Interactive individual soldier equipment for overhauling mining truck and fault diagnosis method

Abstract

The invention discloses interactive individual equipment for overhauling a mining truck and a fault diagnosis method, and belongs to the technical field of intelligent operation and maintenance of industrial equipment. The core of the scheme is that a device CAN bus fault code is obtained through a physical interface, dynamic semantic retrieval is carried out through a device knowledge vector library pre-built in local, then real-time data and retrieved related knowledge are fused and input into a generative AI model deployed in an edge computing device, and finally a natural language overhaul suggestion is generated and presented. The method has the main effects that automatic and intelligent conversion from a fault code to a maintenance scheme is realized, the diagnosis efficiency is improved to a minute level from a traditional dozens of minutes of manual consultation, the maintenance problem in mine, underground and other off-line environments is solved through complete localization deployment, the dependence on deep experts is remarkably reduced, and the maintenance efficiency is improved. And the standardization and the safety of the maintenance operation are improved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation and maintenance and fault diagnosis technology of industrial equipment, and specifically relates to an interactive individual equipment for the maintenance of mining trucks. Background Technology

[0002] Large industrial equipment such as mining trucks have complex structures, and their condition monitoring and fault diagnosis heavily rely on experienced maintenance personnel. Currently, the following problems commonly exist: Fault codes are difficult to understand: Fault codes output by the device's CAN bus system (such as "SPN 5246 FMI 18") are mostly alphanumeric codes conforming to specific standards. These codes themselves do not contain direct semantic information and cannot intuitively reflect the specific faulty component, fault mechanism, severity level, and potential impact, making it difficult for inexperienced maintenance personnel to quickly and accurately understand their meaning.

[0003] Inefficient knowledge retrieval: Complete troubleshooting knowledge is usually contained in unstructured PDF documents (such as maintenance manuals) spanning hundreds of pages. During on-site troubleshooting, relying on manual browsing and flipping through such documents to locate solutions corresponding to specific fault codes is cumbersome, time-consuming, and leads to low troubleshooting efficiency. It is also easy to miss key operating steps or important safety warnings, and may even lead to operational risks.

[0004] High reliance on expert experience: Accurate fault diagnosis and localization often depend on the long-term, unspoken personal experience accumulated by senior maintenance engineers. This "tacit knowledge" is difficult to standardize, document, and quickly share within the team. Once a senior engineer is absent, it can easily lead to long training cycles, high costs, and significant fluctuations in the overall diagnostic level, thus hindering the stability and improvement of the team's overall capabilities.

[0005] On-site network environment limitations: Typical operating scenarios such as mines, open-pit mines, and underground mines often face the challenge of poor network signal coverage or complete offline operation, which makes it impossible for intelligent diagnostic solutions that rely on large cloud models for calculation and analysis to operate effectively in practical applications.

[0006] In existing technologies, expert systems based on preset rule bases can handle some known faults, but they lack flexibility and adaptability, and cannot cope with complex faults beyond the rules. Solutions that merely digitize manuals and support keyword matching lack deep semantic understanding capabilities and cannot generate comprehensive diagnostic decisions by associating context. Therefore, there is an urgent need for a solution that can integrate real-time device data with static knowledge and achieve intelligent diagnosis and decision generation in an offline environment; this is precisely the problem that this invention aims to solve. Summary of the Invention

[0007] In view of the shortcomings of the existing technology, the purpose of this invention is to provide an interactive individual equipment for the maintenance of mining trucks.

[0008] The technical solution adopted in this invention is: an interactive individual soldier equipment for the maintenance of mining trucks, the key technical points of which include: An edge computing device that serves as a hardware platform; A physical interface adapter connected to the edge computing device for connecting to the CAN bus of the mining equipment to obtain data; and A fault diagnosis system is set up and runs on the edge computing device; The fault diagnosis system includes: The data acquisition and parsing module is used to acquire and parse CAN bus data frames through the physical interface adapter to extract fault codes and their context parameter data. The local device knowledge vector database is constructed by parsing and semantically slicing unstructured maintenance documents and then vectorizing them. The query generation module is used to generate retrieval queries based on the parsed fault information; A generative AI model, deployed locally on the edge computing device, receives relevant knowledge fragments retrieved from the local device's knowledge vector database based on the retrieval query, and combines this with real-time fault data to generate natural language troubleshooting suggestions; and A human-computer interaction interface, integrated into the edge computing device, is used to display the maintenance suggestions.

[0009] In the above scheme, the process of constructing vectorized unstructured maintenance documents through parsing and semantic slicing includes the following steps: The unstructured maintenance document was parsed using a document parsing tool to extract the text, tables, and metadata information. Based on the inherent logical structure of the document, the extracted text is segmented by chapter, subheading and natural paragraph to form knowledge slices with independent semantics; The knowledge slices are converted into numerical vectors using a text embedding model and stored in a local vector database to construct the local device knowledge vector database.

[0010] In the above scheme, the process of generating a retrieval query statement based on the parsed fault information is as follows: the parsed fault code is used as the core keyword, and the context parameter data parsed from its data frame is integrated to generate a structured query statement containing fault semantic information. The cosine similarity algorithm is used to perform dynamic knowledge retrieval in the local device knowledge vector database.

[0011] In the above scheme, the similarity threshold set in the search query is above 0.7.

[0012] In the above scheme, the specific process of generating natural language maintenance suggestions includes: combining real-time fault data, knowledge fragments from retrieval, and instructions output in a prescribed format by a preset requirement model acting as a maintenance expert, into a structured prompt word; The structured prompt is input into a locally deployed generative AI model; the model infers based on the prompt and generates natural language text containing fault interpretation, troubleshooting steps, required tools, and safety precautions as the troubleshooting suggestion.

[0013] A method for diagnosing mining equipment faults using interactive individual equipment for maintenance of mining trucks, the key technical points of which include the following steps: S1: Real-time data acquisition and fault code parsing: Connect to the CAN bus of the mining equipment through the physical interface adapter, listen to and parse data frames in real time, and extract fault codes and their context parameter data; S2: Fault Information Triggering and Query Generation: The parsed fault code is used as a trigger signal to automatically generate a structured retrieval query; S3: Dynamic Knowledge Retrieval: Input the generated query into the pre-built local device knowledge vector database for retrieval, and output knowledge fragments related to the current fault; S4: Multi-source information fusion and intelligent decision generation: Combine real-time fault data, retrieved knowledge fragments and preset reasoning task instructions into structured prompt words, input them into a locally deployed generative AI model, and generate natural language maintenance suggestions. S5: Maintenance suggestion presentation: Output the generated maintenance suggestions and display them on the human-machine interface of the edge computing device.

[0014] The beneficial effects of the present invention are as follows: Compared with the prior art, the present invention has the following significant technical effects: Intelligent diagnostic process: It realizes the fully automatic conversion from abstract fault codes to specific maintenance solutions, and improves the intelligence level of diagnostic decision-making from rule-based keyword matching to generative decision-making based on semantic understanding.

[0015] Improved diagnostic efficiency: Experimental results show that this solution can reduce the average time for fault location and solution finding from 45 minutes required by traditional manual review to less than 5 minutes.

[0016] Standardization of maintenance operations: Integrate expert experience into AI models to output standardized and regulated maintenance procedures, reduce reliance on personal experience, and improve the quality and safety of maintenance operations.

[0017] Offline system deployment: The entire solution runs on local edge devices, completely eliminating dependence on cloud networks and ensuring system availability in the absence of or under weak network conditions. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a general flowchart of an embodiment of the present invention; Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the following description is provided in conjunction with the accompanying drawings. Figure 1 The present invention will be further described in detail below with reference to specific embodiments.

[0021] This embodiment provides an interactive individual soldier equipment for maintaining mining trucks, which specifically includes the following steps: S1: Steps for real-time data acquisition and fault code parsing; The system connects to the CAN bus of the mining equipment via a physical interface (such as OBD-II), monitors and parses data frames in real time, and performs protocol decoding and semantic mapping on the raw data according to a pre-set J1939 or ISO 15765 protocol database. From this, it extracts fault codes (DTCs) representing the equipment status and their contextual parameter data. For example, if the fault code reflects an engine-related fault, the contextual parameter data must include engine speed, torque, temperature, and other data.

[0022] S2: Fault information triggering and query generation steps; Using the fault code parsed in step S1 as the initial trigger signal, one or more structured retrieval queries are automatically generated. These queries may include the fault code itself, as well as contextual information parsed from its data frame, including numerical values ​​and timestamps.

[0023] S3: Dynamic knowledge retrieval steps; The query generated in step S2 is input into a pre-built local device knowledge vector database for retrieval. This database constructs a knowledge base by semantically slicing unstructured maintenance manual PDF documents and then vectorizing the semantic slices using BGE or a similar embedded model. For example, in this embodiment, the official maintenance manual (PDF format) for mining trucks is parsed using a text parsing tool (such as PyMuPDF or Docling). The parsed text is intelligently sliced ​​by chapter and paragraph, and then converted into 768-dimensional vectors using a text embedding model (such as BGE or Sentence-Transformers) and stored in a local vector database (such as ChromaDB or Milvus).

[0024] During retrieval, an approximate nearest neighbor algorithm is used, and a cosine similarity threshold is set, such as 0.7 or 0.8. It is recommended to set the threshold above 0.7 to ensure the relevance and accuracy of the retrieval results. This step outputs several knowledge fragments that are most relevant to the current fault.

[0025] S4: Multi-source information fusion and intelligent decision generation steps; The real-time fault data from S1, the static knowledge retrieved from S3, and the preset inference task instructions are combined into a structured prompt word. An example of an inference instruction used in this embodiment is as follows: You are a mining equipment maintenance expert. Your current mining truck has a 'XX fault' (from fault code analysis). The latest data shows that the values ​​of each item are XX (from parameter data below). After consulting the maintenance manual, the following relevant explanations are found: XX (text from the local equipment knowledge vector database). Please interpret the fault based on the above information and provide detailed inspection and maintenance suggestions.

[0026] The most similar text from the fault code parsing, context parameter data, and local device knowledge vector database is concatenated with the inference task instructions to form a prompt word.

[0027] The prompt is input into an AI model (e.g., qwen2-0_5b) that has been quantized and optimized, has fewer than 10 bytes of parameters (e.g., Qwen-7B or Llama-3-8B), uses its optimized version (e.g., quantized using GGML or using the Llama.cpp engine), and is deployed on a local edge computing device (e.g., an Ubuntu tablet). After comprehensively understanding multi-source information, the model generates natural language troubleshooting suggestions that include fault interpretation, step-by-step troubleshooting procedures, required tools, and safety precautions.

[0028] S5: Maintenance recommendations are presented; The natural language maintenance suggestions generated in step S4 are output and displayed on the human-computer interaction interface of the local computing device for maintenance personnel to execute.

[0029] The effectiveness test of this embodiment: The maintenance personnel connected the tablet computer to the diagnostic interface in the cab of the mining truck via an interface cable.

[0030] Upon launching the application, the program automatically executes step S1, listens for and parses the CAN bus data, and discovers the fault code "SPN5246 FMI 18".

[0031] The program automatically triggers (step S2) and generates a query: "SPN 5246 fault, possible causes and troubleshooting steps".

[0032] The system searches the local knowledge base (step S3) and returns relevant paragraphs about "engine coolant temperature sensor circuit failure", including circuit diagram location and test standards.

[0033] The system automatically generates a Prompt (step S4): "You are a mining truck maintenance expert. The current fault code is: SPN5246 FMI 18. Relevant maintenance manual knowledge is as follows: [Insert retrieved knowledge fragment here...]. Please analyze comprehensively and generate a detailed maintenance suggestion." Input this Prompt into the local generative AI model.

[0034] The generative AI model suggests: "Fault SPN 5246 may be related to the engine coolant temperature sensor. Recommended steps: 1. Disconnect the sensor connector and check for corrosion... 2. Use a multimeter to measure the resistance; the standard value should be... Note: Ensure the engine is cooled before proceeding." Finally, the maintenance plan is displayed on the tablet screen (step S5) to guide the maintenance personnel in completing the work.

[0035] In the complete processing flow of this instance, from connecting the device to displaying the repair plan on the screen, the total time is less than 8 seconds. The average response time of the knowledge base retrieval stage is less than 200 milliseconds, and the inference speed of the locally generative AI model reaches an average of 15 tokens / second. Testing shows that the system's diagnostic accuracy is consistently above 90%.

[0036] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An interactive individual soldier equipment for repairing mining trucks, characterized in that, include: An edge computing device that serves as a hardware platform; A physical interface adapter connected to the edge computing device is used to connect to the CAN bus of the mining equipment to obtain data; as well as A fault diagnosis system is set up and runs on the edge computing device; The fault diagnosis system includes: The data acquisition and parsing module is used to acquire and parse CAN bus data frames through the physical interface adapter to extract fault codes and their context parameter data. The local device knowledge vector database is constructed by parsing and semantically slicing unstructured maintenance documents and then vectorizing them. The query generation module is used to generate retrieval queries based on the parsed fault information; A generative AI model, deployed locally on the edge computing device, receives relevant knowledge fragments retrieved from the local device's knowledge vector database based on the retrieval query, and combines this with real-time fault data to generate natural language troubleshooting suggestions; and A human-computer interaction interface, integrated into the edge computing device, is used to display the maintenance suggestions.

2. The interactive individual equipment for mine truck maintenance according to claim 1, characterized in that, The process of constructing vectorized unstructured maintenance documents through parsing and semantic slicing includes the following steps: The unstructured maintenance document was parsed using a document parsing tool to extract the text, tables, and metadata information. Based on the inherent logical structure of the document, the extracted text is segmented by chapter, subheading and natural paragraph to form knowledge slices with independent semantics; The knowledge slices are converted into numerical vectors using a text embedding model and stored in a local vector database to construct the local device knowledge vector database.

3. The interactive individual soldier equipment for mine truck maintenance according to claim 1, characterized in that, The process of generating a retrieval query statement based on the parsed fault information is as follows: the parsed fault code is used as the core keyword, and the context parameter data parsed from its data frame is integrated to generate a structured query statement containing fault semantic information. The cosine similarity algorithm is used to perform dynamic knowledge retrieval in the local device knowledge vector database.

4. The interactive individual equipment for mine truck maintenance according to claim 3, characterized in that, The similarity threshold set in the search query is above 0.

7.

5. The interactive individual soldier equipment for mine truck maintenance according to claim 1, characterized in that, The specific process of generating natural language troubleshooting suggestions includes: The real-time fault data, the retrieved knowledge fragments, and the instructions output by the preset requirement model as a maintenance expert in a prescribed format are combined into a structured prompt word. The structured prompt is input into a locally deployed generative AI model; the model infers based on the prompt and generates natural language text containing fault interpretation, troubleshooting steps, required tools, and safety precautions as the troubleshooting suggestion.

6. A method for diagnosing mining equipment faults based on the interactive individual equipment for repairing mining trucks as described in any one of claims 1 to 5, characterized in that, Includes the following steps: S1: Real-time data acquisition and fault code parsing: Connect to the CAN bus of the mining equipment through the physical interface adapter, listen to and parse data frames in real time, and extract fault codes and their context parameter data; S2: Fault Information Triggering and Query Generation: The parsed fault code is used as a trigger signal to automatically generate a structured retrieval query; S3: Dynamic Knowledge Retrieval: Input the generated query into the pre-built local device knowledge vector database for retrieval, and output knowledge fragments related to the current fault; S4: Multi-source information fusion and intelligent decision generation: Combine real-time fault data, retrieved knowledge fragments and preset reasoning task instructions into structured prompt words, input them into a locally deployed generative AI model, and generate natural language maintenance suggestions. S5: Maintenance suggestion presentation: Output the generated maintenance suggestions and display them on the human-machine interface of the edge computing device.

Images