Methods for troubleshooting charging equipment and charging systems

By acquiring the operating status and visual evidence data of charging equipment, and using a multimodal diagnostic model for fault diagnosis and converting it into plain language for display, the problem of obscure code and low communication efficiency in charging equipment fault inspection is solved. This enables accurate fault diagnosis and self-resolvement, improving user experience and operation and maintenance efficiency.

CN123066432APending Publication Date: 2026-11-10XIAN LINCHR NEW ENERGY TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611082506.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-21
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing methods for troubleshooting charging equipment suffer from problems such as obscure and difficult-to-understand fault codes, unclear causes of faults, low efficiency in customer service communication, long operation and maintenance response cycles, and numerous disputes over vehicle-charging station responsibility, resulting in poor user experience and waste of operation and maintenance resources.

Method used

By acquiring operational status data and visual evidence data of charging equipment, a multimodal diagnostic model is used to diagnose faults, generate fault diagnosis results, and convert the results into plain language through a natural language generation model. Combined with voice guidance, a self-inspection process is realized, which can resolve simple faults or dispatch work orders on its own.

Benefits of technology

It enables accurate diagnosis of charging equipment faults, improves the self-resolution rate of simple faults, reduces the rate of ineffective maintenance, and enhances user experience and maintenance efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123066432A_ABST
    Figure CN123066432A_ABST
Patent Text Reader

Abstract

This application discloses a method, apparatus, terminal, and medium for fault inspection of charging equipment. The method includes: acquiring operating status data of the charging equipment within a preset time window; in response to a fault event trigger signal, acquiring images of a preset image acquisition area corresponding to the charging equipment to obtain visual evidence data corresponding to the charging equipment; and based on the operating status data and the visual evidence data, performing fault diagnosis on the charging equipment through a preset multimodal diagnostic model to obtain a fault diagnosis result corresponding to the charging equipment. This application aims to achieve accurate fault diagnosis of vehicle charging equipment, thereby improving the self-resolution rate of simple faults in vehicle charging equipment and reducing the rate of ineffective maintenance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method for fault detection of charging equipment and a charging system. Background Technology

[0002] With the rapid development of the new energy vehicle industry, charging equipment, as a key basic supporting infrastructure, has been deployed on a large scale. However, in actual use, the problems of frequent charging failures and poor user experience are becoming increasingly prominent. At present, the industry routinely uses fixed error codes + screen text prompts to display fault information to users (such as displaying "E0201" or "communication timeout").

[0003] However, existing methods for troubleshooting charging equipment have the following shortcomings: Obscure and difficult-to-understand fault codes: Ordinary users cannot understand the technical meaning of hexadecimal / numeric codes such as E0201 and 0x1A3F, and are often at a loss when faced with error messages on the screen; Unclear causes of faults: Even if some text descriptions are displayed (such as "CP signal abnormality"), users still cannot determine whether the fault is due to a problem with the charging equipment itself, a problem with the vehicle's BMS, or improper operation; Low efficiency of customer service communication: When users call customer service, they lack on-site professional knowledge and cannot accurately describe the fault phenomenon. Customer service personnel also lack real-time operational data and cannot remotely locate the problem, resulting in repeated communication with little effect; Long maintenance response cycle: After receiving a repair report, maintenance personnel need to rush to the site blindly. Many simple faults (such as dirty gun contacts, improper insertion of the gun, or discontinuation due to unpaid fees) that could be resolved by the user themselves still require waiting for maintenance personnel to arrive, resulting in a serious waste of maintenance resources and long waiting times for users; Frequent disputes over vehicle-pile responsibility: When charging fails, users tend to think that the pile is broken, while operators suspect that the problem is with the vehicle. Both parties lack objective evidence to determine responsibility, leading to complaints and disputes. Summary of the Invention

[0004] The main purpose of this application is to provide a method for fault diagnosis of charging equipment and a charging system, which aims to achieve accurate fault diagnosis of vehicle charging equipment, thereby improving the self-resolution rate of simple faults of vehicle charging equipment and reducing the rate of ineffective maintenance.

[0005] To achieve the above objectives, this application provides a method for fault detection of charging equipment, applied to a charging system, the method comprising: Acquire the operating status data of the charging device within a preset time window; In response to a fault event trigger signal, images are acquired from a preset image acquisition area corresponding to the charging device to obtain visual evidence data corresponding to the charging device. Based on the operational status data and the visual evidence data, a pre-set multimodal diagnostic model is used to diagnose the charging equipment and obtain the corresponding fault diagnosis results.

[0006] Specifically, the operating status data is a time-series data window, and the preset time window is used to represent the time period from a preset time before the charging device's failure time to the current time; The acquisition of the charging device's operating status data within a preset time window includes: Based on a preset sampling frequency, the built-in sensor group collects the output voltage, output current, internal temperature of the charging device, charging gun lock status data, network link quality data, and vehicle-charging pile communication log data of the charging device. The timing data window is generated based on the output voltage, the output current, the pile temperature, the charging gun lock-up status data, the network link quality data, and the vehicle-pile communication log data.

[0007] Specifically, the preset image acquisition area includes a first area, a second area, and a third area. The first area is used to characterize the area where the charging gun and the vehicle interface cooperate. The second area is used to characterize the overall appearance area of ​​the charging gun body and cable. The third area is used to characterize the human-machine interaction screen display area of ​​the charging device. In response to a fault event trigger signal, the system acquires images of a preset image acquisition area corresponding to the charging device to obtain visual evidence data corresponding to the charging device, including: A first image corresponding to the first region is acquired, wherein the first image is used to characterize the insertion depth of the charging gun, the attitude of the charging gun body, and the state of the cable body. Acquire a second image corresponding to the second region, wherein the second image is used to characterize the condition of the gun body casing and the cable sheath. A third image corresponding to the third region is acquired, wherein the third image is used to characterize the display area of ​​the human-computer interaction screen; The visual evidence data is generated based on the first image, the second image, the third image, the shooting timestamp corresponding to the first image, the shooting timestamp corresponding to the second image, the shooting timestamp corresponding to the third image, and the fault triggering timestamp corresponding to the fault event triggering signal.

[0008] Specifically, the preset multimodal diagnostic model includes an electrical anomaly pattern recognition layer, a visual physical state analysis layer, and a cross-modal joint reasoning layer. The fault diagnosis result includes fault category, fault root cause description data, and fault severity level. The fault root cause description data is used to characterize the judgment result text of the technical root cause of the fault. Based on the operational status data and the visual evidence data, a pre-set multimodal diagnostic model is used to diagnose faults in the charging equipment, resulting in fault diagnosis results for the charging equipment, including: The electrical anomaly pattern recognition layer is used to identify anomalies in the operating status data to obtain anomaly identification results. The visual physical state analysis layer performs image analysis on the visual evidence data to obtain image analysis results. The cross-modal joint inference layer correlates the anomaly identification results and the image analysis results to obtain the fault diagnosis results.

[0009] Specifically, before executing a preset fault self-check strategy based on the fault diagnosis result to obtain the target self-check result, the method further includes: The electrical anomaly pattern recognition layer extracts the CC circuit resistance value of the charging gun contact from the operating status data. The anomaly identification result is determined based on the voltage amplitude of the charging gun contact CP signal, the resistance value of the charging gun contact CC circuit, and the preset electrical threshold in the operating status data. Based on the anomaly identification results, the first image, and the second image, the charging device is fault diagnosed using the preset multimodal diagnostic model to obtain the fault diagnosis results.

[0010] Specifically, after performing fault diagnosis on the charging device based on the operating status data and the visual evidence data using a preset multimodal diagnostic model to obtain the corresponding fault diagnosis result for the charging device, the method further includes: The fault root cause description data is processed by replacing professional terms, rewriting in colloquial language, and adapting to the scene using a preset natural language generation model to obtain colloquial language description text. This colloquial language description text is then displayed on the human-computer interaction screen of the charging device. The colloquial language description text is used to present fault information to non-professionals. Based on the colloquial language description text, an audio message is generated to enable the colloquial language description text to be played through the speaker of the charging device; Based on the fault category and the fault severity level, it is determined whether to execute a first self-test process or a second self-test process. The first self-test process is used to perform a self-test on the charging equipment through voice guidance, and the second self-test process is used to determine whether to transfer to the work order dispatch process based on the self-test results.

[0011] Specifically, the fault categories include self-repairable and non-self-repairable categories, and the fault severity levels include minor, moderate, severe, and critical. The step of determining whether to execute the first self-test process or the second self-test process based on the fault category and the fault severity level includes: If the fault category of the charging device is self-repairable, and the fault severity level of the charging device is minor or moderate, then the first self-test process is executed. If the fault category of the charging device is non-self-repairable, or if the fault severity level of the charging device is severe or critical, then the second self-test process will be executed.

[0012] Specifically, executing the first self-test process includes: Based on the fault diagnosis results, a self-test step sequence is generated by calling a preset self-test knowledge base; The self-test sequence is played through the speaker of the charging device so that the user can operate according to the self-test sequence; The system obtains user feedback information after each step in the self-test sequence through a preset parallel channel corresponding to the charging device, and records the execution status information of the self-test sequence through an interactive state machine. The execution status information includes the number of times the user failed to execute each step in the self-test sequence. Based on the feedback information and the execution status information, the user's self-inspection result is determined; Repeat the above steps multiple times. If the number of self-check failures is not less than the preset failure threshold, then execute the second self-check process.

[0013] Specifically, the execution of the second self-test process includes: Based on the fault diagnosis results, the visual evidence data, and the operating status data, a work order package is generated, wherein the work order package is used to notify professional operation and maintenance personnel to inspect the charging equipment and display the current status of the charging equipment; After the work order package is dispatched, it is determined that the charging equipment enters a state of waiting for maintenance and handling. In response to the repair confirmation signal corresponding to the charging device, it is determined that the charging device has entered the normal charging monitoring state.

[0014] To achieve the above objectives, this application also provides a charging system, including at least two power modules, at least one charging interface, a power distribution device, a main controller, and a charging interface; the main controller is connected to each of the power modules, each of the charging interfaces, and the power distribution device, and is used to execute the charging equipment fault detection method; the power distribution device is connected to each of the power modules and each of the charging interfaces; the charging interface is used to connect a charging gun, and the charging gun is connected to the corresponding host of the charging system via a gun holder on the main body of the charging system.

[0015] This application provides a method, apparatus, terminal, and medium for fault inspection of charging equipment. It first acquires the operating status data of the charging equipment within a preset time window; in response to a fault event trigger signal, it acquires images of a preset image acquisition area corresponding to the charging equipment to obtain visual evidence data corresponding to the charging equipment; based on the operating status data and the visual evidence data, it performs fault diagnosis on the charging equipment using a preset multimodal diagnostic model to obtain the corresponding fault diagnosis result, thereby achieving accurate fault diagnosis of vehicle charging equipment, improving the self-resolution rate of simple faults in vehicle charging equipment, and reducing the rate of ineffective maintenance. Attached Figure Description

[0016] Figure 1 A flowchart illustrating the method provided in the embodiments of this application; Figure 2 This is an overall architecture diagram of the intelligent semantic explanation and voice-guided self-testing system for charging equipment faults provided in the embodiments of this application; Figure 3 This is a schematic diagram of the hardware composition of the intelligent semantic interpretation device for charging equipment faults provided in an embodiment of this application; Figure 4 This is a flowchart of multi-source heterogeneous data fusion acquisition and fault-triggered visual forensics provided in the embodiments of this application; Figure 5 This is a schematic diagram of the charging system provided in an embodiment of this application. Detailed Implementation

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

[0018] The existing methods for troubleshooting charging equipment have the following shortcomings: Obscure and difficult-to-understand fault codes: Ordinary users cannot understand the technical meaning of hexadecimal / numeric codes such as E0201 and 0x1A3F, and are often at a loss when faced with error messages on the screen; Unclear causes of faults: Even if some text descriptions are displayed (such as "CP signal abnormality"), users still cannot determine whether the fault is due to a problem with the charging equipment itself, a problem with the vehicle's BMS, or improper operation; Low efficiency of customer service communication: When users call customer service, they lack on-site professional knowledge and cannot accurately describe the fault phenomenon. Customer service personnel also lack real-time operational data and cannot remotely locate the problem. Repeated communication between the two parties yields little result; Long maintenance response cycle: After receiving a repair report, maintenance personnel have to rush to the site blindly. Many simple faults (such as dirty gun contacts, improper insertion of the gun, or suspension due to unpaid fees) that could be resolved by the user themselves still require waiting for maintenance personnel to arrive, resulting in a serious waste of maintenance resources and long waiting times for users; Frequent disputes over vehicle-pile responsibility: When charging fails, users tend to think that the pile is broken, while operators suspect that the problem is with the vehicle. Both parties lack objective evidence to determine responsibility, leading to complaints and disputes.

[0019] Therefore, embodiments of this application provide a method, apparatus, terminal, and medium for troubleshooting charging equipment to solve practical technical problems.

[0020] In some embodiments, the device may be integrated into an electronic terminal, which may be a terminal, server, or other terminal.

[0021] In some embodiments, the server may also be implemented as a terminal.

[0022] The server can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0023] The terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal and the server can be connected directly or indirectly through wired or wireless communication, which is not limited herein.

[0024] The following sections provide detailed descriptions of each example. It should be noted that the sequence numbers of the following embodiments are not intended to limit the preferred order of the embodiments.

[0025] This application provides a method for fault diagnosis of charging equipment and a charging system, which aims to achieve accurate fault diagnosis of vehicle charging equipment, thereby improving the self-resolution rate of simple faults of vehicle charging equipment and reducing the rate of ineffective maintenance.

[0026] like Figure 1 The specific process of the method may include: S110. Obtain the operating status data of the charging device within a preset time window.

[0027] The operating status data is a time-series data window, and the preset time window is used to represent the time period from a preset time before the charging device's failure time to the current time.

[0028] In some embodiments, obtaining the operating status data of the charging device within a preset time window includes the following steps S111 to S112: S111. Based on a preset sampling frequency, the built-in sensor group collects the output voltage, output current, internal temperature of the charging device, charging gun lock status data, network link quality data, and vehicle-charging station communication log data of the charging device.

[0029] Specifically, the charging equipment terminal collects six-dimensional runtime sequence data in real time through a built-in sensor group, including: output voltage U(t), output current I(t), pile temperature T(t), charging gun lock status data S(t), network link quality data QoS(t), and vehicle-pile communication log data L(t). The preset sampling frequency of the data is not less than 1Hz.

[0030] S112. Based on the output voltage, the output current, the pile temperature, the charging gun lock-up status data, the network link quality data, and the vehicle-pile communication log data, generate the timing data window.

[0031] Specifically, a time-series data window W_data=[U,I,T,S,QoS,L] is formed by using the output voltage U(t), output current I(t), pile temperature T(t), charging gun lock-up status data S(t), network link quality data QoS(t), and vehicle-pile communication log data L(t). The preset time window is used to characterize the time period from a preset time before the charging device failure time to the current time. That is, the window length of the time-series data window can be from 30 seconds before the failure time to the current time.

[0032] S120. In response to the fault event trigger signal, image acquisition is performed on the preset image acquisition area corresponding to the charging device to obtain visual evidence data corresponding to the charging device.

[0033] The preset image acquisition area includes a first area, a second area, and a third area. The first area is used to characterize the area where the charging gun and the vehicle interface cooperate. The second area is used to characterize the overall appearance area of ​​the charging gun body and cable. The third area is used to characterize the human-machine interaction screen display area of ​​the charging device.

[0034] The step of responding to a fault event trigger signal by acquiring images of a preset image acquisition area corresponding to the charging device to obtain visual evidence data corresponding to the charging device includes the following specific implementation process: A first image corresponding to a first region is acquired, wherein the first image is used to characterize the insertion depth of the charging gun, the posture of the charging gun body, and the state of the cable body; a second image corresponding to a second region is acquired, wherein the second image is used to characterize the condition of the gun body shell and the condition of the cable sheath; a third image corresponding to a third region is acquired, wherein the third image is used to characterize the condition of the human-computer interaction screen display area; the visual evidence data is generated based on the first image, the second image, the third image, the shooting timestamp corresponding to the first image, the shooting timestamp corresponding to the second image, the shooting timestamp corresponding to the third image, and the fault trigger timestamp corresponding to the fault event trigger signal.

[0035] In some embodiments, when the charging device detects a fault event (including but not limited to: charging start failure, charging stop midway, over-temperature protection tripping, communication interruption, insulation detection failure, etc.), the system automatically activates the built-in camera module (installed on the upper front of the charging device, with a viewing angle covering the gun mount and operating area), and performs directional image acquisition on three key areas according to a preset shooting strategy. The deviation between the shooting timestamp t_capture and the fault trigger timestamp t_fault is controlled within ±200ms (hardware interrupt triggered), and the total acquisition time for the three images is ≤1.5 seconds. The area where the charging gun meets the vehicle, i.e., the first area (target image: I_interface, i.e., the first image), is the most crucial visual evidence. The camera captures the overall fit of the charging gun after it's inserted into the vehicle's charging dock: insertion depth—the gap between the gun's flange and the vehicle's charging dock panel (normally ≤5mm, larger gaps indicate incomplete insertion); gun posture—whether the gun's axis is parallel to the charging dock's axis, and whether it's tilted / drooping; cable condition—whether the cable is excessively bent (bending radius < 5 times the wire diameter), stretched by external force, or run over by a vehicle wheel. The camera is positioned perpendicular to the charging dock panel, covering the entire gun-to-dock joint.

[0036] The overall appearance of the charging gun and cable, i.e., the second area (target: I_gun_body, i.e., the second image) – supplementary physical evidence. The side of the gun and the near end of the cable are photographed: for cracks, damage, or burn marks on the gun casing; abnormal dirt or oil stains on the grip area; and damage, flattening, or twisting of the cable sheath. The camera is shot from a horizontal angle, covering the entire gun and approximately 50cm of cable.

[0037] The human-computer interaction screen display area, i.e., the third area (shooting target: I_screen, i.e., the third image) – captures the current display of the charging device screen / indicator: fault code number, fault text description, alarm icon (e.g., over-temperature). ,insulation (etc.), indicator light color / blinking mode. The lens is directly facing the screen to ensure that characters are readable and icons are recognizable, serving as an objective record of the machine's self-reported status at the time of the fault. The three captured images are recorded as I_interface, I_gun_body, and I_screen, respectively, and together with the fault trigger timestamp t_fault and the precise capture timestamps of each image, they are packaged into a visual evidence package V_evidence, i.e., the visual evidence data.

[0038] S130. Based on the operating status data and the visual evidence data, the charging device is diagnosed using a preset multimodal diagnostic model to obtain the corresponding fault diagnosis result of the charging device.

[0039] The preset multimodal diagnostic model includes an electrical anomaly pattern recognition layer, a visual physical state analysis layer, and a cross-modal joint reasoning layer. The fault diagnosis results include fault category, fault root cause description data, and fault severity level. The fault root cause description data is used to characterize the judgment result text of the technical root cause of the fault.

[0040] In some embodiments, the step of performing fault diagnosis on the charging device based on the operating status data and the visual evidence data using a preset multimodal diagnostic model to obtain the corresponding fault diagnosis result for the charging device includes the following specific implementation process: The electrical anomaly pattern recognition layer identifies anomalies in the operating status data to obtain anomaly identification results; the visual physical state analysis layer performs image analysis on the visual evidence data to obtain image analysis results; and the cross-modal joint inference layer correlates the anomaly identification results and the image analysis results to obtain the fault diagnosis results.

[0041] Continuing with the above embodiments, the acquired time-series data window W_data and the acquired visual evidence package V_evidence are fed together into the multimodal large model deployed on the terminal side of the charging device, namely the preset multimodal diagnostic model (supporting but not limited to DeepSeek-VL, Qwen-VL, Doubao VL, Qianwen VL, GLM-4V, Kimi-VL, MiniMax-VL, etc., or enterprise-developed lightweight multimodal models). After vertical domain fine-tuning, this model has the following capabilities: abnormal pattern recognition of time-series electrical parameters in the time-series data window (such as voltage drop waveforms, current ripple anomalies, temperature rise curves, etc.); physical state analysis of visual images in the visual evidence package (such as contact oxidation degree assessment, plug-in angle deviation measurement, screen readability judgment, etc.); cross-modal joint reasoning—associating electrical abnormal patterns with visual physical states, and outputting structured fault diagnosis results D_result. Specifically, D_result={fault category C, confidence level Conf, fault root cause description data R_tech, fault severity level Sev}, where fault category C is an element in the five-dimensional classification set: C∈{pile-side hardware failure, vehicle BMS / battery failure, poor contact between gun and vehicle, abnormal network communication, abnormal fee / account}, confidence level Conf∈[0,1], and fault severity level Sev∈{L1 minor, L2 moderate, L3 severe, L4 critical}.

[0042] In some embodiments, before executing a preset fault self-check strategy based on the fault diagnosis result to obtain the target self-check result, the method further includes the following specific implementation process: The electrical anomaly pattern recognition layer extracts the charging gun contact CC circuit resistance value from the operating status data; based on the charging gun contact CP signal voltage amplitude, the charging gun contact CC circuit resistance value, and a preset electrical threshold in the operating status data, the anomaly identification result is determined; based on the anomaly identification result, the first image, and the second image, the charging device is fault-diagnosed using the preset multimodal diagnostic model to obtain the fault diagnosis result.

[0043] Continuing with the above embodiments, since the national standard GB / T 20234 stipulates that the charging gun contacts (CC / CP / PE / DC+ / DC-) are recessed within the insulating shell and their end faces are not exposed, and the charging equipment camera is usually 1-2 meters away from the gun head, making it impossible to clearly see the details of the contacts within the recess, the contact status judgment is reduced to the level of electrical parameters. The CP signal voltage amplitude in the operating status data—that is, the CP signal voltage amplitude of the charging gun contacts—and the CC circuit resistance value extracted by the electrical anomaly pattern recognition layer of the aforementioned multimodal large model—that is, the CC circuit resistance value of the charging gun contacts—are used to replace direct visual judgment. Specifically, when the CC circuit resistance exceeds the normal range of 100Ω~880Ω, it is inferred that the contacts are dirty / oxidized; when the CP signal voltage is abnormal (normal 9V / 6V PWM square wave, amplitude deviation >10%), it is inferred that there is poor contact; when the insulation resistance drops sharply (<500Ω / V), it is inferred that there are foreign objects or moisture between the contacts. The resulting anomaly identification results, together with images I_interface (large gap, skewed posture) and I_gun_body (dirty and damaged casing), form a multimodal joint evidence chain. Even when it is impossible to directly photograph the contact points, it can still achieve a high-confidence judgment of poor contact. The results are directly input into the preset multimodal diagnostic model, and the obtained fault diagnosis results are more reliable and objective than those relying solely on image visual detection.

[0044] In some embodiments, after performing fault diagnosis on the charging device based on the operating status data and the visual evidence data using a preset multimodal diagnostic model to obtain the corresponding fault diagnosis result for the charging device, the method further includes the following specific implementation process: The fault root cause description data is processed by a preset natural language generation model to replace technical terms, rewrite in a colloquial style, and adapt to different scenarios, resulting in plain language description text. This plain language description text is then displayed on the human-computer interaction screen of the charging device, specifically for presenting fault information to non-professionals. Based on this plain language description text, audio is generated to be played through the charging device's speaker. Based on the fault category and severity level, a first self-inspection process or a second self-inspection process is determined. The first self-inspection process uses voice guidance to perform a self-inspection of the charging device, while the second self-inspection process determines whether to proceed to the work order dispatch process based on the self-inspection results.

[0045] In some embodiments, the fault categories include self-healing categories and non-self-healing categories, and the fault severity levels include minor, moderate, severe, and critical.

[0046] Specifically, determining whether to execute the first self-test process or the second self-test process based on the fault category and the fault severity level includes the following specific implementation process: If the fault category of the charging device is self-repairable and the severity level of the fault is minor or moderate, then the first self-test process is executed; if the fault category of the charging device is non-self-repairable, or the severity level of the fault is severe or critical, then the second self-test process is executed.

[0047] Continuing with the above embodiments, the fault root cause description data R_tech is fed into the Natural Language Generation (NLG) module (i.e., the preset natural language generation model) and converted into plain language description text P_plain for non-professional users. The conversion rules include, but are not limited to: technical terminology replacement—replacing "CC signal loop resistance exceeds the allowable range (100Ω~880Ω)" with "the charging gun metal contacts may be dirty or not plugged in tightly"; colloquial rewriting—removing passive voice and technical modifiers, and using simple subject-verb-object sentence structures; scenario adaptation—selecting wording style based on the inferred user profile (such as the presence or absence of a child seat implying a family user, or the time period inferring the user's identity, etc.).

[0048] Continuing with the above embodiments, a graded interaction strategy decision is executed based on the fault severity level Sev and fault category C: if Sev∈{L1,L2} and C belongs to the self-repairable category (poor contact, minor dirt, overdue payment, etc.), the voice-guided self-test mode is entered, i.e., the first self-test process is executed; if Sev∈{L3,L4} or C belongs to the non-self-repairable category (hardware damage, vehicle malfunction, etc.), the automatic work order dispatch mode is entered, i.e., the second self-test process is executed; in addition, for boundary cases (such as Sev=L2 but the user fails the self-test twice in a row), the system is automatically upgraded to the work order dispatch mode, and the TTS (Text-to-Speech) engine is called to synthesize P_plain into audio Audio_out, which is played through the speaker of the charging device, and a simplified text prompt corresponding to the colloquial description text P_plain is simultaneously displayed on the screen—that is, the human-computer interaction screen display area of ​​the charging device.

[0049] In some embodiments, executing the first self-test process includes the following specific implementation steps: Based on the fault diagnosis results, a self-test step sequence is generated by calling a preset self-test knowledge base; the self-test step sequence is played through the speaker of the charging device so that the user can operate according to the self-test step sequence; feedback information from the user after each step in the self-test step sequence is obtained through a preset parallel channel corresponding to the charging device, and the execution status information of the self-test step sequence is recorded through an interactive state machine, wherein the execution status information includes the number of times the user failed to execute each step in the self-test step sequence; based on the feedback information and the execution status information, the user's corresponding self-test result is determined; the above steps are repeated multiple times, and if the number of self-test failures is not less than a preset failure number threshold, the second self-test process is executed.

[0050] Continuing with the above embodiments, for faults determined to be self-repairable, a preset self-test knowledge base (encoded as an expert experience decision tree) is invoked, and a personalized self-test step sequence Step_seq=[step_1,step_2,...,step_n] is dynamically generated based on the fault diagnosis results. Each step_i contains five tuples: step_i = (operation instruction Act_i, expected result Exp_i, judgment condition Cond_i, path Pass_i when satisfied, alternative path Fail_i when failed); The meanings of each tuple are as follows: -Act_i: The specific operation guidance text broadcast by TTS, in plain language for end users, without technical jargon; -Exp_i: The phenomenon that the user is expected to observe or the state that the system should reach after executing Act_i; -Cond_i: The judgment condition expression based on user feedback or sensor readings (e.g., ASR recognition keywords belong to {OK, Okay, Work} or gun lock sensor = LOCKED); -Pass_i: The next step number or self-test success exit when Cond_i is satisfied; -Fail_i: The alternative step number or upgrade work order exit when Cond_i is not satisfied. Act_i is broadcast step by step by TTS, and after each step, user confirmation is required (three channels are used in parallel: voice keywords / screen button clicks / automatic advancement after 15 seconds of timeout). If Cond_i=false, the Fail_i adjustment step or upgrade is performed. The entire self-test is recorded in structured O_log, that is, the execution status information (step sequence + feedback + timestamp + result). If the self-test is successful, the user will be prompted to recharge and the subsequent status will be monitored; if the global failure counter fail_count_global is greater than or equal to 3 times, the system will automatically downgrade and transfer the work order to another task – that is, the second self-test process will be executed.

[0051] In some embodiments, the execution of the second self-test process includes the following specific implementation steps: Based on the fault diagnosis results, the visual evidence data, and the operating status data, a work order package is generated. The work order package is used to notify professional maintenance personnel to inspect the charging equipment and display the current status of the charging equipment. After the work order package is dispatched, it is determined that the charging equipment enters a waiting state for maintenance. In response to the repair confirmation signal corresponding to the charging equipment, it is determined that the charging equipment enters a normal charging monitoring state.

[0052] Continuing with the above embodiments, for faults determined to require maintenance intervention (including those directly identified as complex faults and those from self-test upgrades), a complete work order package (Ticket_pkg) is automatically assembled, containing the following: on-site evidence—collected photos of three areas (I_gun, I_conn, I_screen); diagnostic conclusion—the complete content of the output D_result; operational snapshot—six-dimensional time-series data collected by S1 for 60 seconds before and after the fault, i.e., the operational status data; user operation record—generated O_log (i.e., execution status information, if any); device identifier—charging device number, location information, firmware version, etc. The work order package is encrypted and pushed to the maintenance management platform via the OCPP protocol or MQTT message queue. The platform automatically performs the following actions: automatically assigns maintenance personnel to the corresponding skill group (electrical repair group / network group / settlement group, etc.) according to the fault category C; pushes a mobile notification with a work order summary and on-site photos, so that maintenance personnel can understand the full picture of the fault and carry appropriate tools before departure; establishes a two-way communication channel between maintenance and the charging pile, so that maintenance personnel can retrieve more detailed diagnostic data and operation history through a mobile APP after arriving at the site.

[0053] The method is further illustrated below through another specific embodiment: This application provides a charging equipment fault intelligent semantic explanation and voice-guided self-testing system, such as... Figure 2 As shown, it is divided into four levels: Data Acquisition Layer (Layer 1): Responsible for acquiring raw data, comprising two sub-modules—the sensor array sub-module and the visual forensics sub-module. The sensor array sub-module consists of voltage transformers, current transformers (Hall sensors or shunts), an NTC temperature sensor array (covering key temperature measurement points such as power modules, PCBAs, and gun cables), a gun lock position switch sensor, a network status monitoring module (ping / ICMP detection), and a CAN bus sniffer, respectively acquiring six-dimensional runtime sequence data U / I / T / S / QoS / L. The visual forensics sub-module consists of a wide-angle camera module (recommended viewing angle ≥120°, resolution ≥1080p, with night vision supplementary lighting) and a pan / tilt / electric rotation mechanism (optional), performing directional shooting tasks when a fault is triggered.

[0054] The Intelligent Diagnostic Layer (Layer 2) is the core computing unit, comprising a multimodal large-scale model inference engine, a temporal feature extractor, and a visual feature extractor. The temporal feature extractor preprocesses the six-dimensional data within the sliding window (denoising, normalization, feature engineering) to extract statistical features (mean, variance, peak value, kurtosis, etc.) and temporal / frequency domain features (FFT harmonic components, wavelet decomposition coefficients, etc.). The visual feature extractor performs object detection (YOLO / SSD series), defect segmentation (U-Net / Mask R-CNN series), and OCR text recognition on the three-region image, extracting structured visual feature vectors. The two layers are then concatenated and fed into the multimodal large-scale model for joint inference.

[0055] The Semantic Interaction Layer (Layer 3) is responsible for converting diagnostic results into a user-understandable output format. It includes a Natural Language Generation (NLG) module, a Text-to-Speech (TTS) module, and a Human-Machine Interface (HMI) rendering module. The NLG module performs three conversion operations: terminology replacement, colloquialization, and scene adaptation. The TTS module supports Chinese speech synthesis, with a recommended speech rate of 120-150 words / minute (adjustable) and volume of 65-80 dB(A) (adaptively adjustable to ambient noise). The HMI rendering module synchronously displays simplified graphic prompts on the screen (large font size ≥ 24pt, high-contrast color scheme).

[0056] Application Interaction Layer (Layer 4): Responsible for interfacing with external systems, including the self-inspection process engine (B1) and the work order linkage module (B2). The self-inspection process engine maintains a pre-built knowledge base (in the form of a decision tree / rule base) and dynamically orchestrates personalized self-inspection step sequences. The work order linkage module is responsible for work order package assembly, encrypted transmission (TLS 1.3+AES-256), integration with the operation and maintenance platform (OCPP2.0 / MQTT), and the establishment of a two-way communication channel.

[0057] like Figure 3 As shown, this application provides an apparatus whose hardware composition is as follows: Main control unit: Employs an embedded computing platform with AI inference capabilities. Recommended configurations include, but are not limited to: Rockchip RK3588 (6TOPS NPU computing power), NVIDIA Jetson Orin Nano (40TOPS), Horizon Journey 6 (128 TOPS) or equivalent AIoT chips. The main control unit runs an embedded Linux operating system (Ubuntu Core / Buildroot) and is equipped with a multimodal large model inference runtime (ONNX Runtime / TensorRT / MindSporeLite). The recommended model quantization format is INT8 to balance accuracy and inference latency (target latency for a single diagnostic inference ≤ 3 seconds).

[0058] Sensing units: voltage sampling circuit (accuracy ±0.5%), current sampling circuit (Hall closed-loop sensor, accuracy ±1%), temperature acquisition network (NTC 3950, 8-12 channels covering key points), gun lock status detection circuit (micro switch + optocoupler isolation), network monitoring module (4G Cat.1 module + Wi-Fi 6 module dual-link redundancy), CAN transceiver (ISO 11898-2 compatible, used for monitoring vehicle-to-pile communication).

[0059] Visual Unit: Wide-angle camera module. Recommended installation location: above the front panel of the charging device or next to the human-machine interface screen. Tilt angle: 15°-30° downwards to cover the charging head suspension area and screen area. Lens focal length: equivalent to 2.5-4mm; F-number: ≤2.0 to ensure image quality in low-light environments. Equipped with an IR-CUT dual filter switcher to adapt to changes in day and night lighting. Optional infrared supplementary light array (850nm wavelength, does not affect driver visibility).

[0060] Interaction unit: Speaker (3-5W power, frequency response 300Hz-8kHz, meeting the requirements for clear TTS broadcasting), microphone array (2-4 microphones, used for voice command pickup during self-testing, supporting wake-up recognition of keywords such as "okay", "no", "repeat", etc.), touch screen (≥7 inches, brightness ≥800cd / ㎡, supporting gloves touch mode).

[0061] Storage units: eMMC 32GB (system + model files), LPDDR4 / LPDDR5 4GB-8GB (RAM). An additional EEPROM or secure storage area is provided for storing fault history records and work order caches (recommended capacity to retain the most recent 500 fault records).

[0062] Communication unit: 4G / 5G cellular communication module (for work order upload and maintenance platform integration), Ethernet port (PoE+ power supply optional), Bluetooth / BLE 5.0 module (optional, for near-field pairing and debugging). Units are interconnected via internal buses (I²C / SPI / UART / CAN / USB 3.0). The main control unit controls the camera's rotation / focus and the fill light's on / off state via GPIO and PWM.

[0063] Based on the provided intelligent semantic interpretation and voice-guided self-diagnosis system and device for charging equipment faults, this application can achieve accurate fault diagnosis of vehicle charging equipment through the following specific implementation steps, thereby improving the self-resolution rate of simple faults of vehicle charging equipment and reducing the rate of ineffective maintenance: (I) Multi-source data fusion acquisition and fault-triggered visual forensics: like Figure 4As shown, the system continuously acquires a six-dimensional data stream at a sampling frequency of f_s=10Hz. Each sampling period's data frame Frame_k contains: {U_k, I_k, T_k_array, Sk, QoS_k, CAN_msg_k}, where T_k_array is an array of multi-channel temperature values ​​(e.g., [T_pcb, T_module, T_gun, T_ambient]). Data frames are written to a circular buffer RingBuf (capacity 600 frames = 60 seconds of data). When the fault detector (an independently running lightweight rule engine based on threshold judgment and sliding window confirmation mechanism) detects a fault event FaultEvent, it immediately extracts a data window W_data from RingBuf, centered on t_fault, consisting of the first 30 seconds and the last 30 seconds.

[0064] like Figure 4 As shown, the FaultEvent also serves as the Trigger_cam signal for the camera module. After triggering, the system performs three directional shots in sequence: Shot_1 (t+0.5s): The gimbal rotates to the preset position Pos_gun (pitch angle α=-25°, yaw angle β=0°, aligned with the gun head mount), focusing distance d=0.5-0.8m, and takes a picture of I_gun. The image processing pipeline automatically annotates the gun head boundingbox and crops it to a 224×224 input size.

[0065] Shot_2 (t+2.0s): The gimbal rotates to Pos_conn (α=-35°, β=+15° or -15° adaptive according to the left and right direction of the parking space), and the gun-vehicle connection area I_conn is photographed. The algorithm focuses on detecting the gun insertion depth (the length of the gun head inserted into the base) and the interface gap (the degree of visible light transmission).

[0066] Shot_3 (t+3.5s): The gimbal rotates to Pos_screen (α=0°, β=0° facing the screen), capturing the screen display content I_screen. The OCR engine extracts all text content (Text_OCR) and icon layout (Layout_OCR) from the screen. Alternatively, the main controller can directly capture a screenshot of the touchscreen to obtain a photo, or a combination of both methods can be used. After three shots, the system performs an image quality check (sharpness score > threshold, normal exposure, no severe motion blur). Any image that fails to meet the standards is marked as "low confidence" and noted in the diagnostic report. The final output is V_evidence={I_gun, I_conn, I_screen, Text_OCR, Layout_OCR, t_fault, quality_flags}. The entire process, from fault triggering to evidence collection completion, takes less than 5 seconds.

[0067] (II) AI Multimodal Fault Semantic Understanding Engine: Model input construction: W_data is encoded into a temporal feature vector V_time ∈ R^256 by a temporal encoder TimeEncoder (1D-CNN+BiLSTM hybrid architecture); the three images in V_evidence are encoded into visual feature vectors V_gun, V_conn, V_screen ∈ R^384 each by a visual encoder VisEncoder (based on ViT-base or ResNet-50 backbone). The four feature vectors are concatenated into a joint feature vector V_joint = Concat[V_time,V_gun, V_conn, V_screen] ∈ R^(256+384*3)=R^1408.

[0068] Model core: A multimodal large language model (MLLM) is used as the inference core. Recommended base models include: Qwen2.5-VL-7B-Instruct (Alipay Tongyi Qianwen Visual Language Edition): Strong Chinese language capabilities; 7B parameter quantity suitable for edge deployment; INT4 quantization consumes approximately 4GB of VRAM. DeepSeek-VL2 (DeepSeek Visual Language Model): Excellent long context support and outstanding multimodal reasoning capabilities. InternVL2.5 (Shanghai AI Lab): An excellent open-source, commercially viable multimodal model. GLM-4V (Zhipu Qingyan Visual Edition): Domestically developed and controllable, supporting private deployment. The company developed its own lightweight multimodal model: a dedicated model obtained by fine-tuning the above open-source model on a charging device fault scenario dataset using LoRA / RAG. The model input uses a structured Prompt template: [System] You are a senior expert in diagnosing electric vehicle charging equipment malfunctions. Based on the following operational data and on-site photos, you need to determine the cause of the malfunction.

[0069] [Time Series Data]: Voltage curve: {U_desc}; Current curve: {I_desc}; Temperature: {T_desc}; Gun lock: {S_desc}; Network: {QoS_desc}; Communication log summary: {L_summary} [Photos from the scene]: Charging gun head: [Image I_gun]; Connection interface: [Image I_conn]; Screen display: [Image I_screen]; [Requirement] Please output in the following JSON format: { "category": "Pile fault | Vehicle fault | Poor contact | Network | Fees" "confidence": 0.xx, "root_cause_tech":"...", "severity":"L1|L2|L3|L4", “key_evidence”: [...] } Model output parsing: The text response generated by MLLM is parsed into a structured object D_result using a JSON parser. If the output format is not standard (e.g., MLLM produces an illusion and outputs invalid JSON), a retry mechanism is enabled (up to 3 retries, adjusting the few-shot examples in the Prompt each time). The final output includes the five-dimensional classification result C, confidence Conf, technical root cause R_tech, and severity level Sev.

[0070] (III) Fault semantic transformation and hierarchical interactive output, and the specific mechanisms of semantic transformation and hierarchical decision-making: Semantic transformation pipeline (Tech→Plain Pipeline): R_tech is transformed into P_plain after three levels of transformation. Level 1 — Terminology Dictionary Replacement: Maintain a terminology dictionary (TermDict) for charging equipment, with a range of approximately 200-300 entries. For example: { "CC signal loop resistance exceeds threshold": "Charging gun not plugged in properly or metal contacts are dirty". "BMS communication handshake timed out": "Your vehicle and the pile 'talk' timed out." "Insulation resistance reading is low": "There's a leak somewhere, don't touch it." } Level 2 — NLG Fluency Rewriting: The rough sentences after replacement are fed into a lightweight NLG model (such as a Paraphrase model fine-tuned based on T5-small) to generate expressions of moderate length (15-40 words), colloquial style, and friendly tone. The training corpus for the NLG model comes from manually compiled parallel sentence pairs of "professional description → plain language" (approximately 5000 pairs).

[0071] Level 3 — Scene Adaptive Fine-tuning: Based on the context Context={time period, weather prediction (from temperature / light sensitivity), number of consecutive failures}, the output is polished. For example, reassuring statements such as "Don't worry" are added for nighttime scenes, more explanatory information is given for the first failure, and the description is simplified for repeated similar failures.

[0072] Hierarchical interactive decision matrix: Based on a combined query of Sev and C, the decision table DecisionTable is used to determine the output mode OutputMode: Mode A — Voice-guided self-check: Condition = (Sev∈{L1,L2}) AND (C∈{Poor Contact, Abnormal Fees}) → Execute S5, TTS broadcasts P_plain + self-check guidance voice, screen displays "Guiding you through self-troubleshooting" + progress bar. Mode B — Voice Explanation + Suggestion to Change Dock: Condition = (Sev∈{L2}) AND (C∈{Vehicle Fault}) → Only TTS broadcasts P_plain + "We suggest you try a different dock. If the problem persists at other docks, it may be a problem with your vehicle," with a brief explanation displayed on the screen + navigation to nearby available docks (optional). Mode C — Automatic Work Order Dispatch: Condition = (Sev∈{L3,L4}) OR (C∈{Hardware Failure, Network Abnormality and Duration > 5min}) → Execute S6, TTS broadcasts P_plain + "We have arranged for a technician to come and fix it, they will arrive in approximately X hours", the screen displays the work order number + estimated arrival time. Mode D — Upgrade Mode: Condition = (Number of self-check failures ≥ N_fail) → Force conversion to Mode C regardless of the original category. Mode E — Work Order Level Linkage and Upgrade Mechanism Work order levels are divided into four levels: Level 1 (Minor), Level 2 (Moderate), Level 3 (Severe), and Level 4 (Critical). The mapping and escalation rules between these levels and the fault severity level (Sev) are as follows: Initial Sev / Initial Work Order Level / Processing Strategy / Escalation Conditions / Escalated Level L1 Minor / Level_1 / Voice-guided self-check (S5) / Consecutive failures N=3 times / Level_2 L2 Moderate / Level_2 / Voice-guided self-check (S5) / Consecutive failures N=3 times / Level_3 L3 Critical / Level_3 / Direct work order dispatch (ModeC) / — (No self-check) / — L4 Critical / Level_4 / Direct dispatch + emergency marking / — (No self-check) / — Upgrade triggering rules: A global failure counter `fail_count_global` is maintained in the self-test state machine (accumulated across steps). Whenever a step's `Cond_i` condition is not met, `fail_count_global` is incremented. When `fail_count_global` ≥ `N_fail_max` (default = 3), an upgrade is triggered—the table above is consulted to determine the upgrade level. If it ≥ Level_3, the self-test terminates and switches to Mode C for automatic work order dispatch; if it is still Level_2, the self-test continues but the intensity of the voice prompts at each step is increased (e.g., adding motivational statements such as "Try again, please be more careful this time"). Upgrade events are recorded in `O_log` (trigger time, cumulative number of failures, current step, and upgrade / upgrade level).

[0073] Differentiated processing principle: Minor / moderate faults (L1 / L2) give users the opportunity for voice self-check, reflecting the invention purpose of "user self-service and reduced operation and maintenance costs"; serious / critical faults (L3 / L4) are directly dispatched to ensure safety and response speed, and avoid users wasting self-check time in dangerous fault scenarios.

[0074] TTS speech synthesis: Employs an edge TTS engine (such as Coqui TTS / VITS / SenseVoice, etc.), recommends Chinese acoustic models, supports speech rate (adjustable from 0.8x to 1.5x) and volume (adjustable relative dB), and the TTS output is played through an audio amplifier to drive the speaker.

[0075] (iv) Voice-guided progressive self-test engine, the complete voice-guided self-test process: The structure of the self-inspection knowledge base (KnowledgeBase) uses a hybrid representation of decision trees (DT) and production rules (HR). Each self-healing fault type corresponds to a decision tree. The nodes of the tree represent the inspection steps, the edges represent the judgment results (yes / no / uncertain), and the leaf nodes represent three exit points: {successful repair, maintenance upgrade required, try alternative solutions}. Taking the "poor contact" type as an example, its decision tree contains the following typical step sequence: Taking the "poor contact" fault type as an example, the detailed complete sequence of steps is as follows: Re-insertion Confirmation - Act_1: TTS announces, "Please unplug the charging gun and plug it back in. Push it in firmly until you hear a click to lock it in place." - Exp_1: The screen displays "Connected" or similar confirmation text, and the charging gun's electronic lock is engaged. - Cond_1: The gun lock status sensor returns a value of LOCKED, and the user's voice / click feedback is positive (keywords include {OK, Okay, Plug in, Work}). - Pass_1: → Self-test successful exit (prompts the user to restart charging; the system monitors subsequent status). - Fail_1: → Step 2 Cleaning the nozzle's metal contacts - Act_2: TTS announces, "Please use a clean tissue or soft cloth to wipe the silver metal ring at the very front of the nozzle, wiping back and forth two or three times to remove surface dust or black spots." - Exp_2: User confirms wiping is complete - Cond_2: User voice / click feedback is positive (keywords include {done, wiped, clean}), and Cond_1 is satisfied after re-executing Step 1 (inserting the nozzle) - Pass_2: → Self-test successful exit - Fail_2: → Step 3 Visual inspection and dust removal of the vehicle-side charging socket - Act_3: TTS broadcast: "Please check the charging socket in the vehicle. Is there any dust, leaves, or water stains inside? If so, you can blow on it or gently wipe the metal holes of the socket with a dry cloth." - Exp_3: User confirms there are no obvious foreign objects in the vehicle-side interface. - Cond_3: User voice / click feedback is positive (keywords include {looked, clean, wiped, nothing}), and Cond_1 is satisfied after re-executing Step 1 plug-in operation. - Pass_3: → Self-test successful exit. - Fail_3: → Upgrade exit (fail_count_global cumulative ≥3, automatically switch to S6 work order dispatch mode, work order includes complete self-test log + three-area photos + electrical parameter snapshot). Actual maintenance statistics show that dust accumulation, water ingress, and foreign objects inside the vehicle-side socket are frequent causes of poor contact (accounting for approximately 25% to 30% of poor contact faults). Users can visually inspect and perform simple cleaning themselves, demonstrating full feasibility and practical maintenance value.

[0076] Dynamic Step Generation: While the knowledge base provides standard step templates, in actual execution, MLLM will personalize the steps based on the diagnostic details of S3. For example, if visual evidence shows "obvious ablation marks on the nozzle contact," a safety warning will be added before Step 2: "Observe whether the contact has turned black. If so, do not continue using this device." Step adjustments follow the safety-first principle—any situation involving safety hazards should prioritize terminating the self-test and recommending the equipment to be shut down.

[0077] Multi-round interaction mechanism: Each step is executed in a synchronous loop of "TTS broadcast → waiting for user feedback → judgment → next step". User feedback supports three parallel channels: (a) voice keyword recognition ("Okay", "Alright", "Fine" = positive, "No", "Useless", "Still the same" = negative, "Say it again / Didn't hear clearly" = repetition); (b) touch screen virtual button (a persistent button at the bottom of the screen) "Ready" and " (c) Automatic advancement upon timeout (each step waits for 60 seconds; timeout is considered "uncertain" and the default branch is followed). The interactive state machine records the current position current_step, the historical trajectory path_taken, and the failure count failure_count. When failure_count reaches N_fail_max=3, an upgrade is triggered.

[0078] User confirmation is limited to three channels, which are listened to in parallel according to priority: (a) Voice Keyword Confirmation (Highest Priority): The charging device's microphone continuously picks up audio, and the ASR engine transcribes it into text in real time, matching the keyword set. Positive keyword set PosKey = {Okay, can, works, good, done, plugged in, wiped clean}; Negative keyword set NegKey = {No, doesn't work, still doesn't work, same old story, doesn't work}; Repeat request keyword set RepeatKey = {Say it again, didn't hear clearly, didn't understand, say it again}. The above sets support OTA dynamic expansion during system operation.

[0079] (b) Touchscreen virtual button confirmation: Two large touch buttons are permanently located at the bottom of the screen—" "(Positive) and" have been prepared "Still not working" (negative), button size ≥ 48×48dp to ensure no accidental touch by finger operation.

[0080] (c) Automatic Timeout Advancement: The waiting time for each step is Timeout_step = 15 seconds (configurable system parameter, default 15s). After the timeout, the system automatically announces "We have not received your reply, we will continue to the next step" and jumps to the Fail_i path of that step. Timeout is considered a conservative judgment of uncertainty / incompleteness—compared to the conventional 60-second timeout scheme, 15 seconds significantly shortens the user's waiting period, avoids user churn due to prolonged silence, and ensures safety through the conservative strategy of unconfirmed = failure: it is better to mistakenly escalate the order than to make the user wait indefinitely.

[0081] The structured recording specification for self-inspection logs (O_log) – each self-inspection log entry is a JSON document containing the following fields to ensure that fault debriefing and knowledge base optimization are verifiable: { "session_id": "UUID v4 (unique identifier for self-check session)", "pile_id": "Charging device number", "fault_category": "Fault category C (five-dimensional classification value)", "fault_severity": "Severity level (Sev)" "start_time": "Self-test start timestamp (ISO 8601)", "end_time": "Self-check end timestamp", "final_result": "success | upgraded_to_dispatch | user_abandoned","total_steps": "total number of execution steps", "fail_count_global": "Global cumulative number of failures", "steps": [ { "seq": 1, "act_text": "The original text of the operation instructions broadcast by TTS", "act_start": "Step start timestamp", "feedback_type": "voice | button | timeout", "feedback_raw": "Original user feedback (ASR text or button ID, timeout if it expires)", "feedback_parsed": "positive | negative | repeat | timeout", "cond_result": true, "next_step": 2, "act_end": "Step end timestamp", "screenshot_taken": true } ], escalation: { "triggered": true, "trigger_time": "Upgrade trigger timestamp", "fail_count_at_trigger": 3, "from_level": "Level_2", "to_level": "Level_3" } } Log storage and synchronization strategy: O_log is stored locally on the charging device using SQLite, with one record generated for each self-check session. After the self-check is completed (regardless of success / failure / upgrade), the complete O_log is uploaded to the operations and maintenance platform along with the ticket package (Ticket_pkg) or a separate MQTT message (Topic: pile / {pile_id} / selfcheck / log). Successful self-check cases are regularly (e.g., weekly) aggregated and analyzed by the operations and maintenance platform to extract high-frequency success steps and failure patterns, which are then used to optimize the decision tree structure and step wording of the self-check knowledge base (KnowledgeBase).

[0082] (v) Intelligent work order generation and operation and maintenance dispatch linkage, the process of automatic work order generation and operation and maintenance dispatch: The data structure design of the ticket_pkg work order package is as follows: { "ticket_id": "TK-20260427-001234", / / Globally unique ticket number "pile_id": "CP-SHA-PUDONG-042", / / Unique identifier for charging device "fault_time": "2026-04-27T14:23:17+08:00", / / Time of fault occurrence "diagnosis": { / / AI diagnostic conclusion} "category": "Pile hardware failure", "confidence": 0.87, "root_cause_tech": "AC-DC power module IGBT over-temperature (T_module=94.3°C>threshold 85°C)", "root_cause_plain": "A component inside the stud is overheating and needs cooling or replacement." "severity": "L3", "key_evidence": ["Temperature curve abnormally rises", "Cooling fan speed is zero"] }, "evidence": { / / Visual evidence "photo_gun": "base64_encoded_or_url...", / / Photo of the gun barrel "photo_conn": "base64_encoded_or_url...", / / API photo "photo_screen": "base64_encoded_or_url...", / / Screen photo }, "runtime_data": { / / Runtime data snapshot "window_start": "2026-04-27T14:22:47", "window_end": "2026-04-27T14:23:47", "samples": [...], / / 600 frames × 6D data }, "selfcheck_log": {...}, / / Self-check log (if any) "device_info": { / / Basic device information "model": "LC-DC120kW-v3", "firmware": "v2.8.1-build-20260315", "location": {"lat":31.2304, "lng":121.4737, "addr":"B2-A07, Underground Parking Garage, XX, Pudong New Area, Shanghai"} } } Ticket Transmission and Security: The Ticket_pkg file is serialized into JSON and encrypted using the AES-256-GCM algorithm (the key is negotiated with the pre-shared key PSK of the operations and maintenance platform). It is then published via MQTT over TLS 1.3 to the topic Topic=pile / {pile_id} / ticket / new. Message QoS=2 ensures exactly-once delivery. Upon receiving the message, the operations and maintenance platform returns an ACK confirmation. If no ACK is received within the timeout period (default Timeout_ack=30s), the charging device performs backoff and retransmission (exponential backoff, maximum 5 retransmissions).

[0083] Operations and maintenance work order dispatch logic: After receiving a work order, the operations and maintenance platform automatically performs the following actions: (a) Skill group routing: Match the operation and maintenance skill group according to the category field - pile hardware failure → electrical repair group, vehicle failure → transfer to car manufacturer customer service collaboration group, network anomaly → network support group, fee anomaly → settlement customer service group. (b) Personnel scheduling: Among the online maintenance personnel matched with skill groups, sort them based on GPS distance (nearest priority), current load (idle priority), and historical expertise (priority given to those with similar fault repair records), and push the top-1 work orders; (c) Mobile presentation: When maintenance personnel receive a push notification on their mobile APP, they can directly view the work order summary (one-sentence summary + three thumbnails + fault level label) after opening it. Clicking to expand will allow them to view the complete diagnostic report, operation curves and charts and high-resolution on-site photos. (d) Two-way communication: After receiving the work order, the maintenance personnel establish a real-time two-way channel with the charging equipment (bridged via MQTT WebSocket), and can remotely retrieve more diagnostic data and make voice calls with on-site users while en route (relayed via pile-end microphone + speaker). (e) On-site handling and closed-loop: After the maintenance personnel arrive, they check in via the APP → handle the issue → fill in the results → take a photo and upload it → close the work order. The shutdown signal is sent back to the charging equipment side, the fault status at the charging pile is updated to "repaired" and the current alarm is cleared.

[0084] In summary, this application provides a method for fault diagnosis of charging equipment, which can achieve accurate fault diagnosis of vehicle charging equipment, thereby improving the self-resolution rate of simple faults of vehicle charging equipment and reducing the rate of ineffective maintenance.

[0085] To better implement the above methods, embodiments of this application also provide a charging system, such as... Figure 5 As shown, the present invention provides a charging system including at least two power modules 110, a controller 130, a power distribution device 140, and at least one charging interface 120. The power distribution device is connected to the controller, each power module, and each charging interface. The power module 110 is used to convert AC power from the power grid into DC power and supply it to the charging interface. The controller is used to obtain the power demand of each charging interface and generate scheduling instructions based on the connection relationship of the controllable switches in the power distribution device and the power demand. The power distribution device is used to control the opening or closing of the controllable switches according to the scheduling instructions to distribute the output power of each power module to each charging interface.

[0086] In one optional implementation, the charging system provided in this application is an integrated DC charging device. The charging interface 120 is used to connect the charging gun, and the charging gun is connected to the host of the charging system through the gun holder on the main body of the charging system.

[0087] In one optional implementation, the charging system provided in this application is a split-type DC charging device. The charging system also includes multiple charging terminals. The charging interface 120 is used to connect the charging terminals. The charging terminals are set separately from the main body of the charging system. The charging terminals are equipped with a single charging gun or dual charging guns for outputting power to electric vehicles.

[0088] The above provides a detailed description of a charging device fault detection method and charging system provided by the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for troubleshooting charging equipment, characterized in that, Applied to a charging system, the method includes: Acquire the operating status data of the charging device within a preset time window; In response to a fault event trigger signal, images are acquired from a preset image acquisition area corresponding to the charging device to obtain visual evidence data corresponding to the charging device. Based on the operational status data and the visual evidence data, a pre-set multimodal diagnostic model is used to diagnose the charging equipment and obtain the corresponding fault diagnosis results.

2. The method as described in claim 1, characterized in that, The operating status data is a time-series data window, and the preset time window is used to represent the time period from a preset time before the charging device's failure time to the current time. The acquisition of the charging device's operating status data within a preset time window includes: Based on a preset sampling frequency, the built-in sensor group collects the output voltage, output current, internal temperature of the charging device, charging gun lock status data, network link quality data, and vehicle-charging pile communication log data of the charging device. The timing data window is generated based on the output voltage, the output current, the pile temperature, the charging gun lock-up status data, the network link quality data, and the vehicle-pile communication log data.

3. The method as described in claim 1, characterized in that, The preset image acquisition area includes a first area, a second area, and a third area. The first area is used to characterize the area where the charging gun and the vehicle interface cooperate. The second area is used to characterize the overall appearance area of ​​the charging gun body and the cable. The third area is used to characterize the human-machine interaction screen display area of ​​the charging device. In response to a fault event trigger signal, the system acquires images of a preset image acquisition area corresponding to the charging device to obtain visual evidence data corresponding to the charging device, including: A first image corresponding to the first region is acquired, wherein the first image is used to characterize the insertion depth of the charging gun, the attitude of the charging gun body, and the state of the cable body. Acquire a second image corresponding to the second region, wherein the second image is used to characterize the condition of the gun body casing and the cable sheath. A third image corresponding to the third region is acquired, wherein the third image is used to characterize the display area of ​​the human-computer interaction screen; The visual evidence data is generated based on the first image, the second image, the third image, the shooting timestamp corresponding to the first image, the shooting timestamp corresponding to the second image, the shooting timestamp corresponding to the third image, and the fault triggering timestamp corresponding to the fault event triggering signal.

4. The method as described in claim 3, characterized in that, The preset multimodal diagnostic model includes an electrical anomaly pattern recognition layer, a visual physical state analysis layer, and a cross-modal joint reasoning layer. The fault diagnosis result includes fault category, fault root cause description data, and fault severity level. The fault root cause description data is used to characterize the judgment result text of the technical root cause of the fault. Based on the operational status data and the visual evidence data, a pre-set multimodal diagnostic model is used to diagnose faults in the charging equipment, resulting in fault diagnosis results for the charging equipment, including: The electrical anomaly pattern recognition layer is used to identify anomalies in the operating status data to obtain anomaly identification results. The visual physical state analysis layer performs image analysis on the visual evidence data to obtain image analysis results. The cross-modal joint inference layer correlates the anomaly identification results and the image analysis results to obtain the fault diagnosis results.

5. The method as described in claim 4, characterized in that, Before executing a preset fault self-test strategy based on the fault diagnosis result to obtain the target self-test result, the method further includes: The electrical anomaly pattern recognition layer extracts the CC circuit resistance value of the charging gun contact from the operating status data. The anomaly identification result is determined based on the voltage amplitude of the charging gun contact CP signal, the resistance value of the charging gun contact CC circuit, and the preset electrical threshold in the operating status data. Based on the anomaly identification results, the first image, and the second image, the charging device is fault diagnosed using the preset multimodal diagnostic model to obtain the fault diagnosis results.

6. The method as described in claim 4, characterized in that, After obtaining the fault diagnosis result of the charging device by performing fault diagnosis on the charging device using a preset multimodal diagnostic model based on the operating status data and the visual evidence data, the method further includes: By using a preset natural language generation model, the fault root cause description data is processed by replacing professional terms, rewriting in colloquial language, and adapting to the scene to obtain colloquial language description text, so as to display the colloquial language description text on the human-computer interaction screen display area of ​​the charging device. The colloquial language description text is used to display fault information to non-professionals. Based on the colloquial language description text, an audio message is generated to enable the colloquial language description text to be played through the speaker of the charging device; Based on the fault category and the fault severity level, it is determined whether to execute a first self-test process or a second self-test process. The first self-test process is used to perform a self-test on the charging equipment through voice guidance, and the second self-test process is used to determine whether to transfer to the work order dispatch process based on the self-test results.

7. The method as described in claim 6, characterized in that, The fault categories include self-repairable and non-self-repairable categories, and the fault severity levels include minor, moderate, severe, and critical. The step of determining whether to execute the first self-test process or the second self-test process based on the fault category and the fault severity level includes: If the fault category of the charging device is self-repairable, and the fault severity level of the charging device is minor or moderate, then the first self-test process is executed. If the fault category of the charging device is non-self-repairable, or if the fault severity level of the charging device is severe or critical, then the second self-test process will be executed.

8. The method as described in claim 7, characterized in that, The execution of the first self-test process includes: Based on the fault diagnosis results, a self-test step sequence is generated by calling a preset self-test knowledge base; The self-test sequence is played through the speaker of the charging device so that the user can operate according to the self-test sequence; The system obtains user feedback information after each step in the self-test sequence through a preset parallel channel corresponding to the charging device, and records the execution status information of the self-test sequence through an interactive state machine. The execution status information includes the number of times the user failed to execute each step in the self-test sequence. Based on the feedback information and the execution status information, the user's self-inspection result is determined; Repeat the above steps multiple times. If the number of self-check failures is not less than the preset failure threshold, then execute the second self-check process.

9. The method as described in claim 7, characterized in that, The execution of the second self-test process includes: Based on the fault diagnosis results, the visual evidence data, and the operating status data, a work order package is generated, wherein the work order package is used to notify professional operation and maintenance personnel to inspect the charging equipment and display the current status of the charging equipment; After the work order package is dispatched, it is determined that the charging equipment enters a state of waiting for maintenance and handling. In response to the repair confirmation signal corresponding to the charging device, it is determined that the charging device has entered the normal charging monitoring state.

10. A charging system, characterized in that, The system includes at least two power modules, at least one charging interface, a power distribution device, a main controller, and a charging interface; the main controller is connected to each of the power modules, each of the charging interfaces, and the power distribution device, and is used to execute the charging equipment fault detection method as described in any one of claims 1 to 9; the power distribution device is connected to each of the power modules and each of the charging interfaces; the charging interface is used to connect a charging gun, and the charging gun is hung on the corresponding host of the charging system via a gun holder on the main body of the charging system.