Jacket structure safety diagnosis system and method based on time sequence large language model

Abstract

The invention relates to the field of structure health diagnosis, and discloses a jacket structure safety diagnosis system and method based on a time sequence large language model, and the system comprises a multi-mode sensor module, a time sequence data preprocessing module, a time sequence large language model module, a multistage early warning and decision support module, and an intelligent user feedback and adaptive optimization module. The method comprises the following steps: S1, collecting data in real time; s2, carrying out data block processing to generate vectors; s3, converting the vector and adding a prompt prefix; s4, outputting a structure failure probability and high-risk area positioning; s5, outputting a maintenance priority and a cost optimization scheme; s6, providing a maintenance plan and a cost estimation report; and S7, collecting user feedback through an intelligent interface. Through the multi-mode sensor module of the time sequence large language model, multi-parameter time sequence data of the jacket structure are collected in real time, the comprehensive monitoring capability and the data fusion effect are achieved, and therefore the structural health monitoring precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of structural health diagnosis, and in particular to a jacket structure safety diagnosis system and method based on a time-series large language model. Background Art

[0002] Current technologies for monitoring jacket structures rely primarily on a variety of sensor devices to acquire data related to the structural environment. These sensors typically include anemometers, wave radars, and strain gauges, each capable of monitoring different physical parameters. This multimodal sensor arrangement enables the monitoring system to cover a wide range of environmental factors, such as wind speed variations, wave impacts, and temperature changes, providing essential data support for structural safety. Furthermore, traditional data processing methods often require relatively short signal acquisition and analysis times, making preliminary monitoring and assessment possible.

[0003] However, despite the success of existing technologies in monitoring multi-dimensional parameters, several pressing challenges remain. First, the data processing capabilities of numerous sensors often fall short of the real-time processing requirements required in complex situations, leading to potential information loss and data redundancy. Furthermore, the limitations of traditional algorithms in analyzing this data challenge the accuracy of safety assessments. Consequently, existing systems often struggle to fully and timely reflect the true safety status of jackets. Furthermore, the lack of effective user feedback and optimization mechanisms hinders the system's ability to flexibly adapt to ever-changing environments and demands. Summary of the Invention

[0004] In response to the deficiencies of the prior art, the present invention provides a jacket structure safety diagnosis system and method based on a time-series large language model, which solves the deficiencies of the existing jacket structure safety monitoring system in data acquisition, processing efficiency and dynamic optimization.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a jacket structure safety diagnosis system based on a time-series large language model, comprising: Multimodal sensor module for real-time acquisition of multi-parameter time series data of the jacket structure; The time series data preprocessing module is connected to the multimodal sensor module and is used to process the original time series data collected by the sensor in blocks and generate vector representations; The time series large language model module is connected to the time series data preprocessing module to perform sequence modeling on the mixed representation and output the structural failure probability and high-risk component location; A multi-level early warning and decision support module, which is connected to the time series large language model module, is used to generate visual diagnostic reports and trigger hierarchical alarms, outputting maintenance optimization plans and life extension decision recommendations; The intelligent user feedback and adaptive optimization module is connected to the multi-level early warning and decision support module to optimize system performance based on user feedback.

[0006] Preferably, the multimodal sensor module includes: Wave radar unit, used to monitor wave parameters in the sea area around the jacket in real time; Anemometer unit, used to measure wind speed, wind direction and gust characteristics in the area where the jacket is located; Fiber Bragg grating sensor units for measuring strain, temperature, and pressure parameters; An acceleration sensor unit, used to measure the vibration acceleration of the jacket in three-dimensional space; The tilt sensor unit is used to monitor the overall and local tilt angles of the jacket.

[0007] Preferably, the time series data preprocessing module includes: The data block unit is used to process the original time series data into blocks according to the preset time window; Embedding layer generation unit, used to generate vector representations through custom embedding layers.

[0008] Preferably, the time series large language model module includes: A modality alignment module to reprogram the vector representation of time series into a natural language compatible hybrid representation; The prompt prefix module is used to add a learnable task guidance prompt prefix before the input sequence to inject domain knowledge and task instructions; The safety diagnosis module is used to perform sequence modeling of mixed representations based on a pre-trained large language model, outputting structural failure probability and high-risk component location.

[0009] Preferably, the multi-level warning and decision support module includes: A multi-parameter threshold determination unit is used to set dynamic thresholds based on historical data and industry standards to trigger three levels of alarms; Trend prediction unit, used to generate fatigue expansion trend and structural failure probability in the next 30 days; The maintenance optimization unit is used to output prioritized maintenance plans and cost estimation reports based on risk levels and economic constraints.

[0010] Preferably, the intelligent user feedback and adaptive optimization module includes: Intelligent interface unit, used for users to visually receive feedback information and perform operations; Dynamic adjustment mechanism unit, used to optimize algorithms and models based on user feedback to improve system performance.

[0011] The jacket structure safety diagnosis method based on the time series large language model includes the following steps: S1, real-time acquisition of multi-parameter time series data through wave radar, anemometer, fiber Bragg grating sensor, acceleration sensor and tilt sensor; S2. Use the data block unit to block the original time series data according to the preset time window, and generate a vector representation through the embedding layer generation unit; S3, based on the text prototype library, converts the vector into a natural language compatible hybrid representation through the modality alignment module, and adds a learnable task guidance prompt prefix before the input sequence; S4. Use the safety diagnosis module to perform sequence modeling on the hybrid representation based on the pre-trained time-series large language model, and output the structural failure probability and high-risk area location; S5. Based on the failure probability and high-risk coordinates output by the time-series large language model, the multi-level early warning and decision support module generates a three-level alarm and outputs maintenance priorities and cost optimization plans; S6. Generate fatigue growth trends and structural failure probabilities for the next 30 days using the trend prediction unit, and provide prioritized maintenance plans and cost estimation reports through the maintenance optimization unit. S7. Collect user feedback through an intelligent interface and optimize algorithms and models based on dynamic adjustment mechanisms to improve system performance and user experience.

[0012] The present invention provides a jacket structure safety diagnosis system and method based on a time-series large language model. It has the following beneficial effects: 1. This invention utilizes a multimodal sensor module with a time-series large language model to collect multi-parameter time series data from the jacket structure in real time, achieving comprehensive monitoring capabilities and data fusion. Compared to existing single or traditional sensor solutions, this invention overcomes the limitations of monitoring blind spots and significantly improves the accuracy of structural health monitoring, making jacket safety assessments more reliable and comprehensive, truly achieving a shift from passive monitoring to active early warning.

[0013] 2. This invention achieves new levels of system intelligence and personalization by responding to user behavior in real time and continuously evolving based on feedback. Compared to existing technologies that are fixed and unable to adapt to user needs, this technical solution achieves a high degree of integration between information technology and user experience. This enables the system to not only adapt to environmental changes but also proactively capture subtle changes in user needs, ensuring continued superior performance in complex and changing situations.

[0014] 3. This invention achieves efficient and in-depth information extraction through intelligent data segmentation and vector representation technology. This approach demonstrates exceptional noise immunity when processing complex dynamic data, significantly overcoming the redundancy and information loss issues common in existing technologies. This significantly improves the accuracy and effectiveness of data analysis, laying a solid foundation for subsequent risk assessment and decision support. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a system module architecture diagram of the present invention; Figure 2 Schematic diagram of a multimodal sensor module of the present invention; Figure 3 Schematic diagram of the time series data preprocessing module of the present invention; Figure 4 This is a schematic diagram of the temporal large language model module of the present invention; Figure 5 Schematic diagram of the multi-level warning and decision support module of the present invention; Figure 6 Schematic diagram of the intelligent user feedback and adaptive optimization module of the present invention; Figure 7 The figure is a flow chart of the method steps of the present invention. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] Please see the attached Figure 1 The embodiment of the present invention provides a jacket structure safety diagnosis system and method based on a time-series large language model, including: Multimodal sensor module for real-time acquisition of multi-parameter time series data of the jacket structure; The time series data preprocessing module is connected to the multimodal sensor module and is used to process the original time series data collected by the sensor in blocks and generate vector representations; The time series large language model module is connected to the time series data preprocessing module to perform sequence modeling on the mixed representation and output the structural failure probability and high-risk component location; A multi-level early warning and decision support module, which is connected to the time series large language model module, is used to generate visual diagnostic reports and trigger hierarchical alarms, outputting maintenance optimization plans and life extension decision recommendations; The intelligent user feedback and adaptive optimization module is connected to the multi-level early warning and decision support module to optimize system performance based on user feedback.

[0018] Please see the attached Figure 2 , the multimodal sensor module includes: Wave radar unit, used to monitor wave parameters in the sea area around the jacket in real time; Anemometer unit, used to measure wind speed, wind direction and gust characteristics in the area where the jacket is located; Fiber Bragg grating sensor units for measuring strain, temperature, and pressure parameters; An acceleration sensor unit, used to measure the vibration acceleration of the jacket in three-dimensional space; The tilt sensor unit is used to monitor the overall and local tilt angles of the jacket.

[0019] Specifically, the sensors in the multimodal sensor module can collect key environmental and structural parameters of the jacket structure in real time, providing comprehensive multi-parameter time series data. These sensors are logically connected to form an integrated data acquisition network.

[0020] First, the wave radar unit is used to monitor wave parameters in the sea area surrounding the jacket, recording relevant information such as wave height and wave period. The anemometer unit detects wind speed, wind direction, and gust characteristics, providing important data on the impact of climatic conditions on the jacket. The fiber grating sensor unit is specifically designed to measure structural strain, temperature, and pressure, using fiber optic technology to improve measurement accuracy and reliability. The acceleration sensor unit is responsible for detecting the vibration acceleration of the jacket in three-dimensional space and analyzing the dynamic impact of environmental factors on the structure. In addition, the tilt sensor unit monitors the overall and local tilt angles of the jacket to assess structural stability, which is then integrated into a multimodal data set.

[0021] The mathematical model of time series data is analyzed using the following formula: ; Where, Indicates time A multimodal data set at each moment, For wave monitoring data, is the wind speed data, Measured data for fiber optic sensors, is the acceleration data, The inclination measurement data.

[0022] In summary, the multimodal sensor module has comprehensive sensing capabilities and good data integration, and can provide accurate real-time data support for the safety diagnosis of the jacket structure, ensuring the effectiveness, reliability and scientificity of the system.

[0023] Please see the attached Figure 3 , the time series data preprocessing module includes: The data block unit is used to process the original time series data into blocks according to the preset time window; Embedding layer generation unit, used to generate vector representations through custom embedding layers.

[0024] Specifically, the time series data preprocessing module includes a data segmentation unit and an embedding layer generation unit. The data segmentation unit is used to segment the original time series data into blocks according to preset time windows, so that the data in each time period can be analyzed independently. The preset time window can be set according to specific application requirements, such as every 10 seconds, every minute, or longer. The mathematical model of this process can be expressed as: ; Where, Indicates in A multimodal data set within a time window, For the The central moment of a time period is usually divided according to a predetermined sampling frequency, such as every 10 seconds, every minute, etc. is the total number of time windows, reflecting the data collection scope within the entire monitoring period.

[0025] Then, the data set is generated by the embedding layer Convert it into a corresponding vector representation to improve the computational efficiency and accuracy of subsequent sequence modeling. The relationship of this process can be expressed as: ; Where, Represents the vector representation after processing by the embedding layer, which is convenient for machine learning models to calculate and analyze. This vector is the original multimodal data converted into a form suitable for model input, usually with a higher dimension. Represents a vectorized function designed to efficiently embed multimodal data into a high-dimensional space that can be used in machine learning models.

[0026] In terms of physical structure, the data chunking unit and the embedding layer generation unit are connected via an internal data bus to ensure smooth data transmission. The data chunking unit first receives raw time series data from the multimodal sensor module. The processed data results are then fed into the embedding layer generation unit to form a vector representation.

[0027] The effectiveness of this module lies in its ability to integrate input data from various sensors, providing diverse information outputs. This provides rich contextual data for the subsequent large time-series language model module, thereby supporting accurate structural failure probability assessment and risk component location. This processing flow improves overall system performance, resulting in greater data validity and accuracy.

[0028] In summary, the implementation of the time series data preprocessing module not only ensures efficient data processing and multimodal fusion, but also lays a solid foundation for the subsequent intelligent diagnosis module.

[0029] Please see the attached Figure 4 , the time series large language model module includes: A modality alignment module to reprogram the vector representation of time series into a natural language compatible hybrid representation; The prompt prefix module is used to add a learnable task guidance prompt prefix before the input sequence to inject domain knowledge and task instructions; The safety diagnosis module is used to perform sequence modeling of mixed representations based on a pre-trained large language model, outputting structural failure probability and high-risk component location.

[0030] Specifically, the time-series large language model module mainly includes a modal alignment module, a prompt prefix module, and a security diagnosis module. These modules are connected through data channels to ensure smooth transmission and processing of data flows.

[0031] The modality alignment module receives the vector representations sent from the time series data preprocessing module. The function of this module is to reprogram the vector representations of these time series into a natural language compatible hybrid representation. , so that the subsequent model can be used effectively. The mathematical representation of this process is: ; Where, Is a mapping function, which represents the vector Convert to a hybrid representation suitable for time series modeling .

[0032] Next, the prompt prefix module adds a learnable task-guided prompt prefix to the input sequence , whose expression is: ; Where, Represents the mixed representation that is finally input to the safety diagnosis module, Indicates a splicing operation. Task guidance prompt prefix Aims to inject domain knowledge and task instructions to improve model accuracy and processing efficiency.

[0033] Finally, the security diagnosis module uses the pre-trained temporal large language model to represent the mixture Perform sequence modeling. The output of this module is the probability of structural failure and high-risk component positioning , which can be expressed as: ; Where, Indicates the probability of output structure failure through the model's reasoning process and high-risk component locations that require special attention .

[0034] During implementation, the modal alignment module, the prefix hint module, and the safety diagnosis module form a complete processing chain. Each module receives and processes data from the previous module through an interface. First, a modal alignment model is established. The model input is then enhanced using the prefix hint algorithm. Finally, the output is the structural failure probability and component location, forming a complete processing flow.

[0035] The design of this time-series large language model module ensures effective data conversion and fully utilized information, enabling accurate assessment of jacket structural conditions and rapid identification of high-risk areas. This comprehensive analysis capability provides crucial decision-making support for subsequent multi-level early warning and decision support modules, possessing significant practical application value. Leveraging its powerful sequence modeling capabilities, the module can process complex time series data, effectively improving overall system performance and meeting the requirements for high-reliability structural safety diagnostics.

[0036] Please see the attached Figure 5 , the multi-level warning and decision support module includes: A multi-parameter threshold determination unit is used to set dynamic thresholds based on historical data and industry standards to trigger three levels of alarms; Trend prediction unit, used to generate fatigue expansion trend and structural failure probability in the next 30 days; The maintenance optimization unit is used to output prioritized maintenance plans and cost estimation reports based on risk levels and economic constraints.

[0037] Specifically, this disclosure details the implementation of a multi-level warning and decision support module within a jacket structure safety diagnostic system based on a time-series large language model. This module is designed to provide efficient and accurate decision support for jacket safety operations and maintenance based on the output of the time-series large language model.

[0038] The multi-level early warning and decision support module mainly includes a multi-parameter threshold judgment unit, a trend prediction unit and a maintenance optimization unit. These units are connected through data paths to ensure smooth transmission and processing of information.

[0039] The multi-parameter threshold determination unit is responsible for setting dynamic thresholds based on historical data and industry standards in order to conduct real-time evaluation of various monitoring data. This unit outputs the structural failure probability from the time series large language model module. and high-risk component positioning Setting dynamic thresholds The calculation of the threshold can be expressed as: ; Where, and is a constant used to adjust the weight. Trigger three levels of alarms: normal, warning, and critical.

[0040] The trend prediction unit then analyzes the monitoring data and historical data to generate fatigue expansion trends for the next 30 days. and structural failure probability This unit uses a time series prediction model, and the core algorithm used is the ARIMA or LSTM model. Its output is the predicted fatigue expansion trend and structural failure probability, expressed as: ; ; Where, For the collected time series data, is the probability of structural failure, To predict the fatigue expansion trend, and They are trend prediction and failure probability calculation functions respectively.

[0041] The maintenance optimization unit receives the output from the trend prediction unit, combines the risk level with the economic constraints, and generates a prioritized maintenance plan and cost estimation report based on the output failure probability and high-risk components. The formulation can be expressed as: ; Where, and is the weight constant used to adjust maintenance schedule and cost, is the direct cost of repair. The optimization result is a priority repair list, which facilitates the implementation of effective maintenance operations to reduce the risk of possible structural failure.

[0042] During implementation, the multi-parameter threshold determination unit first sets dynamic thresholds based on pre-set historical data and industry standards. The trend prediction unit then analyzes the data and outputs a forecast of future risks. Finally, the maintenance optimization unit combines this information to develop targeted maintenance and cost optimization plans.

[0043] This multi-level early warning and decision support module is designed to provide real-time structural condition assessment and decision support. It accurately identifies high-risk jacket conditions, proactively issues warnings, and generates maintenance recommendations to reduce the risk of structural failure and enhance operational safety and economic efficiency. By integrating and sharing information across various units, the module effectively improves overall system performance, demonstrates excellent adaptability and practicality, and provides strong technical support for jacket safety management.

[0044] Please see the attached Figure 6 , the intelligent user feedback and adaptive optimization module includes: Intelligent interface unit, used for users to visually receive feedback information and perform operations; Dynamic adjustment mechanism unit, used to optimize algorithms and models based on user feedback to improve system performance.

[0045] Specifically, the intelligent user feedback and adaptive optimization module mainly includes an intelligent interface unit and a dynamic adjustment mechanism unit, which are interconnected through data channels to achieve effective information interaction.

[0046] The intelligent interface unit is responsible for providing a visual platform for users to interact with the system, where users can receive feedback and perform operations. This unit will display system status, alarm information, and recommended maintenance plans, and allow users to enter feedback data. User Feedback Including evaluation of system suggestions and feedback on alarm logic, expressed as: ; Where, Indicates the system status, Indicates the user's feedback action, Indicates the response to the alarm. is a function describing user feedback, For user feedback.

[0047] Dynamic adjustment mechanism unit based on user feedback Optimize the system's algorithms and models to improve the system's accuracy and responsiveness. Specifically, the optimization process can be achieved through the following steps: First, collect user feedback And conduct data analysis to extract key information and demand changes. Then, use the feedback information to train and adjust the existing model parameters And algorithm settings, expressed as: ; Where, It is an increment adjusted based on user feedback. are the existing model parameters.

[0048] Finally, the updated models and algorithms are put back into use, forming a closed-loop feedback mechanism through which the system can be continuously improved and optimized to maintain efficient operation.

[0049] In practice, the intelligent user feedback and adaptive optimization module operates as follows: First, user feedback data is collected through the intelligent interface unit. This collected user feedback data is then sent to the dynamic adjustment mechanism unit for analysis. Finally, based on the analysis results, the algorithm and model parameters are optimized to improve overall system performance.

[0050] This module's design effectively collects user experience and suggestions, making the system more adaptable. This continuous optimization process enhances the jacket structure safety diagnostic system's real-time monitoring and decision-making capabilities, ensuring efficiency and accuracy under various conditions and providing strong support for safety maintenance. It also enables efficient interaction between users and the system, ensuring the safety diagnostic system can operate efficiently and accurately in complex environments.

[0051] Please see the attached Figure 7 The jacket structure safety diagnosis method based on the time series large language model includes the following steps: S1, real-time acquisition of multi-parameter time series data through wave radar, anemometer, fiber Bragg grating sensor, acceleration sensor and tilt sensor; S2. Use the data block unit to block the original time series data according to the preset time window, and generate a vector representation through the embedding layer generation unit; S3, based on the text prototype library, converts the vector into a natural language compatible hybrid representation through the modality alignment module, and adds a learnable task guidance prompt prefix before the input sequence; S4. Use the safety diagnosis module to perform sequence modeling on the hybrid representation based on the pre-trained time-series large language model, and output the structural failure probability and high-risk area location; S5. Based on the failure probability and high-risk coordinates output by the time-series large language model, the multi-level early warning and decision support module generates a three-level alarm and outputs maintenance priorities and cost optimization plans; S6. Generate fatigue growth trends and structural failure probabilities for the next 30 days using the trend prediction unit, and provide prioritized maintenance plans and cost estimation reports through the maintenance optimization unit. S7. Collect user feedback through an intelligent interface and optimize algorithms and models based on dynamic adjustment mechanisms to improve system performance and user experience.

[0052] The above content is written based on the system content part. The technical details are the same and will not be repeated here.

[0053] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A jacket structure safety diagnosis system based on a time-series large language model is characterized by: include: Multimodal sensor module for real-time acquisition of multi-parameter time series data of the jacket structure; The time series data preprocessing module is connected to the multimodal sensor module and is used to process the original time series data collected by the sensor in blocks and generate vector representations; The time series large language model module is connected to the time series data preprocessing module to perform sequence modeling on the mixed representation and output the structural failure probability and high-risk component location; A multi-level early warning and decision support module, which is connected to the time series large language model module, is used to generate visual diagnostic reports and trigger hierarchical alarms, outputting maintenance optimization plans and life extension decision recommendations; The intelligent user feedback and adaptive optimization module is connected to the multi-level early warning and decision support module to optimize system performance based on user feedback.

2. The jacket structure safety diagnosis system based on the time series large language model according to claim 1 is characterized in that: The multimodal sensor module comprises: Wave radar unit, used to monitor wave parameters in the sea area around the jacket in real time; Anemometer unit, used to measure wind speed, wind direction and gust characteristics in the area where the jacket is located; Fiber Bragg grating sensor units for measuring strain, temperature, and pressure parameters; An acceleration sensor unit, used to measure the vibration acceleration of the jacket in three-dimensional space; The tilt sensor unit is used to monitor the overall and local tilt angles of the jacket.

3. The jacket structure safety diagnosis system based on the time series large language model according to claim 1 is characterized in that: The time series data preprocessing module includes: The data block unit is used to process the original time series data into blocks according to the preset time window; Embedding layer generation unit, used to generate vector representations through custom embedding layers.

4. The jacket structure safety diagnosis system based on the time series large language model according to claim 1 is characterized in that: The time series large language model module includes: A modality alignment module to reprogram the vector representation of time series into a natural language compatible hybrid representation; The prompt prefix module is used to add a learnable task guidance prompt prefix before the input sequence to inject domain knowledge and task instructions; The safety diagnosis module is used to perform sequence modeling of mixed representations based on a pre-trained large language model, outputting structural failure probability and high-risk component location.

5. The jacket structure safety diagnosis system based on the time series large language model according to claim 1 is characterized in that: The multi-level early warning and decision support module includes: A multi-parameter threshold determination unit is used to set dynamic thresholds based on historical data and industry standards to trigger three levels of alarms; Trend prediction unit, used to generate fatigue expansion trend and structural failure probability in the next 30 days; The maintenance optimization unit is used to output prioritized maintenance plans and cost estimation reports based on risk levels and economic constraints.

6. The jacket structure safety diagnosis system based on the time series large language model according to claim 1 is characterized in that: The intelligent user feedback and adaptive optimization module includes: Intelligent interface unit, used for users to visually receive feedback information and perform operations; Dynamic adjustment mechanism unit, used to optimize algorithms and models based on user feedback to improve system performance.

7. A jacket structure safety diagnosis method based on a time series large language model, according to a jacket structure safety diagnosis system based on a time series large language model according to any one of claims 1 to 6, characterized in that: The following steps are involved: S1, real-time acquisition of multi-parameter time series data through wave radar, anemometer, fiber Bragg grating sensor, acceleration sensor and tilt sensor; S2. Use the data block unit to block the original time series data according to the preset time window, and generate a vector representation through the embedding layer generation unit; S3, based on the text prototype library, converts the vector into a natural language compatible hybrid representation through the modality alignment module, and adds a learnable task guidance prompt prefix before the input sequence; S4. Use the safety diagnosis module to perform sequence modeling on the hybrid representation based on the pre-trained time-series large language model, and output the structural failure probability and high-risk area location; S5. Based on the failure probability and high-risk coordinates output by the time-series large language model, the multi-level early warning and decision support module generates a three-level alarm and outputs maintenance priorities and cost optimization plans; S6. Generate fatigue growth trends and structural failure probabilities for the next 30 days using the trend prediction unit, and provide prioritized maintenance plans and cost estimation reports through the maintenance optimization unit. S7. Collect user feedback through an intelligent interface and optimize algorithms and models based on dynamic adjustment mechanisms to improve system performance and user experience.

Images