Fault diagnosis report generation method and related equipment
By locally deploying the initial large model and knowledge base, and optimizing the model and report templates based on user feedback, we resolved network latency and security issues caused by cloud-based reliance, and achieved efficient and secure fault diagnosis report generation.
Patent Information
- Application Number
- CN202510901179.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-30
- Publication Date
- 2025-09-26
AI Technical Summary
Existing fault diagnosis report generation methods rely on cloud processing, which has problems such as poor network stability, low security and poor user experience.
By locally deploying the initial large model and knowledge base, and combining user feedback to optimize the model and report templates, a fault diagnosis report that meets user needs is generated.
It improves the efficiency, security, and user experience of fault diagnosis report generation, ensures that data is processed locally, and avoids network delays and data leakage.
Smart Images
Figure CN120704933A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present application relate to the field of fault diagnosis report generation, and more specifically, to a fault diagnosis report generation method, a fault diagnosis report generation apparatus, a fault diagnosis report generation device, a computer-readable storage medium, and a computer program product containing instructions. Background Art
[0002] Existing fault diagnosis report generation methods primarily rely on networked backend analysis systems, uploading vehicle fault data to the cloud for processing and returning diagnostic results as interface data. While this approach leverages the cloud's powerful computing power and abundant data resources, it also has some significant limitations. First, poor network stability, such as in remote areas, can lead to report delays or even inability to generate reports, reducing repair efficiency and impacting user experience. Second, data uploaded to the cloud is susceptible to leakage and tampering, posing security risks.
[0003] In summary, the fault diagnosis report generation efficiency of the existing technology is low, the security is low, and the user experience is poor. Summary of the Invention
[0004] The embodiments of the present application provide a fault diagnosis report generation method, a fault diagnosis report generation device, a fault diagnosis report generation equipment, a computer-readable storage medium, and a computer program product containing instructions, which can generate fault diagnosis reports while improving the security, efficiency, accuracy, and personalized display of fault diagnosis reports.
[0005] In a first aspect, an embodiment of the present application provides a method for generating a fault diagnosis report, comprising:
[0006] Inputting initial fault information into a locally deployed initial large model, the initial large model determining, from an initial knowledge base, initial fault diagnosis knowledge items whose matching degree with the initial fault information reaches a preset matching degree threshold, and obtaining, based on the initial fault diagnosis knowledge items, a predicted fault diagnosis result of the initial fault information output by the initial large model;
[0007] generating, based on the initial fault information and the predicted fault diagnosis result of the initial fault information, and displaying the initial fault diagnosis result report on an interface according to an initial report format template, and obtaining target feedback information from a user regarding the initial fault diagnosis result report, wherein the target feedback information represents user feedback information on report display content and report display format, and the feedback information on report display content includes preference feedback information and / or correction feedback information;
[0008] Adjusting the initial large model based on the feedback information of the report display content to obtain a locally deployed target large model, and adjusting the initial report format template based on the feedback information of the report display format to obtain a target report format template;
[0009] Inputting target fault information into the target macro model, the target macro model determining, from a target knowledge base, target fault diagnosis knowledge items whose matching degree with the target fault information reaches a preset matching degree threshold, and obtaining, based on the target fault diagnosis knowledge items, a predicted fault diagnosis result of the target fault information output by the target macro model;
[0010] According to the target report format template, a target fault diagnosis result report is generated based on the target fault information and the predicted fault diagnosis result of the target fault information and is displayed on the interface.
[0011] In a second aspect, an embodiment of the present application provides a fault diagnosis report generating device, comprising:
[0012] an initial large model processing unit, configured to input initial fault information into a locally deployed initial large model, having the initial large model determine, from an initial knowledge base, initial fault diagnosis knowledge items whose matching degree with the initial fault information reaches a preset matching degree threshold, and obtain, based on the initial fault diagnosis knowledge items, a predicted fault diagnosis result for the initial fault information output by the initial large model;
[0013] an acquisition unit, configured to generate and display an initial fault diagnosis result report on an interface based on the initial fault information and the predicted fault diagnosis result of the initial fault information in accordance with an initial report format template, and to acquire target feedback information from a user regarding the initial fault diagnosis result report, wherein the target feedback information represents user feedback information on report display content and report display format, and the feedback information on report display content includes preference feedback information and / or correction feedback information;
[0014] a large model adjustment unit, configured to adjust the initial large model based on the feedback information of the report display content to obtain a locally deployed target large model, and adjust the initial report format template based on the feedback information of the report display format to obtain a target report format template;
[0015] a target large model processing unit, configured to input target fault information into the target large model, having the target large model determine, from a target knowledge base, target fault diagnosis knowledge items whose matching degree with the target fault information reaches a preset matching degree threshold, and obtain, based on the target fault diagnosis knowledge items, a predicted fault diagnosis result of the target fault information output by the target large model;
[0016] A generating unit is configured to generate and display a target fault diagnosis result report on an interface based on the target fault information and the predicted fault diagnosis result of the target fault information in accordance with the target report format template.
[0017] In a third aspect, an embodiment of the present application provides a fault diagnosis report generating device, comprising:
[0018] CPU, memory, input and output interfaces, wired or wireless network interfaces, and power supply;
[0019] The memory is a short-term storage memory or a persistent storage memory;
[0020] The central processing unit is configured to communicate with the memory and execute instructions in the memory to perform the aforementioned fault diagnosis report generating method.
[0021] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which includes instructions. When the instructions are executed on a computer, the computer executes the aforementioned fault diagnosis report generation method.
[0022] In a fifth aspect, an embodiment of the present application provides a computer program product comprising instructions, which, when the computer program product is run on a computer, enables the computer to execute the aforementioned fault diagnosis report generation method.
[0023] It can be seen from the above technical solution that the embodiment of the present application has the following advantages: through the locally deployed initial large model and knowledge base, the model and report template are optimized in combination with user feedback, and finally a target fault diagnosis report that better meets user needs is generated, thereby improving the efficiency, security and user experience of fault diagnosis report generation. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 A flowchart of a method for generating a fault diagnosis report disclosed in an embodiment of the present application;
[0025] Figure 2 A schematic structural diagram of a fault diagnosis report generating device disclosed in an embodiment of the present application;
[0026] Figure 3 This is a structural diagram of a fault diagnosis report generating device disclosed in an embodiment of the present application. DETAILED DESCRIPTION
[0027] Embodiments of the present application provide a fault diagnosis report generation method, a fault diagnosis report generation device, a fault diagnosis report generation equipment, a computer-readable storage medium, and a computer program product containing instructions, which are used to generate fault diagnosis reports while improving the safety, efficiency, accuracy, and personalized display of fault diagnosis reports.
[0028] See also Figure 1 , Figure 1 This is a flow chart of a method for generating a fault diagnosis report disclosed in an embodiment of the present application, the method comprising:
[0029] 101. Input the initial fault information into the locally deployed initial large model, and the initial large model determines the initial fault diagnosis knowledge items whose matching degree with the initial fault information reaches a preset matching degree threshold from the initial knowledge base, and obtains the predicted fault diagnosis result of the initial fault information output by the initial large model based on the initial fault diagnosis knowledge items.
[0030] In an optional implementation, the method for inputting the initial fault information into the locally deployed initial large model may be to first pre-process the initial fault information and then input the pre-processed initial fault information into the locally deployed initial large model.
[0031] Specifically, preprocessing can involve cleaning, formatting, and labeling the collected raw information to remove noise and invalid information, and to identify the fault type and location. Initial fault information consists of vehicle operating information, fault codes, maintenance records, and driving behavior information collected in real time from the vehicle's sensor network and diagnostic interfaces (such as the OBD interface). The fault type refers to the type of fault, such as engine failure, transmission failure, or brake system failure, and is a classification of the nature of the fault. The fault location refers to the specific location of the fault, such as the engine cylinder, transmission gear set, or brake disc. It identifies the specific location of the fault within the device or system. A fault code is a code used to identify a specific fault, typically generated by the electronic control system. For example, when the engine check light illuminates, the corresponding fault code might be P0300 (indicating engine misfire). It provides a concise fault identification method that facilitates quick identification of the fault condition. The initial knowledge base includes, but is not limited to, repair manuals, expert experience, maintenance guidelines, repair cases, and repair diagnostics. Initial fault diagnosis knowledge entries refer to content items in the initial knowledge base related to the initial fault information. They contain diagnostic knowledge about the fault, such as the fault phenomenon, possible causes, fault type, and fault location. Predicted fault diagnosis results are the results derived by the initial large-scale model based on the initial fault diagnosis knowledge entries. These include the determination of the fault type, location of the fault location, and possible causes. For example, suppose the initial fault information (including vehicle type and fault code) characterizes a Class A vehicle with underpowered engine and abnormal exhaust emissions, resulting in a black exhaust gas. The initial large-scale model searches the initial knowledge base for highly matching knowledge entries, such as knowledge entries on engine misfire and fuel injection system failure. Based on these knowledge entries, the initial large-scale model predicts that the fault may be an engine misfire and the fault location may be the ignition system or the fuel injector, thus forming a predicted fault diagnosis result. Deploying the large-scale model locally on the diagnostic device enables intelligent diagnosis of fault information and generation of fault diagnosis report results, without relying on a network connection. Furthermore, completing the diagnosis process locally significantly reduces the time required to generate the fault diagnosis report, enabling faster response times and improving user experience.
[0032] 102. According to the initial report format template, an initial fault diagnosis result report is generated and displayed on the interface based on the initial fault information and the predicted fault diagnosis result of the initial fault information, and the user's target feedback information for the initial fault diagnosis result report is obtained. The target feedback information represents the user's feedback information on the report display content and the report display format. The feedback information on the report display content includes preference feedback information and / or correction feedback information.
[0033] In an optional embodiment, the initial fault diagnosis report includes a fault description (such as insufficient engine power and abnormal exhaust emissions), a predicted fault type (such as engine misfire), a predicted fault location (such as the ignition system or fuel injectors), and repair recommendations. The interface allows users to view the diagnostic report, customize the report display format, and provide feedback on the fault diagnosis results (such as thumbs-up, thumbs-down, and corrections). Report content feedback refers to user evaluations and opinions on the report content, including accuracy, detail, and relevance. For example, users may report that the fault analysis in the report is not in-depth enough, or that the repair recommendations are not specific enough. Report format feedback refers to user evaluations and opinions on the report presentation, including the report layout, use of charts, and color scheme. For example, users may report that charts in the report are not clear enough, or that the text size is inappropriate. Preference feedback refers to user preferences for report content or format, indicating which sections or formats the user prefers. For example, users may report that they prefer charts to present data, that they pay more attention to the repair recommendations section, or that they prefer videos to present the fault analysis results. Correction feedback refers to corrections to incorrect or inaccurate fault diagnosis results in a report by the user, providing more accurate or appropriate information. For example, a user may provide a corrected diagnosis instead of an incorrect analysis of the cause and solution for a fault code in a feedback report.
[0034] In an optional embodiment, the generated predicted fault diagnosis results and initial fault diagnosis result report of the initial fault information are stored on a local device to prevent data from being intercepted or tampered with during transmission, thereby ensuring data security and privacy. Furthermore, the stored predicted fault diagnosis results and initial fault diagnosis result report of the initial fault information can be encrypted to ensure data security and privacy.
[0035] 103. Adjust the initial large model based on the feedback information of the report display content to obtain the locally deployed target large model, and adjust the initial report format template based on the feedback information of the report display format to obtain the target report format template.
[0036] In an optional embodiment, the initial large model is continuously adjusted (including but not limited to optimization, fine-tuning, etc.) based on user feedback to improve the accuracy and reliability of diagnosis.
[0037] 104. Input the target fault information into the target macro model, and the target macro model determines the target fault diagnosis knowledge items from the target knowledge base whose matching degree with the target fault information reaches a preset matching degree threshold, and obtains the predicted fault diagnosis result of the target fault information output by the target macro model based on the target fault diagnosis knowledge items.
[0038] In an optional embodiment, the target knowledge base is a knowledge base associated with the target large model after the initial large model is adjusted. It may contain knowledge content optimized according to user feedback, which is more in line with the user's needs for diagnostic reports. It should be understood that the initial knowledge base is the knowledge base used by the initial large model, and the content may be relatively basic or general. The target knowledge base is updated and optimized based on the initial knowledge base and the user's feedback on the initial fault diagnosis result report, so that the knowledge content is more in line with the user's actual needs and preferences. For example, if the user feedback is that some fault diagnosis knowledge items in the initial knowledge base are not in-depth enough in the analysis of the cause of the fault, the target knowledge base will supplement more detailed and accurate fault cause analysis content based on the feedback information.
[0039] 105. According to the target report format template, a target fault diagnosis result report is generated based on the target fault information and the predicted fault diagnosis result of the target fault information and is displayed on the interface.
[0040] In an optional embodiment, for example, after the initial fault diagnosis result report displays vehicle information, fault description, diagnostic results, fault analysis, maintenance suggestions, cost estimates, and other content, the user can click to hide or turn off the display of a certain functional content in the initial fault diagnosis result report. After the large model receives the user's feedback information, it can automatically adjust to the report format that the user wants to present when subsequently outputting the target fault diagnosis result report, and can personalize the report content (the personalized generated diagnostic report is rich in content and easy to understand), thereby improving the user experience. In this way, the embodiment of the present application optimizes the model and report template through the locally deployed initial large model and knowledge base, combined with user feedback, and ultimately generates a target fault diagnosis report that better meets user needs, thereby improving the accuracy, efficiency, and user experience of fault diagnosis.
[0041] In an optional embodiment, the generated predicted fault diagnosis results and target fault diagnosis result reports of the target fault information are stored in a local device to prevent the data from being intercepted or tampered with during transmission, thereby ensuring the security and privacy of the data.
[0042] In an optional implementation, the initial large model is adjusted based on feedback from report display content to obtain a locally deployed target large model. This includes: utilizing user feedback to construct positive and negative sample pairs for comparative learning. Reports confirmed by users as correct diagnoses are used as positive samples, while reports with incorrect diagnoses are used as negative samples. Using a contrastive learning algorithm, the model is able to better distinguish feature differences between positive and negative samples, thereby optimizing the model's diagnostic capabilities. For example, using a contrastive loss function to increase the distance between positive and negative samples in feature space enables the model to more accurately identify fault features in subsequent diagnosis.
[0043] In an optional embodiment, the initial large model is adjusted through feedback information of the report display content to obtain a locally deployed target large model, including: determining the target features and the target weights of the target features through feedback information of the report display content, and obtaining the locally deployed target large model based on the determined target features and the target weights of the target features.
[0044] Specifically, large models can use feedback from report content as training signals, leveraging machine learning techniques such as supervised learning and reinforcement learning to adjust the model. For example, if a user reports an inaccurate prediction result, the model can adjust internal parameters (such as neural network weights and biases) to ensure that the output more closely matches the user's expectations in subsequent similar situations. This ensures that the model can be optimized based on user needs, enhancing its adaptability and accuracy.
[0045] In an optional embodiment, the target feature and the target weight of the target feature are determined by reporting feedback information of the display content, including: determining a first feature corresponding to the display content that the user likes and a second feature corresponding to the display content that the user dislikes based on the preference feedback information, and increasing the weight of the first feature and decreasing the weight of the second feature, wherein the first feature and the second feature are target features, and the increased weight of the first feature and the decreased weight of the second feature are the target weights of the target features; and / or determining a third feature corresponding to the display content required to correct the fault information and the target weight of the third feature based on the correction feedback information, the third feature is the target feature, and the target weight of the third feature is the target weight of the target feature.
[0046] Specifically, revised fault information includes new and / or old fault information. The third feature corresponding to the new fault information refers to features introduced by the new fault information that were previously under-recognized by the model. For example, if the user indicates that the new fault location is the spark plug (new fault information), then the fault location "spark plug" is a newly-appeared feature (the third feature). This is because the model may not have previously adequately associated the spark plug with this fault type. The third feature corresponding to the old fault information is the revised feature. For example, if the old fault information is about the intake system, the revised feedback will inform the model that the focus should actually be on the ignition system. The "ignition system" feature (the third feature) is now revised and strengthened. This helps the model more accurately identify and associate the correct fault feature in subsequent fault diagnosis. The target weight of a target feature represents the importance of that feature for correct fault diagnosis. A higher weight means the model will pay more attention to that feature when processing similar fault information. For example, if the user's preference feedback indicates that they pay more attention to the repair suggestion section (the first feature) and less attention to the fault code explanation section (the second feature), the model will increase the weight of features related to the repair suggestion and decrease the weight of features related to the fault code explanation. This allows the maintenance recommendations to be more detailed and prominent when generating a fault diagnosis report, while the fault code explanation may be simplified or omitted. Secondly, user correction feedback indicates an inaccurate determination of the fault location in the report. Based on this feedback, the model determines that the fault location (the third feature) requires correction. Assigning accurate target weights to the fault location feature enables the model to more accurately diagnose and display the fault location when generating subsequent reports. By further refining how to determine target features and their weights based on user feedback preferences and correcting this feedback, the model can more accurately meet users' personalized needs, improving the relevance and practicality of diagnostic reports.
[0047] In one optional implementation, target features and their target weights are determined based on feedback from the report display. This includes constructing user profiles based on historical user behavior, preferences, vehicle type, and other information. Different target features and corresponding target weights are then determined for different user profiles. For example, a professional repairman may be more interested in the detailed technical details of fault diagnosis, such as an in-depth interpretation of fault codes and the precise location of the fault. Meanwhile, a regular car owner may be more interested in repair recommendations and cost estimates. For example, suppose two user profiles are constructed: one for a professional repairman and one for a regular car owner. For the professional repairman profile, "in-depth interpretation of fault codes," "precise location of the fault," and "detailed classification of fault types" are determined as target features, with weights of 0.3, 0.3, and 0.2, respectively. For the regular car owner profile, "repair recommendations," "cost estimate," and "a brief description of the fault cause" are determined as target features, with weights of 0.4, 0.3, and 0.2, respectively.
[0048] In an optional embodiment, obtaining the user's target feedback information for the initial fault diagnosis result report includes: obtaining preference feedback information based on the user's review status and preference evaluation information corresponding to different display contents of the initial fault diagnosis result report in the interface, and obtaining correction feedback information based on the correction information of the display content input by the user in the interface. Specifically, for example: when viewing the fault diagnosis report, the user frequently reads the maintenance suggestion section and gives it a positive comment (preference feedback), and at the same time enters correction information in the report, pointing out that the fault type judgment is incorrect (correction feedback). The system obtains the user's preference feedback and correction feedback information accordingly. In this way, clarifying how to obtain preference feedback information and correction feedback information from the user's behavior and input in the interface provides a practical and feasible basis for model optimization, ensuring that feedback information can be effectively used to improve the model and report.
[0049] In an optional embodiment, before inputting the target fault information into the locally deployed target large model, the method further includes: obtaining a trained initial large model, the trained initial large model is obtained by training with the initial training samples, regularly transforming and expanding the initial training samples through the data enhancement algorithm to obtain target training samples, the target training samples include newly added fault information samples of new fault types and corresponding labeled fault diagnosis result samples, regularly retraining the trained initial large model through the target training samples to obtain an optimized initial large model, and obtaining a locally deployed initial large model based on the optimized initial large model. Specifically, the transformation includes but is not limited to adding noise, changing certain parameter ranges of fault characteristics, etc. In this way, by regularly performing data enhancement and retraining on the initial training samples, the initial large model is continuously updated and optimized to adapt to new fault types and diagnostic requirements, and to maintain the performance and accuracy of the model.
[0050] In an optional embodiment, before obtaining the trained initial large model, the method further includes: obtaining local initial training samples, the initial training samples including initial fault information samples of various fault types, and each initial fault information sample is annotated with a corresponding initial fault diagnosis result sample; locally inputting the initial training samples into the initial large model; the initial large model determining, from the initial knowledge base, initial fault diagnosis knowledge item samples whose matching degree with the initial fault information samples reaches a preset matching degree threshold; and obtaining, based on the initial fault diagnosis knowledge item samples, a predicted fault diagnosis result of the initial fault information sample output by the initial large model; when the loss between the predicted fault diagnosis result of the initial fault information sample and the annotated initial fault diagnosis result sample meets the convergence condition, obtaining the trained initial large model, and saving the trained initial large model locally. In this way, the initial large model is ensured to be fully trained and verified locally before deployment, and whether the model training is complete is determined by comparing the loss between the predicted result and the annotated result, thereby ensuring the quality and reliability of the initial large model.
[0051] In one optional embodiment, obtaining a locally deployed target large model based on the optimized initial large model includes: converting high-precision values in the optimized initial large model to low-precision values to obtain a quantized initial large model; removing unimportant weights or connections in the quantized initial large model to obtain a pruned initial large model, wherein the pruned initial large model serves as the locally deployed target large model. In this way, the optimized initial large model is lightweighted through quantization and pruning techniques, making it more suitable for efficient operation on local devices, reducing storage and computing resource requirements, and improving model operation efficiency.
[0052] For further information, see Figure 2An embodiment of the fault diagnosis report generating device in the embodiment of the present application includes:
[0053] an initial large model processing unit, configured to input initial fault information into a locally deployed initial large model, having the initial large model determine, from an initial knowledge base, initial fault diagnosis knowledge items whose matching degree with the initial fault information reaches a preset matching degree threshold, and obtain, based on the initial fault diagnosis knowledge items, a predicted fault diagnosis result for the initial fault information output by the initial large model;
[0054] an acquisition unit, configured to generate and display an initial fault diagnosis result report on an interface based on the initial fault information and the predicted fault diagnosis result of the initial fault information in accordance with an initial report format template, and to acquire target feedback information from a user regarding the initial fault diagnosis result report, wherein the target feedback information represents user feedback information on report display content and report display format, and the feedback information on report display content includes preference feedback information and / or correction feedback information;
[0055] a large model adjustment unit, configured to adjust the initial large model based on the feedback information of the report display content to obtain a locally deployed target large model, and adjust the initial report format template based on the feedback information of the report display format to obtain a target report format template;
[0056] a target large model processing unit, configured to input target fault information into the target large model, having the target large model determine, from a target knowledge base, target fault diagnosis knowledge items whose matching degree with the target fault information reaches a preset matching degree threshold, and obtain, based on the target fault diagnosis knowledge items, a predicted fault diagnosis result of the target fault information output by the target large model;
[0057] A generating unit is configured to generate and display a target fault diagnosis result report on an interface based on the target fault information and the predicted fault diagnosis result of the target fault information in accordance with the target report format template.
[0058] The large model adjustment unit is specifically used to determine the target feature and the target weight of the target feature through feedback information of the report display content, and obtain the locally deployed target large model based on the determined target feature and the target weight of the target feature.
[0059] The large model adjustment unit is specifically used to determine the first feature corresponding to the display content that the user likes and the second feature corresponding to the display content that the user does not like based on the preference feedback information, and increase the weight of the first feature and reduce the weight of the second feature, wherein the first feature and the second feature are the target features, and the increased weight of the first feature and the reduced weight of the second feature are the target weights of the target features; and / or determine the third feature corresponding to the display content required to correct the fault information and the target weight of the third feature based on the correction feedback information, the third feature is the target feature, and the target weight of the third feature is the target weight of the target feature.
[0060] The acquisition unit is specifically used to obtain the preference feedback information based on the user's review status and preference evaluation information corresponding to the different display contents of the initial fault diagnosis result report in the interface, and to obtain the correction feedback information based on the correction information of the display content input by the user in the interface.
[0061] The acquisition unit is also used to acquire the trained initial large model, which is obtained by training with initial training samples. The initial training samples are regularly transformed and expanded by a data enhancement algorithm to obtain target training samples. The target training samples include newly added fault information samples of new fault types and corresponding labeled fault diagnosis result samples. The trained initial large model is regularly retrained with the target training samples to obtain an optimized initial large model, so as to obtain a locally deployed initial large model based on the optimized initial large model.
[0062] The acquisition unit is further used to acquire local initial training samples, wherein the initial training samples include initial fault information samples of various fault types, and each initial fault information sample is annotated with a corresponding initial fault diagnosis result sample. The initial training samples are input into the initial large model locally, and the initial large model determines, from the initial knowledge base, initial fault diagnosis knowledge item samples whose matching degree with the initial fault information samples reaches a preset matching degree threshold, and obtains the predicted fault diagnosis result of the initial fault information sample output by the initial large model based on the initial fault diagnosis knowledge item samples. When the loss between the predicted fault diagnosis result of the initial fault information sample and the annotated initial fault diagnosis result sample meets the convergence condition, the trained initial large model is obtained, and the trained initial large model is saved locally.
[0063] The acquisition unit is specifically used to convert high-precision values in the optimized initial large model into low-precision values to obtain a quantized initial large model, and remove unimportant weights or connections in the quantized initial large model to obtain a pruned initial large model, where the pruned initial large model is the locally deployed target large model.
[0064] For further information, see Figure 3 , an embodiment of the fault diagnosis report generating device in the embodiment of the present application includes:
[0065] CPU 301, memory 305, input / output interface 304, wired or wireless network interface 303 and power supply 302;
[0066] The memory 305 is a temporary storage memory or a permanent storage memory;
[0067] The CPU 301 is configured to communicate with the memory 305 and execute the instructions in the memory 305 to perform the aforementioned Figure 1 The method in the embodiment shown.
[0068] The embodiment of the present application also provides a computer-readable storage medium, which includes instructions. When the instructions are executed on a computer, the computer executes the aforementioned Figure 1 The method in the embodiment shown.
[0069] The present application also provides a computer program product comprising instructions, which, when executed on a computer, causes the computer to execute the aforementioned Figure 1 The method in the embodiment shown.
[0070] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.
[0071] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0072] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.
[0073] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.
[0074] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.
[0075] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), disk or optical disk, and other media that can store program code.
Claims
1. A method for generating a fault diagnosis report, characterized in that: include: Inputting initial fault information into a locally deployed initial large model, the initial large model determining, from an initial knowledge base, initial fault diagnosis knowledge items whose matching degree with the initial fault information reaches a preset matching degree threshold, and obtaining, based on the initial fault diagnosis knowledge items, a predicted fault diagnosis result of the initial fault information output by the initial large model; generating, based on the initial fault information and the predicted fault diagnosis result of the initial fault information, and displaying the initial fault diagnosis result report on an interface according to an initial report format template, and obtaining target feedback information from a user regarding the initial fault diagnosis result report, wherein the target feedback information represents user feedback information on report display content and report display format, and the feedback information on report display content includes preference feedback information and / or correction feedback information; Adjusting the initial large model based on the feedback information of the report display content to obtain a locally deployed target large model, and adjusting the initial report format template based on the feedback information of the report display format to obtain a target report format template; Inputting target fault information into the target macro model, the target macro model determining, from a target knowledge base, target fault diagnosis knowledge items whose matching degree with the target fault information reaches a preset matching degree threshold, and obtaining, based on the target fault diagnosis knowledge items, a predicted fault diagnosis result of the target fault information output by the target macro model; According to the target report format template, a target fault diagnosis result report is generated based on the target fault information and the predicted fault diagnosis result of the target fault information and is displayed on the interface.
2. The method according to claim 1, characterized in that The adjusting the initial large model by using feedback information of the report display content to obtain a locally deployed target large model includes: Determining a target feature and a target weight of the target feature through feedback information of the content displayed in the report; Based on the determined target features and the target weights of the target features, the locally deployed target macro model is obtained.
3. The method according to claim 2, characterized in that The determining of the target feature and the target weight of the target feature through the feedback information of the report display content includes: Determining, based on the preference feedback information, a first feature corresponding to display content preferred by the user and a second feature corresponding to display content disliked by the user, and increasing a weight of the first feature and decreasing a weight of the second feature, wherein the first feature and the second feature are the target features, and the increased weight of the first feature and the decreased weight of the second feature are the target weights of the target features; and / or The third feature corresponding to the display content required to correct the fault information and the target weight of the third feature are determined according to the correction feedback information, the third feature being the target feature, and the target weight of the third feature being the target weight of the target feature.
4. The method according to claim 1, wherein The obtaining of target feedback information from the user regarding the initial fault diagnosis result report includes: Obtaining the preference feedback information based on the user's review status and preference evaluation information corresponding to different displayed contents of the initial fault diagnosis result report in the interface; The correction feedback information is obtained based on the correction information of the display content input by the user in the interface.
5. The method according to any one of claims 1 to 4, characterized in that Before inputting the target fault information into the locally deployed target large model, the method further includes: Obtaining a trained initial large model, where the trained initial large model is obtained by training with the initial training sample; Regularly transforming and expanding the initial training samples through a data enhancement algorithm to obtain target training samples, wherein the target training samples include newly added fault information samples of new fault types and corresponding labeled fault diagnosis result samples; The trained initial large model is retrained periodically using the target training samples to obtain an optimized initial large model, and a locally deployed initial large model is obtained based on the optimized initial large model.
6. The method according to claim 5, characterized in that Before obtaining the trained initial large model, the method further includes: Acquire local initial training samples, wherein the initial training samples include initial fault information samples of various fault types, and each initial fault information sample is annotated with a corresponding initial fault diagnosis result sample; locally inputting the initial training sample into an initial large model, the initial large model determining, from an initial knowledge base, initial fault diagnosis knowledge item samples whose matching degree with the initial fault information sample reaches a preset matching degree threshold, and obtaining a predicted fault diagnosis result of the initial fault information sample output by the initial large model based on the initial fault diagnosis knowledge item samples; When the loss between the predicted fault diagnosis result of the initial fault information sample and the marked initial fault diagnosis result sample meets the convergence condition, a trained initial large model is obtained and the trained initial large model is saved locally.
7. The method according to claim 5, characterized in that The locally deployed target large model is obtained based on the optimized initial large model, including: Converting high-precision values in the optimized initial large model into low-precision values to obtain a quantized initial large model; Unimportant weights or connections in the quantized initial large model are removed to obtain a pruned initial large model, where the pruned initial large model is the locally deployed target large model.
8. A fault diagnosis report generating device, characterized in that: include: an initial large model processing unit, configured to input initial fault information into a locally deployed initial large model, having the initial large model determine, from an initial knowledge base, initial fault diagnosis knowledge items whose matching degree with the initial fault information reaches a preset matching degree threshold, and obtain, based on the initial fault diagnosis knowledge items, a predicted fault diagnosis result for the initial fault information output by the initial large model; an acquisition unit, configured to generate and display an initial fault diagnosis result report on an interface based on the initial fault information and the predicted fault diagnosis result of the initial fault information in accordance with an initial report format template, and to acquire target feedback information from a user regarding the initial fault diagnosis result report, wherein the target feedback information represents user feedback information on report display content and report display format, and the feedback information on report display content includes preference feedback information and / or correction feedback information; a large model adjustment unit, configured to adjust the initial large model based on the feedback information of the report display content to obtain a locally deployed target large model, and adjust the initial report format template based on the feedback information of the report display format to obtain a target report format template; a target large model processing unit, configured to input target fault information into the target large model, having the target large model determine, from a target knowledge base, target fault diagnosis knowledge items whose matching degree with the target fault information reaches a preset matching degree threshold, and obtain, based on the target fault diagnosis knowledge items, a predicted fault diagnosis result of the target fault information output by the target large model; A generating unit is configured to generate and display a target fault diagnosis result report on an interface based on the target fault information and the predicted fault diagnosis result of the target fault information in accordance with the target report format template.
9. A fault diagnosis report generating device, characterized in that: include: central processing unit and memory; The memory is a short-term storage memory or a persistent storage memory; The central processing unit is configured to communicate with the memory and execute instructions in the memory to perform the method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that The computer-readable storage medium comprises instructions, which, when executed on a computer, enable the computer to perform the method according to any one of claims 1 to 7.