Stereo garage fault analysis system and method based on large language model

Abstract

The invention discloses a stereo garage fault analysis system and method based on a large language model (LLM), and the method comprises the steps: generating a private large language model through the fine adjustment of a low-rank adaptation (LoRA) method of the large language model, and constructing an equipment maintenance knowledge base through knowledge graph generation software; when the system runs, equipment internet-of-things signals are subjected to semantic processing, classified according to attributes and features and correspondingly input into the local questioning sub-module and the global questioning sub-module to be processed, and the whole process from description of equipment running states to automatic giving of fault root cause analysis results and fault solutions is achieved. According to the invention, the efficiency and accuracy of fault diagnosis are improved, the maintenance cost is reduced, and comprehensive fault root cause analysis results and fault solutions are provided.

Description

Technical Field

[0001] The present invention relates to the field of fault diagnosis, and in particular to a system and method for analyzing faults of a stereoscopic car park based on a large language model. Background Art

[0002] In modern industry, complex equipment systems such as stereo garages are prone to failures during operation. These failure points are numerous and complex in combination, making it difficult to quickly locate the root cause of the failure, which in turn affects the timely repair of the failure. The traditional fault diagnosis model has the following main defects: (1) Models based on reasoning or logical analysis are complex and costly to build, and their application in complex systems is limited; (2) Traditional machine learning models require a large amount of labeled data and have poor interpretability; (3) The existing direct application of LLM has the problem of "hallucination" caused by insufficient domain knowledge; (4) Complex faults require comprehensive analysis based on multi-source heterogeneous data. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention proposes a system and method for analyzing faults in a stereoscopic car park based on a large language model.

[0004] The specific technical solutions are as follows: A fault analysis system for a stereo garage based on a large language model, comprising: a private LLM module, an equipment maintenance knowledge base module, an IoT and edge computing module, and a retrieval and reasoning module; The private LLM module is used to fine-tune the pre-trained large language model to generate a private large language model suitable for intelligent stereo garage fault analysis, and is used to output fault root cause analysis results and fault solutions; The equipment maintenance knowledge base module is used to construct an equipment maintenance knowledge base based on a knowledge graph through a knowledge graph generation software, which contains knowledge in the field of intelligent stereo garages; The IoT and edge computing module is used to collect daily operation data and fault information of the intelligent stereo garage, upload it to the fault analysis cloud platform, and convert it into semantic information, i.e., problems; The retrieval and reasoning module includes a question classification submodule, a local questioning submodule and a global questioning submodule; the question classification submodule is used to classify questions according to their attributes and characteristics, input questions that can be concluded by retrieval and summary into the local questioning submodule for processing, and input questions that need to evaluate the impact and analyze the root cause to give a conclusion into the global questioning submodule for processing; The local question submodule is used to perform neighbor search on text segments, entities, relationships and graph communities in the equipment maintenance knowledge base according to the questions, and screen out the content with the highest similarity and relevance scores, and organize and output them through a private large language model; The global question submodule performs retrieval based on the Map-Reduce architecture. First, a neighbor search is performed on the graph community in the equipment maintenance knowledge base according to the question, and the graph community whose similarity reaches the set threshold is taken as the relevant community; the community summary of the relevant community is input into the private large language model to generate an intermediate result; then the intermediate result is grouped according to the similarity and relevance score with the question, and after being summarized through the private large language model, it is integrated into the fault root cause analysis and fault solution.

[0005] Furthermore, in the private LLM module, fine-tuning the pre-trained large language model includes the following steps: (1.1) Data preprocessing: The maintenance manual and troubleshooting methods of the intelligent parking garage are converted into input-output question-answer pairs, and data cleaning, redundant information removal and standardization are performed; (1.2) Using the preprocessed data, the pre-trained large language model is fine-tuned through the LoRA method to obtain a private large language model.

[0006] Furthermore, the knowledge graph generation software selects GraphRAG, and building the equipment maintenance knowledge base includes the following steps: (2.1) Text segmentation: After unstructured processing of structured historical maintenance records, they are segmented into blocks according to pre-defined block sizes to obtain multiple text blocks; (2.2) Use a private large language model to identify entities from each text block and extract the relationship between entities; (2.3) Community clustering: An entity and its corresponding attribute information are recorded as a node. Based on the relationship between entities obtained in (2.2), the nodes are clustered into graph communities, and a private large language model is used to generate community summaries. (2.4) constructing a knowledge graph including nodes, relationships, and graph communities, and storing the knowledge graph as an equipment maintenance knowledge base; the equipment maintenance knowledge base is plugged into the private large language model to obtain a fault analysis model.

[0007] Furthermore, in the step (2.2), entities are identified from the text block by performing named entity recognition, and the relationship between entities is extracted by using a relationship extraction algorithm; and in the step (2.3), node clustering is performed using the Leiden algorithm.

[0008] Furthermore, the PLC controller of the intelligent stereoscopic parking garage is the target device, the IoT box is connected to the serial port or Ethernet port of the target device, and the IoT box supports the industrial communication protocol to obtain the operating data and fault information of the target device; Edge computing is performed on the IoT box: For non-critical operation data, an abnormal data trigger mechanism is set. Only when a specific error code or abnormal situation is detected, the IoT box will publish the non-critical operation data within a fixed time before and after the fault to the fault analysis cloud platform; for critical operation data, the IoT box will publish it to the fault analysis cloud platform in real time; In the fault analysis cloud platform, data cleaning and format conversion are performed to convert the data into semantic information that can be processed by a large language model.

[0009] Furthermore, the Internet of Things box uses a data terminal unit Internet of Things box, which corresponds to the RS485 interface connected to the target device and supports Modbus or DLT protocol.

[0010] A method for analyzing a fault in a stereoscopic car park based on a large language model is implemented based on the stereoscopic car park fault analysis system based on a large language model, and includes the following steps: S1: In the private LLM module, based on the maintenance manual and standard fault handling method of the intelligent stereo garage, the LoRA method is used to fine-tune the pre-trained large language model to obtain a private large language model suitable for fault analysis of the intelligent stereo garage; S2: In the equipment maintenance knowledge base module, the GraphRAG method is used to build an equipment maintenance knowledge base based on the knowledge graph, and the domain knowledge of the intelligent stereo garage is used as a database plug-in outside the private large language model to form a fault analysis model; S3: Through the IoT and edge computing modules, IoT data collection and processing of daily operation and fault information of the smart parking garage are carried out, and semantic information is output to the retrieval and reasoning modules; S4: Take the semantic information output by the IoT and edge computing modules as questions, and classify the questions into simple retrieval questions or complex analysis questions according to their attributes through the question classification submodule; the simple retrieval questions are questions that can give conclusions through retrieval and summary, and are processed by the local questioning submodule; the complex analysis questions require impact assessment and root cause analysis to give conclusions, and are processed by the global questioning submodule.

[0011] Furthermore, the S3 specifically includes the following steps: The IoT box is connected to the serial port or Ethernet port of the target device and obtains the operating data and fault information of the target device through the industrial communication protocol; The IoT box transmits the collected data to the fault analysis cloud platform, where the data is recorded and secondary processed, including data cleaning and format conversion; the secondary processed data is converted into semantic information that can be processed by a large language model; During the above transmission process, edge computing is performed on the IoT box side: for non-critical operation data, an abnormal data trigger mechanism is set. Only when a specific error code or abnormal situation is detected, the IoT box will publish the non-critical operation data within a fixed time before and after the fault to the fault analysis cloud platform; for critical operation data, the IoT box will publish it to the fault analysis cloud platform in real time.

[0012] Furthermore, the characteristics of the simple search type questions include: ① querying a single clear fault code processing method; ② querying the standard maintenance process or parts replacement method; ③ querying equipment parameters or configuration information; ④ the problem description is simple and direct, without the need to associate multiple factors; The characteristics of the complex analysis type problems include: ① containing multiple related fault codes; ② involving mutual influence or cascading failures between devices; ③ repeated failures; ④ the need to analyze the impact of environmental factors on the failure; ⑤ the problem description contains words that require in-depth analysis, and the words that require in-depth analysis include: non-standard, root cause, and correlation.

[0013] The beneficial effects of the present invention are: The fault analysis system for the stereoscopic parking garage constructed by the present invention is based on a large language model and graph retrieval enhanced generation. By combining a private large language model, an equipment maintenance knowledge base and Internet of Things data processing technology, it realizes the whole process from describing the equipment operation status to automatically generating fault identification, root cause analysis results and fault solutions, thereby improving the efficiency and accuracy of fault diagnosis, especially for complex faults that are difficult to solve through traditional rules or logical reasoning; and solves the "hallucination" problem caused by insufficient knowledge. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is an architecture diagram of a three-dimensional parking garage fault analysis system generated based on a large language model and graph retrieval enhancement in the present invention.

[0015] Figure 2 It is a schematic diagram of the IoT and edge computing architecture and model connection in the present invention.

[0016] Figure 3 It is a partial questioning flow chart of retrieval and reasoning of the fault analysis model in the present invention.

[0017] Figure 4 It is a global questioning flow chart of retrieval and reasoning of the fault analysis model in the present invention. DETAILED DESCRIPTION

[0018] The present invention will be described in detail below according to the accompanying drawings and preferred embodiments, and the purpose and effect of the present invention will become more clear. The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] like Figure 1 and Figure 2 As shown, a stereo garage fault analysis system based on a large language model includes: a private LLM module, an equipment maintenance knowledge base module, an Internet of Things and edge computing module, and a retrieval and reasoning module.

[0020] Private LLM module: used to fine-tune the pre-trained large language model to enhance model performance, generate a private large language model (hereinafter referred to as private LLM) suitable for intelligent stereo garage fault analysis, so that it can output indefinite length descriptive fault root cause analysis and fault solution; the pre-trained large language model uses existing open source large language models, including Qwen2.5-32B-Instruct, Gemma2-27B-instruct, Llama3-70B-Instruct, etc. The specific fine-tuning process includes: (1.1) Data preprocessing: The maintenance manual and troubleshooting methods of the intelligent parking garage are converted into input-output question-answer pairs, that is, question-answer pairs containing "input" and "output" in JSON format. The question-answer pairs are first cleaned, and then redundant information such as irrelevant characters, punctuation marks, and stop words are removed. The text is then standardized and the data is converted into a JSON format suitable for the input model.

[0021] As shown in Table 1 below, some question-answer pair examples are given, where the input is a question and the output is a descriptive fault root cause analysis (i.e., the possible cause of the problem) and a fault solution (including the location that needs to be checked) of indefinite length.

[0022] Table 1 Question-answer pair examples

[0023] (1.2) Combined with the above preprocessed data, the LoRA method is used to fine-tune the pre-trained large language model to obtain a private LLM. LoRA is a technology used to adjust the parameters of a pre-trained large language model to adapt to new tasks. It adapts to intelligent stereo garage fault analysis by introducing additional low-rank matrices instead of directly modifying the original weight matrix; this method can reduce the demand for computing resources while maintaining the performance of the model. In this embodiment, the training parameters are set as follows: epoch number = 5, batch size batch_size = 8, and learning rate learning_rate = 0.0003.

[0024] Equipment maintenance knowledge base module: used to build an equipment maintenance knowledge base based on the knowledge graph through knowledge graph generation software (GraphRAG is used in this embodiment), so as to plug the domain knowledge of the intelligent stereo garage into the private LLM as a database, which can effectively reduce the illusion of a large model and cover as much knowledge related to fault handling generated in the daily operation of the equipment as possible. The equipment maintenance knowledge base and the private LLM constitute a fault analysis model.

[0025] The specific process of building a knowledge graph-based equipment maintenance knowledge base through GraphRAG includes: (2.1) Text segmentation: The structured historical maintenance records are processed into unstructured form, and then the unstructured maintenance records are segmented into blocks according to a predefined block size to obtain multiple text blocks. In this embodiment, the predefined block size is 500 characters.

[0026] (2.2) Entity Relationship Extraction: Use private LLM to identify entities from each text block and extract the relationship between each entity. In this embodiment, named entity recognition (NER) is performed to identify entities, and a relationship extraction algorithm is used to extract the relationship between entities.

[0027] (2.3) Community clustering: An entity and its corresponding attribute information are recorded as a node. Based on the extracted entity relationships (i.e., the relationships between nodes), the Leiden algorithm is used to cluster the nodes into graph communities, and a community summary is generated by a private LLM. To generate a community summary, the graph community features must first be extracted. The graph community features include topological features (i.e., graph community size, internal connection density, central nodes, connectivity between graph communities, etc.), semantic features (i.e., extracting high-frequency entities within the graph community and the distribution of entity types, so as to quickly determine the topic tendency of the community), and relationship features (i.e., the relationships between entities within the community); then, the features are structured and the graph community features are converted into text templates or JSON format; then, community summaries are generated, the feature information is combined in natural language, and redundant information is filtered out to achieve a summary of the content of each community.

[0028] (2.4) Knowledge storage: A knowledge graph with 120,000 nodes and 350,000 relationships is constructed and stored in the LanceDB vector database as the equipment maintenance knowledge base of the private LLM.

[0029] IoT and edge computing module: used to collect daily operation data and fault information of the smart parking garage, and convert these operation data into semantic information that can be answered by a large language model. It specifically includes the following main contents: (3.1) Hardware connection: Use a data transfer unit (DTU) or remote terminal unit (RTU) IoT box to connect to the serial port or Ethernet port of the target device (i.e., the PLC controller of the intelligent parking garage) to transmit the data collected by the target device to the fault analysis cloud platform. In this embodiment, the IoT box uses a DTU IoT box, which is connected to the RS485 interface of the target device.

[0030] (3.2) Protocol conversion: The IoT box supports multiple industrial communication protocol parsing, thereby obtaining real-time operating data and fault information of the target device based on physical connection. This embodiment supports Modbus-TCP or DLT645-2007 protocol.

[0031] (3.3) Edge computing strategy: Edge computing is performed on the IoT box to save data traffic from the target device to the server. Specifically, the IoT box first determines whether the operating data is critical based on the preset operating parameter classification. For critical operating data, the IoT box publishes it to the MQTT server (MQTT Broker) in the fault analysis cloud platform through an interface; for non-critical operating data, an abnormal data trigger mechanism is set. Only when a specific error code or abnormal situation is detected (in this embodiment, the error code defined by a certain three-dimensional parking garage equipment manufacturer shall prevail. When the E80 "broken rope detection action" or E99 "lift switch error" fault code is detected), the IoT box uploads the data 30 seconds before and after the fault to the fault analysis cloud platform through the interface.

[0032] (3.4) In the fault analysis cloud platform, the data from the MQTT Broker is cleaned and formatted, and the data is converted into semantic information that can be processed by a large language model, namely, a semantic fault description.

[0033] Retrieval and reasoning module: The semantic fault description output by the IoT and edge computing modules is used as a problem, classified according to the problem attributes, and intelligently selects the reasoning method of local query or global query, and performs retrieval and reasoning in combination with the fault analysis model. The retrieval and reasoning module is divided into the following functions: problem classification submodule, local query submodule, and global query submodule; the specific functions implemented by each module are as follows.

[0034] Question classification submodule: Since local questions and global questions have a significant difference in the resource consumption of the system, this embodiment designs a question classification submodule, uses private LLM to judge the attributes and characteristics of the question, and classifies the question into different submodules (i.e., local question submodule or global question submodule) for processing to shorten the system response time. Specifically, the private LLM classifies the questions that can be concluded through retrieval summary as simple retrieval questions (retrieval_query) based on the attributes and characteristics of the question (the result is obtained by querying the private LLM once through the prompt word template), and classifies them into the local question submodule for processing; and classifies the questions that require impact assessment and root cause analysis to give a conclusion as complex analysis questions (complex_query), and classifies them into the global question submodule for processing. The following is an example of a prompt word template for private LLM to judge question classification: "System prompt: You are the problem classification expert of the intelligent three-dimensional parking garage fault analysis system. You are responsible for judging the complexity of the fault problem and deciding which query method to use to handle it.

[0035] Task: Please determine whether the following fault information is a simple search type problem that can be solved through a simple search, or a complex analysis type problem that requires in-depth root cause analysis.

[0036] Judgment criteria: The characteristics of simple search questions include: ① querying a single clear fault code handling method; ② querying the standard maintenance process or parts replacement method; ③ querying equipment parameters or configuration information; ④ the problem description is simple and direct, without the need to associate multiple factors. The characteristics of complex analysis questions include: ① containing multiple related fault codes; ② involving mutual influence or cascading failures between devices; ③ repeated faults; ④ the need to analyze the impact of environmental factors (such as temperature and humidity) on the fault; ⑤ the problem description contains words such as "non-standard", "root cause", "association", etc. that require in-depth analysis.

[0037] Question category: {question text}. Please answer only 'retrieval_query' or 'complex_query', no explanation or additional content." Local question submodule: Figure 3 As shown in the figure, after receiving the input semantic fault description information, in the equipment maintenance knowledge base, the original text segment (i.e., the text segment generated in the first step of the equipment maintenance knowledge base generation process), entity, relationship and graph community are searched for neighbors respectively, and the four groups of neighbor search results are ranked (Ranking) according to the similarity and relevance scores, and filtered (Filtering) from top to bottom according to the context window size of the private LLM to obtain the highest ranked original text, entity, relationship and community, and finally the filtered content is provided to the private LLM for language organization and output.

[0038] Global question submodule: Figure 4 As shown in the figure, after receiving the input semantic fault description information, the Map-Reduce architecture (Map-Shuffle-Reduce) is used for retrieval. The architecture includes three stages: mapping (Map), intermediate process (Shuffle), and reduction (Reduce), as shown below.

[0039] ① Map stage: Based on the input semantic fault description information, a neighbor search is performed on the graph community in the equipment maintenance knowledge base, that is, the similarity between the semantic fault description information and the community summary of each graph community is calculated. If the similarity is greater than or equal to the set threshold, the graph community is selected as the relevant community, and prompt words are constructed for each relevant community; the community summary of the relevant community is used as input, and intermediate results are generated with the help of private LLM. These intermediate results are generally partial answers related to the query, or further refinement of the content of the relevant community.

[0040] At this stage, a prompt word template similar to the following is generated: “System prompt: You are a fault analysis expert for stereo garages. Please answer questions related to the fault based on the community information provided.

[0041] Task: Analyze the content related to '{user question summary}' in the following community information, extract key information and generate partial answers.

[0042] Community information: {community summary}, {community key entities and relationships}.

[0043] Requirements: ① Answer only based on the community information provided; ② Extract the causes and solutions related to the fault '{fault code or phenomenon}'; ③ If the community information contains multiple related cases, please summarize the commonalities and differences; ④ The output format should include two parts: 'Possible causes' and 'Suggested measures'; ⑤ The answer length should be controlled within 200 words.

[0044] User question: {Original question}." ②Shuffle stage: sort and group the intermediate results of each relevant community according to their similarity and relevance scores with the question. The similarity and relevance scores of the intermediate results in the same group are similar. This process can bring together answers from different graph communities but with similar topics, preparing for the next reduction.

[0045] ③Reduce phase: The grouped intermediate results obtained in the Shuffle phase are summarized and executed by the private LLM, integrating the partial answers of each relevant community into a coherent and comprehensive final answer, namely the root cause analysis and fault solution. The final answer is obtained by synthesizing the answers of all relevant communities, which ensures that even in the face of complex global problems, a comprehensive and context-rich answer can be provided.

[0046] Local questions mainly use semantic token neighbor search to search the equipment maintenance knowledge base. The processing speed is fast, and the content provided to the private LLM is related to the literal meaning of the question. Global questions traverse all communities and call the private LLM multiple times to summarize the intermediate results, and finally integrate and output the most relevant content. Although the processing speed is slow, it can provide more comprehensive answers and is suitable for integrating indirect information to deeply analyze complex faults.

[0047] Based on the above-mentioned three-dimensional parking garage fault analysis system based on a large language model, this embodiment further proposes a three-dimensional parking garage fault analysis method based on a large language model, comprising the following steps: S1: In the private LLM module, based on the maintenance manual and standard fault handling method of the intelligent stereo garage, the LoRA method is used to fine-tune the pre-trained large language model to obtain a private LLM suitable for fault analysis of the intelligent stereo garage.

[0048] S2: In the equipment maintenance knowledge base module, the GraphRAG method is used to build an equipment maintenance knowledge base based on the knowledge graph, and the domain knowledge of the intelligent stereo garage is used as a database plug-in outside the private LLM to form a fault analysis model, which effectively reduces the large model illusion.

[0049] The equipment maintenance knowledge base is constructed by the GraphRAG method, which is implemented through the following sub-steps: S2.1: The structured historical maintenance records are unstructured. These unstructured maintenance record texts are cut into pre-defined block sizes, each of which contains a certain number of text units, such as sentences or paragraphs. This block-by-block approach breaks down large-scale text data into smaller parts that are easier to manage and process.

[0050] S2.2: For each text block, named entity recognition is performed through private LLM to identify specific entities in the text, such as names, places, organizations, etc. Subsequently, the semantic relationships between these entities are analyzed through the relationship extraction algorithm to extract the relationship information between the entities.

[0051] S2.3: The entities and relations extracted from the text blocks are identified and clustered by the Leiden algorithm to form multiple graph communities. Then, the private LLM is used to summarize the content of each community and generate a community summary for fast retrieval. The above entities, relations, and graph communities form an equipment maintenance knowledge base, which is plugged into the private LLM to form a fault analysis model.

[0052] S3: Through the IoT and edge computing modules, IoT data collection and processing of daily operation and fault information of the smart parking garage are carried out, and semantic fault descriptions are output to the retrieval and reasoning modules. This step specifically includes: IoT box connection: Use a DTU or RTU IoT box to connect to the serial port or Ethernet port of the target device to achieve physical connection between the target device and the IoT box, providing a basis for data collection. In this embodiment, the IoT box uses a DTU type FBOX.

[0053] Protocol data acquisition: Based on the physical connection, real-time operation and error signals are obtained from PLC, inverter, HMI and other devices in the intelligent parking garage through industrial communication protocols such as Modbus and DLT. These protocols are selected based on their wide application and reliability in the field of industrial automation.

[0054] Data transmission: The collected data is transmitted to the fault analysis cloud platform through the IoT gateway and finally reaches the MQTT Broker. The MQTT protocol is suitable for IoT environments due to its lightweight and low bandwidth consumption, especially in scenarios where data transmission is limited.

[0055] Data recording and processing: The server backend program of the fault analysis cloud platform receives data from the MQTT Broker and records and processes it. These secondary processes include data cleaning and format conversion. The processed data is converted into semantic information that can be understood and answered by the big model.

[0056] Edge computing: In order to effectively manage data traffic and reduce the burden on the server side, edge computing is implemented on the IoT box (implemented in the "Publish" step). Edge computing allows data processing to be performed near the data source, thereby reducing the amount of data transmitted to the fault analysis cloud platform. Specifically, for non-critical data, a condition-triggered collection strategy is adopted. Only when an error code or abnormal situation is detected, the IoT box will collect and send the data at that time and the previous period to the fault analysis cloud platform.

[0057] S4: Take the semantic fault description output by the IoT and edge computing modules as questions, and classify the questions into simple retrieval questions or complex analysis questions according to their attributes through the question classification submodule. Simple retrieval questions can be concluded by simply summarizing the retrieval, and are processed by the local questioning submodule; complex analysis questions require impact assessment and root cause analysis to give conclusions, and are processed by the global questioning submodule. Finally, the final answer is output, including fault root cause analysis and fault solution.

[0058] Specifically, the problem classification submodule receives the semantic fault description output by the IoT and edge computing modules, and performs classification judgment by calling the private LLM. A special prompt word template is designed, including the instruction "Please determine whether the following fault information needs to be evaluated and the root cause analyzed", and additional judgment criteria such as "contains multiple fault codes", "involves mutual influence between devices" or "repeated faults" and other complex situation characteristics. After receiving this prompt word and fault information, LLM outputs the classification result "simple retrieval type question (retrieval_query)" or "complex analysis type question (complex_query)". According to the classification results, simple retrieval type questions are guided to the local question submodule, and complex analysis type questions are guided to the global question submodule to achieve reasonable allocation of resources, ensure quick response to simple problems, and in-depth analysis of complex problems.

[0059] In this embodiment, 3825 fault records of the intelligent stereoscopic garage from 2018 to 2023 are collected and used as a sample data set. The present invention is used to perform fault analysis on the sample data set. The fault analysis using local questions has an average response time of 1.2 seconds and an accuracy of 89.7%; the fault analysis using global questions has an average response time of 8.5 seconds and an accuracy of 93.4%. Compared with the existing expert system-based fault analysis system, the efficiency of the present invention in performing root cause analysis of faults is improved by 47%, and the speed of generating the corresponding solution is increased by 62%. In summary, the present invention can achieve fast and accurate fault analysis and solution generation, improve the maintenance efficiency of the intelligent stereoscopic garage, and reduce maintenance costs.

[0060] Those skilled in the art can understand that the above are only preferred examples of the invention and are not intended to limit the invention. Although the invention is described in detail with reference to the above examples, those skilled in the art can still modify the technical solutions recorded in the above examples or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, etc. made within the spirit and principle of the invention shall be included in the protection scope of the invention.

Claims

1. A three-dimensional parking garage fault analysis system based on a large language model, characterized in that: include: Private LLM module, equipment maintenance knowledge base module, IoT and edge computing module, retrieval and reasoning module; The private LLM module is used to fine-tune the pre-trained large language model to generate a private large language model suitable for intelligent stereo garage fault analysis, and is used to output fault root cause analysis results and fault solutions; The equipment maintenance knowledge base module is used to construct an equipment maintenance knowledge base based on a knowledge graph through a knowledge graph generation software, which contains knowledge in the field of intelligent stereo garages; The IoT and edge computing module is used to collect daily operation data and fault information of the intelligent stereo garage, upload it to the fault analysis cloud platform, and convert it into semantic information, i.e., problems; The retrieval and reasoning module includes a question classification submodule, a local questioning submodule and a global questioning submodule; the question classification submodule is used to classify questions according to their attributes and characteristics, input questions that can be concluded by retrieval and summary into the local questioning submodule for processing, and input questions that need to evaluate the impact and analyze the root cause to give a conclusion into the global questioning submodule for processing; The local question submodule is used to perform neighbor search on text segments, entities, relationships and graph communities in the equipment maintenance knowledge base according to the questions, and screen out the content with the highest similarity and relevance scores, and organize and output them through a private large language model; The global question submodule performs retrieval based on the Map-Reduce architecture. First, a neighbor search is performed on the graph community in the equipment maintenance knowledge base according to the question, and the graph community whose similarity reaches the set threshold is taken as the relevant community; the community summary of the relevant community is input into the private large language model to generate an intermediate result; The intermediate results are then grouped according to their similarity and relevance scores with the questions, and aggregated through a private large language model to form fault root cause analysis and fault solutions.

2. The large language model-based three-dimensional parking garage fault analysis system according to claim 1 is characterized in that: In the private LLM module, fine-tuning the pre-trained large language model includes the following steps: (1.1) Data preprocessing: The maintenance manual and troubleshooting methods of the intelligent parking garage are converted into input-output question-answer pairs, and data cleaning, redundant information removal and standardization are performed; (1.2) Using the preprocessed data, the pre-trained large language model is fine-tuned through the LoRA method to obtain a private large language model.

3. The large language model-based three-dimensional parking garage fault analysis system according to claim 1 is characterized in that: The knowledge graph generation software is GraphRAG, and building the equipment maintenance knowledge base includes the following steps: (2.1) Text segmentation: After unstructured processing of structured historical maintenance records, they are segmented into blocks according to pre-defined block sizes to obtain multiple text blocks; (2.2) Use a private large language model to identify entities from each text block and extract the relationship between entities; (2.3) Community clustering: An entity and its corresponding attribute information are recorded as a node. Based on the relationship between entities obtained in (2.2), the nodes are clustered into graph communities, and a private large language model is used to generate community summaries. (2.4) constructing a knowledge graph including nodes, relationships, and graph communities, and storing the knowledge graph as an equipment maintenance knowledge base; the equipment maintenance knowledge base is plugged into the private large language model to obtain a fault analysis model.

4. The large language model-based three-dimensional parking garage fault analysis system according to claim 3 is characterized in that: In the step (2.2), entities are identified from the text block by performing named entity recognition, and the relationship between entities is extracted by using a relationship extraction algorithm; in the step (2.3), node clustering is performed using the Leiden algorithm.

5. The large language model-based three-dimensional parking garage fault analysis system according to claim 1 is characterized in that: The PLC controller of the intelligent stereo garage is the target device, the IoT box is connected to the serial port or Ethernet port of the target device, and the IoT box supports the industrial communication protocol to obtain the operating data and fault information of the target device; Perform edge computing on the IoT box: For non-critical operation data, set up an abnormal data trigger mechanism. Only when a specific error code or abnormal situation is detected, the IoT box will publish the non-critical operation data within a fixed time before and after the fault to the fault analysis cloud platform; For critical operating data, the IoT box publishes it to the fault analysis cloud platform in real time; In the fault analysis cloud platform, data cleaning and format conversion are performed to convert the data into semantic information that can be processed by a large language model.

6. The large language model-based three-dimensional parking garage fault analysis system according to claim 5 is characterized in that: The IoT box uses a data terminal unit IoT box, which corresponds to the RS485 interface connected to the target device and supports Modbus or DLT protocol.

7. A method for analyzing a fault in a stereoscopic parking garage based on a large language model, implemented based on the system for analyzing a fault in a stereoscopic parking garage based on a large language model according to any one of claims 1 to 6, characterized in that: The following steps are involved: S1: In the private LLM module, based on the maintenance manual and standard fault handling method of the intelligent stereo garage, the LoRA method is used to fine-tune the pre-trained large language model to obtain a private large language model suitable for fault analysis of the intelligent stereo garage; S2: In the equipment maintenance knowledge base module, the GraphRAG method is used to build an equipment maintenance knowledge base based on the knowledge graph, and the domain knowledge of the intelligent stereo garage is used as a database plug-in outside the private large language model to form a fault analysis model; S3: Through the IoT and edge computing modules, IoT data collection and processing of daily operation and fault information of the smart parking garage are carried out, and semantic information is output to the retrieval and reasoning modules; S4: Take the semantic information output by the IoT and edge computing modules as questions, and classify the questions into simple retrieval questions or complex analysis questions according to their attributes through the question classification submodule; the simple retrieval questions are questions that can give conclusions through retrieval and summary, and are processed by the local questioning submodule; the complex analysis questions require impact assessment and root cause analysis to give conclusions, and are processed by the global questioning submodule.

8. The method for analyzing faults in a stereo garage based on a large language model according to claim 7 is characterized in that: The S3 specifically includes the following steps: The IoT box is connected to the serial port or Ethernet port of the target device and obtains the operating data and fault information of the target device through the industrial communication protocol; The IoT box transmits the collected data to the fault analysis cloud platform, where the data is recorded and secondary processed, including data cleaning and format conversion; the secondary processed data is converted into semantic information that can be processed by a large language model; During the above transmission process, edge computing is performed on the IoT box side: for non-critical operation data, an abnormal data trigger mechanism is set. Only when a specific error code or abnormal situation is detected, the IoT box will publish the non-critical operation data within a fixed time before and after the fault to the fault analysis cloud platform; for critical operation data, the IoT box will publish it to the fault analysis cloud platform in real time.

9. The method for analyzing faults in a stereo garage based on a large language model according to claim 7, characterized in that: The characteristics of the simple search type questions include: ① querying a single clear fault code processing method; ② querying the standard maintenance process or parts replacement method; ③ querying equipment parameters or configuration information; ④ the problem description is simple and direct, without the need to associate multiple factors; The characteristics of the complex analysis type problems include: ① containing multiple related fault codes; ② involving mutual influence or cascading failures between devices; ③ repeated failures; ④ the need to analyze the impact of environmental factors on the failure; ⑤ the problem description contains words that require in-depth analysis, and the words that require in-depth analysis include: non-standard, root cause, and correlation.

Images