Industrial system automatic fault diagnosis method based on large language model

Abstract

The invention discloses an industrial system automatic fault diagnosis method based on a large language model. According to the method, a three-layer mapping system of industrial data, natural language description and knowledge reasoning is constructed, field multi-source sensor data are subjected to semantic conversion, and a quantitative calculation model based on a large language model is constructed based on historical data and logs. And a fault case is matched in real time with the help of a retrieval-enhancement generation technology to serve as a reference, the fault case and abnormal information are input into a knowledge reasoning model based on a large language model together, a structured logical reasoning chain is generated, and a diagnosis conclusion containing candidate faults, cause analysis and disposal suggestions is further output. Meanwhile, through user feedback and a reinforcement learning mechanism, the model and the knowledge base are adaptively updated, the defects of traditional static rules and expert experience are effectively overcome, the accuracy, interpretability and robustness of fault detection are remarkably improved, and the method adapts to complex and changeable working condition requirements.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial intelligent diagnosis, and particularly relates to an industrial system automated fault diagnosis method based on a large language model. Background Art

[0002] In the field of industrial fault diagnosis, traditional expert systems mainly rely on domain experts to manually write and maintain a rule base. For example, inferences are made through logics such as "parameter A is abnormally high + parameter B is abnormally low → fault X". However, such systems often face bottlenecks such as high rule base maintenance costs, insufficient logic coverage, and difficulty in coping with dynamic working conditions when extended to multi-scenario and multi-variable working conditions. At the same time, the judgment method based on manual experience also has obvious subjectivity problems in practice: operators or engineers may give different diagnostic conclusions for the same fault phenomenon, making it difficult to ensure the consistency and accuracy of the results. With the continuous increase in the scale of industrial systems and monitoring variables, the fault diagnosis method relying on individual experience is increasingly difficult to meet the requirements of complex scenarios. With the rapid development of LLM (Large Language Model) technology, the knowledge base and logical reasoning ability formed by its learning from a large number of literatures and cases provide a new path for fault diagnosis. However, there is a lack of a unified mapping relationship between sensor coding and calculation parameters in industrial systems and natural language descriptions, resulting in difficulty for LLM to intuitively understand multi-source data; at the same time, traditional static thresholds are difficult to accurately quantify the degree of parameter abnormality, and it is urgent to construct a dynamic normal range based on historical data, process adjustments, and environmental factors. In addition, LLM can only rely on inherent static knowledge and may not be able to timely absorb newly emerging fault information and cases during long-term operation, which may lead to misjudgment or missed judgment. In view of this, how to effectively combine industrial data with the semantic reasoning ability of LLM and form an intelligent diagnosis framework that can be updated quickly and dynamically adapt to different working conditions has become a technical problem that needs to be solved urgently at present. Summary of the Invention

[0003] The present invention proposes an automated diagnosis method based on a three-layer mapping of industrial data-natural language-knowledge reasoning. This method constructs an industrial system operation dataset and a log library, establishes a bidirectional mapping between industrial variables and natural language, dynamically constructs the normal range of variables using a quantitative calculation model, and outputs abnormal information. By means of the Retrieval-Augmented Generation (RAG) technology, fault cases are retrieved, and based on the abnormal information and the above cases, a knowledge reasoning model generates a complete logical reasoning chain, achieving accurate fault diagnosis of industrial systems. At the same time, combined with user feedback and reinforcement learning algorithms, the knowledge reasoning model is adaptively updated, and new fault cases and fault modes are regularly added to the domain knowledge base to ensure high-precision, real-time response, and adaptive update capabilities in dealing with variable working conditions and new faults. To achieve the above object, the present invention adopts the following technical solutions: An industrial system automated fault diagnosis method based on a large language model, comprising the steps of: S1, constructing an industrial system operation dataset and preprocessing the data in the industrial system operation dataset; S2, constructing a semantic mapping relationship of industrial variables and converting industrial data and natural language descriptions; S3, constructing a quantitative calculation model based on a large language model, and outputting abnormal content based on semantic fusion and comprehensive reasoning; S4, constructing a knowledge reasoning model based on a large language model to diagnose faults in the industrial system; S5, optimizing the knowledge reasoning model based on an expert feedback mechanism, and adaptively updating the knowledge base and the knowledge reasoning model; Further, step S1 includes: Step S11, obtaining the historical operation data of key equipment in the industrial system, external environmental factors, physical parameters, and load characteristics at corresponding times; constructing an industrial system dataset , , wherein represents the measured data of the i-th feature, 1 ≤ i ≤ , represents the real number field; Step S12, preprocessing the data in the industrial system dataset X, and preliminarily screening each column of features based on the 3 criterion, wherein is the mean value of this column of features, is the standard deviation of this column of features, and the observed values that meet the range are denoted as :

[0004] Based on the industrial business scope, the observed values that meet the business scope are recorded as :

[0005] The rated operating range of the equipment is , The obtained data set is expressed as: 。

[0006] Step S13, construct an industrial system log library: The system collects the original log data from the production scheduling system, including process parameter change records, scheduling decisions, equipment status updates, and operator remarks, and cleans, formats, and structures the log data.

[0007] Step S14, construct an industrial system domain knowledge base: Sort out and collect the industrial system failure cases in public literature, industry reports, technical manuals, and historical failure logs, and standardize and structurally store the collected cases to form an industrial system domain knowledge base with multi-dimensional information such as time stamps, failure types, causes, diagnosis, and treatment plans.

[0008] Further, step S2 includes: Based on the multi-source monitoring variables obtained in step S11, construct an industrial variable-natural language description mapping table, match the variables with their standardized natural language descriptions, and establish the mapping relationship between industrial variables and standardized natural languages.

[0009] Further, step S3 constructs a quantitative calculation model based on the large language model to achieve the output of abnormal content, including: Step S31, construct a quantitative calculation model based on the large language model, generate the dynamic normal range of each variable, and preset the reasonable boundaries of the variables; Step S32, based on the dynamic normal range of the variables generated in step S31, compare the real-time monitoring data with the preset reasonable boundaries, and output the specific values of the parameters that exceed the normal range.

[0010] Further, step S4 constructs a knowledge reasoning model based on the large language model to achieve the fault diagnosis of the industrial system, including: Step S41, based on the retrieval-enhanced-generation technology, retrieve in real time from the industrial system domain knowledge base constructed in step S14 the fault cases and diagnostic experiences highly relevant to the current abnormal state; Step S42, based on the large language model, comprehensively analyze the internal relationship between the abnormal state and historical fault cases, generate a logical reasoning chain, and output the fault diagnosis conclusion.

[0011] Further, step S41 includes: Using the word vector encoder of LLaMA Convert the data to be retrieved and each document into word vectors and :

[0012] Calculate the similarity score between the data to be retrieved and each document using cosine similarity , where represents the dot product operation, represents the norm of the vector:

[0013] Select the document with the highest similarity documents to form a set :

[0014] Concatenate the abnormal information with cases and use them together as the input of the large language model : , where represents the concatenation operation.

[0015] Further, adopting a structured output mechanism, the output of the fault diagnosis conclusion includes: Candidate fault set: The candidate fault set includes possible fault types and their confidence scores based on inference chain matching; Fault cause analysis: The fault cause analysis includes showing the logical relationship between abnormal parameters and revealing how they lead to specific faults; Recommended disposal plan: Based on the recommended disposal plan and cases, put forward targeted maintenance suggestions.

[0016] Further, step S5 optimizes the knowledge reasoning model based on the feedback mechanism to realize the adaptive update of the knowledge base and the knowledge reasoning model, including: Step S51, feedback optimization: Establish a structured user feedback collection mechanism, record the evaluation scores of the operators on the diagnosis conclusion, and use the feedback data as the input quantity to perform reinforcement learning through the GRPO algorithm to optimize the strategy of the knowledge reasoning model.

[0017] Step S52, iterative update of the knowledge base cases: Based on expert feedback, the latest industry literature, equipment maintenance manuals and expert reports, regularly add new fault cases and fault modes to the domain knowledge base.

[0018] Further, step S51 includes: In one update, the GRPO algorithm first samples from the old policy for the same exception information to sample diagnostic results , perform a within-group comparison on the evaluation scores of the diagnostic conclusions, and calculate the normalized advantage value of each answer relative to the within-group mean . On this basis, the objective function of GRPO is written as:

[0019] where is the probability ratio of the old and new policies at the th token, is the advantage value obtained by scoring and normalizing the within-group answers, and min jointly limit the update amplitude, is the hyperparameter of the clipping operation, is to control the coefficient of the divergence constraint strength, is the reference policy.

[0020] An industrial system automation fault diagnosis system based on a large language model continuously collects key system device data in the industrial system as input, and outputs corresponding diagnostic results based on the industrial system automation fault diagnosis method based on the large language model.

[0021] The advantages of the present invention are as follows: First, by constructing an industrial variable semantic mapping mechanism, multi-source on-site data (such as sensor code TF, calculation parameter O2_Level) is converted into natural language descriptions (such as "furnace outlet temperature", "air oxygen content") to solve the semantic gap between industrial data and the language model. Second, based on the large language model, semantic fusion of recent historical data and the industrial system log library is performed, the normal fluctuation range of parameters is dynamically inferred, and the deviation amplitude of abnormal data is output in real time; at the same time, relying on the RAG technology, fault cases highly relevant to the current anomaly are retrieved, and combined with logical reasoning to generate structured diagnostic results, and the output includes a candidate fault set, fault cause analysis, and recommended disposal solutions. Finally, based on the user feedback results, the large language model is updated through reinforcement learning, and the fault cases and fault patterns in the knowledge base are updated regularly, so as to continuously improve the accuracy and robustness of fault detection and diagnosis. Brief Description of the Drawings

[0022] Figure 1 is the flowchart of the industrial system automation fault diagnosis method based on the large language model according to an embodiment of the present invention; Figure 2Schematic diagram of the industrial system automation fault diagnosis method based on large language model according to an embodiment of the present invention. Detailed implementation manners

[0023] The technical solutions of the embodiments of the present invention will be explained and described below. However, the following embodiments are only the preferred embodiments of the present invention, not all of them. Based on the embodiments in the implementation manners, other embodiments obtained by those skilled in the art without creative efforts all fall within the protection scope of the present invention.

[0024] Next, the present invention will be further described in conjunction with Figure 1 and Figure 2 and the detailed implementation manners.

[0025] For an embodiment of the industrial system automation fault diagnosis method based on large language model of the present invention, the construction process and principle can be referred to Figure 1 and Figure 2 as shown, and specifically includes the following steps: S1. Construct an industrial system operation data set, collect the operation data of the industrial system and perform data preprocessing; S2. Construct an industrial variable semantic mapping relationship to convert industrial data and natural language descriptions; S3. Construct a quantitative calculation model based on large language model, and output abnormal content based on semantic fusion and comprehensive reasoning; S4. Construct a knowledge reasoning model based on large language model to diagnose faults in the industrial system; S5. Optimize the knowledge reasoning model based on the expert feedback mechanism, and adaptively update the knowledge base and the knowledge reasoning model.

[0026] In an embodiment of the present invention, step S1 includes the following steps: Step S11. The method for constructing the industrial system operation data set is as follows: Obtain the historical operation data of each key device (such as boilers, pumps, compressors, heat exchangers, etc.) in the industrial system, as well as the external environmental factors (such as environmental temperature, humidity, wind speed, etc.), equipment physical parameters (such as material characteristics, heat transfer coefficient, design rated power), and load characteristics (such as overall system load changes, key parameter fluctuations) at the corresponding time. Organize them according to the time series to form an industrial system state data set at multiple time points. Assume that in the sampling task, the sampling time length is , and a total of types of characteristic data are collected, then the industrial system data set is denoted as: , where represents the measured data of the i-th feature, 1 ≤ i ≤ , represents the real number field.

[0027] Step S12, the data preprocessing includes: First, based on the 3 criterion (i.e., under the assumption of normal distribution, values exceeding the mean ± 3 times the standard deviation are regarded as anomalies) to preliminarily screen each column of features , where is the mean of this column of features, is the standard deviation of this column of features, and the observed values within the range can be denoted as :

[0028] At the same time, combined with industrial business knowledge, such as the rated working range of equipment , further identify potential abnormal data points. Thus, the observed values within the business scope can be denoted as :

[0029] The finally obtained data set can be expressed as: .

[0030] Adopt the local time window mean filling method to fill in the missing values. The specific method is: Select the data of 10 effective time steps before and after the time point where the missing value is located to calculate the mean as the filling value. If there are still missing data within the window, only use the valid data for calculation to maintain the time continuity of the data and improve the accuracy and robustness of fault diagnosis.

[0031] Step S13, the method for constructing the industrial system log library is: The system collects the original log data from the production scheduling system, covering information such as process parameter change records, scheduling decisions, equipment status updates, and operator remarks, and performs cleaning, formatting, and structuring processing on the collected logs.

[0032] Step S14, the method for constructing the industrial system domain knowledge base is: By systematically collecting and organizing industrial system fault cases from public literature, industry reports, technical manuals, and historical fault logs, construct a fault case knowledge base D, and each case . The described fault cases cover the fault descriptions, cause analyses, diagnostic methods, and disposal measures of various equipment faults (such as boilers, pumps, compressors, heat exchangers, etc.), including both fault examples reported in academic research and abnormal events recorded during actual operation. The collected cases are processed through standardization and stored in a structured manner to form a database with multi-dimensional information such as time stamps, fault types, causes, diagnostic and treatment plans.

[0033] Step S2 includes the following steps: For the multi-source monitoring variables obtained in step S11, construct an industrial variable-natural language description mapping table. Match the variables with their standardized natural language descriptions. For example, map the variable "TF" to "furnace outlet temperature" and "O2_Level" to "oxygen content in air" to ensure the generality and consistency of variable names. Secondly, for variables with different physical units and numerical dimensions, formulate a normalization expression strategy to reduce the impact of unit constraints on natural language conversion. For example, for pressure variables, they can be uniformly converted to the unit of "MPa" and the normal range reference value can be attached so that the large language model can accurately understand the physical meaning of the variables.

[0034] Step S3 includes the following steps: Step S31, the construction method of the quantitative calculation model based on the large language model is as follows: Construct a quantitative calculation model based on the large language model to realize semantic fusion and comprehensive reasoning of multi-source industrial data, and dynamically generate the "normal range" of each variable. Specifically, first select the historical monitoring data in the recent days (such as 90 days), combine the text data in the industrial system log library, and conduct unified modeling of multiple factors such as seasonal load changes, equipment aging, and operation strategy adjustments to infer the reasonable fluctuation range of the variable. For example, in a steel mill, in the initial stage of equipment commissioning, although temperature fluctuations exist but are concentrated in the range of 1450–1460 °C; in the stable aging stage of the equipment, due to long-term high load and environmental temperature rise, the system automatically expands the normal range to 1455–1475 °C. Compared with the traditional fixed range method, this dynamic range can more flexibly capture the phased characteristics, effectively reducing false alarms or missed alarms caused by ignoring factors such as process adjustments and environmental changes, thus providing guarantee for subsequent fault detection.

[0035] Step S32, the abnormal determination method is: Based on the dynamic normal range constructed in step S31, compare the real-time monitoring data with the preset reasonable boundary, and output the specific value of the parameter that exceeds the normal range. Specifically, when the actual value of a certain parameter is lower than the normal lower limit or exceeds the normal upper limit, the system automatically calculates the deviation amount (in percentage form) between it and the boundary to intuitively quantify the degree of abnormality. Finally, store the abnormality and the corresponding variable name in text form, such as "The furnace outlet temperature is 7% lower than the normal range and the fuel flow rate is 10% higher than the normal upper limit".

[0036] Step S4 includes the following steps: Step S41, fault case retrieval based on RAG: This step relies on the abnormal information recorded in step S32 , Using Retrieval-Augmented Generation (RAG) technology, real-time retrieve failure cases and diagnostic experiences highly relevant to the current abnormal state from the industrial domain knowledge base D constructed in step S13.

[0037] Specifically, first, use the word vector encoder of Llama to convert the data to be retrieved and each document into word vectors and :

[0038] Next, calculate the similarity scores between the data to be retrieved and each document using cosine similarity , where represents the dot product operation, represents the norm of the vector:

[0039] Select the documents with the highest similarity to form a set :

[0040] Subsequently, splice the abnormal information with these cases and use them together as the input to the LLM : , where represents the splicing operation.

[0041] Step S42, generation of fault diagnosis conclusion: After inputting into the knowledge inference model, relying on its profound knowledge reserve and rigorous logical reasoning mechanism, the model comprehensively analyzes the internal relationship between the abnormal information and historical fault cases, thereby generating a detailed logical reasoning chain. This reasoning chain not only reveals the causal relationship between abnormal indicators but also enhances the interpretability of the fault, facilitating the tracing and verification of each key node in the fault evolution. For example, when the input information includes data such as "the furnace outlet temperature is 5% higher and the oxygen content drops by 10%", the knowledge inference model will combine its own knowledge and the input similar cases to generate a logical reasoning chain of "abnormal temperature increase + insufficient oxygen → decrease in combustion efficiency → may lead to coking".

[0042] Meanwhile, by prompting the knowledge inference model to adopt a structured output mechanism, the diagnostic results are broken down into three core parts: (1) Candidate fault set: possible fault types and their confidence scores based on inference chain matching; (2) Fault cause analysis: showing the logical relationships between abnormal parameters and revealing how they lead to specific faults; (3) Recommended disposal solutions: presenting targeted maintenance suggestions in combination with cases. For example, for the diagnosis of "furnace coking risk (confidence 92%)", it is recommended to "adjust the combustion ratio and check the carbon deposition situation at the nozzles". This strategy that combines cross-modal reasoning and structured output not only significantly improves the accuracy and comprehensiveness of diagnosis but also enhances the transparency and credibility of the fault analysis process.

[0043] Step S5 includes the following steps: Step S51, User feedback collection and proximal policy optimization algorithm (Proximal Policy Generalized Reward Policy Optimization, GRPO) feedback optimization: Establish a structured user feedback collection mechanism to record the evaluation scores of operators on the diagnostic conclusions. These feedback data will be used for reinforcement learning through the GRPO algorithm to optimize the inference strategy of the LLM.

[0044] Specifically, in one update, the GRPO algorithm first samples from the old policy for the same abnormal information and samples out diagnostic results , and uses the evaluation scores of operators on the diagnostic conclusions for within-group comparison to calculate the normalized advantage value of each answer relative to the within-group mean . On this basis, the objective function of GRPO can be written as:

[0045] where is the probability ratio of the old and new policies at the th token, is the advantage value obtained by scoring and normalizing the within-group answers, and min together limit the update amplitude, is the hyperparameter of the clipping operation, is the coefficient that controls the divergence constraint strength, is the reference policy (set to LLaMA-32B).

[0046] Step S52, iterative update of the knowledge base cases: Based on expert feedback, the latest industry literature, equipment maintenance manuals, and expert reports, the system regularly adds new fault cases and fault modes to the domain knowledge base. Thereby identifying and integrating previously uncovered fault modes (such as newly emerging equipment failures or anomalies caused by process adjustments), continuously expanding and correcting the content of the knowledge base to enhance the LLM's understanding and generalization capabilities for specific industrial scenarios.

[0047] As described above, only the specific embodiments of the present invention are provided, but the protection scope of the present invention is not limited thereto. Those skilled in the art should understand that the present invention includes, but is not limited to, the content described in the above specific embodiments. Any modification that does not deviate from the functional and structural principles of the present invention will be included within the scope of the claims.

Claims

1. An industrial system automation fault diagnosis method based on a large language model, characterized in that, Including the steps: S1. Build an industrial system operation dataset and preprocess the data in the industrial system operation dataset; S2. Build an industrial variable semantic mapping relationship to convert industrial data and natural language descriptions; S3. Build a quantitative calculation model based on a large language model, and output abnormal content based on semantic fusion and comprehensive reasoning; S4. Build a knowledge reasoning model based on a large language model to diagnose industrial system faults; S5. Optimize the knowledge reasoning model based on an expert feedback mechanism, and adaptively update the knowledge base and the knowledge reasoning model.

2. The industrial system automation fault diagnosis method based on a large language model according to claim 1, wherein, The step S1 includes: Step S11: Obtain the historical operation data of key equipment in the industrial system, external environmental factors, physical parameters, and load characteristics at corresponding times; construct an industrial system dataset , , wherein represents the measured data of the i-th feature, 1 ≤ i ≤ , represents the real number field; Step S12, preprocess the data in the industrial system dataset X, and perform preliminary screening on each column of features based on 3 criteria, where is the mean of the feature values of this column, is the standard deviation of the feature values of this column, and record the observed values that meet the range as : ​ Based on the industrial business scope, the observed values that meet the business scope are denoted as : The rated operating range of the device is , The obtained data set It is expressed as: ； Step S13. Build an industrial system log library: The system collects raw log data from the production scheduling system, including process parameter change records, scheduling decisions, equipment status updates, and operator remarks, and cleans, formats, and structures the log data; Step S14. Build an industrial system domain knowledge base: Sort and collect industrial system fault cases in public literature, industry reports, technical manuals, and historical fault logs, and after standardizing and structuring the collected cases, form an industrial system domain knowledge base with multi-dimensional information such as time stamps, fault types, causes, diagnosis, and treatment plans.

3. The industrial system automation fault diagnosis method based on a large language model according to claim 2, wherein, The step S2 includes: Based on the multi-source monitoring variables obtained in step S11, build an industrial variable-natural language description mapping table, match the variables with their standardized natural language descriptions, and establish an industrial variable and standardized natural language mapping relationship.

4. The industrial system automation fault diagnosis method based on a large language model according to claim 3, wherein, The step S3 builds a quantitative calculation model based on a large language model to achieve abnormal content output, including: Step S31. Build a quantitative calculation model based on a large language model to generate the dynamic normal range of each variable and preset the reasonable boundaries of the variables; Step S32. Based on the dynamic normal range of the variables generated in step S31, compare the real-time monitoring data with the preset reasonable boundaries, and output the specific values of the parameters that exceed the normal range.

5. The industrial system automation fault diagnosis method based on a large language model according to claim 4, characterized in that, The step S4 builds a knowledge reasoning model based on a large language model to achieve the fault diagnosis of the industrial system, including: Step S41. Based on the retrieval-enhanced-generation technology, retrieve in real time from the industrial system domain knowledge base built in step S14 the fault cases and diagnostic experiences highly relevant to the current abnormal state; Step S42. Based on the large language model, comprehensively analyze the internal relationship between the abnormal state and historical fault cases, generate a logical reasoning chain, and output a fault diagnosis conclusion.

6. The industrial system automation fault diagnosis method based on a large language model according to claim 5, wherein The step S41 includes: Using the word vector encoder of LLaMA Convert the data to be retrieved and each document into word vectors and : Calculate the similarity score between the data to be retrieved and each document using cosine similarity , where represents the dot product operation, represents the magnitude of the vector: Select the document with the highest similarity to form a set : Combine the exception information with a case and use them together as the input to the large language model : , wherein represents a splicing operation.

7. An industrial system automation fault diagnosis method based on a large language model according to claim 5 or 6, characterized in that, Adopt a structured output mechanism to output the fault diagnosis conclusion, including: Candidate fault set: The candidate fault set includes possible fault types and their confidence scores matched based on the reasoning chain; Fault cause analysis: The fault cause analysis includes showing the logical relationship between abnormal parameters and revealing how they lead to specific faults; Recommended disposal plan: The recommended disposal plan puts forward targeted maintenance suggestions in combination with cases.

8. The industrial system automation fault diagnosis method based on a large language model according to claim 7, characterized in that, The step S5 optimizes the knowledge reasoning model based on the feedback mechanism to achieve the adaptive update of the knowledge base and the knowledge reasoning model, including: Step S51, Feedback Optimization: Establish a structured user feedback collection mechanism to record the evaluation scores of operators on the diagnostic conclusions, and use the feedback data as input to perform reinforcement learning through the GRPO algorithm to optimize the strategy of the knowledge inference model; Step S52, Iterative Update of Knowledge Base Cases: Based on expert feedback, the latest industry literature, equipment maintenance manuals, and expert reports, regularly add new fault cases and fault modes to the domain knowledge base.

9. The industrial system automation fault diagnosis method based on a large language model according to claim 8, wherein The said step S51 includes: The GRPO algorithm first starts from the old policy in one update for the same exception information sample out diagnostic results , make a within-group comparison of the evaluation scores of the diagnostic conclusions, and calculate the normalized advantage value of each answer relative to the within-group mean , on this basis, the objective function of GRPO is written as: ， where is the probability ratio of the new and old policies at the -th token, is the advantage value obtained by scoring and normalizing the in-group answers, together with min limits the update amplitude, is the hyperparameter of the clipping operation, is to control the coefficient of the divergence constraint strength, is the reference policy.

10. An industrial system automated fault diagnosis system based on a large language model, characterized in that, Continuously collect key system device data in the industrial system as input, and output corresponding diagnostic results based on any one of the industrial system automation fault diagnosis methods based on large language models in claims 1-9.

Images