Steel production equipment fault detection method and device and storage medium

Abstract

The embodiment of the invention provides a steel production equipment fault detection method and device and a storage medium, and the method comprises the steps: obtaining historical equipment fault maintenance data, real-time equipment data and user problems; based on an SBert model, according to a soft prompt optimization technology and the historical equipment fault maintenance data, generating a semantic embedding vector; on the basis of an unsupervised clustering algorithm, according to a preset cluster number and a preset enhanced prompt template, clustering each semantic embedding vector to generate a new enhanced prompt template; based on a large language model, generating an equipment fault knowledge graph according to the new enhanced prompt template and the historical equipment fault maintenance data; and generating a fault detection result according to the real-time equipment data, the equipment fault knowledge graph and the user problem. Therefore, the accuracy and flexibility of fault detection are improved.

Description

TECHNICAL FIELD

[0001] The present document relates to the technical field of industrial intelligence and equipment fault diagnosis, and particularly relates to a steel production equipment fault detection method and device and a storage medium. BACKGROUND

[0002] Steel production equipment is usually large and complex, and once the real fault cannot be timely investigated and solved, it may cause the production line to stop working, resulting in loss of manpower and material resources, and small problems that are not solved in time may lead to greater faults, threatening the safety of workers, affecting production efficiency and the use time of equipment.

[0003] The prior art mainly forms fixed rules through expert knowledge or collects fault data through sensors and manually labels fault types, and then trains based on traditional machine learning models to help users identify fault types.

[0004] However, it is difficult to find the real cause from multiple fault factors based on the existing machine learning model, resulting in incomplete fault detection and affecting the efficiency of fault diagnosis; when a new type of fault occurs, it is difficult to match the rules in traditional machine learning, resulting in inaccurate or lagging detection results, making it difficult to respond to sudden situations in actual production processes in time; and it is difficult to extract fault diagnosis rules from unstructured data, resulting in knowledge waste. SUMMARY

[0005] In view of the above scheme, the present application aims to provide a steel production equipment fault detection method, device and storage medium to solve at least one of the above technical problems.

[0006] In a first aspect, one or more embodiments of the present specification provide a steel production equipment fault detection method, comprising:

[0007] Obtaining historical equipment fault maintenance data, real-time equipment data and user problems;

[0008] Based on the SBert model, generating a semantic embedding vector according to the soft prompt optimization technology and the historical equipment fault maintenance data;

[0009] Based on an unsupervised clustering algorithm, clustering each semantic embedding vector according to a preset cluster number and a preset enhanced prompt template to generate a new enhanced prompt template;

[0010] Based on a large language model, generating an equipment fault knowledge graph according to the new enhanced prompt template and the historical equipment fault maintenance data;

[0011] Generating a fault detection result according to the real-time equipment data, the equipment fault knowledge graph and the user problems.

[0012] Further, further comprising,

[0013] Obtaining user maintenance correction suggestions;

[0014] Updating the SBert model according to the user maintenance correction suggestions.

[0015] Further, according to the historical equipment failure maintenance data, extracting first entity data and first relationship data;

[0016] Based on the soft prompt optimization technology, according to the first entity data and the first relationship data, constructing triple data;

[0017] Based on the loss function, according to the triple data, training the SBert model to obtain an optimized SBert model;

[0018] Based on the optimized SBert model, according to the first entity data and the first relationship data, generating semantic embedding vectors.

[0019] Further, based on a large language model, according to the historical equipment failure maintenance data, selecting an enhanced prompt template;

[0020] According to the historical equipment failure maintenance data and the enhanced prompt template, extracting second entity data and second relationship data;

[0021] According to the second entity data and the second relationship data, generating a device failure knowledge graph.

[0022] In a second aspect, the embodiments of the present application provide a steel equipment failure detection device, comprising,

[0023] A data collection module is configured to obtain historical equipment failure maintenance data, real-time equipment data, and user questions;

[0024] An optimization module is configured to generate semantic embedding vectors based on a SBert model, a soft prompt optimization technology, and the historical equipment failure maintenance data;

[0025] An enhanced prompt template generation module is configured to cluster each semantic embedding vector based on a preset cluster number and a preset enhanced prompt template to generate a new enhanced prompt template using an unsupervised clustering algorithm;

[0026] A knowledge graph construction module is configured to generate a device failure knowledge graph based on a large language model, the new enhanced prompt template, and the historical equipment failure maintenance data;

[0027] A diagnosis module is configured to generate a failure detection result based on the real-time equipment data, the device failure knowledge graph, and the user questions.

[0028] Further, a human-computer interaction module is further included,

[0029] Obtaining a user maintenance correction suggestion;

[0030] Updating the SBert model according to the user maintenance correction suggestion.

[0031] Further, the optimization module is configured to:

[0032] According to the historical equipment failure maintenance data, extracting first entity data and first relationship data;

[0033] Based on a soft prompt optimization technology, constructing triple data according to the first entity data and the first relationship data;

[0034] Based on a loss function, training the SBert model according to the triple data to obtain an optimized SBert model;

[0035] Based on the optimized SBert model, generating semantic embedding vectors according to the first entity data and the first relationship data.

[0036] Further, the knowledge graph construction module is configured to:

[0037] Based on a large language model, selecting an enhanced prompt template according to the historical equipment failure maintenance data;

[0038] According to the historical equipment failure maintenance data and the enhanced prompt template, extracting second entity data and second relationship data;

[0039] According to the second entity data and the second relationship data, generating an equipment failure knowledge graph.

[0040] In a third aspect, an embodiment of the present application provides a computing device, including a memory, a processor, and computer instructions stored on the memory and executable on the processor, wherein the processor executes the computer instructions to implement the steps of the steel production equipment failure detection method in any one of the first aspect.

[0041] In a fourth aspect, an embodiment of the present application provides a storage medium for storing computer executable instructions, wherein the computer executable instructions are executed to implement the steps of the steel production equipment failure detection method in any one of the first aspect.

[0042] Compared with the prior art, the present application can at least achieve the following technical effects:

[0043] The application can find the most representative enhanced prompt template from a large amount of historical equipment maintenance data by using the optimized SBert model, significantly reducing the difficulty and workload of manual design prompts, and improving the efficiency and quality of prompt generation; at the same time, the use of enhanced prompt templates can enable large language models to automatically generate more accurate device fault knowledge graphs, significantly reducing the cost and cycle of manual construction and maintenance of knowledge graphs, and efficiently processing massive historical equipment fault maintenance data, realizing the large-scale accumulation and utilization of equipment faults. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the one or more embodiments of the present specification or the prior art, the drawings needed to be used in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments described in the present specification, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0045] Figure 1 A steel production equipment fault detection method flow chart is provided for one or more embodiments of the present specification.

[0046] Figure 2 A steel production equipment fault detection device structure schematic diagram is provided for one or more embodiments of the present specification. DETAILED DESCRIPTION

[0047] In order to enable those skilled in the art to better understand the technical solutions in the one or more embodiments of the present specification, the technical solutions in the one or more embodiments of the present specification will be described clearly and completely below in conjunction with the drawings in the one or more embodiments of the present specification. Obviously, the described embodiments are only some embodiments of the present specification, not all embodiments. Based on the one or more embodiments of the present specification, all other embodiments obtained by those skilled in the art without creative labor should belong to the protection scope of the present document.

[0048] Steel manufacturing is a core technology of heavy industry, involving large equipment such as rolling mills, continuous casting machines, heating furnaces, etc. These devices run in harsh environments of high temperature, high pressure and high load, and faults occur frequently (such as bearing overheating, gear wear, etc.). Accurate fault detection can ensure safe operation of equipment and guarantee production efficiency.

[0049] However, when introducing artificial intelligence technology (such as machine learning) for detection, existing steel production equipment is mostly based on training a machine learning model for a single fault type. When multiple faults occur concurrently or faults are intertwined (for example, a bearing simultaneously exhibits temperature abnormalities, increased vibration, and abnormal noise, or multiple sensor signal abnormalities indicated by "sensor signal abnormality superposition"), it is difficult to effectively distinguish whether the multiple representations are caused by a single root fault or the result of multiple independent or mutually coupled faults acting together, and it is difficult to accurately trace the root cause. When the actual fault symptom combination or the fault type learned by the model does not completely match, it is easy to cause misjudgment, missed judgment, or ineffective classification. These systems generally lack continuous learning and adaptive updating mechanisms, that is, they cannot automatically optimize or update their internal diagnostic logic or model parameters based on newly generated fault data, maintenance experience, or operation feedback, lack the ability to dynamically discover new knowledge and associations, that is, it is difficult to actively identify emerging feature associations or causal relationships that may indicate new or evolving fault modes from multiple source data, lack the ability to autonomously generate new diagnostic rules or patterns, that is, cannot form new diagnostic rules or fault patterns based on newly acquired information to cope with changing equipment states and production environments, thereby affecting the accuracy and timeliness of fault diagnosis.

[0050] During the operation and maintenance of steel equipment, maintenance personnel will generate operation and maintenance logs, operator records, maintenance reports, and other text materials during maintenance, which contain a large amount of detailed and critical fault clues (for example, "operator record 'after replacing the lubricating oil recently, only under high load working conditions, grinding sound appears'" or "technician observes that the surface of component X has a slight discoloration, and there is a slight burnt smell near Y motor"). However, existing technologies are difficult to directly process and utilize unstructured text. Expert systems based on rules directly rely on a set of "IF-THEN" rules defined by humans in advance. These rules are usually derived from limited expert experience and historical fault cases, and the knowledge exists in a static form. The system is difficult to identify new faults or unencountered fault modes outside the rule base, and knowledge updating and expansion need to rely on human intervention. It is unable to automatically extract subtle clues, event timing relationships, context backgrounds, and other deep semantic information from text that can be discerned by human experts, resulting in high costs for updating fault rules and lagging response to fault updates when new faults occur, limiting the comprehensiveness and depth of fault detection.

[0051] To solve the above technical problems, the present application provides a steel production equipment fault detection method, as shown in Figure 1 The specific steps are as follows:

[0052] Step S1, obtaining historical equipment fault maintenance data, real-time equipment data, and user questions.

[0053] In the embodiments of the present application, diversified historical equipment failure maintenance data related to steel industry equipment failures are collected and organized, such as equipment maintenance reports, troubleshooting guidelines, operator logs, fault phenomenon descriptions, and fragments about fault diagnosis in equipment manuals. Then, real-time equipment data are collected through a sensor network, including real-time sensor data (vibration, temperature, pressure, etc.) of the equipment and real-time data such as equipment operation logs, which are obtained from an equipment management system (MES) or manually input. The real-time equipment data are then preprocessed, and the data are aligned according to timestamps. For example, the real-time equipment data include a. sensor monitoring data:

[0054] temperature data: "Equipment ID: RM001, collection time: 2023-11-15 09:30:05, monitoring point: main motor bearing seat, temperature value: 85°C";

[0055] vibration data: "Equipment ID: RM001, collection time: 2023-11-15 09:30:10, monitoring point: gear box output shaft, vibration amplitude_X direction: 1.2g, main peak vibration frequency: 50Hz";

[0056] pressure data: "Equipment ID: CS002, collection time: 2023-11-15 09:31:00, monitoring point: main oil way of hydraulic system, pressure value: 5.5MPa";

[0057] current / voltage data: "Equipment ID: HF003, collection time: 2023-11-15 09:32:15, monitoring point: A-phase current of heating unit, current value: 150A".

[0058] b. equipment status and event logs:

[0059] running status: "Equipment ID: RM001, record time: 2023-11-15 09:29:50, current state: normal operation, load rate: 80%";

[0060] key event: "Equipment ID: CS002, record time: 2023-11-15 09:30:55, event description: trigger emergency stop signal";

[0061] control system alarm information: "Equipment ID: HF003, record time: 2023-11-15 09:32:20, alarm code: PLC_ALM_E404, alarm content: motor overload protection".

[0062] c. Production and process parameters: "Equipment ID: RM001, Record Time: 2023-11-15 09:30:00, Current Production Steel Grade: HRB400, Rolling Procedure Number: RP005, Rolling Speed Setting: 5 m / s".

[0063] Obtain the maintenance question that the user wants to know and the current operation status and key parameters of the equipment (such as timestamp, equipment identification, etc.). For example, a. Direct fault inquiry: "Please analyze whether the temperature of the main motor bearing seat of rolling mill RM001 is abnormal at 85°C, and determine the possible cause." b. Analysis request based on text description (may imply diagnostic requirements): User submits operator record text: "Operator log: 14:20, periodic abnormal noise is emitted from the transmission side of rolling mill RM001, the sound is dull and accompanied by slight vibration. 14:25, it is found that the oil temperature near the reduction gearbox has slightly increased." (The system needs to understand the implied diagnostic request). c. Fault prediction or hypothetical scenario analysis request: "What equipment components may fail if the cooling water system flow rate of rolling mill RM001 decreases by 20%?" and other questions.

[0064] Step S2, based on the SBert model, generating a semantic embedding vector according to the soft prompt optimization technology and the historical equipment fault maintenance data.

[0065] In the embodiments of the present application, the first entity data and the first relationship data are extracted according to the historical equipment fault maintenance data; based on the soft prompt optimization technology, the first entity data and the first relationship data are used to construct a triple data; based on the loss function, the SBert model is trained according to the triple data to obtain an optimized SBert model; and based on the optimized SBert model, the first entity data and the first relationship data are used to generate a semantic embedding vector.

[0066] Specifically, the historical equipment failure data is divided into four parts, including a training set, a test set, a validation set, and a failure data set; the training set, the test set, and the validation set are labeled to determine entity data and relationship data; and the SBert (Sentence-BERT, sentence embedding) model is pre-trained using the labeled training set. For example, core entities and relationships related to equipment failure are automatically extracted from historical equipment failure data. Entities: equipment component units: "rolling mill bearing", "main transmission system", "hydraulic pump", "sensor", and the like. Performance characterization / failure symptoms: "abnormal temperature rise", "vibration frequency out of limit", "pressure fluctuation", "abnormal noise", "sudden current surge", and the like. Fault state / fault mode: "bearing overheating", "gear wear", "lubrication failure", "electrical short circuit", "control unit malfunction", and the like. Detection tool / diagnostic method: "infrared thermal imager", "vibration analyzer", "oil detection", "logic diagnostic process", and the like. The above entities are recognized and defined to exist in various semantic relationships to reveal the mechanism, propagation path, and influencing factors of failure occurrence. Typical relationship types: causal relationship: ("insufficient lubrication", "leads to", "bearing overheating"), ("component aging", "triggers", "performance degradation"). Performance relationship / symptom association: ("bearing wear", "manifests as", "abnormal vibration"), ("abnormal temperature rise", "associated symptoms", "oil degradation"). Composition relationship / hierarchical relationship: ("rolling mill bearing", "is a component of", "rolling mill main transmission system"), ("sensor A", "belongs to", "monitoring subsystem B"). Fault diagnosis association: the phenomenon "vibration exceeds threshold X and temperature exceeds threshold Y" may indicate "bearing failure". These relationships are formally modeled to ensure accurate expression in the knowledge graph. The extracted entities are used as nodes (Node) and the relationships between entities are used as edges (Edge) to organize a network-like graph structure (Graph Structure). This graph-based representation changes the field knowledge from implicit and scattered to explicit and interconnected, facilitating efficient knowledge storage, retrieval, reasoning, and updating by computers.

[0067] Based on the soft prompt optimization technology, a trainable prompt embedding [PROMPT] Token is introduced, and anchor, positive and negative triplets are constructed using the training set. The SBert model is fine-tuned using a contrastive learning loss function to generate an optimized SBert model. The soft prompt optimization technology for SBert model tuning process is divided into three stages, including preparation stage, configuration stage and training stage. First, in the preparation stage: define and initialize the learnable prompt token; 1, add special token ([PROMPT]): add a new and special token to the tokenizer of the pre-trained SBert model, and this example uses [PROMPT]. If the token does not exist, add it and adjust the size of the word embedding layer of the model accordingly. 2, initialize the embedding vector: initialize an embedding vector for the new [PROMPT] token. This is done by taking the mean of existing word embeddings and adding a small amount of random noise to ensure that it is unique in the embedding space and coordinated with other tokens. Second, in the configuration stage: learn by isolating the prompt token; 1, freeze model parameters: freeze all parameters in the SBert model except the embedding vector corresponding to the [PROMPT] token (i.e. set requires_grad = False). 2, unfreeze prompt token embedding: ensure that the embedding vector of the [PROMPT] token is trainable (i.e. its requires_grad = True). 3, gradient control: register a gradient hook to the weights of the word embedding layer. This hook function ensures that only the gradient of the [PROMPT] token corresponding position will be retained and used for updating during backpropagation, and the gradient of all other word vectors is set to zero. In this way, the focus of training is completely focused on optimizing this soft prompt token. Then, in the training stage: adapt to specific tasks by optimizing the prompt token; 1, build training data: combine the [PROMPT] token with a fixed task instruction (in this example, "Extract entities and relations from the text:") to form a prefix, such as "[PROMPT]Extract entities and relations from the text:". Combine this soft prompt containing prefix with the actual training text (positive sample) and a negative sample into a training instance (InputExample) for contrastive learning. 2, define the loss function: use a loss function suitable for the task.In this task, MultiTaskLoss is used, which combines CosineSimilarityLoss and TripletLoss, aiming to let the model learn to distinguish similar and dissimilar samples (under the guidance of soft prompts).3、Model training: using the prepared training data and loss function to fine-tune the SBert model. During the training process, due to the effect of parameter freezing and gradient hook, only the embedding vector of the [PROMPT] mark will be updated. The model learns to adjust this vector so that it can guide the SBert model to generate better sentence embeddings for a specific task (here, the similarity judgment related to entity and relation extraction). Finally, after the above optimization process, the embedding vector of the [PROMPT] mark becomes a learned, task-specific "soft prompt", and after testing the set, the validation set is verified to produce an "optimized SBert model".

[0068] Finally, the fault data set is input to the optimized SBert model piece by piece (for example, a text paragraph or sentence describing each fault); the optimized SBert model processes each piece of input text to generate a corresponding low-dimensional dense semantic embedding vector.

[0069] The SBert model fine-tuned in this application (especially its learned [PROMPT] embedding) is then used to: generate semantic embedding vectors for new input text (also preceded by "[PROMPT] Extract entities and relations from the text:"). The vector can capture the core semantic information of the text, so that semantically similar texts are close in vector space. These specially optimized embeddings are used to retrieve the most similar training samples and identify relevant domain patterns (clustering analysis). Finally, this information is used to construct more rich and instructive "enhanced prompts" for downstream large language models (qwen2-72b or deepseek) to improve entity and relation extraction performance. In short, the core of soft prompt optimization is to learn one or more continuous vectors (i.e. [PROMPT] embedding) that can guide the pre-trained model to perform better on a specific task as part of the model input, without the need for large-scale parameter updates to the model body.

[0070] Step S3, based on an unsupervised clustering algorithm, clustering each semantic embedding vector according to a preset cluster number and a preset enhanced prompt template to generate a new enhanced prompt template.

[0071] Specifically, a set of semantic vectors is generated using each semantic embedding vector; the set of semantic vectors is input into an unsupervised clustering algorithm to cluster the semantic embedding vectors; and by setting a suitable number of clusters K (the value of K can be determined according to task requirements or through clustering evaluation indicators), semantically similar fault description texts are grouped into the same cluster. Each cluster represents a collection of texts with similar semantic characteristics or fault modes (for example, one cluster may mainly contain descriptions about "bearing lubrication failure causing overheating", and another cluster may be about "communication abnormality of the electrical control system"). Then, one or more texts that best express the core semantics of each text cluster are selected as "representative examples", such as calculating the vector of all texts in the cluster, and then finding the text with the smallest average distance to all other text vectors in the cluster as the representative example. Then, using the pre-set enhanced prompt template and the representative example, structured natural language prompt content is constructed, including: task instructions (Task Description) and context examples (Few-Shot Examples).

[0072] The task instructions (Task Description) are used to clearly inform the subsequent large language model of the specific task it needs to perform, such as "Please extract the defined device entity, fault state, and their relationship from the following text and output it in JSON format" or "Please diagnose the most likely fault cause and provide suggestions based on the following fault phenomenon description and related data".

[0073] The context examples (Few-Shot Examples) are used to select "representative examples" as context learning examples. Each example should include an input example: the representative example text itself, which shows the typical style of the text to be processed. The output example: the standard output format and content expected to be generated by the large language model corresponding to the input example (for example, for entity relationship extraction tasks, provide corresponding JSON structured data; for diagnosis tasks, provide standard diagnosis conclusions and suggestions).

[0074] Finally, the prompt content containing task instructions and high-quality context examples is integrated into one or more "enhanced prompt templates". These templates can contain placeholders for filling in new texts to be processed. The generated enhanced prompt templates are stored for the knowledge graph construction module or the large language model diagnosis module to call and instantiate when processing new fault data or user requests.

[0075] The enhanced prompt template generation method proposed in the application can automatically discover and induce typical fault description patterns and semantic categories from a large number of field texts by combining the powerful semantic representation capability of the optimized SBert model and the unsupervised learning characteristics of K-Means clustering. Further, by selecting representative examples to construct prompt templates containing explicit task instructions and high-quality context examples, the guiding effect of subsequent large language models (such as Deepseek) is significantly improved. This not only reduces the complexity and workload of manually designing and debugging prompts, achieving partial automation of prompt engineering, but more importantly, provides large language models with highly domain-related, structured and instructive inputs, thereby effectively improving their accuracy, consistency and overall performance in tasks such as steel equipment fault knowledge extraction and fault diagnosis.

[0076] Step S4, based on the large language model, generating a device fault knowledge graph according to the new enhanced prompt template and the historical device fault maintenance data.

[0077] In the embodiments of the application, based on the large language model, an enhanced prompt template is selected according to the historical device fault maintenance data; second entity data and second relationship data are extracted according to the historical device fault maintenance data and the enhanced prompt template; and a device fault knowledge graph is generated according to the second entity data and the second relationship data.

[0078] Specifically, first, a segment of current text data to be processed is selected from the fault data set in the historical device fault maintenance data as the input for this extraction. Second, an enhanced prompt template is selected according to the text data to be processed. Third, the current text to be processed and the enhanced prompt template are combined to form a complete enhanced prompt instance (including task description, few-shot example and current text to be processed); then the complete enhanced prompt instance is input into the pre-trained large language model, and the large language model performs semantic understanding and information extraction on the current text to be processed according to the guidance of the received task description and few-shot example (especially the mapping relationship between the input text and the output JSON in the example). Then, the structured output is obtained to obtain the entity and the relationship between the entities. Finally, the JSON result output by the large language model is checked and necessary post-processing (such as format conversion, redundancy removal, etc.) is performed; then the extracted structured entity and relationship information are imported into the graph database (such as Neo4j) according to the storage specification of the knowledge graph (such as the form of triplets), realizing the incremental construction or updating of the knowledge graph.

[0079] For example, randomly extract a piece of text data from historical equipment failure maintenance data: during the shutdown maintenance process caused by the overheating of the main drive system bearing of the rolling mill, it was found that the bearing temperature of the rolling mill abnormally rose to 120℃, accompanied by abnormal vibration. After inspection, it was confirmed that the bearing roller had serious wear and the bearing needed to be replaced immediately and the lubrication system pressure needed to be adjusted.

[0080] Select an enhanced prompt template for the text data to be processed. The template is designed specifically for entity and relationship extraction tasks, and its core components include:

[0081] a. Clear task description: Clearly state the task to be performed to the large language model. For example: "You are a specialized steel equipment failure entity and relationship extraction model. Please carefully analyze the following provided text and extract relevant information according to the predefined entity types (such as: component unit, performance characterization, failure state, detection tool) and relationship types (such as: component failure, performance failure, detection tool, composition). Please output the extraction results in a structured JSON format, with the specific format referring to the following example."

[0082] The task description should clearly indicate the definition of entities and relationships and the expected output format.

[0083] b. Few-shot examples: Provide one or more complete "input-output" example pairs to demonstrate the execution of the task and the expected results. Each example contains:

[0084] Input example text: A piece of domain text similar to the "current text to be processed".

[0085] Output example JSON: For this input example text, manually annotated or pre-extracted entity and relationship results in the expected JSON format. For example, the JSON example provided in the patent disclosure contains "text", "entity", and "relation" fields. These examples help the large language model understand the specific manifestations of entities and relationships and their context in the text.

[0086] Instantiated prompt: Fill in the corresponding positions in the enhanced prompt template with the "current text to be processed" to form a complete "enhanced prompt instance" ready for input into the large language model.

[0087] Combine the enhanced prompt template and the text data to be processed and input into the large language model. The expected JSON output for the aforementioned input text example should include the identified "rolling mill bearing", "lubrication system" entities and their "entity failure", "performance failure" relationships.

[0088] The application can enable a large language model to efficiently extract valuable structured entity and relationship information from unstructured fault text. By providing relevant few-shot examples in the prompt, the in-context learning ability of the large language model is activated, allowing the model to quickly adapt to specific terminology, entity types, and relationship patterns in the steel equipment fault field without the need for large-scale parameter fine-tuning, improving the accuracy and domain relevance of the extraction. Clear task descriptions and formatted output examples guide the large language model to produce more standardized and consistent extraction results, facilitating subsequent knowledge fusion and application. Ultimately, the automatic construction of a knowledge graph significantly reduces the cost and cycle of manual construction and maintenance of a knowledge graph, enabling efficient processing of massive historical equipment fault maintenance data and realizing the large-scale accumulation and utilization of equipment faults.

[0089] Step S5, generating a fault detection result according to the real-time equipment data, the equipment fault knowledge graph, and the user question.

[0090] Specifically, the user question is input into the large language model for fault diagnosis. During the diagnosis process, real-time sensor equipment data, user input text descriptions, and relevant historical cases, fault mechanisms, component relationships retrieved from the knowledge graph, and other multi-source heterogeneous information are dynamically integrated to provide a more comprehensive diagnostic context for the large language model, enabling more comprehensive and accurate fault judgment, and ultimately generating a more accurate fault detection result.

[0091] The application combines the powerful generalization reasoning ability of the large language model and the comprehensive domain background knowledge obtained from the knowledge graph, enabling a deeper analysis and judgment of complex fault phenomena. Even if the fault mode has not been explicitly learned, reasonable inferences can be made based on existing knowledge and logic, improving the diagnostic ability for complex and unknown faults.

[0092] Further, a user maintenance correction suggestion is obtained; and the SBert model is updated according to the user maintenance correction suggestion.

[0093] Specifically, the fault cause, evidence chain (such as sensor data, knowledge graph relationship), and input box for collecting correction opinions are displayed through a graphical interface, and the correction opinions are fed back to the SBert model to update the knowledge graph and optimize the prompt.

[0094] The application embodiment provides a steel equipment fault detection device, as shown in Figure 2 The device comprises,

[0095] A data collection module is configured to obtain historical equipment fault maintenance data, real-time equipment data, and user questions.

[0096] an optimization module configured to generate semantic embedding vectors based on an SBert model, according to a soft prompt optimization technique and the historical equipment failure maintenance data;

[0097] an enhanced prompt template generation module configured to generate new enhanced prompt templates by clustering each of the semantic embedding vectors based on an unsupervised clustering algorithm, according to a preset cluster number and a preset enhanced prompt template;

[0098] a knowledge graph construction module configured to generate an equipment failure knowledge graph based on a large language model, according to the new enhanced prompt templates and the historical equipment failure maintenance data;

[0099] a diagnosis module configured to generate a failure detection result based on the real-time equipment data, the equipment failure knowledge graph, and the user question.

[0100] Further, the system further comprises a human-computer interaction module,

[0101] obtain a user maintenance correction suggestion;

[0102] update the SBert model according to the user maintenance correction suggestion.

[0103] Further, the optimization module is configured to:

[0104] extract first entity data and first relationship data from the historical equipment failure maintenance data;

[0105] construct triple data based on a soft prompt optimization technique, according to the first entity data and the first relationship data;

[0106] train the SBert model based on a loss function, according to the triple data, to obtain an optimized SBert model;

[0107] generate semantic embedding vectors based on the optimized SBert model, according to the first entity data and the first relationship data.

[0108] Further, the knowledge graph construction module is configured to:

[0109] select enhanced prompt templates based on a large language model, according to the historical equipment failure maintenance data;

[0110] extract second entity data and second relationship data from the historical equipment failure maintenance data and the enhanced prompt templates;

[0111] generate an equipment failure knowledge graph based on the second entity data and the second relationship data.

[0112] The embodiment of the present application provides a computing device, comprising a memory, a processor and computer instructions stored in the memory and executable on the processor, and the processor executes the computer instructions to implement the steps of the steel production equipment fault detection method in any of the above embodiments.

[0113] The present application provides a storage medium for storing computer executable instructions, which implement the steps of the steel production equipment fault detection method in any of the above embodiments when executed.

[0114] It should be noted that the embodiments of the storage medium in the present specification and the embodiments of the steel equipment fault detection method in the present specification are based on the same inventive concept, and therefore the specific implementation of the embodiments can be referred to the foregoing implementation of the corresponding steel equipment fault detection method, and the repeated parts will not be described herein.

[0115] The above describes specific embodiments of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the acts or steps recited in the claims can be performed in an order other than the order in which they are recited and still achieve desirable results. In addition, the processes depicted in the figures do not necessarily require the particular order shown, or sequential order, to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous or possible.

[0116] In the 1930s, it was clear to distinguish whether an improvement in a technology was in hardware (e.g., improvement in circuit structure of diodes, transistors, switches, etc.) or in software (e.g., improvement in method flow). However, as technology has evolved, many improvements in method flow today can be considered as direct improvements in hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented by a hardware entity module. For example, a programmable logic device (PLD) (e.g., a field programmable gate array (FPGA)) is an integrated circuit whose logic function is determined by user programming of the device. A digital system is "integrated" on a PLD by the designer programming it, rather than by asking a chip manufacturer to design and fabricate a custom integrated circuit chip. Moreover, instead of manually fabricating an integrated circuit chip, this programming is now mostly implemented by "logic compiler" software, which is similar to software compilers used in program development, and the original code to be compiled is written in a specific programming language, called a hardware description language (HDL), of which there are many, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc., the most commonly used being VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. It should be clear to those skilled in the art that, by simply logically programming a method flow in one of the above hardware description languages and programming it into an integrated circuit, a hardware circuit implementing the logical method flow can be easily obtained.

[0117] The controller can be implemented in any suitable way, for example, the controller can take the form of a microprocessor or processor and a computer readable medium storing computer readable program code, e.g. software or firmware, executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller and an embedded microcontroller, examples of which include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20 and Silicone Labs C8051F320, the memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that, in addition to implementing the controller purely in terms of computer readable program code, it is possible to implement the controller to perform the same functions using logic gates, switches, application specific integrated circuits, programmable logic controllers and embedded microcontrollers, by logically programming the method steps. The controller can thus be considered a hardware component, and the means for performing the various functions comprised therein can be considered structures within the hardware component. Alternatively, or even additionally, the means for performing the various functions can be considered both software modules which implement the method and structures within the hardware component.

[0118] The systems, apparatuses, modules or units illustrated by the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0119] For the sake of description, the above apparatuses are described in various units by functions respectively. Of course, the functions of each unit can be implemented in the same or multiple software and / or hardware in the implementation of the embodiments of the present specification.

[0120] Those skilled in the art will appreciate that one or more embodiments of the specification can be provided as a method, system or computer program product. Therefore, one or more embodiments of the specification can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the specification can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROMs, optical storage devices, etc.) containing computer usable program code.

[0121] The specification is presented with reference to flow diagrams and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the specification. It will be understood that each block of the flow diagrams and / or block diagrams, and combinations of blocks in the flow diagrams and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing element or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 one or more flow or multiple flows and / or blocks Figure 1 one or more flow or multiple flows and / or blocks

[0122] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 one or more flow or multiple flows and / or blocks Figure 1 one or more flow or multiple flows and / or blocks

[0123] The computer program instructions can also be loaded into a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 one or more flow or multiple flows and / or blocks Figure 1 one or more flow or multiple flows and / or blocks

[0124] In one typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0125] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) and / or cache memory, non-volatile memory, such as read-only memory (ROM), EPROM, and / or flash memory, etc. The memory is an example of computer readable media.

[0126] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0127] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusion, such that processes, methods, articles or devices that include a series of elements not only include those elements, but also include other elements not explicitly listed, or inherent to such processes, methods, articles or devices. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or device that includes the element.

[0128] One or more embodiments of the specification can be described in the general context of computer-executable instructions being executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of the specification can also be practiced in a distributed computing environment, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0129] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other, and each embodiment focuses on the difference from other embodiments. In particular, for system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiment.

[0130] The above merely provides the example of the present document and is not intended to limit the present document. For those skilled in the art, the present document can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present document shall be included in the scope of claims of the present document.

Claims

1. A method for detecting faults in steel production equipment, characterized in that: include: Obtain historical equipment failure and repair data, real-time equipment data, and user issues; Based on the SBert model, a semantic embedding vector is generated according to the soft prompt optimization technology and the historical equipment failure maintenance data; Based on an unsupervised clustering algorithm, clustering the semantic embedding vectors according to a preset number of clusters and a preset enhancement prompt template to generate a new enhancement prompt template; Based on the large language model, generating an equipment failure knowledge graph according to the new enhanced prompt template and the historical equipment failure maintenance data; A fault detection result is generated based on the real-time device data, the device fault knowledge graph, and the user question.

2. The method according to claim 1, characterized in that The method further comprises, Obtain user maintenance and correction suggestions; The SBert model is updated according to the user maintenance correction suggestion.

3. The method according to claim 1, characterized in that Based on the SBert model, the semantic embedding vector is generated according to the soft prompt optimization technology and the historical equipment failure maintenance data, including: Extracting first entity data and first relationship data based on the historical equipment failure maintenance data; constructing triple data according to the first entity data and the first relationship data based on a soft prompt optimization technology; Based on the loss function, the SBert model is trained according to the triple data to obtain an optimized SBert model; Based on the optimized SBert model, a semantic embedding vector is generated according to the first entity data and the first relationship data.

4. The method according to claim 1, wherein Based on the large language model, according to the new enhanced prompt template and the historical equipment failure maintenance data, the equipment failure knowledge graph is generated, including: Based on the large language model, an enhanced prompt template is selected according to the historical equipment failure maintenance data; extracting second entity data and second relationship data according to the historical equipment failure maintenance data and the enhanced prompt template; Generate a device failure knowledge graph based on the second entity data and the second relationship data.

5. A steel equipment fault detection device, characterized in that: include, Data collection module, used to obtain historical equipment failure and repair data, real-time equipment data and user problems; An optimization module, configured to generate a semantic embedding vector based on the SBert model, the soft prompt optimization technique, and the historical equipment failure maintenance data; An enhanced prompt template generation module is used to cluster the semantic embedding vectors based on an unsupervised clustering algorithm according to a preset number of clusters and a preset enhanced prompt template to generate a new enhanced prompt template; A knowledge graph construction module, configured to generate an equipment failure knowledge graph based on a large language model, the new enhanced prompt template, and the historical equipment failure maintenance data; The diagnosis module is used to generate a fault detection result based on the real-time device data, the device fault knowledge graph and the user question.

6. The device according to claim 5, characterized in that The device also includes a human-computer interaction module, Obtain user maintenance and correction suggestions; The SBert model is updated according to the user maintenance correction suggestion.

7. The device according to claim 5, characterized in that The optimization module is configured to: Extracting first entity data and first relationship data based on the historical equipment failure maintenance data; constructing triple data according to the first entity data and the first relationship data based on a soft prompt optimization technology; Based on the loss function, the SBert model is trained according to the triple data to obtain an optimized SBert model; Based on the optimized SBert model, a semantic embedding vector is generated according to the first entity data and the first relationship data.

8. The device according to claim 5, characterized in that The knowledge graph construction module is configured as follows: Based on the large language model, an enhanced prompt template is selected according to the historical equipment failure maintenance data; extracting second entity data and second relationship data according to the historical equipment failure maintenance data and the enhanced prompt template; Generate a device failure knowledge graph based on the second entity data and the second relationship data.

9. A computing device comprising a memory, a processor, and computer instructions stored in the memory and executable on the processor, wherein: When the processor executes the computer instructions, the steps of the steel production equipment fault detection method according to any one of claims 1 to 4 are implemented.

10. A storage medium for storing computer-executable instructions, characterized in that: When the computer executable instructions are executed, the steps of the steel production equipment fault detection method according to any one of claims 1 to 4 are implemented.

Images