Large language model-based cutting and winding integrated machine fault diagnosis method and system

Abstract

The application discloses a large language model-based cutting and winding integrated machine fault diagnosis method and system, relates to the technical field of equipment fault diagnosis, and comprises a table data intelligent processing module, a knowledge base construction module, a diagnosis reasoning module and a user interaction module; the table data intelligent processing module performs automatic fault mode labeling and description standardization on a cutting and winding integrated machine fault diagnosis case base provided by a user; a domain knowledge graph construction module integrates a standardized table database and automatically constructs a domain knowledge graph; a diagnosis reasoning module performs fault diagnosis reasoning based on a large language model and the knowledge graph; and a user interaction module provides a multi-modal user interaction interface and supports visual presentation of fault diagnosis. Therefore, the large language model-based cutting and winding integrated machine fault diagnosis method and system can improve the intelligent level of fault diagnosis, reduce the cost of knowledge base construction and expansion, and enhance the ease of use and maintainability of the system.

Description

Technical Field

[0001] This invention relates to the field of equipment fault diagnosis technology, and in particular to a fault diagnosis method and system for a cutting and rolling machine based on a large language model. Background Technology

[0002] With the rapid development of the new energy industry and the continuous growth in energy storage demand, lithium batteries have been widely used due to their high energy density, long cycle life, and environmental performance. In the lithium battery production process, the laser die-cutting and winding machine (hereinafter referred to as the "die-cutting and winding machine") is one of the key pieces of equipment, playing a crucial role in forming the tabs of the electrode roll and winding it into square cells. The stable operation of the die-cutting and winding machine is of great significance to the manufacturing efficiency and product quality of lithium batteries. However, the fault diagnosis and maintenance of the die-cutting and winding machine currently face multiple challenges. Its complex structure increases the difficulty of fault diagnosis technology. Existing diagnostic methods still mainly rely on the experience of maintenance personnel, resulting in low diagnostic efficiency and difficulty in quickly locating the root cause of the fault. Furthermore, although a large amount of relevant data has been accumulated, due to the complex data structure, lack of classification system, and insufficient standardization of description, this data has not been fully explored and effectively utilized, making it difficult to achieve systematic and structured preservation and inheritance, leading to the continuous loss of enterprise knowledge assets.

[0003] To address the issues of insufficient intelligence and low level of data knowledge in fault diagnosis, researchers have proposed a knowledge graph-based equipment fault diagnosis method. This method leverages the powerful data organization, representation, and management capabilities of knowledge graphs to transform multi-source, heterogeneous data into knowledge serving intelligent industrial production and providing decision support for maintenance workers. For example, the invention patent "A Method for Constructing a Knowledge Graph for Fault Diagnosis of a Turnout Switch Machine" uses a bidirectional long short-term memory neural network model and a linear chain conditional random field model for named entity recognition, thereby constructing a knowledge graph for information retrieval. The invention patent "A Method for Diagnosing Concurrent Faults of Transformers Based on Graph Convolutional Neural Networks and Knowledge Graphs" fully models concurrent faults to identify the equipment or components experiencing concurrent faults.

[0004] However, existing technologies still have some shortcomings: First, existing knowledge graph-based device fault diagnosis models largely rely on domain experts manually managing datasets for each task and creating specific models for training. This method is not only time-consuming and labor-intensive but also difficult to maintain in the long term. Second, tabular data is widely used in fault diagnosis knowledge management, and existing technologies have limited research on methods for constructing knowledge graphs based on tabular processing. Furthermore, there is currently a lack of a method and system for end-to-end fault diagnosis of integrated roll-to-roll cutters. Summary of the Invention

[0005] The purpose of this invention is to provide a fault diagnosis method and system for a cutting and rolling machine based on a large language model, which realizes an end-to-end process from tabular data processing to diagnostic suggestion generation, reduces manual dependence, significantly improves diagnostic efficiency and accuracy, and realizes proactive fault diagnosis services.

[0006] To achieve the above objectives, this invention provides a fault diagnosis method for a paper-slicing integrated machine based on a large language model, comprising the following steps:

[0007] S1. Collect various data from the slitting and rolling machine, unify the format and units and integrate them into a table framework, obtain a fault case library of the slitting and rolling machine, divide the equipment structure and define fault modes, and build a domain dictionary;

[0008] S2. For the two sub-tasks of fault mode labeling and description standardization, a special prompt word template is designed. Combined with rule constraints, it guides the large language model to complete the task, realizes the automated processing of tabular data, and builds a standardized fault diagnosis knowledge base.

[0009] In the subtask of fault mode labeling, different labeling strategies were designed, and corresponding fault modes were obtained according to predefined rules. In the subtask of description normalization, a two-layer large language model optimizer-evaluator was constructed. The optimizer uses fault modes as the analysis unit to perform data extraction and serialization representation, semantic grouping and standardization, and optimization output and result writing operations. The evaluator receives the input and output of the optimizer, performs binary scoring according to predefined evaluation dimensions, and feeds back the evaluation results to the optimizer.

[0010] S3. Extract triples from the fault diagnosis knowledge base obtained in step S2 using the knowledge extraction method, design a top-down knowledge framework to construct a domain knowledge graph, store it in the Neo4j database, and connect it with the large language model.

[0011] S4. The large language model is used as a recommender for fault diagnosis suggestions. It receives real-time fault alarm information from the roll cutting machine and supports users to input fault phenomena in a multimodal manner. It performs reasoning by searching the domain knowledge graph and web page queries to provide diagnostic conclusions and solutions.

[0012] In one possible implementation, in step S1, the system structure or functional modules of the slitting and rolling machine are abstracted, and the equipment structure is divided into three layers: modules, components, and parts; the fault mode F is represented as: Mapping method f: M→F, C→F, P→F, where module M represents an independent and indivisible system function in the device, component C represents the structural unit of the module subclass, and part P represents the specific physical component or function of the component subclass.

[0013] In one possible implementation, in step S1, the constructed domain dictionary includes a volume control component dictionary, an alarm dictionary, a stop word dictionary, a dictionary of commonly used words and their synonym representations;

[0014] The component dictionary of the integrated slitting and rolling machine includes the equipment hierarchy structure and its mapping relationship with fault modes, while the alarm dictionary includes the description of operating alarms and its mapping relationship with fault modes.

[0015] In one possible implementation, in step S2, the large language model acts as a pipeline supervisor for fault mode labeling, and performs fault mode labeling through three strategies: entity recognition based on domain dictionary, fault mode recognition based on text similarity, and agent-driven by the large language model.

[0016] Among them, entity recognition based on domain dictionary includes text filtering, disambiguation, and word segmentation based on a pre-built domain dictionary, obtaining fault phrase sequences, matching fault patterns using a forward maximum matching algorithm, and establishing a correction rule base to correct common matching errors.

[0017] Fault pattern recognition based on text similarity includes extracting feature vectors of fault phenomena and operational fault alarm texts using the TF-IDF algorithm, calculating the similarity between fault phenomena and operational fault alarm texts in the alarm dictionary of the integrated machine using cosine similarity, setting a similarity threshold, and mapping the corresponding fault patterns in combination with the domain dictionary.

[0018] The agent driven by the large language model includes labeling rules based on user-inputted fault modes, generating labeled code based on the large language model using React prompt word strategy, and testing and debugging through external tools. Data is filtered, judged and automatically labeled in a provided table, and users review and provide feedback on the results to guide the agent to continuously learn and optimize. At the same time, the labeling rules and code are stored in a correction rule library to support subsequent reuse.

[0019] In one possible implementation, in step S2, for the description normalization subtask, an optimizer-evaluator based on a two-layer large language model is constructed, including:

[0020] The optimizer uses failure modes as the analysis unit to extract all rows of data from the target column and divides them into several text blocks for batch processing. The data is then serialized using a preset prompt word template.

[0021] Based on context learning capabilities, the large language model semantically groups the extracted text content and standardizes and unifies semantically similar but worded descriptions within each group.

[0022] The optimizer has long short-term memory, which stores the original data as well as the reasoning steps, intermediate results and normalization results of each processing step. When it receives feedback from the evaluator, the optimizer adjusts its generation strategy through a reflection mechanism, re-analyzes and optimizes. When all evaluation dimensions meet the standards, a new column will be created and the optimized content corresponding to the original content will be written.

[0023] The optimizer also accumulates expertise from historical processing by integrating an iterative feedback mechanism and performs memory optimization based on the Ebbinghaus forgetting curve.

[0024] The evaluator receives the input and output of the optimizer. In the preset iteration rounds, it evaluates the quality of the optimizer's output using a binary scoring method in combination with predefined evaluation dimensions. When the score does not meet the standard, it feeds back the unqualified content to the optimizer and proposes specific modification suggestions. The above process is repeated until all dimensions meet the standard or the loop round is reached.

[0025] In one possible implementation, the evaluation dimensions include completeness, semantic consistency, simplicity, understandability, format conformity, and illusion detection.

[0026] In one possible implementation, step S3 includes periodically updating the domain knowledge graph. For each newly labeled fault mode, the corresponding data of the fault diagnosis knowledge base is extracted and converted into a computable vector form using vectorization technology. The cosine similarity is used to calculate the correlation between the newly labeled fault mode and the corresponding data of the fault diagnosis knowledge base. It is then determined whether there is any updated content. When new content exists, the updated content is integrated into the existing fault diagnosis knowledge base.

[0027] In one possible implementation, in step S4, for the fault phenomenon input by the user, possible fault modes are identified based on the entity recognition based on the domain dictionary, the fault mode recognition based on text similarity, and the offline rules accumulated in step S1 using the proxy strategy driven by the large language model. Simultaneously, the input text is aligned with entities in the domain knowledge graph using the m3e embedding model. By calculating cosine similarity, the top k nodes with the highest scores are selected as candidate nodes, and the corresponding fault modes are further queried, as follows:

[0028]

[0029] In the formula, Sim(E q E t E represents the similarity score. q E t These are knowledge graph entities and user query entities, respectively.

[0030] Considering that the information obtained by the above retrieval methods may contain noise and irrelevant details, an LLM (Local Knowledge Graph) is introduced as a retrieval evaluator. The evaluator judges the relevance between candidate fault modes and their corresponding fault phenomenon lists and the user's query description. Relevance is divided into two confidence levels: "correct" and "incorrect." Fault modes judged as relevant are retained, while irrelevant ones are discarded, and the final filtered list of fault modes is output. For the identified fault modes, related historical fault cases are further retrieved to achieve knowledge reuse. If no information exactly matching the user's query is found in the domain knowledge graph, supplementary information is actively retrieved from external web pages through prompts, further recommending potential fault causes and solutions.

[0031] This invention also provides a fault diagnosis system for a paper-slicing and wrapping machine based on a large language model, used to implement the aforementioned fault diagnosis method for paper-slicing and wrapping machines based on a large language model. The system includes an intelligent table data processing module, a domain knowledge graph construction module, a diagnostic reasoning module, and a user interaction module.

[0032] The aforementioned intelligent table data processing module uses a large language model as a pipeline monitor for fault mode labeling, which is used to automatically label fault modes in the fault diagnosis case library of the roll cutting machine provided by the user; at the same time, an optimizer-evaluator based on a two-layer large language model is constructed to process the files provided by the user, standardize the description of the target column, and build a standardized fault diagnosis knowledge base.

[0033] The domain knowledge graph construction module consists of a pattern layer related to equipment and a pattern layer related to faults. The two pattern layers interact through fault patterns. It is used to integrate the tabular database after the standardization of the roll cutting machine. It uses Python to build an extraction function to extract triples and store them in the Neo4j database. Then, it uses LangChain's vectorstores to connect the Neo4j database with the large language model and regularly performs incremental updates to the domain knowledge graph.

[0034] The diagnostic reasoning module uses a large language model as a recommender for fault diagnosis suggestions. When a corresponding case exists in the constructed domain knowledge graph, a recommendation is generated directly; otherwise, the device information related to the fault is identified, and the cause of the fault and the solution are inferred through network search.

[0035] The user interaction module is used to provide a multimodal user interaction interface and support the visualization of fault diagnosis.

[0036] In one possible implementation, the optimizer-evaluator based on a two-layer large language model includes:

[0037] The optimizer generates the initial output, while the evaluator provides feedback to the optimizer. Entity alignment and knowledge fusion are achieved through a two-way loop, and the file processing results and process explanations are returned.

[0038] The optimizer uses the large language model as a professional assistant for database normalization management, performing data extraction and serialization representation, semantic grouping and standardization, optimizing output and writing results, and self-evolutionary tasks;

[0039] The evaluator takes the large language model as the manager of database normalization management, receives input and output from the optimizer, and evaluates the output quality of the professional assistant in a preset iteration cycle by using binary scoring combined with predefined evaluation dimensions, and proposes specific modification suggestions.

[0040] Therefore, the present invention employs the above-mentioned fault diagnosis method and system for integrated slitting and rolling machines based on a large language model, and has the following technical effects:

[0041] (1) This invention designs an automatic knowledge graph construction method from tabular data, including automated fault mode labeling, entity alignment and knowledge fusion, which improves the efficiency, flexibility and accuracy of knowledge base construction, reduces labor costs and human errors, and lowers the cost of knowledge base construction and expansion.

[0042] (2) This invention forms a complete solution from data processing to diagnostic suggestions, while supporting the tracing of knowledge sources, assisting maintenance personnel in guided troubleshooting, shortening downtime, reducing reliance on human experience, and improving the level of intelligence in fault diagnosis.

[0043] (3) The system provided by the present invention supports users to interact with the system through natural language, which reduces the operation threshold and improves the user experience. At the same time, with the continuous interaction of users and the dynamic update of the knowledge graph, the model is iteratively updated, which improves the accuracy of understanding business needs and the quality of answers, and enhances the ease of use and maintainability of the system.

[0044] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0045] Figure 1 This is a schematic diagram illustrating the fault mode classification of the integrated slitting and rolling machine in an embodiment of the fault diagnosis method and system based on a large language model.

[0046] Figure 2 This is a schematic diagram of the mode layer of the integrated slitting and rolling machine in an embodiment of the fault diagnosis method and system for integrated slitting and rolling machines based on a large language model;

[0047] Figure 3This is a schematic diagram of the design and development process of a large language model for tabular data processing in an embodiment of a fault diagnosis method and system for a paper-slicing and paper-cutting machine based on a large language model.

[0048] Figure 4 This is a schematic diagram of the fault diagnosis reasoning process in an embodiment of a fault diagnosis method and system for a roll-to-roll cutter based on a large language model;

[0049] Figure 5 This is a schematic diagram of the fault diagnosis method for a slitting and rolling machine based on a large language model in an embodiment of the fault diagnosis method and system for a slitting and rolling machine. Detailed Implementation

[0050] The present invention will be explained in more detail through the following embodiments. The purpose of disclosing the present invention is to protect all changes and modifications within the scope of the present invention. The present invention is not limited to the following embodiments.

[0051] Please see Figures 1 to 3 This invention provides a fault diagnosis method for a paper-cutting integrated machine based on a large language model, comprising the following steps:

[0052] S1. By collecting and integrating data related to the slitting and rolling machine, a fault case library is obtained, fault modes are defined, and a domain dictionary is constructed, thus initially forming the data foundation for fault diagnosis, as detailed below:

[0053] Data collection: Extensive data collection is carried out on all aspects of the slitting and rolling machine. The data sources include (1) Equipment user manual: covering the main functions, related parameters and performance of the equipment and its sub-modules, the models of the main components, common faults and handling methods, equipment operation instructions and maintenance specifications; (2) Operation data: software system provided by the equipment manufacturer, whose interface includes functions such as operation fault alarm, working condition parameter setting, alarm history record; (3) Historical maintenance work order: records the equipment repair process, including equipment information, fault phenomenon, fault cause, maintenance measures, maintenance time and maintenance personnel; (4) Equipment fault maintenance analysis report: detailed root cause analysis is carried out for high-frequency or high-hazard impact faults, including equipment information, maintenance time, fault phenomenon, fault cause, maintenance description, preventive measures, maintenance personnel and spare parts consumption.

[0054] Data Integration and Preliminary Preprocessing: The collected data is integrated into a unified table framework, including six basic elements: equipment information, fault symptoms, fault troubleshooting steps, fault causes, repair plans, and preventive measures. A unified format and units are used, and regular expressions are employed to remove erroneous, missing, duplicate, and invalid table rows, resulting in a fault case library for the roll-to-roll cutter. Invalid rows typically include excessively short text descriptions of the fault symptoms, or fault causes and repair plans consisting of single sentences containing meaningless words such as "not found," "unidentified," or "no abnormalities."

[0055] Defining Equipment Structure and Failure Modes: The system structure or functional modules of the slitting and winding machine are abstracted, dividing it into a three-layer structure of modules, components, and parts. Failure mode F can be represented as: Mapping method f: M→F, C→F, P→F. Module M represents an independent and indivisible system function in the equipment, component C represents the structural unit of the module subclass, and part P represents the specific physical component or function of the component subclass. Modules include 10 categories: positive and negative electrode feeding, positive and negative electrode unwinding, positive and negative electrode laser, positive electrode cutting and adhesive application, upper and lower diaphragms, winding, adhesive application, unloading, unloading machine, and overall control system. Components and parts differ according to the characteristics of each module.

[0056] Construct a domain dictionary: including a component dictionary for the slitting and rolling machine, an alarm dictionary, a stop word dictionary, and a dictionary of commonly used words and their synonyms. The component dictionary for the slitting and rolling machine includes the equipment hierarchy structure and its mapping to fault modes, while the alarm dictionary includes operational alarm descriptions and their mapping to fault modes.

[0057] In one embodiment, a sensor network provides critical data for detecting and diagnosing faults in the lithium battery production line. Fault text descriptions derived from the sensor data are used to understand the operational health of various industrial equipment. For example, fault downtime descriptions are filtered, equipment hierarchical structures are segmented using the methods described above, and fault modes are abstracted and categorized. See also... Figure 1 For the error message "AJGQAlm.W16.b01 Negative Electrode Laser Cutting Point-like Reinforcing Rib Cylinder Reached the End of Service Life!", the equipment hierarchy is: module "Negative Electrode Laser", component "Cylinder", and part "Point-like Reinforcing Rib Cylinder". For the error message "CJGQAlm.W10.b02 Positive Electrode Laser Cutting Strip-like Reinforcing Rib Cylinder Not Returning or Sensor Failure", the equipment hierarchy is: module "Positive Electrode Laser", component "Cylinder", and part "Strip-like Reinforcing Rib Cylinder". The corresponding fault modes for both are "Positive and Negative Electrode Laser - Cylinder - Reinforcing Rib Cylinder". The equipment hierarchy, sensor text, and their correspondence with fault modes are compiled into a component dictionary and an alarm dictionary for the integrated cutting and rolling machine, stored in tabular form, as shown in Tables 1 and 2.

[0058] Table 1 Alarm Dictionary

[0059]

[0060] Table 2. Dictionary of Components for a Cutting and Rolling Machine

[0061]

[0062] S2. Compile a fault case library for the integrated slitting and winding machine, including fault symptoms, causes, and repair solutions. After data preprocessing, fault mode labeling and standardized descriptions are performed. Please refer to [link / reference]. Figure 3 This embodiment develops a large language model for tabular data processing. For the two sub-tasks of fault mode labeling and description standardization, a special prompt word template is designed. Combined with rule constraints, the large language model is guided to complete the task, realize the automated processing of tabular data, and thus build a standardized fault diagnosis knowledge base.

[0063] The prompt word template is designed with eight core elements to ensure that the LLM can accurately understand the task objective and generate the required output:

[0064] Prompt={T, R, TS, I, TD, FA, Rea, IO};

[0065] Here, time (T) records the current time; role (R) defines the LLM's role, capabilities, task context, code of conduct, and next instructions, helping the LLM understand its tasks and limitations; toolset (TS) describes the external tools the LLM can use; I represents the received input; final answer (FA) specifies the content and format of the output for accurate identification of user intent; task decomposition (TD) summarizes and automatically decomposes the user's task requirements; Rea is the reasoning strategy adopted by the LLM to complete the task; and intermediate output (IO) is the intermediate result generated by the LLM during the iteration process, used by the LLM to learn from feedback.

[0066] For fault mode labeling, three strategy functions are used to implement fault mode labeling:

[0067] First, the entity recognition method based on the domain dictionary includes (1) word segmentation: combining the domain dictionary to segment the fault phenomenon and obtain the fault phrase sequence; (2) component matching: using the positive maximum matching algorithm to match the fault phrase with the component dictionary of the cutting and rolling machine; (3) mapping fault mode: according to the mapping relationship between the equipment hierarchy and the fault mode, the corresponding fault mode is obtained; (4) error correction: establishing a correction rule base to correct common matching errors; (5) result storage: creating a new column in the table to store the annotation results. For example, extract "parts" and "part mode" from the component dictionary of the cutting and rolling machine, remove duplicates and store them in a txt file as a custom dictionary for jieba word segmentation. Given the fault description "negative electrode strip reinforcing rib rubber roller wear", the positive maximum matching is used to obtain the word segmentation result "negative electrode / strip reinforcing rib / rubber roller / wear". According to the component dictionary of the cutting and rolling machine, the fault mode corresponding to "strip reinforcing rib" is "positive and negative electrode laser-cylinder-reinforcing rib cylinder".

[0068] Second, the fault mode recognition method based on text similarity includes: (1) extracting feature vectors of fault phenomena and operation fault alarm texts using the TF-IDF algorithm; (2) similarity calculation: calculating the similarity between fault phenomena and operation fault alarm texts in the alarm dictionary of the integrated winding and cutting machine using cosine similarity; (3) threshold judgment: if the similarity is greater than the preset threshold, the corresponding fault mode is obtained according to the mapping relationship between the alarm description and the fault mode; (4) result storage: creating a new column in the table to store the labeling results. For example, given the fault description "normal winding alarm: positive pole four-stage correction motor exceeds the set lower limit value", calculate its TF-IDF vector and cosine value similarity with "processed alarm description" in the alarm dictionary, and obtain that the description is similar to "positive pole four-stage correction motor exceeds the set upper limit value" in the alarm dictionary, and obtain the corresponding fault mode "winding-motor-four-stage correction motor", then label the fault mode as "winding-motor-four-stage correction motor".

[0069] Third, fault mode annotation is performed based on a large language model-driven agent. Users are required to provide files, annotation instructions, and rules. The large language model then uses a React strategy to generate code to manipulate the files for annotation. This method can handle common colloquial expressions. Specifically, it includes: (1) Code generation and execution: Annotated code is generated based on the React prompt word strategy and tested and debugged using external tools (such as PythonREPLTool). New columns are created in the table to store the annotation results; (2) Continuous learning and optimization: Users review the results. If the review is unsatisfactory, feedback is provided to guide the agent to regenerate. If the review is satisfactory, the annotation rules and code are stored in the rule base for subsequent reuse. For example, given the following annotation instructions and rules: "Apply the given rules to the [Fault Phenomenon] column and execute the corresponding annotation in the [Strategy 3 Annotation Result] column. The rules are independent of each other, and multiple annotations are allowed. Each value is concatenated with an English comma. It should be noted that: [OR] relationship means that the presence of any keyword will trigger the annotation; [AND] relationship means that the specified keyword must be present simultaneously. The following are the annotation rules: If it contains ([positive electrode] or [negative electrode]) and [CCD], then annotate [positive / negative electrode laser-CCD-laser cut CCD]; if it contains [overhang], then annotate [winding-CCD-head and tail length]."

[0070] Based on the fault modes labeled by the above strategy, the final fault mode is obtained by judging through predefined rules and a new column is created for storage. Among them, the predefined rules include: (1) Consistency principle: if the labeling results of multiple methods are consistent and unique, the result is directly adopted; (2) Majority principle: for multiple results, weighted voting is used to determine the final fault mode; (3) Similarity inference: for fault phenomena that have not been successfully labeled, the semantic similarity between them and existing labeled samples is calculated and inferred labeling is performed.

[0071] For the description normalization subtask, an optimizer-evaluator based on a two-layer large language model is constructed. Its core task is for the user to provide the file to be processed and specify the target columns that need to be normalized. The optimizer generates the initial output, and the evaluator provides feedback to it. Entity alignment and knowledge fusion are achieved through bidirectional loops, and finally the file processing results and process explanation are returned.

[0072] The optimizer defines the large language model as a "professional assistant for database normalization management." Its core tasks include: (1) Data extraction and serialization representation: Extracting all rows of data from the target column and serializing the data using preset prompt word templates. Serialization methods include, but are not limited to, JSON, Markdown, CSV, DFLoader, and other formats. (2) Semantic grouping and standardization: Based on context learning capabilities, the large language model performs semantic grouping on the extracted text content. Within each group, descriptions that are semantically similar but have different wording are further standardized and unified. (3) Optimization output and writing results: The optimizer has long short-term memory and can store the original data as well as the reasoning steps, intermediate results, and normalization results in each processing process. When the evaluator provides feedback on the evaluation results, the optimizer adjusts its generation strategy through a reflection mechanism and re-analyzes and optimizes. When all evaluation dimensions meet the standards, a new column will be created and the optimized content corresponding to the original content will be written. (4) Self-evolution: Through an integrated iterative feedback mechanism, it can accumulate professional knowledge from historical processing and improve its processing capabilities in similar tasks. Memory optimization based on the Ebbinghaus forgetting curve improves the processing capacity of long-span information.

[0073] The evaluator defines the large language model as the "supervisor" and receives the input and output of the optimizer. Its task is to evaluate the output quality of the "professional assistant" in a preset iteration cycle by using a binary scoring method combined with predefined evaluation dimensions. If the score does not meet the standard, the "supervisor" will provide feedback on the unqualified content to the "professional assistant" and make specific modification suggestions. The above process is repeated until all dimensions meet the standard or the cycle cycle is reached.

[0074] The evaluation dimensions include: (1) Completeness: All original data are correctly preserved or classified, and no valid information or cases are omitted; (2) Semantic consistency: All identical or similar descriptions use uniform terminology; (3) Conciseness: The descriptions are concise, redundant information is removed, and no important content is lost; (4) Understandability: There are no ambiguous or obscure expressions; (5) Format standardization: The data is output according to the specified structure, the hierarchy is clear, and the information is complete; (6) Illusion detection: All information can be found in the original data, and no new information that cannot be derived from the original data is generated.

[0075] S3. Extract triples from the fault diagnosis knowledge base obtained in step S2 using the knowledge extraction method, design a top-down knowledge framework to construct a domain knowledge graph, store it in the Neo4j database, and connect it with the large language model.

[0076] This embodiment designs a schema layer consisting of two interacting sublayers: a device-related schema layer and a fault-related schema layer. These layers interact through fault patterns, employing a top-down approach to construct the domain knowledge graph. The device-related schema layer describes the actual device structure, including three entity categories: modules, components, and parts. The fault-related schema layer consists of fault entities extracted from the knowledge base, including six categories: fault patterns, phenomena, causes, troubleshooting steps, maintenance plans, and preventative measures. The attributes of each entity category are defined as the names of the corresponding instances. Since fault patterns occur on corresponding faulty components, the attributes also include the names of modules, components, and parts. The relationships between entity categories are defined as follows: "Module - Contains - Component", "Module - Contains - Part", "Component - Contains - Part", "Module - Contains - Fault Pattern", "Component - Contains - Fault Pattern", "Part - Contains - Fault Pattern", "Fault Pattern - Exists - Fault Phenomenon", "Fault Cause - Leads to - Fault Phenomenon", "Fault Cause - Leads to - Fault Pattern", "Fault Phenomenon - Exists - Fault Troubleshooting Steps", "Fault Cause - Suggestion - Maintenance Plan", and "Fault Cause - Suggestion - Preventative Measures".

[0077] Guided by the knowledge organization architecture of the schema layer, an extraction function is constructed using Python to extract triples from the fault diagnosis knowledge base obtained in step S2, construct a domain knowledge graph, and store it in the Neo4j database. Then, LangChain's vectorstores are used to connect the Neo4j graph database with the large language model, integrating them into a system. In one embodiment, the update of the domain knowledge graph includes (1) for new table data, using the fault pattern annotation method, obtaining the fault pattern corresponding to each record; (2) for each fault pattern, extracting the same column content of the knowledge base and the new table, converting it into a computable vector form through vectorization technology, calculating the correlation between the new table data and the fault diagnosis knowledge base content through cosine similarity, and determining whether there is new content; (3) if there is updated content, integrating the new data with the content of the knowledge base, and updating the integrated new content to the existing fault diagnosis knowledge base, thereby realizing the periodic incremental update of the knowledge graph.

[0078] S4, please see Figure 4 and Figure 5The system uses a large language model as a recommender for fault diagnosis suggestions, receives real-time fault alarm information from the roll-to-roll cutter, and supports users to input fault phenomena in a multimodal manner. It then uses domain knowledge graph retrieval and web page queries to perform reasoning, providing diagnostic conclusions and solutions, as detailed below:

[0079] User interaction and intent understanding: When the roll-to-roll cutter malfunctions, the large language model can receive and generate an analysis report in real time; at the same time, it supports users to conduct fault diagnosis questions and answers through natural language text and voice descriptions, maintains continuous dialogue with users, and records users' historical interaction content in order to combine the context for accurate fault reasoning.

[0080] Knowledge Graph Query and Reasoning: The large language model is defined as a "fault diagnosis expert for integrated paper cutting machines." Given input, it first uses a self-questioning strategy combined with m3e vector embedding to understand user intent and automatically generates structured query Cypher statements for the knowledge graph. The retrieved triples are further filtered and summarized to generate comprehensive diagnostic conclusions. If no information matching the user's query is found in the knowledge graph, it proactively retrieves supplementary information from external web pages based on the abnormal fault symptoms in the input fault descriptions, further recommending potential fault causes and solutions.

[0081] Visualization and Interpretation of Diagnostic Results: Based on queries and reasoning results from domain knowledge graphs or external web pages, fault diagnosis conclusions are generated, including multimodal visualization and natural language interpretation. Multimodal visualization presents the diagnostic results in the form of charts, flowcharts, or tree diagrams, highlighting key fault nodes and influencing relationships, and providing detailed information on key nodes. Natural language interpretation involves the system converting the diagnostic results and reasoning process into natural language descriptions for easy user understanding.

[0082] This invention also provides a fault diagnosis system for a paper-slicing and wrapping machine based on a large language model, used to implement the aforementioned fault diagnosis method for paper-slicing and wrapping machines based on a large language model. The system includes an intelligent table data processing module, a domain knowledge graph construction module, a diagnostic reasoning module, and a user interaction module, as follows:

[0083] The intelligent processing module for tabular data uses a large language model as a pipeline monitor for fault mode annotation, which is used to automatically annotate the fault mode of the user-provided integrated roll cutting machine fault diagnosis case library; at the same time, an optimizer-evaluator based on a two-layer large language model is built to process the user-provided files, standardize the description of the target columns, and build a standardized fault diagnosis knowledge base.

[0084] The domain knowledge graph construction module consists of two pattern layers: one related to equipment and the other to faults. These two pattern layers interact through fault patterns. The module integrates the standardized tabular database of the roll-to-roll cutter. It uses Python to build an extraction function to extract triples and store them in the Neo4j database. Then, it uses LangChain's vectorstores to connect the Neo4j database with the large language model and performs incremental updates to the domain knowledge graph on a regular basis.

[0085] Diagnostic reasoning module: The large language model is used as a recommender for fault diagnosis suggestions. When there are corresponding cases in the constructed domain knowledge graph, recommendations are generated directly; otherwise, the device information related to the fault is identified and the cause of the fault and solution are inferred through network search.

[0086] User interaction module: Provides a multimodal user interaction interface and supports the visualization of fault diagnosis.

[0087] Therefore, the fault diagnosis method and system for the integrated cutting and rolling machine based on the large language model described above can improve the intelligence level of fault diagnosis, reduce the cost of knowledge base construction and expansion, and enhance the ease of use and maintainability of the system.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A fault diagnosis method for a slitting and rolling machine based on a large language model, characterized in that, Includes the following steps: S1. Collect various data from the slitting and rolling machine, unify the format and units and integrate them into a table framework, obtain a fault case library of the slitting and rolling machine, divide the equipment structure and define fault modes, and build a domain dictionary; S2. For the two sub-tasks of fault mode labeling and description standardization, a special prompt word template is designed. Combined with rule constraints, it guides the large language model to complete the task, realizes the automated processing of tabular data, and builds a standardized fault diagnosis knowledge base. Within the subtask of fault mode labeling, three strategy functions are used to implement fault mode labeling, specifically: The entity recognition method based on the domain dictionary includes: word segmentation: segmenting fault phenomena into words using the domain dictionary to obtain fault phrase sequences; component matching: matching fault phrases with the component dictionary of the integrated cutting and rolling machine using the forward maximum matching algorithm; mapping fault modes: obtaining the corresponding fault modes based on the mapping relationship between the equipment hierarchy and fault modes; error correction: establishing a correction rule base to correct common matching errors; and result storage: creating a new column in the table to store the annotation results. The text similarity-based fault pattern recognition method includes: extracting feature vectors of fault phenomena and operational fault alarm texts using the TF-IDF algorithm; similarity calculation: calculating the similarity between fault phenomena and operational fault alarm texts in the alarm dictionary of the integrated roll-to-roll machine using cosine similarity; threshold judgment: if the similarity is greater than a preset threshold, the corresponding fault pattern is obtained according to the mapping relationship between alarm description and fault pattern; result storage: creating a new column in the table to store the annotation results. Fault mode annotation is performed using an agent driven by a large language model. Users are required to provide files, annotation instructions, and rules. The large language model then uses a React strategy to generate code to manipulate the files for annotation. Specifically, this includes: Code generation and execution: Annotated code is generated based on a React prompt word strategy and tested and debugged using external tools. A new column is created in a table to store the annotation results; Continuous learning and optimization: Users review the results. If the review is unsatisfactory, feedback is provided to guide the agent to regenerate. If the review is satisfactory, the annotation rules and code are stored in a rule base for future reuse; The given rules are applied to the fault phenomenon column, and the corresponding annotations are executed in the strategy three annotation result column. Rules are independent of each other, and multiple annotations are allowed, with each value separated by a comma. Based on the fault modes labeled using the above strategy, the final fault mode is determined by predefined rules and stored in a new column. The predefined rules include: consistency principle: if the labeling results of multiple methods are consistent and unique, then the result is directly adopted; majority principle: for cases with multiple results, weighted voting is used to determine the final fault mode. In the subtask of description normalization, a two-layer large language model optimizer-evaluator is constructed. The optimizer, based on the user-provided file to be processed and the target columns to be normalized, uses failure modes as the analysis unit to perform data extraction and serialization representation, semantic grouping and standardization, and optimization output and result writing operations. The evaluator receives the input and output of the optimizer, performs binary scoring according to predefined evaluation dimensions, and feeds back the evaluation results to the optimizer. Specifically: The optimizer uses failure modes as the analysis unit to extract all rows of data from the target column and divides them into several text blocks for batch processing. The data is then serialized using a preset prompt word template. Based on context learning capabilities, the large language model semantically groups the extracted text content and standardizes and unifies semantically similar but worded descriptions within each group. The optimizer has long short-term memory, which stores the original data as well as the reasoning steps, intermediate results and normalization results of each processing step. When it receives feedback from the evaluator, the optimizer adjusts its generation strategy through a reflection mechanism, re-analyzes and optimizes. When all evaluation dimensions meet the standards, a new column will be created and the optimized content corresponding to the original content will be written. The optimizer also accumulates expertise from historical processing by integrating an iterative feedback mechanism and performs memory optimization based on the Ebbinghaus forgetting curve. The evaluator receives the input and output of the optimizer. In a preset iteration cycle, it evaluates the quality of the optimizer's output using a binary scoring method in combination with predefined evaluation dimensions. When the score is not up to standard, the evaluator feeds back the unqualified content to the optimizer and makes specific modification suggestions. The above process is repeated until all dimensions meet the standard or the cycle is completed. The evaluation dimensions include completeness, semantic consistency, conciseness, understandability, format standardization, and illusion detection. S3. Extract triples from the fault diagnosis knowledge base obtained in step S2 using the knowledge extraction method, design a top-down knowledge framework to construct a domain knowledge graph, store it in the Neo4j database, and connect it with the large language model. S4. The large language model is used as a recommender for fault diagnosis suggestions. It receives real-time fault alarm information from the roll cutting machine and supports users to input fault phenomena in a multimodal manner. It performs reasoning by searching the domain knowledge graph and external web page queries to provide diagnostic conclusions and solutions.

2. The fault diagnosis method for a paper-cutting integrated machine based on a large language model according to claim 1, characterized in that, In step S1, the system structure or functional modules of the slitting and rolling machine are abstracted, and the equipment structure is divided into three layers: modules, components, and parts; the fault mode F is represented as: The mapping method f is: M→F, C→F, P→F, where module M represents an independent and indivisible system function in the device, component C represents the structural unit of the module subclass, and part P represents the specific physical component or function of the component subclass.

3. The fault diagnosis method for a slitting and rolling machine based on a large language model according to claim 1, characterized in that, In step S1, the constructed domain dictionary includes a component dictionary of the integrated slitting and rolling machine, an alarm dictionary, a stop word dictionary, a dictionary of commonly used words and their synonym representations; The component dictionary of the integrated slitting and rolling machine includes the equipment hierarchy structure and its mapping relationship with fault modes, while the alarm dictionary includes the description of operating alarms and its mapping relationship with fault modes.

4. The fault diagnosis method for a slitting and rolling machine based on a large language model according to claim 1, characterized in that, In step S2, for the description normalization subtask, an optimizer-evaluator based on a two-layer large language model is constructed. The optimizer generates the initial output, and the evaluator provides feedback to the optimizer. Entity alignment and knowledge fusion are achieved through bidirectional loop, and the file processing results and process explanation are returned.

5. The fault diagnosis method for a paper-cutting integrated machine based on a large language model according to claim 1, characterized in that, Step S3 includes periodically updating the domain knowledge graph. For each newly labeled fault mode, the corresponding data of the fault diagnosis knowledge base is extracted and converted into a computable vector form through vectorization technology. The cosine similarity is used to calculate the correlation between the newly labeled fault mode and the corresponding data of the fault diagnosis knowledge base. It is then determined whether there is any updated content. When there is new content, the updated content is integrated into the existing fault diagnosis knowledge base.

6. The fault diagnosis method for a paper-cutting integrated machine based on a large language model according to claim 1, characterized in that, In step S4, for the fault phenomenon input by the user, based on the entity recognition based on the domain dictionary, the fault pattern recognition based on text similarity, and the offline rules accumulated in step S1 by the proxy strategy driven by the large language model, candidate fault patterns are identified. Then, the evaluator is used to retain relevant fault patterns based on the correlation between the candidate fault patterns and the user's query description, and the filtered fault pattern list is output. Finally, the corresponding fault causes and solutions are generated by querying through the domain knowledge graph or external web pages.

7. A fault diagnosis system for a slitting and rolling machine based on a large language model, used to implement the fault diagnosis method for a slitting and rolling machine based on a large language model as described in any one of claims 1-6, characterized in that, It includes a module for intelligent processing of tabular data, a module for constructing a domain knowledge graph, a module for diagnostic reasoning, and a module for user interaction: The intelligent processing module for table data uses a large language model as a pipeline monitor for fault mode labeling, which is used to automatically label fault modes in the fault diagnosis case library of the roll cutting machine provided by the user; at the same time, an optimizer-evaluator based on a two-layer large language model is constructed to process the files provided by the user, standardize the description of the target column, and build a standardized fault diagnosis knowledge base. The domain knowledge graph construction module consists of a pattern layer related to equipment and a pattern layer related to faults. The two pattern layers interact through fault patterns. It is used to integrate the tabular database after the standardization of the roll cutting machine. It uses Python to build an extraction function to extract triples and store them in the Neo4j database. Then, it uses LangChain's vectorstores to connect the Neo4j database with the large language model and regularly performs incremental updates to the domain knowledge graph. The diagnostic reasoning module uses the large language model as a recommender for fault diagnosis suggestions. When the constructed domain knowledge graph contains corresponding cases, it directly generates recommendations. Otherwise, identify the equipment information currently related to the fault, and infer the cause of the fault and solution through network search; The user interaction module is used to provide a multimodal user interaction interface and support the visualization of fault diagnosis.

8. The fault diagnosis system for a cutting and rolling machine based on a large language model according to claim 7, characterized in that, The constructed optimizer-evaluator includes: The optimizer generates the initial output, while the evaluator provides feedback to the optimizer. Entity alignment and knowledge fusion are achieved through a two-way loop, and the file processing results and process explanations are returned. The optimizer uses the large language model as a professional assistant for database normalization management, performing data extraction and serialization representation, semantic grouping and standardization, optimizing output and writing results, and self-evolutionary tasks; The evaluator takes the large language model as the manager of database normalization management, receives input and output from the optimizer, and evaluates the output quality of the professional assistant in a preset iteration cycle by using binary scoring combined with predefined evaluation dimensions, and proposes specific modification suggestions.

Images