Dry bulk cargo large-scale equipment fault intelligent question and answer method and system
By building an intelligent question-and-answer system for large dry bulk equipment faults and utilizing natural language processing and knowledge graph technology, we can quickly identify faults and generate diagnostic reports, solving maintenance problems caused by equipment complexity and improving maintenance efficiency and port operation stability.
Patent Information
- Application Number
- CN202510762991.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-09
- Publication Date
- 2025-10-10
AI Technical Summary
Fault diagnosis and maintenance of large equipment in dry bulk ports rely on experienced workers. Inexperienced workers find it difficult to complete troubleshooting and repairs independently. The complexity of the equipment also makes maintenance difficult, and the traditional manual mode cannot meet the needs of modern port operations.
Natural language processing technology and knowledge graph are used to build an intelligent question-answering system for large dry bulk equipment faults. The fault entity recognition model and question intent recognition model are used to parse user questions, generate Cypher query statements to retrieve fault knowledge in the knowledge graph, and generate readable fault diagnosis reports.
It enables inexperienced maintenance personnel to quickly diagnose and repair large equipment failures in dry bulk ports, reducing equipment downtime and improving the continuity and operational stability of port loading and unloading operations.
Smart Images
Figure CN120763282A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of intelligent question and answer systems, and in particular to an intelligent question and answer method and system for dry bulk cargo large equipment faults. BACKGROUND
[0002] Currently, the fault diagnosis and maintenance of dry bulk cargo port large equipment face many challenges, and the current maintenance work highly depends on experienced workers for fault diagnosis. Due to the complex structure of dry bulk cargo port large equipment, only maintenance workers who have accumulated experience for many years can efficiently and accurately determine the fault point and cause of the equipment, and take appropriate maintenance measures, while workers with insufficient experience often have difficulty in independently completing fault troubleshooting and maintenance work when facing complex equipment failures. At the same time, the complexity of large equipment further exacerbates the difficulty of maintenance, and these equipment are composed of numerous precision components, and the fault point and cause of each component can be diverse, involving mechanical, electronic, hydraulic and other fields. In the face of such diversity and complexity, maintenance personnel must have extensive knowledge and rich practical experience to be able to handle it easily. In addition, it takes a long period of time and high cost to train a qualified maintenance worker, including professional technical training, practical experience accumulation, etc. Under the background of continuous growth of global trade volume, the shutdown of dry bulk cargo port large equipment will cause the interruption of dry bulk cargo port loading and unloading operations, and in turn affect the ship's in and out port activities, so the dry bulk cargo port large equipment needs more efficient and reliable equipment maintenance means to ensure the stability and continuity of the entire port operation, and the traditional manual dependence mode has been difficult to fully meet the operation needs of modern ports. SUMMARY
[0003] According to the above technical problems, an intelligent question and answer method and system for dry bulk cargo large equipment faults are provided. The present application mainly uses natural language processing technology and knowledge graph to establish an intelligent question and answer system for typical faults of dry bulk cargo port large equipment, to realize the intelligent management and application of typical fault knowledge of dry bulk cargo port large equipment.
[0004] The technical means adopted by the present application are as follows:
[0005] An intelligent question and answer method for dry bulk cargo large equipment faults, comprising:
[0006] Using a fault entity recognition model and a question intent recognition model to analyze the user input question, extract fault entity information, and identify the corresponding question intent;
[0007] According to the fault entity information and the question intent, matching the corresponding Cypher statement template, and generating a Cypher query statement;
[0008] Retrieving related fault knowledge in the knowledge graph according to the generated Cypher query statement;
[0009] Use a large language model to grammatically restructure the retrieved fault knowledge and generate a readable fault diagnosis report;
[0010] The generated fault diagnosis report is sent to the user interface, providing relevant maintenance suggestions and operation guidelines.
[0011] Furthermore, the fault entity information includes: equipment location, fault point and fault characterization; the BIO annotation method is used to manually annotate the sample data to obtain training data, and the training data is used to train the fault entity recognition model; the manual annotation method is used to perform intent annotation on the question data to obtain training data, and the training data is used to train the question intent recognition model;
[0012] The fault entity recognition model decomposes and encodes the training data and inputs the decomposition results into a linear classification layer. The linear classification layer identifies entities and obtains the labeling results of each character. Characters of the same category are combined into words, which are used as the category classification results corresponding to the entity to obtain different entities. Through cross-entropy loss function optimization, the context information captured by self-attention is used to evaluate the classification results of entity types, thereby realizing the training of the fault entity recognition model.
[0013] The question intent recognition model uses the Tongyi Qianwen model to generate multiple different types of sentences, including inquiries about basic equipment information, inquiries about fault representations, and inquiries about fault repair processes; the pre-trained model BERT is combined with the deep learning classifier LSTM to perform text classification tasks, and the training data is used to adjust the parameters of the question intent recognition model to realize the training of the question intent recognition model.
[0014] Furthermore, generating a Cypher query statement specifically includes:
[0015] According to the fault entity information and the question intent, a corresponding preset Cypher statement template is matched. The Cypher statement template reserves a fault entity information slot for filling in the fault entity information extracted from the user question to form a corresponding Cypher statement; the identified entity information is filled into the entity information slot reserved in the Cypher statement query template to obtain the corresponding Cypher query statement.
[0016] Furthermore, the knowledge graph is constructed based on the Neo4j graph database and is used to store equipment components, fault locations, fault representations and maintenance processes; the fault knowledge is output in the form of a dictionary.
[0017] Furthermore, the fault diagnosis report is sent to the user interaction interface using JSON format.
[0018] The application also comprises a dry bulk cargo large equipment fault intelligent question and answer system based on a dry bulk cargo large equipment fault intelligent question and answer method, comprising a semantic recognition module, a query statement matching module, a knowledge graph query module, an answer generation and optimization module and a system interaction module, wherein:
[0019] The semantic recognition module parses the user input question by using a fault entity recognition model and a question intention recognition model, extracts fault entity information, and identifies the corresponding question intention.
[0020] The query statement matching module matches the corresponding Cypher statement template according to the fault entity information and the question intention, and generates a Cypher query statement.
[0021] The knowledge graph query module retrieves related fault knowledge in the knowledge graph according to the generated Cypher query statement.
[0022] The answer generation and optimization module uses a large language model to reorganize the retrieved fault knowledge in syntax and generates a readable fault diagnosis report.
[0023] The system interaction module sends the generated fault diagnosis report to the user interaction interface and provides related maintenance suggestions and operation guidelines.
[0024] Compared with the prior art, the application has the following advantages:
[0025] The dry bulk cargo large equipment fault intelligent question and answer method and system provided by the application use natural language processing technology to quickly identify the intention of maintenance personnel and the fault representation described by the maintenance personnel, use the efficient organization and fast retrieval of the knowledge graph to return related fault entity information in time, and can enable inexperienced maintenance personnel to quickly perform fault diagnosis and maintenance when facing different dry bulk cargo port large equipment fault representations, thereby reducing the interruption time of dry bulk cargo port maintenance work caused by large equipment downtime.
[0026] The application uses a natural language processing model combined with a knowledge graph and a large language model to decompose the question into fault point positions and fault representations and the like entity information, automatically matches a Cypher query statement in combination with intention recognition, realizes fast retrieval of fault entity information and reply, integrates the basic information and typical fault knowledge of multiple sources of dry bulk cargo port large equipment, and enables efficient organization and fast dissemination and application of diversified fault diagnosis knowledge accumulated by previous maintenance personnel.
[0027] Based on the above reasons, the application can be widely promoted in the field of intelligent question and answer systems. BRIEF DESCRIPTION OF DRAWINGS
[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings described below are only some embodiments of the present application, and the ordinary skilled in the art can obtain other drawings according to these drawings without any creative effort.
[0029] Figure 1 The flow chart of the intelligent question and answer method for large equipment failure of dry bulk cargo in the present application.
[0030] Figure 2 The flow chart of the large equipment failure entity recognition model training of dry bulk cargo port in the embodiments of the present application.
[0031] Figure 3 The function module composition diagram of the intelligent question and answer system for large equipment failure of dry bulk cargo port in the embodiments of the present application.
[0032] Figure 4 The reasoning process diagram of the intelligent question and answer method for large equipment of dry bulk cargo port in the embodiments of the present application. DETAILED DESCRIPTION
[0033] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the drawings and in combination with the embodiments.
[0034] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings of the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. The description of the at least one exemplary embodiment is actually only illustrative, but not as any limitation on the present application and its application or use. Based on the embodiments in the present application, all other embodiments obtained by the ordinary skilled in the art without any creative effort are within the scope of protection of the present application.
[0035] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit the exemplary embodiments according to the present application. As used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise, and it should also be understood that when the terms "comprise" and / or "include" are used in the specification, there is a feature, step, operation, device, component and / or combination thereof.
[0036] Unless otherwise specified, the relative arrangement of the parts and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present invention. At the same time, it should be clear that, for ease of description, the sizes of the various parts shown in the accompanying drawings are not drawn according to actual proportional relationships. The technology, methods and equipment known to those of ordinary skill in the relevant art may not be discussed in detail, but in appropriate cases, the technology, methods and equipment should be considered as a part of the specification. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a limitation. Therefore, other examples of the exemplary embodiments can have different values. It should be noted that similar numbers and letters represent similar items in the following drawings, and therefore, once an item is defined in an accompanying drawing, it does not need to be further discussed in subsequent drawings.
[0037] like Figure 1 As shown, the present invention provides an intelligent question-answering method for large-scale dry bulk equipment faults, comprising:
[0038] Utilize the fault entity recognition model and question intent recognition model to parse user input questions, extract fault entity information, and identify the corresponding question intent;
[0039] In specific implementation, as a preferred embodiment of the present invention, the fault entity information includes: equipment location, fault point and fault characterization; the BIO labeling method is used to manually label the example data to obtain training data, and the training data is used to train the fault entity recognition model; the manual labeling method is used to perform intent labeling on the question data to obtain training data, and the training data is used to train the question intent recognition model; in implementation, a supervised training method is used to train the model.
[0040] The fault entity recognition model decomposes and encodes the training data and inputs the decomposition results into the linear classification layer. The linear classification layer identifies the entity and obtains the labeling results of each word. The words of the same category are combined into words as the category classification results corresponding to the entity to obtain different entities. Through the optimization of the cross entropy loss function, the context information captured by self-attention is used to evaluate the classification results of the entity type and realize the training of the fault entity recognition model. The training process is as follows: Figure 2 shown.
[0041] Before data labeling, the table type data needs to be processed. The device part, fault point, and fault characterization attributes are used to splice each row of table data into a complete sentence, such as "the steel structure of the whole machine steel structure will appear structure loss, deformation, crack, and loose problems". After the sentence is spliced, LabelStudio is used for fault entity text data labeling, including: using conda to create a new virtual environment, installing Django dependency library for Label Studio data labeling. Use the command pip install label-studio to install LabelStudio, and use the command label-studio start to start the labeling tool and select text labeling.
[0042] In implementation, the entity recognition task uses the BIO mode for labeling, and the labeled data is exported in json format. In the construction phase of the fault entity recognition model, the Chinese pre-trained BERT model (chinese-macbert-base) is used to process the training sentences. The BERT model decomposes the sentence and passes the decomposition result to the linear classification layer, which classifies each word into ["O", "B-mac_name", "I-mac_name", "B-error_loc", "I-error_loc", "B-des", "I-des"]. The model is optimized by the cross-entropy loss function, which uses self-attention to capture the context information and determine the type of fault entity at each position. F1Score is used as the evaluation indicator in the evaluation stage.
[0043] The question intent recognition model uses the general-purpose Q&A model to generate multiple types of sentences, including asking for basic information about the device, asking for fault characterization, and asking for fault repair process. The pre-trained BERT model is combined with the deep learning classifier LSTM to perform text classification tasks. The model parameters are adjusted using training data to train the question intent recognition model.
[0044] According to the fault entity information and the question intent, the corresponding Cypher sentence template is matched, and the Cypher query statement is generated. The user input question is sent to the server-side monitored Flask interface in json format. The model first decomposes the natural language text and encodes it into a form that can be received by the deep learning model, and uses the deep learning model as a classifier to identify key information such as entities and semantics in the sentence. If the model cannot correctly identify the corresponding fault entity information or the intent of the question, the user is prompted to re-output without continuing the following steps.
[0045] In specific implementation, as a preferred embodiment of the present invention, generating a Cypher query statement specifically includes:
[0046] Based on the fault entity information and question intent, the corresponding preset Cypher statement template is matched. The Cypher statement template reserves a fault entity information slot for filling in the fault entity information extracted from the user question to form the corresponding Cypher statement; the identified entity information is filled into the entity information slot reserved in the Cypher statement query template to obtain the corresponding Cypher query statement.
[0047] Retrieve relevant fault knowledge in the knowledge graph based on the generated Cypher query statement;
[0048] In specific implementations, as a preferred embodiment of the present invention, a knowledge graph is constructed based on a Neo4j graph database to store equipment components, fault locations, fault characterizations, and repair procedures. Fault knowledge is output as a dictionary. Constructed Cypher statements are used to query the knowledge graph to obtain the corresponding fault knowledge, which is ultimately returned as a dictionary.
[0049] The typical failure data for large dry bulk port equipment in the knowledge graph includes possible failure points for large dry bulk port equipment and the corresponding failure symptoms at these failure points. In the original data, the failure points and failure symptoms corresponding to equipment parts form a one-to-many relationship. After preprocessing using the Pandas library, a one-to-one correspondence between the three is finally obtained. The typical failure data for large dry bulk port equipment also includes multiple archive tables for large dry bulk port equipment, which contain information on each large dry bulk port equipment and parameter information corresponding to the different components of each equipment. Multiple maintenance process manuals for large dry bulk port equipment contain the maintenance process steps for each failure symptom corresponding to each failure point of each equipment.
[0050] The collected typical fault entity information of large equipment in dry bulk ports (including the archive table, fault entity information and maintenance process of large equipment in dry bulk ports) is sorted out for the construction of knowledge graphs and the generation of model training data.
[0051] When constructing a knowledge graph based on typical failures of large equipment in dry bulk ports, it is necessary to logically organize the existing data and design a knowledge graph structure. Equipment names, fault locations, fault characteristics, and repair processes are considered entities, or points in the knowledge graph. The relationships between different entities are considered edges. Based on this structure, the knowledge graph is constructed.
[0052] Both knowledge graph construction and querying require writing corresponding Cypher statements. First, Cypher statements are written based on the knowledge graph structure designed in the previous step, importing entity names and relationships between entities to build a complete knowledge graph. Based on a large language model, entity and relationship triples of maintenance processes from the maintenance manual for large equipment in dry bulk ports are extracted. Using a variety of methods, a knowledge graph based on typical fault knowledge of large equipment in dry bulk ports is ultimately constructed. Finally, Cypher statement templates are preset for subsequent knowledge graph queries. Different query intents correspond to different Cypher statement templates. The Cypher statement templates reserve slots for fault entity information, which are used to fill in the fault entity information extracted from the user's question to form the corresponding Cypher statement.
[0053] The retrieved fault knowledge is grammatically reorganized using a large language model to generate a readable fault diagnosis report; the returned dictionary is input into the Tongyi Qianwen large model, which performs semantic polishing to reorganize the scattered fault knowledge into natural language with more appropriate semantics.
[0054] In specific implementation, as a preferred embodiment of the present invention, fault diagnosis reports are sent to the user interface in JSON format. This implementation utilizes the Tongyi Qianwen qwen2.5-coder:3b model, downloaded and deployed locally using ollama. Input information is directly accessed through the local interface to call the large model. Before use, enter the command line: ollama run qwen2.5-coder:3b to run the large model.
[0055] The generated fault diagnosis report is sent to the user interface, providing relevant maintenance suggestions and operation guidelines.
[0056] The present invention also includes an intelligent question-answering system for dry bulk cargo large equipment faults based on the intelligent question-answering method for dry bulk cargo large equipment faults, including: a semantic recognition module, a query statement matching module, a knowledge graph query module, an answer generation and optimization module and a system interaction module, such as Figure 3 As shown, where:
[0057] The semantic recognition module uses the fault entity recognition model and the question intent recognition model to parse the user input question, extract the fault entity information, and identify the corresponding question intent;
[0058] The query statement matching module matches the corresponding Cypher statement template based on the fault entity information and question intent, and generates a Cypher query statement;
[0059] The knowledge graph query module retrieves relevant fault knowledge in the knowledge graph based on the generated Cypher query statement; in this module, users can add, modify and delete knowledge graphs, update the data in the knowledge graph according to actual conditions, and query the knowledge graph for fault entity information.
[0060] The answer generation and optimization module uses a large language model to grammatically restructure the retrieved fault knowledge and generate a readable fault diagnosis report;
[0061] The system interaction module sends the generated fault diagnosis report to the user interaction interface and provides relevant maintenance suggestions and operation guidelines.
[0062] Example
[0063] like Figure 4 As shown, this embodiment provides a method for completing the user question reasoning process by using the intelligent question answering method for large dry bulk equipment failure in the present invention.
[0064] Accept user questions. The user interface passes the user's questions to the server, which processes the questions.
[0065] Fine-tune the model's fault entity recognition and question intent recognition capabilities. Using a supervised model fine-tuning method, we fine-tune the natural language processing model using labeled fault entity recognition sentences and question intent recognition sentences. This builds a model suitable for identifying fault entities and question intent for large equipment in dry bulk ports, improving the model's accuracy in identifying typical fault entity information for large equipment in dry bulk ports.
[0066] The model receives the processed question information and matches it with the corresponding Cypher statement. The model identifies the fault entity information in the question, which may include the device name, fault location, and fault representation. Combined with the identified question intent, the model matches the pre-written Cypher statement template, populates the identified entity information into the fault entity slot in the template, and generates the corresponding Cypher statement.
[0067] Use the matched and completed Cypher statement to query the fault entity information. Use the generated Cypher statement to query the knowledge graph and return the corresponding fault knowledge.
[0068] The query results and questions are reorganized and polished through the big model. The fault knowledge found is input into the big model, which then prompts the big model to organize the language. No additional supplementary answers are required from the big model, and the fault knowledge is reorganized into more fluent language.
[0069] Answer output: The server sends the organized language to the user interface, completing the entire question-answering process.
[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. An intelligent question-answering method for large dry bulk equipment failure, characterized in that: include: Utilize the fault entity recognition model and question intent recognition model to parse user input questions, extract fault entity information, and identify the corresponding question intent; Based on the fault entity information and question intent, the corresponding Cypher statement template is matched and a Cypher query statement is generated. Retrieve relevant fault knowledge in the knowledge graph based on the generated Cypher query statement; Use a large language model to grammatically restructure the retrieved fault knowledge and generate a readable fault diagnosis report; The generated fault diagnosis report is sent to the user interface, providing relevant maintenance suggestions and operation guidelines.
2. The intelligent question-answering method for large dry bulk equipment faults according to claim 1 is characterized in that: The fault entity information includes: equipment location, fault point and fault characterization; the sample data is manually annotated using the BIO annotation method to obtain training data, and the training data is used to train the fault entity recognition model; the question data is manually annotated using the intent annotation method to obtain training data, and the question intent recognition model is trained using the training data; The fault entity recognition model decomposes and encodes the training data and inputs the decomposition results into a linear classification layer. The linear classification layer identifies entities and obtains the labeling results of each character. Characters of the same category are combined into words, which are used as the category classification results corresponding to the entity to obtain different entities. Through cross-entropy loss function optimization, the context information captured by self-attention is used to evaluate the classification results of entity types, thereby realizing the training of the fault entity recognition model. The question intent recognition model uses the Tongyi Qianwen model to generate multiple different types of sentences, including inquiries about basic equipment information, inquiries about fault representations, and inquiries about fault repair processes; the pre-trained model BERT is combined with the deep learning classifier LSTM to perform text classification tasks, and the training data is used to adjust the parameters of the question intent recognition model to realize the training of the question intent recognition model.
3. The intelligent question-answering method for large dry bulk equipment faults according to claim 1 is characterized in that: Generating a Cypher query statement specifically includes: According to the fault entity information and the question intent, a corresponding preset Cypher statement template is matched. The Cypher statement template reserves a fault entity information slot for filling in the fault entity information extracted from the user question to form a corresponding Cypher statement; the identified entity information is filled into the entity information slot reserved in the Cypher statement query template to obtain the corresponding Cypher query statement.
4. The intelligent question-answering method for large dry bulk equipment faults according to claim 1 is characterized in that: The knowledge graph is constructed based on the Neo4j graph database and is used to store equipment components, fault locations, fault representations, and repair processes; the fault knowledge is output in the form of a dictionary.
5. The intelligent question-answering method for large dry bulk equipment faults according to claim 1 is characterized in that: The fault diagnosis report is sent to the user interface in JSON format.
6. An intelligent question-answering system for dry bulk cargo equipment failures implemented based on the intelligent question-answering method for dry bulk cargo equipment failures according to any one of claims 1 to 5, characterized in that: include: Semantic recognition module, query statement matching module, knowledge graph query module, answer generation and optimization module and system interaction module, among which: The semantic recognition module uses the fault entity recognition model and the question intention recognition model to parse the question input by the user, extract the fault entity information, and identify the corresponding question intention; The query statement matching module matches the corresponding Cypher statement template according to the fault entity information and the question intention, and generates a Cypher query statement; The knowledge graph query module retrieves relevant fault knowledge in the knowledge graph according to the generated Cypher query statement; The answer generation and optimization module uses a large language model to grammatically restructure the retrieved fault knowledge and generate a readable fault diagnosis report; The system interaction module sends the generated fault diagnosis report to the user interaction interface and provides relevant maintenance suggestions and operation guidelines.