An intelligent work order processing method, system, device and medium based on a large model
By processing work order data through a large model, combined with multimodal parsing and weighted fusion, the problems of relying on experience and wasting resources in work order processing have been solved, enabling accurate fault recommendation and parts management, and improving the experience for enterprises and customers.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENZHEN FENXIANG INTERNET TECH CO LTD
- Filing Date
- 2026-05-08
- Publication Date
- 2026-06-05
AI Technical Summary
The existing work order processing system relies on engineers' experience, resulting in a low first-time repair rate, serious resource waste, a lack of scientific basis for parts management, and low accuracy of keyword matching methods, making it impossible to achieve integrated intelligent scheduling.
We employ large-scale models for semantic and multimodal parsing, combined with weighted fusion of multiple data sources, to generate structured fault features and spare parts scheduling instructions. This enables accurate fault solution recommendations and automatic spare parts prediction, and we design a self-service interception mechanism to handle simple faults.
It improved the accuracy of fault location and solution recommendation, enabled dynamic configuration of parts on demand, reduced enterprise operating costs and customer waiting time, and improved after-sales service efficiency.
Smart Images
Figure CN122153015A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of computer data processing and artificial intelligence technology, and is applicable to enterprise field service and after-sales support scenarios. In particular, it relates to an intelligent work order processing method, system, equipment and medium based on a large model. Background Technology
[0002] In after-sales service for machinery and IT infrastructure, on-site work order processing is a core component. Currently, companies face three main pain points in work order processing: First, work order processing is highly dependent on the individual experience of engineers. Newly hired or inexperienced engineers struggle to handle complex faults, resulting in a low first-time repair rate, which severely impacts customer satisfaction, while the company incurs high training costs.
[0003] Secondly, parts management lacks a scientific basis. Engineers often request parts based on experience, which can easily lead to inventory backlogs or shortages of parts at critical moments, increasing operating costs and affecting repair efficiency.
[0004] Finally, many work orders actually involve simple issues (such as improper operation or configuration errors) that customers could resolve themselves with guidance. However, existing processes typically require work orders to be routed to engineers for processing, resulting in wasted resources, long customer wait times, and a poor customer experience.
[0005] In existing technologies, some work order systems use keyword matching to recommend historical solutions. However, due to the diversity of natural language descriptions, the accuracy of keyword matching is low, and invalid interference words cannot be eliminated. In addition, existing systems usually separate solution recommendation from parts management, failing to achieve integrated intelligent scheduling. Summary of the Invention
[0006] The purpose of this application is to provide a method, system, device and medium for intelligent work order processing based on a large model, which aims to utilize the semantic understanding and multimodal parsing capabilities of the large model to achieve accurate recommendation of fault solutions, automatic prediction and scheduling of spare parts, and simple self-service fault interception, thereby comprehensively improving the efficiency of after-sales service.
[0007] The first aspect of this invention discloses an intelligent work order processing method based on a large model, the method comprising: Step S1: Obtain the work order data submitted by the user, the work order data including text description data and image data of the equipment site; Step S2: Call the large language model to perform semantic parsing on the text description data, and call the visual large language model to perform abnormal feature recognition on the image data, extracting and fusing to obtain structured fault features; Step S3: Convert the fault features into feature vectors and perform similarity retrieval in a vector knowledge base containing multiple types of data sources to obtain related multi-source candidate solution knowledge; Step S4: Prioritize and weight the retrieved multi-source candidate solutions, combine them with the fault features to construct structured prompt words, and input them into the large language model to generate the target fault handling solution; Step S5: Match the parts consumption records of similar historical work orders based on the fault characteristics, run the parts prediction algorithm in conjunction with the product manual requirements, predict the required parts type and quantity, and generate the corresponding parts scheduling instructions.
[0008] Furthermore, in step S2, the extraction and fusion of structured fault features includes: By filtering emotional language in the text description data using a large language model, product model, fault code, and preliminary fault symptoms are extracted. By using a visual large language model to identify circuit board burn marks or status indicator features in the image data, the location of the fault can be determined. The extracted multi-dimensional features are encapsulated into a standardized JSON format to generate the structured fault features that include product, fault location, and fault phenomenon elements.
[0009] Furthermore, in steps S3-S4, the multiple data sources include a product manual knowledge base, an employee experience knowledge base, and a historical work order fault database; the priority-weighted fusion of the retrieved multi-source candidate solution knowledge includes: A three-tiered priority weight is assigned to the multi-source candidate solution knowledge from different data sources: the first weight corresponding to the product manual knowledge base is the highest, the second weight corresponding to the employee experience knowledge base is the next highest, and the third weight corresponding to the historical work order fault database is the lowest. Based on the weighted filtering, low-scoring knowledge with conflicts is filtered out, and high-scoring knowledge is combined in priority order to serve as the context for reasoning in the large language model.
[0010] Furthermore, in step S4, the construction of structured prompt words based on the fault characteristics includes: The system role settings, task definitions, weighted and fused context background, and mandatory output format requirements are assembled into the structured prompt words according to the preset template. The task definition limits the large language model to output a processing solution from a technical perspective, which includes fault root cause analysis, step-by-step troubleshooting actions, and safety warnings.
[0011] Furthermore, in step S5, the running of the parts prediction algorithm to predict the required parts type and quantity includes: Retrieve the target parts consumption data for the N most recent historical work orders of the same type; Calculate the variance of the quantity of the target component consumed in historical work orders; If the variance is within the preset threshold range, it indicates that the consumption of parts is stable, and the standard maintenance cycle specified in the product manual is used for prediction first. If the variance exceeds the preset threshold range, it indicates that there is a significant difference in the consumption of spare parts. In this case, the average value of the consumption data of the most recent N historical work orders of the same type is taken as the benchmark for the recommended quantity.
[0012] Furthermore, the method also includes a front-end work order interception step: During the process of the user entering the work order data, a lightweight model is invoked in real time to perform vector retrieval on the initially extracted fault features; If the similarity between the fault characteristics and known simple faults is greater than the set interception threshold, a self-service troubleshooting guide will be pushed to the front-end page in real time; if feedback confirming the solution is received from the user, the loop will be closed and the generation of the work order will be intercepted.
[0013] Furthermore, the invocation of the large language model is achieved through the underlying deployed large model proxy module, which performs the following operations: Convert the non-standard API interfaces of different large language model service providers into a unified JSON protocol format; A unified token authentication and key routing and distribution mechanism is used to achieve load balancing for multi-model requests. The request vector is cached using a semantic caching mechanism. When the similarity between the fault vector of a new request and a recently processed fault vector exceeds a preset cache threshold, the target fault handling solution in the cache is returned directly. Server-Sent Events technology is used to stream the target fault handling solution.
[0014] A second aspect of this invention discloses an intelligent work order processing system based on a large model, wherein the system employs the method described in any one of the first aspects above, and the system comprises: The data acquisition and parsing module is used to acquire work order data submitted by users, which includes text and images, and call a multimodal large language model to extract structured fault features. The knowledge retrieval and weighting module is used to convert the fault features into feature vectors for similarity retrieval and to perform priority weighting and fusion of the acquired multi-source candidate solution knowledge. The solution generation module is used to construct structured prompt words input into a large language model and generate target fault handling solutions; The scheduling and interception module is used to predict the type and quantity of parts based on the parts prediction algorithm and generate parts scheduling instructions, as well as to perform self-service troubleshooting and interception at the front-end order stage.
[0015] A third aspect of this invention discloses an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the intelligent work order processing method based on a large model according to any one of the first aspects of this disclosure.
[0016] A fourth aspect of this invention discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a large-model-based intelligent work order processing method according to any one of the first aspects of this disclosure.
[0017] Compared with the prior art, this application has the following significant advantages: 1. Significantly improve the accuracy of fault location and solution recommendation: By introducing a large visual model to process on-site images and combining it with a large language model to filter out redundant emotional information in the text, standardized "product-location-phenomenon" structured features are generated, which completely solves the pain point of traditional keyword retrieval being interfered with by invalid words.
[0018] 2. Overcoming knowledge conflicts and ensuring solution reliability: An innovative three-level weighted integration mechanism of "product manual - employee experience - historical work orders" was designed. This not only makes full use of the accumulated after-sales experience, but also ensures that the final solution generated by the large model absolutely complies with the original manufacturer's safety specifications, avoiding erroneous repair operations caused by AI "illusion".
[0019] 3. Enable on-demand dynamic allocation of inventory resources: Through a differentiated parts prediction algorithm (automatically switching between standard period and historical average based on variance threshold), accurately predict the parts required for each failure and connect with the system to automatically generate allocation instructions, realizing the pre-positioning and personalized scheduling of parts, effectively reducing inventory backlog and eliminating the embarrassment of "no parts available".
[0020] 4. Significantly reduce enterprise operating costs and customer waiting time: The upfront large model interception mechanism can respond to self-service troubleshooting solutions in seconds during the user order submission stage, keeping a large number of simple faults out of the system; the underlying large model proxy architecture (cache interception and token routing) effectively reduces the concurrent pressure and financial costs of enterprises calling large model APIs. Attached Figure Description
[0021] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0022] Figure 1 A flowchart illustrating an intelligent work order processing method based on a large model, provided for an embodiment of this application; Figure 2 This is a schematic diagram of the fault feature structure and vectorized relationship in the embodiments of this application; Figure 3 This is a diagram illustrating the interaction architecture between the large-scale agent system and the business system in this application embodiment; Figure 4 This is a schematic diagram of the parts management system in an embodiment of this application; Figure 5 A block diagram of an intelligent work order processing system based on a large model is provided for embodiments of this application; Figure 6 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0023] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0024] Before the system is transferred, the system involved in this application will prepare basic data and build a "vector knowledge base". Specifically: The system utilizes a large model to clean historical work orders, extracting five key elements: "Product (P), Fault Location (L), Fault Symptom (H), Fault Cause (R), and Solution (S)," forming a one-to-N tree-like relationship and generating a unique fault code (e.g., using the algorithm M = "1" + MD5(P+L+H+R+S)%1000000). Simultaneously, product manuals and maintenance guides are sliced. The system inputs the string combining "Product + Location + Symptom" into a BERT-like model, transforming it into a high-dimensional feature vector through mean pooling or CLS vector extraction, and storing it in a vector database such as Milvus or Faiss.
[0025] The first aspect of this invention discloses an intelligent work order processing method based on a large model. Figure 1 This is a flowchart of a smart work order processing method based on a large model according to an embodiment of the present invention, such as... Figure 1 As shown, the method provided in this application embodiment specifically includes the following steps: Step S1: Obtain the work order data submitted by the user, the work order data including text description data and image data of the equipment site; When users enter work orders on the client, the data often includes redundant text filled with emotion (such as "My refrigerator stopped cooling after only one year of purchase, and everything inside is broken, please come quickly!") as well as on-site photos taken by the user (such as a flashing red light on the control panel or burnt wires).
[0026] Step S2: Call the large language model to perform semantic parsing on the text description data, and call the visual large language model to perform abnormal feature recognition on the image data, extracting and fusing to obtain structured fault features; The system uses large language models (such as Wenxin Yiyan and GPT-4) to perform semantic parsing on the text, filtering out emotional language and extracting time clues, user requests, and initial fault symptoms. Simultaneously, it uses a visual large language model (VLM) in parallel to deeply analyze image data, identifying abnormal features (such as "main control board") and status indicator features. Finally, the system encapsulates this multi-dimensional information into a standardized JSON format, for example: {"P": "refrigerator", "L": "main control board", "H": "not cooling"}, as structured fault features.
[0027] In step S2, the extraction and fusion of structured fault features includes: By filtering emotional language in the text description data using a large language model, product model, fault code, and preliminary fault symptoms are extracted. By using a visual large language model to identify circuit board burn marks or status indicator features in the image data, the location of the fault can be determined. The extracted multi-dimensional features (product model, fault code, initial fault phenomenon, and fault location) are encapsulated into a standardized JSON format to generate the structured fault features containing product, fault location, and fault phenomenon elements.
[0028] This application features a specially designed pre-interception mechanism for work orders, namely front-end work order interception. The specific steps are as follows: During the process of the user entering the work order data, a lightweight model is invoked in real time to perform vector retrieval on the initially extracted fault features; If the similarity between the fault characteristics and known simple faults is greater than the set interception threshold, a self-service troubleshooting guide will be pushed to the front-end page in real time; if feedback confirming the solution is received from the user, the loop will be closed and the generation of the work order will be intercepted. When a user pauses for more than 2 seconds while inputting a fault description or clicks "Next," the system extracts the aforementioned core features and immediately calls a lightweight model or rule engine on the front end for matching. If the fault is determined to be a simple problem that the user can solve themselves (such as router reconfiguration or washing machine filter cleaning), the system immediately streams a self-troubleshooting guide with pictures and text on the page. If the user follows the guide and clicks "Resolved," the system automatically closes the loop and prevents the creation of a work order; if the user still requires on-site repair, the process continues.
[0029] Step S3: Convert the fault features into feature vectors, and perform similarity retrieval in a vector knowledge base containing multiple types of data sources to obtain related multi-source candidate solution knowledge; such as Figure 2 As shown.
[0030] In step S3, the multiple data sources include a product manual knowledge base, an employee experience knowledge base, and a historical work order fault database; the priority-weighted fusion of the retrieved multi-source candidate solution knowledge includes: A three-tiered priority weight is assigned to the multi-source candidate solution knowledge from different data sources: the first weight corresponding to the product manual knowledge base is the highest, the second weight corresponding to the employee experience knowledge base is the next highest, and the third weight corresponding to the historical work order fault database is the lowest. Based on the weighted filtering, low-scoring knowledge with conflicts is filtered out, and high-scoring knowledge is combined in priority order to serve as the context for reasoning in the large language model.
[0031] For work orders that enter the backend, the system will vectorize the structured JSON features generated in step S101 and calculate the cosine similarity: Set a threshold (e.g., 0.8) and select the top 5 records with the highest similarity. When Similarity > 0.8, they are considered to be the same work order fault.
[0032] Retrieve Top-K candidate knowledge from a vector database containing multiple types of data sources.
[0033] To avoid the "illusion" caused by knowledge conflicts in large models, this application designs a three-level weighted fusion mechanism: (1) First weight (highest): Product manual knowledge base. Contains authoritative parameters from the original manufacturer, with the highest weight, ensuring that the repair solution meets the original manufacturer's safety and compliance requirements.
[0034] (2) Second weight (medium): Employee experience knowledge base. It contains practical tips for avoiding pitfalls summarized by senior engineers.
[0035] (3) Third weight (lowest): Historical work order fault database. The data volume is large but may contain individual errors or noise, and is for reference only.
[0036] The system scores and sorts the data according to the weights mentioned above, filters out conflicting low-quality knowledge, and forms a high-quality contextual background.
[0037] Step S4: Prioritize and weight the retrieved multi-source candidate solutions, combine them with the fault features to construct structured prompt words, and input them into the large language model to generate the target fault handling solution; In step S4, constructing structured prompt words based on the fault characteristics includes: The system role settings, task definitions, weighted and fused context background, and mandatory output format requirements are assembled into the structured prompt words according to the preset template. The task definition limits the large language model to output a processing solution from a technical perspective, which includes fault root cause analysis, step-by-step troubleshooting actions, and safety warnings.
[0038] The system does not directly input the original question into a large model; instead, it assembles structured prompts using engineered templates. The specific prompt comprises four dimensions: (1) Role: Limited to the large model playing the role of "senior senior equipment maintenance engineer".
[0039] (2) Task (Task Definition): Based on the given context, output "root cause of failure, step-by-step troubleshooting actions, required tools, and safety warnings".
[0040] (3) Context (Context Injection): Fill in the three-level weighted knowledge fragment produced in step S3.
[0041] (4) Format: Forces output to conform to a specific Markdown table or JSON node.
[0042] After receiving the standardization instruction, the large model can reliably generate detailed and highly professional best fault handling solutions and push them to the order-receiving engineer.
[0043] Step S5: Match the parts consumption records of similar historical work orders based on the fault characteristics, run the parts prediction algorithm in conjunction with the product manual requirements, predict the required parts type and quantity, and generate the corresponding parts scheduling instructions. In step S5, the running parts prediction algorithm predicts the required parts type and quantity, including: Retrieve the target parts consumption data for the N most recent historical work orders of the same type; Calculate the variance of the quantity of the target component consumed in historical work orders; If the variance is within the preset threshold range, it indicates that the consumption of parts is stable, and the standard maintenance cycle specified in the product manual is used for prediction first. If the variance exceeds the preset threshold range, it indicates that there is a significant difference in the consumption of spare parts. In this case, the average value of the consumption data of the most recent N historical work orders of the same type is taken as the benchmark for the recommended quantity.
[0044] The system simultaneously retrieves the actual parts consumption records of the most recent N similar historical work orders using fault codes. To cope with complex real-world environments, this application designs a differentiated parts prediction algorithm based on variance thresholds: The system calculates the variance of the consumption quantity of the target part in similar historical work orders. If the variance is within a preset threshold range (indicating that the consumption quantity of the part is very stable each time it is damaged), the system prioritizes issuing a prediction command directly using the standard maintenance cycle specified in the product manual.
[0045] If the variance exceeds the preset range (for example, the number of damaged parts in the same batch of equipment under different harsh environments is highly random), the standard manual is no longer applicable. In this case, the algorithm automatically switches logic and takes the mathematical average of the recent historical consumption data as the benchmark.
[0046] After the forecast is completed, the system checks the inventory of warehouses at all levels in conjunction with the daily automatically executed Cron task. If the central warehouse has the goods, a transfer order is directly generated and sent to the corresponding engineer's personal inventory; if there is a shortage, a procurement alert is triggered, such as... Figure 4 As shown.
[0047] Furthermore, the invocation of the large language model is achieved through the underlying deployed large model proxy module, which is used to perform the following operations: Convert the non-standard API interfaces of different large language model service providers into a unified JSON protocol format; A unified token authentication and key routing and distribution mechanism is used to achieve load balancing for multi-model requests. The request vector is cached using a semantic caching mechanism. When the similarity between the fault vector of a new request and a recently processed fault vector exceeds a preset cache threshold, the target fault handling solution in the cache is returned directly. Server-Sent Events technology is used to stream the target fault handling solution.
[0048] See Figure 3 To support high concurrency and save costs, this application innovatively deploys a "Large Model Proxy Module (LLM Proxy)" between the business side and the cloud-based models. This module implements four core functions: 1) Unified Protocol: Convert the non-standard interfaces of major model manufacturers into a unified JSON interaction.
[0049] 2) SSE Streaming Response: Using Server-Sent Events, long solutions are streamed to the front end like a typewriter, eliminating waiting anxiety.
[0050] 3) Route distribution: Unified management of API keys and dynamic routing based on load.
[0051] 4) Semantic Caching: Redis is used to store question-and-answer vectors. When the feature vector of a new work order is extremely similar to the vector of a recently resolved work order in the cache pool (e.g., similarity > 0.95), the large model is not called without consuming tokens. Instead, the cached solution is returned directly, which greatly saves costs and achieves millisecond-level response.
[0052] In addition, once the work order is completed, the system will automatically collect the actual repair plan and consumable data, and feed it back into the standardized fault database to form a data flywheel.
[0053] In summary, the solution proposed in this invention can leverage a large-scale model to address three major pain points in existing technologies: Firstly, it provides engineers with standardized and precise fault solution recommendations, enabling even ordinary engineers to efficiently handle complex work orders; secondly, it enables accurate prediction and automatic scheduling of parts demand based on fault analysis and historical data, achieving on-demand supply of parts, reducing inventory and avoiding parts shortages, thus comprehensively improving the intelligence level and operational efficiency of enterprise after-sales services; finally, it allows for finding solutions to work orders during the customer's work order entry process, eliminating the need to wait for engineers to visit and reducing the burden on enterprises.
[0054] The second aspect of this invention discloses an intelligent work order processing system based on a large model. Figure 5 This is a structural diagram of an intelligent work order processing system based on a large model according to an embodiment of the present invention; as shown below. Figure 5 As shown, the system 100 includes: The data acquisition and parsing module 101 is used to acquire work order data containing text and images submitted by users, and call a multimodal large language model to extract structured fault features. The knowledge retrieval and weighting module 102 is used to convert the fault features into feature vectors for similarity retrieval, and to perform priority weighting and fusion of the acquired multi-source candidate solution knowledge. Solution generation module 103 is used to construct structured prompt words input into a large language model and generate target fault handling solutions; The scheduling and interception module 104 is used to predict the type and quantity of parts based on the parts prediction algorithm and generate parts scheduling instructions, as well as to perform self-service troubleshooting and interception at the front-end order stage.
[0055] A third aspect of this invention discloses an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the intelligent work order processing method based on a large model, as disclosed in any of the first aspects of this invention.
[0056] Figure 6 This is a structural diagram of an electronic device according to an embodiment of the present invention, such as... Figure 6As shown, the electronic device includes a processor, memory, communication interface, display screen, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, Near Field Communication (NFC), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.
[0057] Those skilled in the art will understand that Figure 6 The structure shown is merely a structural diagram of the part related to the technical solution of this disclosure and does not constitute a limitation on the electronic device to which the solution of this application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.
[0058] A fourth aspect of this invention discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a large-model-based intelligent work order processing method according to any one of the first aspects of this invention.
[0059] Please note that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. The above embodiments only illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention application. It should be pointed out that for those skilled in the art, several modifications and improvements can be made without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
[0060] The above are preferred embodiments of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A smart work order processing method based on a large model, characterized in that, The method includes: Step S1: Obtain the work order data submitted by the user, the work order data including text description data and image data of the equipment site; Step S2: Call the large language model to perform semantic parsing on the text description data, and call the visual large language model to perform abnormal feature recognition on the image data, extracting and fusing to obtain structured fault features; Step S3: Convert the structured fault features into feature vectors, and perform similarity retrieval in a vector knowledge base containing multiple types of data sources to obtain related multi-source candidate solution knowledge; Step S4: Prioritize and weight the retrieved multi-source candidate solutions, combine them with the structured fault features to construct structured prompt words, and input them into the large language model to generate the target fault handling solution; Step S5: Match the component consumption records of similar historical work orders with the structured fault characteristics, combine with the preset component maintenance standards, run the component prediction algorithm to predict the required component types and quantities, and generate corresponding component scheduling instructions.
2. The intelligent work order processing method based on a large model according to claim 1, characterized in that, In step S2, the extraction and fusion of structured fault features includes: By filtering emotional language in the text description data using a large language model, product model, fault code, and preliminary fault symptoms are extracted. By using a visual large language model to identify circuit board burn marks or status indicator features in the image data, the location of the fault can be determined. The extracted product model, fault code, preliminary fault symptoms, and fault location are encapsulated into a standardized JSON format to generate the structured fault features containing product, fault location, and fault symptom elements.
3. The intelligent work order processing method based on a large model according to claim 1, characterized in that, In steps S3-S4, the multiple data sources include a product manual knowledge base, an employee experience knowledge base, and a historical work order fault database; the priority-weighted fusion of the retrieved multi-source candidate solution knowledge includes: A three-tiered priority weight is assigned to the multi-source candidate solution knowledge from different data sources: the first weight corresponding to the product manual knowledge base is the highest, the second weight corresponding to the employee experience knowledge base is the next highest, and the third weight corresponding to the historical work order fault database is the lowest. Based on the three-level priority weight filtering, conflicting low-priority knowledge is filtered out, and high-priority knowledge is combined according to priority order to serve as the context background for reasoning in the large language model.
4. The intelligent work order processing method based on a large model according to claim 1, characterized in that, In step S4, structured prompt words are constructed by combining the structured fault features, including: The system role settings, task definitions, weighted and fused context background, and mandatory output format requirements are assembled into the structured prompt words according to the preset template. The task definition limits the large language model to output a processing solution from a technical perspective, which includes fault root cause analysis, step-by-step troubleshooting actions, and safety warnings.
5. The intelligent work order processing method based on a large model according to claim 1, characterized in that, In step S5, the running parts prediction algorithm predicts the required parts type and quantity, including: Retrieve the target parts consumption data for the N most recent historical work orders of the same type; Calculate the variance of the quantity of required parts consumed in similar historical work orders; If the variance is less than or equal to the preset variance threshold, it indicates that the consumption of parts is stable, and the standard maintenance cycle specified in the product manual is used for prediction first. If the variance is greater than the preset variance threshold, it indicates that there is a significant difference in the consumption of spare parts. In this case, the average value of the target spare parts consumption data of the most recent N historical work orders of the same type is taken as the benchmark for the recommended quantity.
6. The intelligent work order processing method based on a large model according to claim 1, characterized in that, The method also includes a front-end work order interception step: During the process of the user entering the work order data, a lightweight model is invoked in real time to perform vector retrieval on the initially extracted fault features; If the similarity between the initially extracted fault features and known simple faults is greater than the set interception threshold, a self-service troubleshooting guide will be pushed to the front-end page in real time; if feedback confirming the solution is received from the user, the loop will be closed and the generation of the work order will be intercepted.
7. A method for intelligent work order processing based on a large model according to any one of claims 1 to 6, characterized in that, The invocation of the large language model is achieved through the underlying deployed large model proxy module, which performs the following operations: Convert the non-standard API interfaces of different large language model service providers into a unified JSON protocol format; A unified token authentication and key routing and distribution mechanism is used to achieve load balancing for multi-model requests. The request vector is cached using a semantic caching mechanism. When the similarity between the fault vector of a new request and the fault vector that has been processed within a set time period exceeds a preset cache threshold, the target fault handling solution in the cache is returned directly. Server-Sent Events technology is used to stream the target fault handling solution.
8. A smart work order processing system based on a large model, characterized in that, The system employs the method described in any one of claims 1 to 7, the system comprising: The data acquisition and parsing module is used to acquire work order data containing text and images submitted by users, and call the large language model and visual large language model to extract structured fault features. The knowledge retrieval and weighting module is used to convert the structured fault features into feature vectors for similarity retrieval, and to perform priority weighting and fusion of the acquired multi-source candidate solution knowledge. The solution generation module is used to construct structured prompt words input into a large language model and generate target fault handling solutions; The scheduling and interception module is used to predict the type and quantity of parts based on the parts prediction algorithm and generate parts scheduling instructions, as well as to perform self-service troubleshooting and interception at the front-end order stage.
9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the intelligent work order processing method based on a large model according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the intelligent work order processing method based on a large model according to any one of claims 1 to 7.