An urban rail fault root cause automatic analysis method and system based on an industry knowledge graph and a large model

Abstract

The application discloses a kind of urban rail fault root cause automatic analysis method and system based on industry knowledge graph and big model.The method extracts knowledge from unstructured historical maintenance records and technical manuals automatically, constructs the knowledge graph of urban rail operation and maintenance;Using fine-tuned vertical field large language model, combined with real-time monitoring data and graph search results, automatically generate natural language diagnostic report containing fault root cause, risk level, maintenance scheme and spare parts list.The application effectively solves the pain points of difficulty in digitizing expert experience, dependence on manual consultation of manuals in urban rail operation and maintenance, significantly improves the intelligent level, accuracy and efficiency of fault diagnosis, and realizes the decision-making closed-loop support of operation and maintenance.

Description

Technical Field

[0001] This invention relates to the technical field of urban rail transit operation and maintenance, specifically to an intelligent diagnostic method and system that utilizes natural language processing, knowledge graph construction technology, and vertical domain large language model (LLM) to digitally mine unstructured maintenance records of urban rail transit equipment and realize automatic fault root cause analysis and generation of handling solutions. Background Technology

[0002] With the rapid expansion of urban rail transit networks, the amount of data involved in operation and maintenance is growing rapidly. Current urban rail operation and maintenance faces the following major pain points:

[0003] 1. Data is severely unstructured: Historical maintenance records and fault reports are mostly in the form of free text, with expert experience implied in them, which is difficult for computers to understand and reuse directly, resulting in "rich data but poor knowledge".

[0004] 2. Diagnosis relies on human experience: After an alarm occurs, maintenance staff need to manually consult thick technical manuals or rely on personal experience to make a judgment, which is inefficient and easily affected by differences in the skill level of personnel, and lacks standardized handling suggestions.

[0005] 3. Lack of closed-loop decision support: Existing monitoring systems mostly remain at the "alarm" level and cannot directly provide a complete decision report that includes root cause analysis, risk level, maintenance plan and spare parts list.

[0006] While some rule-based or traditional machine learning-based fault diagnosis methods exist in current technologies, they still fall short in handling complex semantic understanding, fusing multi-source heterogeneous data, and generating highly interpretable natural language reports. Recently (2025-2026), companies such as Beijing Subway and CRRC Times Electric have begun exploring diagnostic technologies combining knowledge graphs and large models. However, there is still significant room for innovation in deep vertical fine-tuning and dynamic knowledge graph update mechanisms across all urban rail transit specialties (vehicles, signaling, power supply, etc.). Summary of the Invention

[0007] The purpose of this invention is to provide an automatic root cause analysis method for urban rail transit faults based on industry knowledge graphs and large models, aiming to solve the above-mentioned pain points and realize the transformation from "passive alarm" to "proactive intelligent decision-making".

[0008] The present invention adopts the following technical solution:

[0009] An automated root cause analysis method for urban rail transit faults based on industry knowledge graphs and large models includes the following steps:

[0010] S1. Acquisition and preprocessing of multi-source heterogeneous data;

[0011] S2, Construction of a dedicated knowledge graph for urban rail transit operation and maintenance;

[0012] S3, fine-tuning of large language models in vertical domains and construction of inference engines;

[0013] S4. Automatic root cause analysis and report generation for faults;

[0014] S5, Human-Computer Feedback and Graph Self-Evolution.

[0015] Step S1: Acquisition and Preprocessing of Multi-Source Heterogeneous Data

[0016] Real-time acquisition of multi-source data from the urban rail system, including:

[0017] Static data: Maintenance manuals, technical standards, drawings and documents (PDF / Word format);

[0018] Dynamic data: Real-time monitoring of sensor data, SCADA system logs, and TCMS (Train Control and Management System) messages;

[0019] Historical data: Past fault reports, repair work orders, and expert experience records (unstructured text).

[0020] Cleaning, chunking, and noise removal are performed on unstructured text; time alignment and outlier filtering are performed on dynamic data.

[0021] Step S2: Construction of a dedicated knowledge graph for urban rail transit operation and maintenance

[0022] Automated construction of domain knowledge graphs by leveraging the entity extraction and relation recognition capabilities of large language models:

[0023] Ontology layer definition: Defines the ontology schema for "equipment-component-fault phenomenon-fault cause-maintenance measures-required spare parts-safety risks".

[0024] Knowledge extraction: A fine-tuned NER (Named Entity Recognition) model is used to extract entities from historical text, and a relation extraction model is used to construct triples (head entity, relation, tail entity).

[0025] Graph Fusion and Storage: Extracted knowledge is stored in a graph database (such as Neo4j) and mapped with the metadata of real-time monitoring data to form a dynamically updated knowledge graph for operation and maintenance.

[0026] Step S3: Fine-tuning of the large language model for the vertical domain and construction of the inference engine

[0027] Base model selection: A general open-source large language model is selected as the base.

[0028] Domain Fine-tuning (SFT): Construct a high-quality instruction fine-tuning dataset containing "fault description - root cause analysis - remediation plan", and use LoRA (low-rank adaptation) technology to fine-tune the base model so that it can master urban rail professional terminology, logical reasoning ability and report generation standards.

[0029] Retrieval Enhancement Generation (RAG) Mechanism: Establish a vector index library to transform subgraph information in the knowledge graph into natural language context.

[0030] Step S4: Automatic Root Cause Analysis and Report Generation

[0031] When a real-time alarm or a manually entered fault description is received:

[0032] Intent recognition and graph retrieval: Parse the fault description, retrieve relevant fault modes, historical cases and device topology neighbor nodes in the knowledge graph, and obtain a set of candidate root causes.

[0033] Multimodal data fusion: The retrieved map knowledge (structured) and real-time monitoring data (numerical) are combined to construct a Prompt.

[0034] Large Model Inference: Input a finely tuned large model, which combines graph knowledge with real-time data status to perform logical reasoning and output:

[0035] Root cause analysis of failures: the most probable causes of failures and their derivation process;

[0036] Risk level assessment: Classification based on the scope of the failure's impact and safety standards (e.g., high / medium / low);

[0037] Recommended repair plan: Specific operating procedures and safety precautions;

[0038] Required spare parts list: Automatically generate a BOM table by associating spare parts entities in the associated diagram.

[0039] Step S5: Human-computer feedback and graph self-evolution

[0040] The maintenance staff confirms or corrects the generated reports, and the system feeds back the feedback data (corrected root causes, actual spare parts used) to the training set, regularly updates the knowledge graph and optimizes the parameters of the large model, so as to achieve continuous evolution of the system.

[0041] As can be seen from the above technical solution, compared with the prior art, the present invention discloses an automatic method and system for analyzing the root causes of urban rail transit faults based on industry knowledge graphs and large models, which has the following beneficial effects:

[0042] (1) Knowledge digitization and reuse: Transforming expert experience scattered in unstructured text into structured knowledge graphs solves the problems of knowledge loss and difficulty in retrieval.

[0043] (2) Accurate and efficient diagnosis: Combining the logical constraints of knowledge graphs with the semantic understanding capabilities of large models, the accuracy and speed of root cause analysis of faults are significantly improved, and the reliance on human experience is reduced.

[0044] (3) Decision-making closed-loop support: Directly generate complete natural language reports containing causes, risks, solutions and spare parts, significantly shortening fault response time (MTTR).

[0045] (4) Adaptive evolution: Through the human-machine feedback mechanism, the system can continuously optimize itself as maintenance data accumulates, adapting to the maintenance needs of new vehicle models and new equipment. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0047] Figure 1 This invention provides an automatic root cause analysis method and system structure diagram for urban rail transit faults based on industry knowledge graphs and large models. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described examples are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Reference Figure 1 As shown, this invention discloses an automatic root cause analysis method for urban rail transit faults based on industry knowledge graphs and large models, including the following steps:

[0050] S1. Acquisition and preprocessing of multi-source heterogeneous data;

[0051] S2, Construction of a dedicated knowledge graph for urban rail transit operation and maintenance;

[0052] S3, fine-tuning of large language models in vertical domains and construction of inference engines;

[0053] S4. Automatic root cause analysis and report generation for faults;

[0054] S5, Human-Computer Feedback and Graph Self-Evolution.

[0055] The invention will now be described in further detail with reference to specific application scenarios.

[0056] Example 1: Fault Diagnosis of Traction System of a Urban Rail Vehicle

[0057] Data input: During train operation, the TCMS system reports an alarm for "traction inverter module overheating", accompanied by voltage fluctuation data.

[0058] Knowledge Graph Search: The system automatically searches the knowledge graph for nodes related to "traction inverter" and "overheating," finding that in similar historical cases, 80% of the causes were "cooling fan filter blockage," 15% were "temperature sensor drift," and 5% were "power module aging." The graph also links to the standard maintenance manual section for this inverter model and the required spare part "cooling fan filter (model XYZ)."

[0059] Large model inference:

[0060] The system combines real-time data (current wind speed is low, temperature rise curve is steep) with map search results to form a Prompt.

[0061] The fine-tuned large model analysis suggests that, combined with the low real-time wind speed data, the most likely cause is "clogged cooling fan filter" (92% confidence level).

[0062] Report Generation: The system automatically generates the following report:

[0063] [Root cause of the problem]: It is highly suspected that the cooling fan filter is clogged, resulting in poor heat dissipation.

[0064] [Risk Level]: Medium (Continued operation may cause the module to burn out; it is recommended to limit the speed or clear passengers at the next stop).

[0065]

Repair Solution

[0066] [Spare Parts List]: Cooling Fan Filter (Inventory Location: Shelf 3, Area A, Quantity: 5).

[0067] Feedback Optimization: After on-site confirmation by maintenance personnel that the filter is clogged, the system clicks "Confirm" in the system. The system then enhances the feature weights of this case in the graph, improving the speed of diagnosing similar faults in the future.

[0068] The various examples in this specification are described in a progressive manner, with each example focusing on the differences from other embodiments. Similar or identical parts between examples can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0069] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An automatic root cause analysis method for urban rail transit faults based on industry knowledge graphs and large models, characterized in that, include: Acquire multi-source heterogeneous data in the field of urban rail transit operation and maintenance, including unstructured historical maintenance records, technical manuals, and real-time equipment monitoring data; Natural language processing technology is used to extract entities and relationships from the unstructured data to construct a knowledge graph for urban rail transit operation and maintenance that includes entities such as equipment, faults, causes, measures, and spare parts. Supervised fine-tuning of a pre-trained large language model based on an urban rail transit corpus yields a vertical domain diagnostic model. When a fault trigger signal is received, a multi-hop search is performed on the knowledge graph based on the fault characteristics to obtain the associated candidate root causes and handling knowledge. The candidate root causes, treatment knowledge, and real-time monitoring data are integrated to construct prompt words, which are then input into the vertical domain diagnostic model to generate a natural language diagnostic report that includes root cause analysis, risk level assessment, recommended maintenance solutions, and a list of required spare parts.

2. The method according to claim 1, characterized in that, The construction of a dedicated knowledge graph for urban rail transit operation and maintenance specifically includes: Define the schema for urban rail transit operation and maintenance, covering professional domains such as rolling stock, signaling, and power supply; A few-shot learning strategy assisted by a large language model is used to extract triples from unstructured text; Establish a mapping relationship between map entities and real-time monitoring data points to realize the association between static knowledge and dynamic data.

3. The method according to claim 1, characterized in that, The generation of the natural language diagnostic report also includes: By using Retrieval Enhanced Generation (RAG) technology, the subgraph structure retrieved from the knowledge graph is transformed into natural language context; Introducing a chain-of-thought mechanism into the prompts guides the large model to demonstrate the reasoning process from phenomenon to root cause.

4. The method according to claim 1, characterized in that, It also includes the self-evolutionary step: Receive user feedback on the generated diagnostic report, including confirmation, correction, or supplementary actual repair results; The entity attributes or relation weights in the knowledge graph are updated based on feedback information, and the corrected data is added to the fine-tuning dataset. The vertical domain diagnostic model is incrementally trained on a regular basis.

5. An automatic root cause analysis system for urban rail transit faults based on industry knowledge graphs and large models, used to implement the method described in any one of claims 1-4, characterized in that, include: The data acquisition and preprocessing module is used for multi-source data access and cleaning. The knowledge graph construction and management module is used for ontology definition, knowledge extraction, storage, and dynamic updating. The model training and inference module includes a large language model for vertical domains and the RAG retrieval engine; The report generation and human-computer interaction module is used to output diagnostic reports and collect user feedback.

6. The system according to claim 5, characterized in that, The system is deployed on cloud servers or edge computing nodes at the urban rail network level, supporting concurrent fault diagnosis requests from multiple lines and multiple vehicle types.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method described in any one of claims 1 to 4.

Images