage characteristics by using a pre-trained wear detection model, and outputting a real-time wear degree evaluation result. And after the test period is finished, integrating the multi-modal data and the evaluation result to generate a three-dimensional visual detection report, positioning a wear concentration position, marking an abnormal wear point and quantifying the structural safety level. According to the invention, the problems of incomplete test and inaccurate evaluation in the prior art are solved, and the product safety and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of safety testing of high-voltage line mirror frames, and in particular to a method and system for testing a split-type safety link structure of a high-voltage line mirror frame. Background Art

[0002] Ultrafine ion glasses act on the eyes through ultrafine ion technology, improving the internal environment of the eyes and alleviating the symptoms of visual fatigue caused by excessive eye use. With the widespread application of ultrafine ion glasses in the medical field, the reliability of the split safety link structure of the high-voltage wire frame has become key. However, traditional testing methods often only focus on single-dimensional performance evaluation, such as only performing plug-in tests or only observing the appearance of the structure, which cannot fully reflect the comprehensive performance of the link structure during actual use. In addition, the existing technology has limited ability to identify potential risks such as interface wear and structural fatigue damage, making it difficult to detect and warn in the early stages, thereby increasing the risk of product failure during use. Therefore, it is particularly important to develop a testing method and system that can comprehensively and accurately evaluate the reliability of the split safety link structure of the high-voltage wire frame. Summary of the Invention

[0003] In view of the above-mentioned technical deficiencies, the purpose of the present invention is to provide a method and system for testing the split-type safety link structure of a high-voltage line mirror frame, so as to solve the problems of incomplete reliability testing and inaccurate evaluation of the split-type link structure of a high-voltage line mirror frame in the prior art.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for testing a split-type safety link structure of a high-voltage wire mirror frame, the method comprising: Step S100: performing a preset number of plugging and unplugging tests and bending tests on the split link structure of the high-voltage line mirror frame by a robotic arm, wherein the plugging and unplugging test controls the robotic arm to perform linear reciprocating motion along the axis of the interface, and the bending test controls the robotic arm to perform reciprocating bending motion within a preset angle range; Step S200: collecting pressure data of the interface during the plug-in test and deformation image data of the link structure during the bending test in real time, wherein the pressure data is obtained by a pressure sensor integrated at the end of the robotic arm, and the deformation image data is obtained by an image sensor; Step S300: Input the pressure data and deformation image data into the pre-trained wear detection model to analyze the interface wear characteristics caused by the plugging and unplugging process and the structural fatigue damage characteristics caused by the bending process, and output a real-time wear degree assessment result; Step S400: After completing the test cycle, the multimodal data and the wear degree assessment results within the test cycle are integrated to generate a test report, which locates the wear concentration location, marks the abnormal wear points and quantifies the structural safety level.

[0005] Preferably, in a possible embodiment of the first aspect, the split link structure includes a plug-in male connector and a female connector, wherein a spring pin built into the male connector forms an electrical connection with a conductive groove of the female connector; an elastic snap fastener is nested on the outside of the male connector, and its protrusion is mechanically interlocked with a limiting groove of the female connector; and an insulating sleeve covers the joint between the male connector and the female connector.

[0006] Preferably, in a possible implementation of the first aspect, the wear detection model includes an input layer, a dual-channel convolution layer, a multimodal fusion layer, and an output layer; The input layer normalizes the pressure data sequence and deformation image data respectively; The dual-channel convolution layer uses parallel-designed temporal and spatial convolution kernels. The temporal convolution kernel extracts pressure fluctuation features, while the spatial convolution kernel identifies image crack features. The multimodal fusion layer correlates wear correlations through an adaptive feature weighting module and a temporal-spatial dual-stream analysis module; The output layer outputs the wear level through a fully connected layer and a Softmax classifier.

[0007] Preferably, in a possible implementation of the first aspect, the adaptive feature weighting module performs the following operations: Calculate the pressure characteristic weight coefficient:

[0008] in is the energy entropy of pressure data, is the image structure entropy; Generate fused feature vector :

[0009] and are the pressure feature vector and image feature vector respectively.

[0010] Preferably, in a possible implementation of the first aspect, the time-domain-spatial-domain dual-stream analysis module includes: The time domain analysis flow uses the LSTM network to extract the time-dependent features of the pressure series:

[0011] The spatial analysis flow uses the ResNet-18 network to extract image spatial damage features:

[0012] Finally, the damage coefficient is output through the gated fusion mechanism:

[0013] in is the time step The fusion feature vector of For the LSTM network, For the ResNet-18 network, 、 are network parameters, is the Sigmoid function, for The transpose of .

[0014] Preferably, in a possible implementation of the first aspect, step S400 is specifically: After completing the preset number of test cycles, the pressure data from the plug-in test and the deformation image data from the bending test are integrated; Generate a 3D visual inspection report based on the wear assessment results, locate the concentrated wear area of ​​the interface through the thermal map, and mark the spatial distribution coordinates of the structural fatigue cracks; Combined with the historical safety database, the current wear level is matched with the remaining life threshold, and the structural integrity assessment conclusion including the quantitative safety factor and the location mark of the abnormal wear point is output.

[0015] Preferably, in a possible implementation manner of the first aspect, the security level quantification adopts a dynamic weight function:

[0016] in is the crack density factor, which is determined by multiplying the number of cracks per unit area by the average crack length. The pressure attenuation factor is the weighted sum of the peak attenuation rate of the pressure data and the fluctuation variance. is the spatial association weight coefficient, which is calculated by the following formula:

[0017] In the formula is the image gradient norm, is the differential entropy of pressure data, Calibrate parameters for the device.

[0018] In a second aspect, the present invention provides a high-voltage line mirror frame split safety link structure testing system, the system comprising: A test execution module, which uses a robotic arm to perform a preset number of plug-in and pull-out tests and bending tests on the split link structure of the high-voltage line mirror frame. The plug-in and pull-out test controls the robotic arm to perform linear reciprocating motion along the axis of the interface, and the bending test controls the robotic arm to perform reciprocating bending motion within a preset angle range. The data acquisition module is used to collect real-time pressure data of the interface during the plug-in test and deformation image data of the link structure during the bending test. The pressure data is obtained by the pressure sensor integrated at the end of the robotic arm, and the deformation image data is obtained by the image sensor. The wear analysis module is used to input pressure data and deformation image data into a pre-trained wear detection model to analyze the interface wear characteristics caused by the plugging and unplugging process and the structural fatigue damage characteristics caused by the bending process, and output real-time wear degree assessment results; The report generation module is used to generate a test report after completing the test cycle by integrating the multimodal data and wear assessment results within the test cycle. The test report locates the concentrated wear points, marks abnormal wear points and quantifies the structural safety level.

[0019] The beneficial effects of the present invention are as follows: the present invention uses a robotic arm to perform a preset number of plug-in tests and bending tests, combines pressure sensors and image sensors to collect data in real time, and uses a pre-trained wear detection model to perform in-depth analysis of multimodal data, thereby achieving accurate quantitative evaluation of the wear characteristics of the interface of the split safety link structure of the high-voltage line frame and the structural fatigue damage characteristics.

[0020] This method not only improves the comprehensiveness and accuracy of testing but also identifies potential risks at an early stage, providing strong support for product design optimization and quality control. Furthermore, by generating a 3D visual inspection report that intuitively displays concentrated wear locations and abnormal wear points, it provides clear guidance for repair and replacement, effectively improving product safety and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 A flow chart of a method for testing a split-type safety link structure of a high-voltage wire frame is provided for this application.

[0023] Figure 2 A structural diagram of a high-voltage line mirror frame split safety link structure test system is provided for this application.

[0024] Figure numerals: 1 - test execution module, 2 - data acquisition module, 3 - wear analysis module, 4 - report generation module. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0026] Example 1: Figure 1 As shown, the present invention provides a method for testing a split-type safety link structure of a high-voltage wire frame, comprising: Step S100: Perform a preset number of plug-in and pull-out tests and bending tests on the split link structure of the high-voltage line mirror frame through a robotic arm, wherein the plug-in and pull-out test controls the robotic arm to perform linear reciprocating motion along the axis of the interface, and the bending test controls the robotic arm to perform reciprocating bending motion within a preset angle range.

[0027] In this embodiment, the split link structure of the high-voltage line mirror frame includes a plug-in male connector and a female connector. The built-in spring pin of the male connector forms an electrical connection with the conductive groove of the female connector. The elastic clip is nested on the outside of the male connector, and its protrusion is mechanically interlocked with the limiting groove of the female connector. The insulating sleeve covers the joint between the male connector and the female connector.

[0028] The plug-in and pull-out test simulates the connection and separation operations in actual use through a robotic arm to ensure the electrical connection stability between the male spring pin and the female conductive slot. At the same time, it tests the interlocking reliability of the elastic snap protrusion and the limiting groove to avoid deformation of the spring pin or wear of the conductive slot due to repeated plugging and unplugging, thereby affecting the transmission performance of the ultra-fine ion glasses.

[0029] The male part of the split link structure is fixed at the end of the robotic arm, and a preset number of linear reciprocating motions are performed along the axis of the interface. The motion speed is controlled at 1 time per second, and the preset number of cycles is 50,000, covering the risk of interface wear in long-term use scenarios; the bending test is aimed at fatigue damage of the link structure under bending conditions. The robotic arm grasps the joint area covered by the insulating sleeve and performs reciprocating bending motion within a preset angle range. The angle range is set to The test simulates accidental bending or external impact when wearing glasses. It tests the mechanical interlocking strength of the elastic buckle and the retaining groove, as well as the bending resistance of the insulating sleeve, to prevent structural fatigue-induced fracture or electrical connection failure. During the test, the motion parameters of the robotic arm are adjusted in real time via a closed-loop control system to evaluate the stress distribution of the male spring pins during the plugging and unplugging process of the split-type connector and the deformation response of the female conductive slot during bending.

[0030] Step S200: Real-time acquisition of pressure data of the interface in the plug-in test and deformation image data of the link structure in the bending test, wherein the pressure data is acquired by a pressure sensor integrated at the end of the robotic arm, and the deformation image data is acquired by an image sensor.

[0031] In this embodiment, the pressure sensor adopts a six-dimensional force sensor array, which is arranged at the contact interface between the end effector of the robotic arm and the male fixture to monitor the axial positive pressure and radial shear force of the contact surface between the spring pin and the conductive slot during the plugging and unplugging process in real time. The sampling frequency is set to 1kHz.

[0032] The image sensor uses a high-speed industrial camera with a circularly polarized light source to capture the dynamic deformation sequence of the link structure during the bending test at a frame rate of 500Hz. The camera's optical axis is perpendicular to the joint surface of the male and female connectors. The sub-pixel edge detection algorithm is used to extract the displacement trajectory of the elastic buckle protrusion and the limiting groove in real time, and record the expansion morphology of microcracks on the surface of the insulating sleeve.

[0033] Step S300: Input the pressure data and deformation image data into the pre-trained wear detection model, analyze the interface wear characteristics caused by the plugging and unplugging process and the structural fatigue damage characteristics caused by the bending process, and output the real-time wear degree assessment results.

[0034] In this embodiment, the wear detection model is based on a deep neural network architecture and achieves accurate quantification of interface wear and structural damage through multimodal data fusion. The model input layer first normalizes the real-time pressure data sequence collected by the six-dimensional force sensor at the end of the robot arm and linearly maps it to In the interval, channel separation and grayscale normalization are performed synchronously on the deformed image captured by the image sensor to eliminate light interference.

[0035] The dual-channel convolution layer uses parallel-designed time-domain convolution kernels and spatial-domain convolution kernels. The time-domain convolution kernel acts on the pressure data stream and extracts the pressure fluctuation characteristics of the contact surface between the spring pin and the conductive slot during the plugging and unplugging process through three layers of dilated convolution (with dilation factors of 1, 2, and 4), identifying the periodic pressure attenuation caused by metal fatigue. The spatial convolution kernel uses a 5×5 deformable convolution kernel to process the displacement trajectory of the elastic buckle protrusion and the limiting groove in the deformation image, thereby enhancing the ability to extract the geometric features of microcracks on the surface of the insulating sleeve.

[0036] The multimodal fusion layer dynamically associates the importance weights of pressure and image features through the adaptive feature weighting module, specifically performing the following operations: Calculating the energy entropy of pressure data (FFT spectrum entropy value based on 1024-point sliding window) and image structure entropy (Extract edge gradient histogram entropy through Sobel operator) and generate pressure feature weight coefficient When the insulation sleeve has large cracks, Significantly increased, Reduce and enhance the weight of image features; when there is a sudden change in pressure during the plugging process, Increase the pressure feature. Fusion feature vector By weighted formula Generate, where is the 128-dimensional feature vector output by time domain convolution, It is the 256-dimensional feature tensor output by the spatial convolution.

[0037] The time-domain and space-domain dual-stream analysis module further correlates cross-modal damage features: the time-domain analysis stream is processed using a bidirectional LSTM network. The network structure consists of two hidden layers with 128 units, which extracts the time-dependent characteristics of the pressure sequence in the plug-in test. , capturing the pressure decay pattern corresponding to the increase in contact resistance of the spring pin; the spatial analysis flow uses the ResNet-18 network to extract the spatial damage characteristics of the bending test image , and its residual block outputs the plastic deformation area of ​​the positioning elastic buckle protrusion. is the time step The fusion feature vector of 、 are network parameters, is the Sigmoid function. Finally, the damage coefficient is calculated through the gated fusion mechanism , when the pressure mutation point coincides with the image crack area in time and space, The inner product value increases, Approaching 1 triggers a high-risk warning.

[0038] The model output layer includes a fully connected layer and a Softmax classifier. The fully connected layer compresses the fusion features to 64 dimensions and outputs five levels of wear through the Softmax classifier (0: no wear, 1: slight wear, 2: moderate wear, 3: severe wear, 4: critical failure).

[0039] The wear detection model is trained using multi-stage transfer learning and cross-modal feature alignment. The training data comes from a historical test database of high-voltage wire mirror frames, including 10,000 sets of pressure time series data from plugging and unplugging tests and 10,000 sets of image sequences from bending tests.

[0040] In the data preprocessing stage, the original signal of the pressure sensor is normalized using a sliding window:

[0041] in for The six-dimensional force sensor readings at the moment, and The mean and standard deviation within a 1024-point sliding window are used to eliminate the impact of device range differences. The deformed image data is corrected for shooting angle deviations using a perspective transformation matrix and scaled to a uniform 512×512 pixel resolution using a bicubic interpolation algorithm.

[0042] The model's transfer learning pre-training was conducted in two phases. The first phase involved training a dual-channel convolutional layer on a common industrial connector dataset. The time-domain convolution kernel employed three layers of convolution with increasing dilation rates (dilation factors d = 1, 2, and 4) to learn the ability to extract periodic features of pressure fluctuations. The loss function was:

[0043] in To reconstruct the feature vector, the spatial convolution kernel uses a deformable convolution module to adaptively capture the crack geometric features through offset learning:

[0044] Here This is the offset tensor for the deformable convolution, and the regularization term enhances the stability of deformable modeling. During this phase, the multimodal fusion layer parameters are frozen, focusing on optimizing the underlying feature extractor.

[0045] The second stage uses a proprietary dataset of high-voltage wireframes for domain adaptation fine-tuning. A gradient reversal layer is introduced to achieve cross-device feature alignment:

[0046] The discriminator Distinguish device source domain, generator For feature extraction network, the feature vector is forced to The distribution is consistent across device domains. At the same time, a feature decoupler is embedded at the end of the ResNet-18 network:

[0047] Adopting orthogonality constraint loss Separate equipment-related noise from intrinsic impairment signatures.

[0048] The historical security threshold library is constructed using hierarchical clustering technology. The validation set samples are extracted through the network to obtain 128-dimensional fusion features. Input OPTICS clustering algorithm to generate 5 security level feature clusters. The center vector of each cluster is As a benchmark threshold, and calculate the key statistics of the samples within the cluster: Pressure decay rate ; Crack growth rate ; During inference, the current feature is matched with the cluster center by cosine similarity:

[0049] Combined with the historical failure cycle distribution of matching clusters , calculate the remaining life prediction value .

[0050] After inputting the real-time pressure data and deformation image data into the pre-trained wear detection model, analysis and evaluation are performed according to the following process: First, the model's dual-channel processing architecture simultaneously analyzes multimodal data. The time-domain analysis channel receives real-time pressure sequences from the six-dimensional force sensor at the end of the robotic arm and extracts dynamic pressure characteristics from plug-in / out tests using three layers of dilated convolution kernels. The first layer identifies the impact peak at the moment the spring pin contacts the conductive slot, the second captures the pressure fluctuation attenuation trend caused by periodic plugging and unplugging, and the third layer detects abnormal pressure mutation points caused by metal fatigue. The spatial analysis channel processes deformation images captured by a high-speed camera, using a deformable convolution kernel to enhance adaptability to microscopic features. This not only locates the displacement trajectory offset of the elastic buckle protrusion during the bending test, but also quantifies the extension length and branching morphology of cracks on the insulating sleeve surface. Specifically, pixel-level edge analysis is performed on the stress concentration areas at the edges of the limiting grooves.

[0051] Subsequently, the multimodal fusion layer establishes a cross-domain damage correlation mechanism. The adaptive feature weighting module dynamically assigns weights based on the real-time ratio of image structural entropy to pressure energy entropy. When large-area cracks appear on the insulating sleeve, the contribution of image features to the fusion vector is significantly increased. When abnormal pressure mutations are detected during the plugging and unplugging process, the analysis priority of pressure features is strengthened. The time-domain and spatial-domain dual-stream analysis module further extracts time-dependent patterns in the pressure sequence through a bidirectional LSTM network, identifying the pressure decay period corresponding to the increase in spring pin contact resistance. Simultaneously, the spatial feature tensor output by the ResNet-18 network locates the plastic deformation region of the elastic buckle. The gated fusion mechanism aligns the two types of features in time and space. When the time point of the pressure mutation coincides with the spatial coordinates of the image crack, the high-risk damage coefficient threshold is triggered.

[0052] Finally, the model combines transfer learning with a historical threshold library to generate a quantitative assessment. The cross-device universal features established during the pre-training phase are used to eliminate the impact of range deviations on different test devices. Adversarial training using the gradient reversal layer decouples device-related noise from intrinsic damage characteristics. During online inference, the current fused feature vector is dynamically matched with the historical security database: cosine similarity is used to retrieve the closest security level feature cluster, and the statistical distribution parameters of the cluster's historical failure cycles (such as pressure decay rate) are combined to determine the optimal safety level. , crack growth rate ), outputs real-time assessment results including a five-level wear classification (0 to 4), and simultaneously generates the 3D coordinates of the damage location and the remaining life prediction. During the test, it continuously provides feedback on the risk of crack propagation in the insulation sleeve and early warning of contact failure in the conductive slot.

[0053] Step S400: After completing the test cycle, the multimodal data and the wear degree assessment results within the test cycle are integrated to generate a test report, which locates the wear concentration location, marks the abnormal wear points and quantifies the structural safety level.

[0054] In this embodiment, based on the real-time evaluation results output by the wear detection model, the system extracts the pressure data of the plug-in test and the deformation image data of the bending test within the test cycle.

[0055] The statistical characteristics of pressure data include pressure peak decay rate and fluctuation variance, which are used to calculate the pressure decay factor The deformation image is extracted through the sub-pixel edge detection algorithm to extract the number of cracks per unit area and the average crack length, and the crack density factor is generated. .

[0056] Based on this data set, the system constructs a 3D visual inspection report and uses the OpenGL rendering engine to map the damage characteristics of the male-female joint area to a 3D coordinate system: the interface wear concentration area is located through the thermal layer, and its color gradient is driven by the local wear level value. The red highlight area corresponds to the conductive groove contact surface and elastic buckle protrusion with a wear level greater than or equal to 3; at the same time, the distribution of structural fatigue cracks is marked with a spatial annotation layer, and the coordinates of the crack start and end points (such as the edge coordinates of the limit groove) are accurately calibrated on the 3D model surface. ) and the center coordinates of the micro crack clusters on the surface of the insulation sleeve.

[0057] The security level quantification module executes a dynamic weight function The spatial association weight coefficient is Determined by real-time calculation: Where is the image gradient norm, and the Sobel operator is used to perform convolution operation on the current deformed image to extract the sum of squared edge intensities; of is the differential entropy of pressure data, which is calculated from the entropy difference of adjacent sampling points of the six-dimensional force sensor sequence. The equipment calibration parameter Matching is obtained from the historical security database.

[0058] In this embodiment, the historical safety database is constructed by accumulating historical test data, and contains more than 50,000 test records of the split link structure of the high-voltage wire frame. Each record stores multimodal data features (such as pressure attenuation factor , crack density factor ), wear level labels and corresponding actual failure cycle data. The database adopts a hierarchical index structure based on security level Associating 3D damage coordinates, material fatigue parameters and remaining life distribution model with primary key , supports real-time feature vector matching via cosine similarity and historical cluster centers .

[0059] When a large area crack appears on the insulation sleeve, the image gradient norm increases significantly. The contribution of the crack density factor is close to 1; when the plug-in test detects abnormal pressure attenuation, the differential entropy increases, resulting in Decrease, increase the weight of the pressure attenuation factor.

[0060] The system combines the failure threshold library in the historical safety database (including the corresponding Threshold interval), the currently calculated The value is mapped to the five security levels (Level A: , Grade B: , Grade C: , D-level: , E-level: ), and associated with the historical remaining life distribution of that grade Predict the remaining service life. The final test report includes a quantitative safety factor The system can provide a “pass / fail” safety certification result and repair suggestions for key failure locations based on the wear concentration area coordinates, abnormal wear point location marks, and structural integrity assessment conclusions.

[0061] Embodiment 2: The present invention provides a high-voltage line mirror frame split-type safety link structure testing system, comprising: Test execution module 1 performs a preset number of plug-in and pull-out tests and bending tests on the split link structure of the high-voltage line mirror frame through a robotic arm. The plug-in and pull-out test controls the robotic arm to perform linear reciprocating motion along the axis of the interface, and the bending test controls the robotic arm to perform reciprocating bending motion within a preset angle range.

[0062] Data acquisition module 2 collects in real time the pressure data of the interface in the plug-in test and the deformation image data of the link structure in the bending test. The pressure data is obtained by the pressure sensor integrated at the end of the robotic arm, and the deformation image data is obtained by the image sensor.

[0063] Wear analysis module 3 inputs pressure data and deformation image data into the pre-trained wear detection model, analyzes the interface wear characteristics caused by the plugging and unplugging process and the structural fatigue damage characteristics caused by the bending process, and outputs real-time wear degree assessment results.

[0064] After completing the test cycle, the report generation module 4 integrates the multimodal data and wear degree assessment results within the test cycle to generate a test report. The test report locates the concentrated wear location and quantifies the structural safety level, while marking abnormal wear points.

[0065] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for testing a split safety link structure of a high-voltage wire frame, characterized in that: The method comprises: Step S100: performing a preset number of plugging and unplugging tests and bending tests on the split link structure of the high-voltage line mirror frame by a robotic arm, wherein the plugging and unplugging test controls the robotic arm to perform linear reciprocating motion along the axis of the interface, and the bending test controls the robotic arm to perform reciprocating bending motion within a preset angle range; Step S200: collecting pressure data of the interface during the plug-in test and deformation image data of the link structure during the bending test in real time, wherein the pressure data is obtained by a pressure sensor integrated at the end of the robotic arm, and the deformation image data is obtained by an image sensor; Step S300: Input the pressure data and deformation image data into the pre-trained wear detection model to analyze the interface wear characteristics caused by the plugging and unplugging process and the structural fatigue damage characteristics caused by the bending process, and output a real-time wear degree assessment result; Step S400: After completing the test cycle, the multimodal data and the wear degree assessment results within the test cycle are integrated to generate a test report, which locates the wear concentration location, marks the abnormal wear points and quantifies the structural safety level.

2. A method for testing a split-type safety link structure of a high-voltage wire frame as claimed in claim 1, characterized in that: The split link structure includes a plug-in male connector and a female connector, wherein the spring pin built into the male connector forms an electrical connection with the conductive groove of the female connector; the elastic clip is nested on the outside of the male connector, and its protrusion is mechanically interlocked with the limiting groove of the female connector; and the insulating sleeve covers the joint between the male connector and the female connector.

3. A method for testing a split-type safety link structure of a high-voltage wire frame as claimed in claim 1, characterized in that: The wear detection model includes an input layer, a dual-channel convolution layer, a multimodal fusion layer and an output layer; The input layer normalizes the pressure data sequence and deformation image data respectively; The dual-channel convolution layer uses parallel-designed temporal and spatial convolution kernels. The temporal convolution kernel extracts pressure fluctuation features, while the spatial convolution kernel identifies image crack features. The multimodal fusion layer correlates wear correlations through an adaptive feature weighting module and a temporal-spatial dual-stream analysis module; The output layer outputs the wear level through a fully connected layer and a Softmax classifier.

4. A method for testing a split-type safety link structure of a high-voltage wire frame as claimed in claim 3, characterized in that: The adaptive feature weighting module performs the following operations: Calculate the pressure characteristic weight coefficient: in is the energy entropy of pressure data, is the image structure entropy; Generate fused feature vector : and are the pressure feature vector and image feature vector respectively.

5. A method for testing a split-type safety link structure of a high-voltage wire mirror frame as claimed in claim 4, characterized in that: The time domain-spatial domain dual flow analysis module includes: The time domain analysis flow uses the LSTM network to extract the time-dependent features of the pressure series: The spatial analysis flow uses the ResNet-18 network to extract image spatial damage features: Finally, the damage coefficient is output through the gated fusion mechanism: in is the time step The fusion feature vector of For the LSTM network, For the ResNet-18 network, 、 are network parameters, is the Sigmoid function, for The transpose of .

6. A method for testing a split-type safety link structure of a high-voltage wire mirror frame as claimed in claim 1, characterized in that: The step S400 is specifically as follows: After completing the preset number of test cycles, the pressure data from the plug-in test and the deformation image data from the bending test are integrated; Generate a 3D visual inspection report based on the wear assessment results, locate the concentrated wear area of ​​the interface through the thermal map, and mark the spatial distribution coordinates of the structural fatigue cracks; Combined with the historical safety database, the current wear level is matched with the remaining life threshold, and the structural integrity assessment conclusion including the quantitative safety factor and the location mark of the abnormal wear point is output.

7. A method for testing a split-type safety link structure of a high-voltage wire frame as claimed in claim 6, characterized in that: The security level quantification adopts a dynamic weight function: in is the crack density factor, which is determined by multiplying the number of cracks per unit area by the average crack length. The pressure attenuation factor is the weighted sum of the peak attenuation rate of the pressure data and the fluctuation variance. is the spatial association weight coefficient, which is calculated by the following formula: In the formula is the image gradient norm, is the differential entropy of pressure data, Calibrate parameters for the device.

8. A high-voltage line mirror frame split safety link structure testing system, characterized in that: The system comprises: A test execution module, which uses a robotic arm to perform a preset number of plug-in and pull-out tests and bending tests on the split link structure of the high-voltage line mirror frame. The plug-in and pull-out test controls the robotic arm to perform linear reciprocating motion along the axis of the interface, and the bending test controls the robotic arm to perform reciprocating bending motion within a preset angle range. The data acquisition module is used to collect real-time pressure data of the interface during the plug-in test and deformation image data of the link structure during the bending test. The pressure data is obtained by the pressure sensor integrated at the end of the robotic arm, and the deformation image data is obtained by the image sensor. The wear analysis module is used to input pressure data and deformation image data into a pre-trained wear detection model to analyze the interface wear characteristics caused by the plugging and unplugging process and the structural fatigue damage characteristics caused by the bending process, and output real-time wear degree assessment results; The report generation module is used to generate a test report after completing the test cycle by integrating the multimodal data and wear assessment results within the test cycle. The test report locates the concentrated wear points, marks abnormal wear points and quantifies the structural safety level.

Images