A method and system for testing performance of a vehicle charger

Abstract

The application discloses a kind of vehicle charger performance test method and system, belong to electronic product manufacturing and automation test field, it includes the initial parameter of acquisition charger interface to generate alignment drive instruction;Based on drive instruction, it is connected and double check is carried out to obtain connection result;After successful connection, start multi-channel detection to generate performance data;Finally, intelligent determination is carried out based on performance data and generates different treatment instruction.The application adopts the mode that virtual simulation and machine learning model are combined, through the construction from physical process risk to final quality result Causal Analysis and feedback optimization closed loop, can realize the full-process automation and intelligent control of test process, significantly improve the pass rate, reliability and intelligent level of vehicle charger production test.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of electronic product manufacturing and automated testing, and in particular to a vehicle charger performance testing method and system. BACKGROUND

[0002] As a core component of new energy vehicles, the performance and reliability of vehicle chargers are directly related to the safety of the entire vehicle. In the production and manufacturing process, performance testing is a necessary process to ensure product quality and compliance with factory standards. Traditional performance testing is usually carried out after product assembly is completed on a dedicated test station by measuring key electrical parameters of the charger to verify whether its basic functions meet design requirements.

[0003] With the increase in production, automated equipment has been applied to assembly and testing to improve efficiency. However, existing automation solutions often treat assembly and testing as independent units, resulting in mismatched production cycles and additional logistics transfers. In addition, product alignment and electrical connections during testing often rely on general-purpose fixtures, which have insufficient alignment accuracy and unreliable contact, thereby affecting the accuracy and consistency of test results. Existing technologies also lack a mechanism for automatically linking test results with subsequent labeling, sorting, and other processes. SUMMARY

[0004] To solve the above problems, the present application provides a vehicle charger performance testing method and system, which uses a combination of virtual simulation and machine learning models to build a causal analysis and feedback optimization closed loop from physical process risks to final quality results, enabling full-flow automated and intelligent control of the testing process and significantly improving the straight-through rate, reliability, and intelligence level of vehicle charger production testing.

[0005] The above objectives can be achieved through the following solutions:

[0006] A vehicle charger performance testing method includes collecting spatial pose parameters and electrical characteristic parameters of a vehicle charger interface, calculating and generating spatial fine-tuning driving instructions; performing docking based on the spatial fine-tuning driving instructions and performing consistency verification with a preset docking verification benchmark to obtain a verification result of successful or failed docking; if the verification result feedbacks successful docking, starting a multi-channel performance detection process to generate full-flow performance data; and based on the full-flow performance data and a preset performance judgment threshold, generating and executing differentiated treatment instructions.

[0007] Optionally, the operation and generation of the spatial fine-tuning driving instruction comprises: acquiring real-time three-dimensional coordinates of the vehicle-mounted charger interface, and performing vector operation to obtain a three-dimensional space deviation vector; synchronously collecting contact impedance and signal response state of the vehicle-mounted charger interface to generate electrical characteristic parameters; based on the three-dimensional space deviation vector and the electrical characteristic parameters, calculating X / Y / Z three-axis fine-tuning amount to generate a spatial fine-tuning driving instruction.

[0008] Optionally, the calculation of the X / Y / Z three-axis fine-tuning amount to generate a spatial fine-tuning driving instruction comprises: kinematically solving the three-dimensional space deviation vector to generate a reference driving scheme; virtually simulating the reference driving scheme to obtain predictive collision and stress distribution parameters; based on the predictive collision and stress distribution parameters and the electrical characteristic parameters, optimizing and correcting the reference driving scheme to calculate the X / Y / Z three-axis fine-tuning amount as a correction value, and integrating to generate a spatial fine-tuning driving instruction.

[0009] Optionally, the consistency check with the preset docking verification reference comprises: extracting geometric contour features of the vehicle-mounted charger interface, and performing similarity calculation with the asymmetric structure template in the docking verification reference to generate a structure matching degree parameter; real-time detecting contact impedance and signal response delay of the vehicle-mounted charger interface, and comparing with the electrical compliance threshold range in the docking verification reference to generate an electrical compliance parameter; logically AND judging the structure matching degree parameter and the electrical compliance parameter to generate a check result of docking success or failure.

[0010] Optionally, the generation of the structure matching degree parameter comprises: obtaining historical interface image data and historical standard feature point coordinates to obtain a historical interface image data set; taking interface image data as input and standard feature point coordinates as output, establishing and training a neural network model using the historical interface image data set to obtain an interface geometric feature point recognition model; capturing real-time interface image data, inputting the real-time interface image data into the interface geometric feature point recognition model to obtain key geometric feature point coordinates; performing spatial position comparison and topological relationship analysis on the key geometric feature point coordinates and the asymmetric structure template to calculate and generate a structure matching degree parameter.

[0011] Optionally, the generation of the full-process performance data comprises: applying a simulated load to a power bus of the vehicle-mounted charger and collecting voltage and current responses to obtain basic electrical performance parameters; sending a protocol test instruction to a data bus of the vehicle-mounted charger and capturing the returned signal waveform to obtain protocol interaction data; aggregating and structuring the basic electrical performance parameters and the protocol interaction data to generate full-process performance data.

[0012] Optionally, the generating and executing the differential treatment instruction comprises: performing hierarchical comparison between the full-process performance data and the performance judgment threshold to generate a multi-dimensional performance judgment result; analyzing the multi-dimensional performance judgment result to obtain a treatment level including qualified, re-inspection required or unqualified information; based on the treatment level, constructing an identification instruction for qualified products, a re-inspection prompt instruction for re-inspection required products and an isolation treatment instruction for unqualified products respectively to generate the differential treatment instruction.

[0013] Optionally, the method further comprises: performing unsupervised clustering analysis on the data of the unqualified products to extract common failure characteristics and generate a failure mode feature vector; tracing spatial attitude parameters and electrical characteristic parameters associated with the failure mode feature vector to construct and generate a parameter-failure mode association rule set.

[0014] Optionally, the method further comprises: inputting the real-time collected spatial attitude parameters and electrical characteristic parameters into the parameter-failure mode association rule set for inference to generate a predictive failure risk level; performing fusion calculation on the predictive failure risk level and the predictive collision and stress distribution parameters to generate a feedforward adjustment parameter; and using the feedforward adjustment parameter to dynamically optimize the docking verification benchmark and the performance judgment threshold.

[0015] Based on the same inventive concept, the application also provides a vehicle-mounted charger performance test system, which comprises: an initial parameter sensing and instruction generating module for collecting spatial attitude parameters and electrical characteristic parameters of a vehicle-mounted charger interface, calculating and generating spatial fine-tuning driving instructions; a docking and verification module for performing docking based on the spatial fine-tuning driving instructions and performing consistency verification with a preset docking verification benchmark to obtain a docking success or failure verification result; a multi-channel performance detection module for starting a multi-channel performance detection process to generate full-process performance data if the verification result feedbacks docking success; and a data judgment and differential treatment module for performing judgment based on the full-process performance data and a preset performance judgment threshold to generate and execute differential treatment instructions.

[0016] Compared with the prior art, the application has the following advantages:

[0017] 1. The application seamlessly integrates the automatic alignment of the charger, connection verification, multi-channel performance detection and result-based identification or isolation treatment in a continuous and automated process. This integrated design fundamentally eliminates the material transfer, waiting and beat mismatch problems caused by process separation in traditional solutions, greatly shortens the production cycle and improves the overall manufacturing efficiency.

[0018] 2、The application constructs the association rules of initial parameters and failure modes by deeply mining historical unqualified product data, and based on this, the verification benchmark and determination threshold of subsequent tests are dynamically adjusted in real time and prospectively. This closed-loop self-optimization capability enables the test system to learn from experience, automatically identify and suppress the risk sources leading to high-frequency failure, and realizes the continuous improvement of product quality and the steady improvement of production yield.

[0019] Other features and advantages of the present application will be set forth in the following description, and in part will be apparent from the description, or can be learned by practice of the application. The objects and other advantages of the present application will be realized and attained by the structure particularly pointed out in the written description and claims thereof as well as the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application or in the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0021] Figure 1 is a flowchart of a vehicle charger performance test method according to an embodiment of the present application.

[0022] Figure 2 is a two-dimensional contour map of the mapping relationship between the adaptive gain coefficient and the electrical initial parameters according to an embodiment of the present application.

[0023] Figure 3 is a stress and collision risk coupling analysis diagram in virtual simulation according to an embodiment of the present application.

[0024] Figure 4 is a hierarchical clustering heat map of failure modes and initial parameters according to an embodiment of the present application.

[0025] Figure 5 is a structure diagram of a vehicle charger performance test according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0027] Referring toFigure 1 One embodiment of the present application proposes a vehicle charger performance test method, which adopts a combination of virtual simulation and machine learning model, builds a causal analysis and feedback optimization closed loop from physical process risk to final quality result, can realize full-process automation and intelligent control of the test process, and significantly improves the pass rate, reliability and intelligent level of vehicle charger production test.

[0028] The method of the embodiment specifically includes:

[0029] Collecting the spatial pose parameters and electrical characteristic parameters of the vehicle charger interface, and operating and generating spatial fine-tuning driving instructions;

[0030] Based on the spatial fine-tuning driving instructions, performing docking and consistency checking with the preset docking verification benchmark, obtaining the checking result of successful or failed docking;

[0031] If the checking result feedbacks successful docking, starting a multi-channel performance detection process, and generating full-process performance data;

[0032] Based on the full-process performance data and the preset performance judgment threshold, making a judgment, generating and executing differentiated treatment instructions.

[0033] The combination of virtual simulation and machine learning model can realize full-process automation and intelligent control of the test process by building a causal analysis and feedback optimization closed loop from physical process risk to final quality result, significantly improving the pass rate, reliability and intelligent level of vehicle charger production test.

[0034] Optionally, the operation and generation of spatial fine-tuning driving instructions include:

[0035] Obtaining real-time three-dimensional coordinates of the vehicle charger interface, and performing vector operation to obtain a three-dimensional space deviation vector;

[0036] Specifically, in the step of obtaining real-time three-dimensional coordinates of the vehicle charger interface and performing vector operation to obtain a three-dimensional space deviation vector, one execution unit can call a high-resolution 3D vision sensor to capture the current pose of the interface. The sensor outputs a set of real-time pose data containing X, Y, Z coordinates and rotation components. A central processor performs vector subtraction operation on the real-time pose data and a target pose data representing the ideal docking position preset in the database, thereby accurately calculating the deviation between the two, as shown in the formula:

[0037] ,

[0038] Among them, is the final obtained three-dimensional spatial deviation vector, which represents the direction and distance between the current position and the target position; is the preset target pose vector; is the current pose vector collected by the 3D vision sensor in real time. This deviation vector provides accurate error input for subsequent closed-loop control.

[0039] Synchronously collect the contact impedance and signal response state of the vehicle-mounted charger interface to generate electrical characteristic parameters;

[0040] Specifically, in the step of synchronously collecting the contact impedance and signal response state of the vehicle-mounted charger interface to generate electrical characteristic parameters, a test control module will precisely measure the contact impedance between the terminals of the charger interface using the four-wire method when the probe of the alignment mechanism initially contacts the charger interface. At the same time, the module will send a weak electrical test pulse and measure the response time or waveform setup time of the signal loop. These two measured values, i.e., the real-time measured value of the contact impedance and the real-time measured value of the signal response state, are combined into a data pair to generate electrical characteristic parameters. This electrical characteristic parameter reflects the potential quality of the electrical connection to be established.

[0041] Based on the three-dimensional spatial deviation vector and the electrical characteristic parameters, calculate the X / Y / Z three-axis fine-tuning amount to generate a spatial fine-tuning driving instruction.

[0042] Specifically, in the step of calculating the X / Y / Z three-axis fine-tuning amount based on the three-dimensional spatial deviation vector and the electrical characteristic parameters to generate a spatial fine-tuning driving instruction, an algorithm processing unit uses an adaptive gain control strategy. This strategy first uses the electrical characteristic parameters generated in the previous step to calculate a normalized electrical quality factor, as shown in the formula:

[0043] ,

[0044] wherein, is the electrical quality factor; and are preset weight coefficients; and are real-time measured values; and are allowed maximum threshold values. Subsequently, the algorithm processing unit uses the electrical quality factor to dynamically adjust the proportional gain of a PID controller to generate an adaptive gain coefficient . Finally, combined with the adaptive gain coefficient and the three-dimensional spatial deviation vector, the final spatial fine-tuning driving instruction containing the X, Y, and Z three-axis fine-tuning amounts is calculated, as shown in the formula:

[0045] ,

[0046] wherein, is the final generated spatial fine-tuning driving instruction, as Figure 2 shown in the contour map, intuitively demonstrates how the final adaptive gain coefficient is nonlinearly and multidimensionally dynamically adjusted according to the real-time combined value of the two core electrical parameters, contact impedance and signal response delay, with the high gain region corresponding to the parameter space with better electrical connection quality.

[0047] Exemplarily, when a charger is placed on the test station, a 3D vision sensor captures its interface coordinates as (10.1, 5.2, 20.0), while the preset target docking coordinates are (10.0, 5.0, 20.0). A processor accordingly calculates a three-dimensional spatial deviation vector as (-0.1, -0.2, 0.0). At the same time, a test module measures a higher contact impedance of the interface, resulting in a calculated electrical quality factor of only 0.4. Since the electrical quality factor is lower than the regular level, an algorithm processing unit will accordingly lower the adaptive gain coefficient , so that the physical movement represented by the finally generated spatial fine-tuning driving instruction is slower and gentler. This strategy ensures that the alignment mechanism can push the interface in a cautious manner, so as to overcome the higher initial impedance through more stable contact and finally establish a reliable electrical connection while avoiding damage.

[0048] Optionally, the calculation of the X / Y / Z three-axis fine-tuning amount and the generation of the spatial fine-tuning driving instruction comprise:

[0049] kinematic solving of the three-dimensional spatial deviation vector to generate a reference driving scheme;

[0050] Specifically, in the step of kinematic solving of the three-dimensional spatial deviation vector to generate a reference driving scheme, an inverse kinematics algorithm is executed by a motion planning processor. The purpose of this algorithm is to analyze an end pose deviation described in the Cartesian coordinate system, i.e., the aforementioned three-dimensional spatial deviation vector, into specific angles or distances that each joint of the driving multi-axis alignment mechanism needs to rotate or move. Its mathematical relationship can be abstractly represented as the formula:

[0051] ,

[0052] wherein, is the generated reference driving scheme, which usually takes the form of a set of time series data containing target positions, velocities and accelerations of each joint; represents the inverse kinematics solving function; is the three-dimensional spatial deviation vector. This reference driving scheme represents the theoretically most efficient alignment path without considering any physical constraints or risks.

[0053] virtually simulating the reference driving scheme to obtain predictive collision and stress distribution parameters;

[0054] Specifically, in the step of virtually simulating the reference driving scheme to obtain predictive collision and stress distribution parameters, a simulation engine executes the reference driving scheme in a pre-constructed digital twin model that is accurately mapped to the physical test fixture. The engine utilizes physical simulation techniques such as Finite Element Analysis (FEA) to simulate the interaction of components during the execution of the driving scheme. The output of this simulation process can be represented by the formula:

[0055]

[0056] wherein, is the obtained predictive collision and stress distribution parameters, which is a set containing multiple risk indicators, such as is the predicted collision probability, is the predicted maximum stress value that will be generated on the charger shell or pins; represents the simulation function; is the input reference driving scheme; is the digital twin model of the fixture and the product to be tested, as shown in the figure, which couples and visualizes the predicted stress size and the predicted collision probability in the form of a three-dimensional surface, enabling managers to assess both physical stress and geometric interference in a single view. Figure 3

[0057] According to the predictive collision and stress distribution parameters and the electrical characteristic parameters, the reference driving scheme is optimized and corrected, and the X / Y / Z three-axis fine-tuning amount is calculated as a correction value, and a spatial fine-tuning driving instruction is generated.

[0058] Specifically, in the step of optimizing and correcting the reference driving scheme according to the predictive collision and stress distribution parameters and the electrical characteristic parameters, an optimization decision unit executes a multi-objective optimization algorithm. The algorithm aims to minimize the positioning time while ensuring that the predicted collision probability and the maximum stress do not exceed the preset safety threshold, and will refer to the electrical characteristic parameters to adjust the strategy of the final contact stage. For example, when the predicted maximum stress value is high, the algorithm will reduce the end speed in the reference driving scheme; when the generated electrical characteristic parameters show that the contact quality is poor, the algorithm may add a low-speed, constant-force "nudge" action at the end of the driving scheme. The output of this optimization and correction process is the final spatial fine-tuning driving instruction, which is a driving sequence that has been optimized for safety and quality and can be directly executed. ​​​

[0059] Exemplarily, assuming that the initial position of a to-be-tested charger results in a large three-dimensional space deviation vector, a motion planning processor calculates a reference driving scheme as a high-speed, direct linear movement. However, a simulation engine finds in the virtual simulation of the scheme that the end point of the high-speed movement path will cause a corner of the clamp to generate stress concentration exceeding a safety threshold with the charger shell. At this time, an optimization decision unit will intervene, and according to the high stress risk, automatically optimize and correct the reference driving scheme, and adjust the original high-speed linear movement to an “L”-shaped composite path of fast translation first and then low-speed vertical pressing. The finally generated space fine-tuning driving instruction successfully avoids the manufacturing risk that may cause damage to the appearance of the product while ensuring efficient alignment.

[0060] Optionally, the consistency checking with the preset docking verification reference comprises:

[0061] extracting the geometric contour features of the vehicle-mounted charger interface, and performing similarity calculation with the asymmetric structure template in the docking verification reference to generate a structure matching degree parameter;

[0062] Specifically, in the step of extracting the geometric contour features of the vehicle-mounted charger interface and performing similarity calculation to generate the structure matching degree parameter, an image processing unit first calls a high-resolution industrial camera to capture a real-time image of the interface. Then, a set of key geometric feature point vectors describing the contour of the current interface are extracted through an interface geometric feature point recognition model. The image processing unit then calculates the similarity between the real-time feature vector and the standard feature vector corresponding to the asymmetric structure template preset in the docking verification reference by using a cosine similarity algorithm, and the calculation process is as shown in the formula:

[0063]

[0064] wherein, is the finally generated structure matching degree parameter, and the value range is [-1, 1], and the closer to 1 indicates the higher the geometric matching degree; is the key geometric feature point vector extracted from the real-time image; is the standard feature vector stored in the docking verification reference; represents the Euclidean norm of the calculation vector.

[0065] real-time detection of the contact impedance and signal response delay of the vehicle-mounted charger interface, and comparison with the electrical compliance threshold range in the docking verification reference to generate an electrical compliance parameter;

[0066] ​Specifically, in the step of detecting the contact impedance and signal response delay in real time to generate the electrical compliance parameter, an electrical detection unit is responsible for quantifying the electrical quality of the connection. The unit compares the real-time measured contact impedance value and signal response delay with the electrical compliance threshold range defined in the docking verification benchmark. In order to obtain a normalized parameter, a scoring function as shown in the formula can be used:

[0067] ,

[0068] wherein, is the generated electrical compliance parameter; and are preset weights; a step function outputs 1 when the measured value is within the threshold range, and outputs a value between 0 and 1 when it exceeds the threshold range.

[0069] Logical AND is performed on the structure matching degree parameter and the electrical compliance parameter to generate a docking success or failure verification result.

[0070] Specifically, in the step of performing logical AND on the structure matching degree parameter and the electrical compliance parameter to generate a docking success or failure verification result, a decision logic unit performs a Boolean logic judgment. The unit compares the structure matching degree parameter and the electrical compliance parameter generated in the previous two steps with their respective judgment thresholds, and only when both conditions are met at the same time is the docking considered successful. The judgment logic is as shown in the formula:

[0071] ,

[0072] wherein, is the final generated docking success or failure verification result, which is a Boolean value; and are the parameters generated in the previous steps; and are the preset judgment thresholds for structure matching degree and electrical compliance, respectively.

[0073] Exemplarily, when one charger interface is operated by the alignment mechanism, an image processing unit captures its image and extracts a feature vector, and a structural matching degree parameter calculated by calculation is 0.99. At the same time, an electrical detection unit measures its contact impedance as 30 mΩ and signal response delay as 8 ns, and an electrical compliance parameter calculated by calculation is 0.97. Assuming that the preset structural matching degree judgment threshold is 0.95 and the electrical compliance judgment threshold is 0.90. A decision logic unit makes a judgment according to the formula, since 0.99≥0.95 and 0.97≥0.90, both conditions are established, and therefore the final verification result is “successful docking”, and the subsequent performance test process will be triggered.

[0074] Optionally, the generating the structural matching degree parameter comprises:

[0075] Obtaining historical interface image data and historical standard feature point coordinates to obtain a historical interface image data set;

[0076] Specifically, in the step of obtaining historical interface image data and historical standard feature point coordinates to obtain a historical interface image data set, data collection is performed in a controlled lighting environment. The collection objects are a plurality of standard charger interface samples confirmed by metrology and having perfect physical dimensions. High-resolution industrial cameras are used to take pictures of these samples from different angles and distances to form an original data set containing thousands or even tens of thousands of images. Subsequently, engineering technicians use professional image annotation software to accurately annotate the pixel coordinates of the preset key geometric feature points on each historical interface image according to the computer-aided design drawings of the product. Each image and its corresponding coordinate annotation together form a data pair, and the collection of all data pairs finally constitutes the historical interface image data set for model training.

[0077] Using the historical interface image data set, a neural network model is established and trained to obtain an interface geometric feature point recognition model, with interface image data as input and standard feature point coordinates as output;

[0078] Specifically, in the step of establishing and training the neural network model to obtain the interface geometric feature point recognition model, a convolutional neural network (CNN) architecture such as U-Net or its variants is adopted because it performs well in image segmentation and key point positioning tasks. The training process takes the image data in the historical interface image data set as the input of the model and the corresponding standard feature point coordinates as the expected output of the model. Through a back propagation algorithm, the weight parameters of the network are iteratively optimized to minimize the mean square error loss function between the model predicted coordinates and the real labeled coordinates. The loss function can be represented by the formula:

[0079] ,

[0080] wherein, is the mean square error loss value; is the total number of samples in the training batch; is the key geometric feature point coordinate vector predicted by the model for the i-th sample image; is the real standard feature point coordinate vector corresponding to the i-th sample. When the loss value converges to a sufficiently small preset value, the training is completed, and the solidified network weight parameters constitute the final interface geometric feature point recognition model.

[0081] capturing real-time interface image data, inputting the real-time interface image data into the interface geometric feature point recognition model, and obtaining key geometric feature point coordinates;

[0082] Specifically, in the step of capturing real-time interface image data and inputting the model to obtain key geometric feature point coordinates, when a charger to be tested enters the test station, the industrial camera deployed at the station is triggered to capture one or more frames of real-time images of the charger interface. After preprocessing, these images are input into the interface geometric feature point recognition model trained in the foregoing step as a tensor. The model performs a forward propagation, and the output layer is directly decoded as a set of coordinate values of real-time key geometric feature points. This process is fully automated and usually takes milliseconds.

[0083] spatial position comparison and topological relationship analysis of the key geometric feature point coordinates and the asymmetric structure template, and calculating and generating a structure matching degree parameter.

[0084] Specifically, in the step of comparing and analyzing the key geometric feature point coordinates and the asymmetric structure template to calculate and generate a structure matching degree parameter, an analysis unit compares the real-time key geometric feature point coordinates output by the model in the previous step with the asymmetric structure template stored in the docking verification reference. To comprehensively evaluate the similarity of the overall structure rather than the error of a single point, a cosine similarity algorithm can be used. By flattening the two sets of coordinate points into high-dimensional vectors, the cosine value of the included angle is calculated, and a quantitative structure matching degree parameter ranging from -1 to 1 is obtained. This parameter can robustly evaluate the accuracy of the overall contour, size, and internal topological relationship of the interface.

[0085] Exemplarily, on the production line, a charger to be tested is placed. The top industrial camera takes an interface image of 1024x1024 pixels. The image is sent to a pre-trained U-Net model, and the model quickly outputs the coordinates of 8 key vertices, such as [(102, 305), (410, 306),...]. At the same time, an analysis unit reads the standard asymmetric structure template of the charger from the database, which also defines the ideal coordinates of the 8 vertices. By constructing two 16-dimensional vectors from the coordinates of the two sets of 8 points respectively, and calculating the cosine similarity between them, the final structure matching degree parameter is 0.9992. This high score indicates that the geometric shape of the current charger interface is highly consistent with the standard design.

[0086] Optionally, the generating full-process performance data comprises:

[0087] Applying a simulated load to the power bus of the vehicle charger and collecting voltage and current responses to obtain basic electrical performance parameters;

[0088] Specifically, in the step of applying a simulated load to the power bus of the vehicle charger to obtain basic electrical performance parameters, a programmable DC electronic load integrated in the test tooling is activated. The electronic load simulates different working conditions in the charging process according to a pre-set test sequence, such as no load, 50% rated load and 100% full load, etc. In each load state, a high-precision data acquisition module measures the output voltage and output current of the charger synchronously, and monitors the input power of the charger. A calculation unit can calculate the charging efficiency of the product in real time according to these measured values, as shown in the formula:

[0089] ,

[0090] wherein, is the charging efficiency. These voltage, current and calculated charging efficiency data collected under different loads constitute a series of basic electrical performance parameters reflecting the core power conversion capability of the charger.

[0091] Sending protocol test instructions to the data bus of the vehicle charger and capturing the returned signal waveform to obtain protocol interaction data;

[0092] Specifically, in the step of sending a protocol test instruction to the data bus of the on-board charger to obtain protocol interaction data, a protocol analyzer or a simulated electronic control unit module sends a standard communication handshake or data request instruction to the charger to be tested through a controller area network bus or a local interconnection network bus. A high-speed oscilloscope or a data acquisition card synchronously captures the complete digital waveform of the response signal returned by the charger. A time analysis unit accurately calculates the time difference between the sending of the instruction and the reception of the valid response, i.e., the protocol response delay, as shown in the formula:

[0093] ,

[0094] wherein, is the protocol response delay; is the time stamp of the reception of the valid response signal; is the time stamp of the sending of the test instruction. The response delay and the captured signal waveform data together constitute the protocol interaction data, which are used to evaluate whether the communication function of the charger is normal and the response is timely.

[0095] The base electrical performance parameters and the protocol interaction data are aggregated and structured to generate whole-process performance data.

[0096] Specifically, in the step of aggregating and structuring the base electrical performance parameters and the protocol interaction data to generate whole-process performance data, a data processing module summarizes and formats all the discrete data generated in the previous two steps. The module integrates the base electrical performance parameters such as voltage, current, and charging efficiency under different loads, and the protocol interaction data such as protocol response delay and signal waveform data, into a unified and structured data record, such as a JSON object or an XML file. This complete data record is the whole-process performance data generated finally, which comprehensively describes the performance of the product in this test and provides complete and objective basis for the subsequent quality judgment link.

[0097] For example, when a charger enters the performance test process, an electronic load is first loaded at 50% of its rated power, at which time the output voltage is measured to be 14.4V, the output current is measured to be 20A, and the calculated charging efficiency is 94.5%. These data are recorded as part of the base electrical performance parameters. Next, a CAN bus analyzer sends a request instruction to read the battery temperature to the charger, and receives a correctly formatted response frame after 5 milliseconds. The response time and the waveform data of the response frame are recorded as protocol interaction data. Finally, a data processing module integrates these information into a JSON object. This complete JSON object is the whole-process performance data generated in this test, and is transmitted to the next judgment unit.

[0098] Optionally, the generating and executing differentiated disposal instructions comprises:

[0099] grading and comparing the full-process performance data with the performance judgment threshold to generate a multi-dimensional performance judgment result;

[0100] Specifically, in the step of grading and comparing the full-process performance data with the performance judgment threshold to generate a multi-dimensional performance judgment result, a judgment analysis unit sets multiple threshold intervals, such as an “optimal interval”, an “acceptable interval” and a “failure interval”, for each key performance parameter to achieve fine-grained grading evaluation. The unit compares the measured values in the full-process performance data with these grading thresholds item by item to generate a qualitative rating for each performance parameter. All single ratings are aggregated into a data structure, which collectively constitutes a multi-dimensional performance judgment result. The structure can be represented by the formula:

[0101] ,

[0102] wherein, is the generated multi-dimensional performance judgment result; represents the key name of the nth performance parameter, such as “charging efficiency”; is the rating result corresponding to the performance parameter, such as “optimal” or “failure”. This step provides detailed, multi-dimensional input basis for subsequent comprehensive decision-making.

[0103] analyzing the multi-dimensional performance judgment result to obtain a disposal level containing qualified, re-inspection required or unqualified information;

[0104] Specifically, in the step of analyzing the multi-dimensional performance judgment result to obtain a disposal level, a rule engine is responsible for executing a set of pre-set business logic rules. These rules define how to determine the final disposal level of the product according to the combination of each single rating. For example, the rule can be set as: when all single ratings are “optimal” or “acceptable”, the final disposal level is determined as “qualified”; when there is no “failure” rating but at least one “attention required” rating exists, it is determined as “re-inspection required”; if any rating is “failure”, it is directly determined as “unqualified”. This analysis process can be abstracted as a function as shown in the formula:

[0105] ,

[0106] wherein, is the final output disposal level containing qualified, re-inspection required or unqualified information; represents the judgment function executed by the rule engine; is the input multi-dimensional performance judgment result.

[0107] Based on the treatment level, identification instructions for qualified products, re-inspection prompt instructions for products requiring re-inspection, and isolation treatment instructions for unqualified products are respectively constructed to generate differentiated treatment instructions.

[0108] Specifically, in the step of constructing different instructions based on the treatment level to generate differentiated treatment instructions, one instruction generation module is responsible for converting the abstract treatment level obtained in the previous step into machine instructions that can be directly executed by downstream devices. According to the input treatment level, the module matches and constructs a data packet containing explicit actions, target addresses, and related data. For example, an identification instruction for qualified products can be a data structure containing a target device address "Laser_1" and to-be-printed data "SN12345"; and an isolation instruction for unqualified products can be a data structure containing a target device address "Reject_Bin_A" and a fault code "E05". The finally generated data packet is the differentiated treatment instruction, and its generation process can be represented by the formula:

[0109] ,

[0110] Wherein, is the generated differentiated treatment instruction; represents the instruction construction function; is the input treatment level.

[0111] Exemplarily, after a charger test is completed, its multi-dimensional performance determination result is determined as {voltage stability: "optimal", communication response: "need attention"}. A rule engine parses this determination result as a "re-inspection" treatment level according to internal rules. Subsequently, an instruction generation module receives the "re-inspection" level and immediately constructs a differentiated treatment instruction with the content {action: "route", destination: "re-inspection_station_02", reason_code: "CAN_TIMEOUT"}. The instruction is sent to the production line control system, and a physical sorting baffle guides the charger to the second re-inspection station according to the instruction action, so that the technical personnel can perform targeted inspection.

[0112] Optionally, the method further comprises:

[0113] Performing unsupervised cluster analysis on the data with the treatment level of unqualified products, extracting common failure features, and generating a failure mode feature vector;

[0114] Specifically, in the step of performing unsupervised clustering analysis on the data of the products with the disposition level of unqualified products to generate the failure mode feature vector, a data analysis module periodically collects the full-process performance data of all products determined as unqualified from the historical database. The module aggregates the high-dimensional performance data using an unsupervised clustering algorithm, such as K-Means Clustering. The goal of the algorithm is to minimize the sum of squared distances of each data point to the centroid of the cluster to which it belongs, thereby automatically classifying products with similar failure behaviors into a category. The objective function can be expressed as the formula:

[0115] ,

[0116] wherein, is the total intra-cluster sum of squares of the clustering algorithm; is the preset number of failure mode clusters; represents the i-th failure mode cluster; is the full-process performance data vector of a certain unqualified product; is the centroid of the i-th cluster. The centroid is a common failure feature extracted, which mathematically represents a typical failure mode and is defined as a specific failure mode feature vector.

[0117] The spatial attitude parameters and electrical characteristic parameters associated with the failure mode feature vector are traced back to construct and generate a parameter-failure mode association rule set.

[0118] Specifically, in the step of tracing back the associated initial parameters to construct and generate the parameter-failure mode association rule set, a rule mining engine performs a data mining task. For each failure mode cluster clustered in the previous step, the engine traces back the initial parameters, including spatial attitude parameters and electrical characteristic parameters, collected at the initial stage of testing for all products in the cluster. This constitutes a large historical pairing data set of "(initial parameter combination) - (failure mode)". Then, the engine trains the data set using an association rule mining algorithm, such as the Apriori algorithm, to discover frequently occurring "item sets" and generate association rules in the form of the formula:

[0119] ,

[0120] wherein, is a generated association rule; represents a combination of initial parameters leading to failure, such as "contact impedance > 150 mΩ" and "Z-axis deviation > 0.2 mm"; is the failure mode frequently associated with the condition combination, such as "communication timeout failure"; support of the rule, representing the frequency of the combination in the total samples; confidence of the rule, representing the probability that the result also occurs when the condition occurs. All valid rules that satisfy the minimum support and confidence thresholds together constitute the final parameter-failure mode association rule set, as shown in Figure 4 the figure not only visually shows the association strength between a specific initial parameter combination and a specific failure mode through the color block depth, but also automatically reveals the similarity between different failure modes and the internal association between different risk sources through the tree cluster diagram on both sides, providing deeper insights for the aggregation analysis of root causes.

[0121] Exemplarily, a data analysis module analyzes 500 non-conforming product data accumulated in the past week, and finds three main failure clusters through the K-means clustering algorithm, one of which is identified as the "voltage drop too large under high load" failure mode. Subsequently, a rule mining engine traces the initial test parameters of the 150 products belonging to this cluster and runs the Apriori algorithm. The algorithm mines a high-confidence rule: if a product's "initial contact resistance" is at a high level and the Y-axis component of its "three-dimensional spatial deviation vector" is negative, then the product has an 85% probability of "voltage drop too large under high load" failure. This rule is automatically stored in the parameter-failure mode association rule set, providing valuable knowledge for subsequent analysis.

[0122] Optionally, the method further comprises:

[0123] inputting the real-time collected spatial posture parameters and electrical characteristic parameters into the parameter-failure mode association rule set for inference to generate a predictive failure risk level;

[0124] Specifically, when a new round of testing begins and the initial parameters are collected in real time, a risk inference engine will immediately call the parameter-failure mode association rule set. The engine takes the real-time collected spatial posture parameters and electrical characteristic parameters as input and performs high-speed query and matching in the rule set. If the current parameter combination hits one or more high-confidence association rules, the engine will calculate a comprehensive and quantitative predictive failure risk level according to the confidence scores of these rules. This process can be abstracted as the formula:

[0125] ,

[0126] wherein, is the generated predictive failure risk level, which can be a value between 0 and 1, and the higher the value, the greater the predicted failure probability; represents the risk inference function; is an initial parameter set containing real-time spatial posture and electrical characteristics; is a parameter-failure mode correlation rule set.

[0127] fusing the predictive failure risk level and the predictive collision and stress distribution parameters to generate feedforward adjustment parameters;

[0128] Specifically, in the step of fusing the predictive failure risk level and the predictive collision and stress distribution parameters to generate feedforward adjustment parameters, a decision fusion unit is responsible for balancing two different dimensions of prospective risks: the “predictive collision and stress distribution parameters” representing the safety of the physical process, and the “predictive failure risk level” representing the product quality trend generated in the previous step. This unit employs a multi-objective decision function to weight and fuse the two risks to generate a set of feedforward adjustment parameters for active intervention. The calculation process can be represented by the formula:

[0129]

[0130] wherein, is the final generated set of feedforward adjustment parameters; and are the preset weights of quality risk and physical process risk, respectively; and are functions that map various risk levels or parameters to specific adjustment values; and are the two input risk data, respectively.

[0131] using the feedforward adjustment parameters to dynamically optimize the docking verification benchmark and the performance judgment threshold.

[0132] Specifically, in the step of dynamically optimizing the test benchmark and threshold using the feedforward adjustment parameters, a parameter configuration module receives the feedforward adjustment parameters generated in the previous step. This module adjusts the relevant standards in this round of testing according to these parameters, which is one-time, dynamic, and personalized. For example, if the feedforward adjustment parameters indicate that the voltage stability judgment needs to be tightened, the module will temporarily adjust the voltage fluctuation allowed range in the performance judgment threshold from ±1% to ±0.5%. Similarly, if it is indicated that the physical verification standard needs to be adjusted, the relevant parameters in the docking verification benchmark will be modified. In this way, the test standards are no longer fixed, but are dynamically optimized for each product to be tested.

[0133] ​Exemplarily, one to-be-tested charger enters the station, and the initial parameters collected in real time match a high-confidence rule in the parameter-failure mode association rule set, indicating that the parameter combination is highly related to "communication protocol test failure". A risk inference engine generates a predictive failure risk level of up to 0.9 based on this. At the same time, virtual simulation shows that the physical risk of this alignment process is very low. After comprehensive judgment of a decision fusion unit, it is considered that the quality risk is the main contradiction of this test, and a set of feedforward adjustment parameters aiming at "strictly checking the communication function" is generated. After receiving the instruction, a parameter configuration module temporarily optimizes and tightens the qualified standard of "CAN bus response delay" in the performance judgment threshold from "<=10ms" to "<=5ms" in this test. This makes the "tolerance" of the communication performance of this round of test lower, so that it can capture potential defects warned by initial parameters with a higher probability.

[0134] Based on the same inventive concept, the present application also provides a vehicle-mounted charger performance test system, as shown in the accompanying drawings, the system comprises: Figure 5

[0135] An initial parameter perception and instruction generation module is configured to collect the spatial pose parameters and electrical characteristic parameters of the vehicle-mounted charger interface, and to calculate and generate spatial fine-tuning driving instructions;

[0136] A docking and verification module is configured to perform docking based on the spatial fine-tuning driving instructions, and to perform consistency verification with a preset docking verification reference to obtain a docking success or failure verification result;

[0137] A multi-channel performance detection module is configured to start a multi-channel performance detection process to generate full-process performance data if the verification result feedback indicates docking success;

[0138] A data judgment and differential treatment module is configured to judge based on the full-process performance data and a preset performance judgment threshold, and to generate and execute differential treatment instructions.

[0139] It should be noted that the functional division and information interaction between the above-mentioned various modules are logical, and can be integrated in the same software platform or distributed in physical implementation. The connection between them represents data flow and control flow, and aims to cooperatively achieve the building energy consumption dynamic optimization goal of the present application. The above-mentioned is only an exemplary embodiment of the present application, and cannot limit the protection scope of the present application.​

Claims

1. A method for testing the performance of an on-board charger, characterized in that, The method includes: Collect the spatial attitude parameters and electrical characteristic parameters of the on-board charger interface, calculate and generate spatial fine-tuning drive commands; The docking is performed based on the space fine-tuning drive command, and a consistency check is performed with the preset docking verification benchmark to obtain the verification result of successful or failed docking. If the verification result indicates successful integration, the multi-channel performance testing process will be initiated to generate full-process performance data. Based on the full-process performance data and the preset performance judgment threshold, a judgment is made, and a differentiated handling instruction is generated and executed. The method further includes: Unsupervised clustering analysis is performed on data of products with a preset disposal level of non-conforming to extract common failure features and generate failure mode feature vectors; Tracing the spatial attitude parameters and electrical characteristic parameters associated with the failure mode feature vector, constructing and generating a parameter-failure mode association rule set; The method further includes: The real-time collected spatial attitude parameters and electrical characteristic parameters are input into the parameter-failure mode association rule set for inference, generating a predictive failure risk level; The predicted failure risk level is fused with the preset predicted collision and stress distribution parameters to generate feedforward adjustment parameters; The feedforward adjustment parameters are used to dynamically optimize the docking verification benchmark and the performance judgment threshold.

2. The method for testing the performance of an on-board charger according to claim 1, characterized in that, The calculation and generation of spatial fine-tuning drive instructions includes: The real-time three-dimensional coordinates of the vehicle charger interface are obtained and vector calculations are performed to obtain the three-dimensional spatial deviation vector. The contact impedance and signal response status of the on-board charger interface are collected synchronously to generate electrical characteristic parameters; Based on the three-dimensional spatial deviation vector and the electrical characteristic parameters, the X / Y / Z three-axis fine-tuning amount is calculated, and spatial fine-tuning drive commands are generated.

3. The method for testing the performance of an on-board charger according to claim 2, characterized in that, The calculation of X / Y / Z three-axis fine-tuning amounts and the generation of spatial fine-tuning drive commands include: The kinematics of the three-dimensional spatial deviation vector are calculated to generate a reference driving scheme; The baseline driving scheme was virtually simulated to obtain predictive collision and stress distribution parameters; Based on the predicted collision and stress distribution parameters and the electrical characteristic parameters, the reference drive scheme is optimized and corrected, and the X / Y / Z three-axis fine-tuning amount is calculated as the correction value. The spatial fine-tuning drive command is then integrated and generated.

4. The method for testing the performance of an on-board charger according to claim 1, characterized in that, The consistency verification with the preset docking verification benchmark includes: Extract the geometric contour features of the vehicle charger interface and perform similarity calculation with the asymmetric structure template in the docking verification benchmark to generate structural matching parameters; The contact impedance and signal response delay of the on-board charger interface are detected in real time and compared with the electrical compliance threshold range in the docking verification benchmark to generate electrical compliance parameters. The structural matching parameter and the electrical compliance parameter are logically ANDed to determine whether the docking was successful or failed, and a verification result is generated.

5. The method for testing the performance of an on-board charger according to claim 4, characterized in that, The generated structure matching degree parameters include: Obtain historical interface image data and historical standard feature point coordinates to obtain a historical interface image dataset; Using interface image data as input and standard feature point coordinates as output, a neural network model is established and trained using the historical interface image dataset to obtain an interface geometric feature point recognition model. Capture real-time interface image data and input the real-time interface image data into the interface geometric feature point recognition model to obtain the coordinates of key geometric feature points; The coordinates of the key geometric feature points are compared with the coordinates of the asymmetric structure template in terms of spatial position and topological relationship, and structural matching degree parameters are calculated and generated.

6. The method for testing the performance of an on-board charger according to claim 1, characterized in that, The generated full-process performance data includes: A simulated load is applied to the power bus of the on-board charger, and the voltage and current responses are collected to obtain basic electrical performance parameters. Send a protocol test command to the data bus of the vehicle charger and capture the returned signal waveform to obtain the protocol interaction data; The basic electrical performance parameters and the protocol interaction data are aggregated and structured to generate full-process performance data.

7. The method for testing the performance of an on-board charger according to claim 3, characterized in that, The generation and execution of differentiated processing instructions include: The full-process performance data is compared with the performance judgment threshold in a hierarchical manner to generate multi-dimensional performance judgment results. The multi-dimensional performance judgment results are analyzed to obtain the handling level, which includes information on whether the performance is qualified, requires re-inspection, or is unqualified. Based on the aforementioned disposal levels, identification instructions for qualified products, re-inspection reminder instructions for products requiring re-inspection, and isolation disposal instructions for unqualified products are constructed respectively, generating differentiated disposal instructions.

8. A vehicle charger performance testing system, applied to the vehicle charger performance testing method as described in any one of claims 1-7, characterized in that, The system includes: The initial parameter sensing and command generation module is used to collect the spatial attitude parameters and electrical characteristic parameters of the on-board charger interface, calculate and generate spatial fine-tuning drive commands; The docking and verification module is used to perform docking based on the space fine-tuning drive command, and to perform consistency verification with the preset docking verification benchmark to obtain the verification result of successful or failed docking. The multi-channel performance testing module is used to initiate the multi-channel performance testing process and generate full-process performance data if the verification result indicates successful docking. The data judgment and differentiated processing module is used to make judgments based on the full-process performance data and preset performance judgment thresholds, and generate and execute differentiated processing instructions.

Images