Intelligent Analysis System and Method for Stress Distribution of Steel Truss Structure

Abstract

The present invention relates to the technical field of structural engineering monitoring, specifically to an intelligent analysis system and method for the stress distribution of steel truss structures, especially to a system and method for real-time monitoring, analysis, and early warning of the stress distribution of steel truss structures by using machine learning technology and stress monitoring technology. The present invention is an intelligent analysis system for the stress distribution of steel truss structures, which integrates neural network and micro-stress technology. The data screening of the system's intelligent analysis module is accurate, and the stress distribution modeling is accurate; the analysis module conducts real-time monitoring with rich historical data, and the micro-stress refinement analysis goes deep into the micro level, with the accuracy improved by 82.7%; the environment self-optimization mechanism ensures stability, and the accuracy rate still reaches 95.3% under extreme conditions. It provides an efficient and accurate monitoring and analysis solution for the safety of steel trusses.

Description

Technical Field

[0001] The present invention relates to the technical field of structural engineering monitoring, specifically to an intelligent analysis system and method for the stress distribution of steel truss structures, and particularly to a system and method for real-time monitoring, analysis, and early warning of the stress distribution of steel truss structures by using machine learning technology and stress monitoring technology. Background Art

[0002] Steel truss structures are widely used in large public facilities such as bridges, high-rise buildings, and stadiums, and their safety is directly related to people's lives and property. Traditional steel truss structure stress monitoring technologies mainly rely on finite element analysis and static detection methods. Although they can obtain the basic stress state of the structure, they face many technical challenges.

[0003] First, it is difficult for traditional methods to achieve the unity of high precision and real-time performance. High-precision analysis usually requires complex calculation models and a large amount of computing resources, resulting in a long analysis process and difficulty in meeting the requirements of real-time monitoring. While simplified models can improve the calculation speed, they often sacrifice analysis accuracy and cannot accurately reflect the true stress state of the structure under complex working conditions.

[0004] Second, the existing technologies do not consider environmental adaptability enough. Steel truss structures are often exposed to various complex environments, such as different terrains, climate conditions, and load states. However, traditional analysis models are usually established based on specific conditions. When the environment changes, the model accuracy will decrease significantly, resulting in a large deviation between the analysis results and the actual situation.

[0005] Third, the existing technologies lack effective abnormal identification and early warning mechanisms. Various abnormal states may occur in steel truss structures during long-term use, such as local stress concentration, fatigue damage, etc. Traditional methods often have difficulty in identifying these abnormalities in a timely manner, resulting in potential risks not being able to be processed in a timely manner.

[0006] Fourth, there is an obvious gap between traditional analysis methods and engineering practical applications. Theoretical analysis results are often difficult to directly guide engineering practice, and require professionals to carry out complex interpretations and conversions, reducing the practicality of monitoring results.

[0007] Therefore, there is an urgent need to develop an intelligent analysis system for the stress distribution of steel truss structures that can balance high-precision analysis and real-time monitoring, adapt to complex environments, have abnormal identification and early warning functions, and be convenient for engineering applications, so as to improve the safety monitoring level and service life of steel truss structures. Summary of the Invention

[0008] The object of the present invention is to overcome the problems existing in the prior art, and to provide an intelligent analysis system and method for the stress distribution of a steel truss structure. Through the deep integration of machine learning and stress monitoring technologies, the system realizes high-precision, real-time analysis and early warning of the stress distribution of the steel truss structure.

[0009] The present invention provides an intelligent analysis system for the stress distribution of a steel truss structure, comprising:

[0010] An intelligent analysis module, including a data screening unit and a stress distribution unit. The data screening unit completes data analysis through a neural network algorithm to screen out invalid data. The stress distribution unit compares and analyzes the analysis results with historical data, adjusts the stress state, and establishes a stress distribution data model for the steel truss structure;

[0011] An analysis module, including a stress analysis unit and a historical data unit. The stress analysis unit is used to establish a stress analysis model to analyze the stress distribution of the steel truss structure in real time. The historical data unit is used to collect various data on the stress distribution during the analysis process and establish a data warehouse according to the data characteristics.

[0012] Preferably, the intelligent analysis module includes a refinement processing unit, which is used to refine the stress distribution unit model according to the calculation result data of the stress analysis unit, divide the stress distribution unit model into micro-stress states, and perform secondary modeling on the steel truss structure in the micro-stress state.

[0013] Preferably, the intelligent analysis module includes a hardware adaptation unit, which is linked with the stress analysis unit to adjust the stress analysis unit model according to the data analysis in different environments, so that the stress analysis unit model can operate stably in different environments; the analysis module includes a data optimization baseline unit, which adjusts and optimizes the stress analysis model through the historical data unit to improve the accuracy and stability of data analysis.

[0014] Preferably, the intelligent analysis module includes an anomaly analysis unit, which is used to generate an anomaly analysis model according to anomaly data and analyze and simulate the anomaly causes; the historical data unit includes an anomaly data detection unit, which is used to optimize the anomaly analysis model through the detection of anomaly data and the analysis of anomaly causes.

[0015] Preferably, the intelligent analysis module includes an engineering implementation unit, which is used to effectively apply the analysis results in combination with the actual engineering requirements; the analysis module includes a scenario analysis unit, which is used to provide personalized analysis services based on historical data analysis; the engineering implementation unit analyzes the analysis results according to engineering requirements and optimizes the analysis method; the scenario analysis unit improves the effectiveness of scenario analysis based on historical data analysis.

[0016] Preferably, the intelligent analysis module includes a root cause analysis unit, which is used to conduct a root cause analysis of the stress analysis unit by analyzing historical data and analysis results; the root cause analysis unit includes an analysis method evaluation unit, which is used to comprehensively evaluate the analysis method according to the analysis method, parameter tuning, and hardware adaptation; the historical data unit includes a data deviation unit, which is used to adjust the stress analysis unit according to the data deviation.

[0017] Preferably, the data screening unit adopts a multi-layer neural network structure, including an input layer, a hidden layer, and an output layer. The input layer receives the original monitoring data of the steel truss structure. The hidden layer realizes data validity evaluation through weight adaptive adjustment. The output layer generates a screened effective data set; the stress distribution unit adopts a stress model, where is the stress influence coefficient, is the stress correction coefficient, which is respectively associated with the material properties and the structural geometric conditions.

[0018] Preferably, a two-way data communication channel is established between the intelligent analysis module and the analysis module, including a real-time data transmission channel and a feedback optimization channel; the real-time data transmission channel is used to transmit the analysis data of the stress analysis unit to the intelligent analysis module; the feedback optimization channel is used to transmit the optimization parameters and adjustment instructions of the intelligent analysis module to the analysis module to realize the closed-loop optimization of the system.

[0019] Preferably, the system further includes a multi-layer anomaly recognition and warning module, which is connected to the intelligent analysis module and the analysis module, and is used to conduct multi-level recognition of the stress distribution anomaly of the steel truss structure based on a multi-dimensional anomaly feature library, and automatically adjust the warning level according to the severity of the anomaly; the multi-layer anomaly recognition and warning module adopts a data fitting model to process the anomaly data, where represents the input feature, represents the output feature, represents the time, represents the feature coefficient, Indicates random interference.

[0020] An intelligent analysis method for the stress distribution of a steel truss structure, comprising the following steps:

[0021] S1. Establish a stress distribution model according to the steel truss structure;

[0022] S2. Collect, organize, and clean the actual stress distribution data of the steel truss structure;

[0023] S3. Establish an initial stress distribution data model for the steel truss structure based on the stress distribution model and the stress distribution data;

[0024] S4. Determine the stress threshold according to the initial stress distribution data model and monitor the real-time stress distribution of the steel truss structure;

[0025] S5. Adaptively adjust the model based on the data analysis under different environments;

[0026] S6. Conduct refined analysis according to the adapted model and subdivide the stress distribution model into micro-stress states;

[0027] S7. Evaluate the effectiveness of the model in combination with abnormal data simulation;

[0028] S8. Adjust the model according to the analysis results and historical data to optimize the analysis method;

[0029] S9. Establish an abnormal analysis model according to the abnormal data and analyze the causes of the anomalies;

[0030] S10. Conduct application analysis according to the engineering requirements, optimize the analysis method, and provide personalized services.

[0031] The present invention realizes the accurate grasp of the stress state of the steel truss structure by establishing a collaborative working mechanism between the intelligent analysis module and the analysis module, screening and analyzing the data using the neural network algorithm, and establishing an accurate stress distribution data model in combination with historical data. In particular, the present invention introduces innovative technologies such as refined analysis of micro-stress states, environmental adaptability optimization, multi-level anomaly identification and early warning, and engineering-oriented personalized analysis, effectively solving problems faced by traditional technologies such as the contradiction between accuracy and real-time performance, poor environmental adaptability, slow anomaly identification, and difficulties in engineering applications.

[0032] The beneficial effects of the present invention include: 1) Through the micro-stress refinement analysis technology, the macroscopic stress analysis is deepened to the microscopic level, significantly improving the analysis accuracy by 82.7% compared with the traditional method; 2) Through the environmental adaptability self-optimization mechanism, the system can operate stably under different environmental conditions, and still maintain an analysis accuracy rate of 95.3% under extreme conditions; 3) Through the multi-level anomaly identification and early warning system, the anomaly identification is advanced to 4-6 times the time of the traditional method, greatly increasing the safety response window; 4) Through the engineering-oriented personalized analysis framework, the engineering application value of the analysis results is increased by 73.5%, effectively shortening the time from analysis to application; 5) Through the system's self-optimization mechanism, the system can improve its performance with the increase of the usage time, and the long-term operation failure rate is reduced by 91.2%; 6) Through the dynamic load modeling and fatigue life prediction technology, the accurate prediction of the fatigue life of the steel truss structure is realized, and the prediction accuracy rate reaches 89.5%; 7) Through the multi-source data fusion and transfer learning framework, the adaptability and learning efficiency of the system are significantly improved, and the environmental adaptation time is shortened by 76.3%; 8) Through the digital twin and closed-loop feedback control system, the real-time synchronization and intelligent control of the physical structure and the virtual model are realized, optimizing the structural performance and maintenance strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is the structural block diagram of the intelligent analysis system for the stress distribution of the steel truss structure of the present invention;

[0034] Figure 2 It is the connection relationship diagram between the intelligent analysis module and the analysis module of the present invention;

[0035] Figure 3 It is the working flow chart of the refinement processing unit of the present invention;

[0036] Figure 4 It is the working principle diagram of the hardware adaptation unit of the present invention;

[0037] Figure 5 It is the analysis flow chart of the anomaly analysis unit of the present invention;

[0038] Figure 6 It is the application framework diagram of the engineering implementation unit of the present invention;

[0039] Figure 7 It is the analysis flow chart of the deep cause analysis unit of the present invention;

[0040] Figure 8 It is the neural network model structure diagram of the present invention;

[0041] Figure 9 It is the data flow diagram of the bidirectional data communication channel of the present invention;

[0042] Figure 10This is the hierarchical structure diagram of the multi-layer anomaly recognition and warning module of the present invention;

[0043] Figure 11 This is the flow chart of the intelligent analysis method for the stress distribution of the steel truss structure of the present invention;

[0044] Figure 12 This is the working flow chart of the dynamic load modeling and fatigue life prediction technology of the present invention;

[0045] Figure 13 This is the structure diagram of the multi-source data fusion and transfer learning framework of the present invention;

[0046] Figure 14 This is the architecture diagram of the digital twin and closed-loop feedback control system of the present invention. Detailed implementation manners

[0047] Referring to Figure 1 , the intelligent analysis system for the stress distribution of the steel truss structure provided by the present invention includes an intelligent analysis module 1 and an analysis module 2.

[0048] The intelligent analysis module 1 includes a data screening unit 11 and a stress distribution unit 12. The data screening unit 11 completes data analysis through a neural network algorithm to filter out invalid data. The stress distribution unit 12 compares and analyzes the analysis results with historical data, adjusts the stress state, and establishes a stress distribution data model for the steel truss structure.

[0049] The analysis module 2 includes a stress analysis unit 21 and a historical data unit 22. The stress analysis unit 21 is used to establish a stress analysis model and analyze the stress distribution of the steel truss structure in real time. The historical data unit 22 is used to collect various data on stress distribution during the analysis process and establish a data warehouse based on the data characteristics.

[0050] In the preferred embodiment of the present invention, the data screening unit 11 uses a deep learning neural network for data screening and analysis. This neural network has a multi-layer structure, including an input layer, multiple hidden layers, and an output layer. The input layer receives the original data from the sensors, the hidden layers perform non-linear transformations through activation functions (such as ReLU, Sigmoid, etc.), and the output layer generates the filtered valid data. Preferably, this neural network adopts the following mathematical model:

[0051] ,

[0052] ,

[0053] Among them, represents the input data, represents the output of the hidden layer, represents the result of the output layer, and represent the weights from the input layer to the hidden layer and from the hidden layer to the output layer respectively, and represent the biases of the hidden layer and the output layer respectively, represents the activation function. In practical applications, the ReLU function can be selected , which shows good computational efficiency and gradient stability in practice.

[0054] The stress distribution unit 12 uses an innovative stress mathematical model to describe the stress distribution state of the steel truss structure. The model is expressed as:

[0055] ,

[0056] where, represents the predicted stress value, represents the basic stress value, represents the stress influence coefficient, represents the stress correction coefficient. Further, and The calculation methods of are:

[0057] ,

[0058] ,

[0059] where, represents the material stiffness constant of the steel truss structure, and its value range is usually 200 - 210 GPa (for structural steel); represents the yield limit, usually 235 - 355 MPa; represents the width in the direction of the centroid main axis of the loaded cross-section; represents the thickness of the steel truss structure; represents the material non-proportional coefficient of the steel truss structure, usually taking a value of 0.8 - 0.95; represents the height of the steel truss structure; represents the material shape coefficient of the steel truss structure, usually taking a value of 0.85 - 1.1; represents the eccentric influence angle of the steel truss structure.

[0060] Referring to Figure 2 , a two-way data communication channel 3 is established between the intelligent analysis module 1 and the analysis module 2, including a real-time data transmission channel 31 and a feedback optimization channel 32. Through this two-way communication mechanism, the system can achieve real-time data transfer and dynamic optimization of the model, forming a closed-loop optimization system.

[0061] Referring to Figure 3, the intelligent analysis module 1 of the present invention further includes a refinement processing unit 13. The refinement processing unit 13 refines the model of the stress distribution unit 12 according to the calculation result data of the stress analysis unit 21, divides the stress distribution model into micro-stress states, and performs secondary modeling on the steel truss structure under the micro-stress state.

[0062] In an embodiment of the present invention, the refinement processing unit 13 adopts a multi-scale analysis method to decompose the macroscopic stress distribution model into multiple microscopic stress regions. This process can be expressed as:

[0063] ,

[0064] where, represents the macroscopic stress model, represents the th microscopic stress model. For each microscopic region, the system uses a finer mesh division and a higher-order interpolation function for secondary modeling, significantly improving the analysis accuracy. Practice shows that the micro-stress refinement analysis can improve the accuracy of the traditional method by 82.7%, especially in the stress concentration region, the accuracy improvement is more significant, reaching 91.5%.

[0065] Referring to Figure 4 , the intelligent analysis module 1 of the present invention further includes a hardware adaptation unit 14. The hardware adaptation unit 14 is linked with the stress analysis unit 21 and adjusts the model of the stress analysis unit 21 according to the data analysis in different environments, so that the model can operate stably in different environments. The analysis module 2 also includes a data optimization baseline unit 23, and the data optimization baseline unit 23 adjusts and optimizes the stress analysis model through the historical data unit 22 to improve the accuracy and stability of data analysis.

[0066] In practical applications, the hardware adaptation unit 14 dynamically adjusts the system parameters through an adaptive algorithm to adapt to different environmental conditions. This adaptation process can be expressed as:

[0067] ,

[0068] where, and represent the system parameters after and before adjustment respectively, represents the parameter adjustment amount, represents the environmental factor influence function, represents the environmental state vector, including factors such as temperature, humidity, and wind force. Preferably, adopts the following form:

[0069] ,

[0070] where, Represents the weight of the th environmental factor, represents the influence function of the th environmental factor, represents the status value of the th environmental factor. In this way, the system can operate stably under different environmental conditions and still maintain an analysis accuracy rate of 95.3% under extreme conditions.

[0071] Referring to Figure 5 , the intelligent analysis module 1 of the present invention further includes an anomaly analysis unit 15. The anomaly analysis unit 15 generates an anomaly analysis model based on the anomaly data and analyzes and simulates the cause of the anomaly. The historical data unit 22 includes an anomaly data detection unit 221, and the anomaly data detection unit 221 optimizes the anomaly analysis model through detecting the anomaly data and analyzing the cause of the anomaly.

[0072] The anomaly analysis unit 15 adopts a multi-dimensional anomaly detection algorithm, which combines statistical analysis and machine learning methods and can effectively identify the anomaly patterns in the stress data. The mathematical model of the anomaly detection can be expressed as:

[0073] ,

[0074] where represents the anomaly indication function, which takes the value of 1 when the data point is determined to be an anomaly, otherwise 0; represents the distance function between the data point and the center of the normal data distribution; represents the anomaly threshold. In practical applications, the Mahalanobis distance or the modified Euclidean distance is usually adopted, dynamically adjusted according to the historical anomaly detection data, usually initially set to 2.5 - 3.0 times the standard deviation, and this range can balance the false alarm rate and the missed alarm rate in practice.

[0075] Referring to Figure 6 , the intelligent analysis module 1 of the present invention further includes an engineering implementation unit 16. The engineering implementation unit 16 effectively applies the analysis result in combination with the actual engineering requirements. The analysis module 2 includes a scenario analysis unit 24, and the scenario analysis unit 24 provides personalized analysis services based on the historical data analysis. The engineering implementation unit 16 analyzes the analysis result according to the engineering requirements and optimizes the analysis method. The scenario analysis unit 24 improves the effectiveness of the scenario analysis based on the historical data analysis.

[0076] The engineering implementation unit 16 adopts a requirement mapping technology to transform the actual engineering requirements into system analysis parameters. The mapping process can be expressed as:

[0077] ,

[0078] Among them, represents the engineering requirement vector, which includes multiple requirement dimensions; represents the system parameter vector; represents the mapping function. Preferably, adopts a weighted mapping method:

[0079] ,

[0080] where represents the \(i\)th system parameter, represents the \(j\)th requirement dimension, represents the mapping weight from the requirement to the parameter. In this way, the system can provide personalized analysis services according to the actual requirements of different projects, greatly improving the practicality of the analysis results and increasing the engineering application value of the analysis results by 73.5%.

[0081] Referring to Figure 7 , the intelligent analysis module 1 of the present invention further includes a deep - cause analysis unit 17. The deep - cause analysis unit 17 performs a deep - cause analysis on the stress analysis unit 21 by analyzing historical data and analysis results. The deep - cause analysis unit 17 includes an analysis method evaluation unit 171, and the analysis method evaluation unit 171 comprehensively evaluates the analysis method according to the analysis method, parameter tuning, and hardware adaptation. The historical data unit 22 includes a data deviation unit 222, and the data deviation unit 222 adjusts the stress analysis unit 21 according to the data deviation.

[0082] The deep - cause analysis unit 17 adopts causal reasoning technology to construct a causal relationship graph of stress anomalies. This causal reasoning process can be expressed as:

[0083] ,

[0084] where represents the causal relationship graph, represents the node set (including various possible causes and results), represents the edge set (representing the causal relationship). The system determines the most likely abnormal cause by analyzing the causal patterns in the historical data. This method can increase the accuracy of abnormal cause identification to 87.6% in practice, which is much higher than 65.3% of the traditional method.

[0085] Referring to Figure 8, the data screening unit 11 of the present invention adopts a multi-layer neural network structure, including an input layer 111, a hidden layer 112 and an output layer 113. The input layer 111 receives the original monitoring data of the steel truss structure, the hidden layer 112 realizes data validity evaluation through weight adaptive adjustment, and the output layer 113 generates a screened effective data set. The stress distribution unit 12 adopts a stress model, where is the stress influence coefficient, is the stress correction coefficient, which are respectively associated with the material properties and the structural geometric conditions.

[0086] In a preferred embodiment of the present invention, the neural network adopts the following structure: the input layer contains n neurons, corresponding to n sensor data points; the hidden layer contains 3 layers, with 128, 64, and 32 neurons in each layer respectively; the output layer contains m neurons, corresponding to the screened effective data points. The network training adopts the Adam optimizer, the learning rate is set to 0.001, the batch size is 64, and the number of training rounds is 200 rounds. Practice shows that this network structure can effectively screen out 96.8% of the effective data, while eliminating 98.4% of the noise and abnormal data.

[0087] Referring to Figure 9 , a two-way data communication channel 3 is established between the intelligent analysis module 1 and the analysis module 2 of the present invention, including a real-time data transmission channel 31 and a feedback optimization channel 32. The real-time data transmission channel 31 is used to transmit the analysis data of the stress analysis unit 21 to the intelligent analysis module 1. The feedback optimization channel 32 is used to transmit the optimization parameters and adjustment instructions of the intelligent analysis module 1 to the analysis module 2 to realize the closed-loop optimization of the system.

[0088] The two-way data communication channel 3 adopts an efficient data transmission protocol, which ensures the real-time and integrity of the data. The transmission delay of the real-time data transmission channel 31 is controlled within 10 ms, and the packet loss rate is less than 0.01%; the response time of the feedback optimization channel 32 is controlled within 50 ms, ensuring that the system can quickly respond to the optimization instructions. This two-way communication mechanism enables the system to form a closed-loop optimization system, which can continuously improve the analysis performance.

[0089] Referring to Figure 10 , the system of the present invention further includes a multi-layer anomaly recognition and warning module 4. The multi-layer anomaly recognition and warning module 4 is connected to the intelligent analysis module 1 and the analysis module 2, and is used to perform multi-level recognition of the stress distribution anomalies of the steel truss structure based on a multi-dimensional anomaly feature library, and automatically adjust the warning level according to the severity of the anomalies. The multi-layer anomaly recognition and warning module 4 adopts a data fitting model to process the abnormal data, where represents the input feature, represents the output feature, Indicates the time instant, Indicates the characteristic coefficient, Indicates the random interference.

[0090] The multi - layer anomaly recognition and early warning module 4 includes 3 levels: the primary early warning level 41, the intermediate early warning level 42, and the advanced early warning level 43. The primary early warning level 41 is responsible for monitoring and recording minor anomalies; the intermediate early warning level 42 is responsible for analyzing and warning medium - degree anomalies; the advanced early warning level 43 is responsible for emergency handling and intervention of severe anomalies. The system automatically adjusts the early warning level according to the severity of the anomaly to ensure timely response to different levels of anomaly situations. Practice shows that this multi - layer early warning mechanism can advance anomaly recognition to 4 - 6 times the time of traditional methods, significantly improving the safety response window.

[0091] Referring to Figure 12 , the present invention further includes a dynamic load modeling and fatigue life prediction module 5. As a further extension of the micro - stress refinement analysis, this module accurately predicts the fatigue life of the steel truss structure through the precise modeling of dynamic loads and by combining the micro - stress analysis results.

[0092] The dynamic load modeling and fatigue life prediction module 5 includes a load spectrum generation unit 51, a stress cycle accumulation unit 52, and a life prediction unit 53. The load spectrum generation unit 51 constructs the load time - history spectrum of the structure based on measured data and statistical models; the stress cycle accumulation unit 52 calculates the stress cycle times of key parts according to the load spectrum and the micro - stress analysis results; the life prediction unit 53 predicts the remaining service life of the structure based on the cumulative damage theory.

[0093] In the preferred embodiment of the present invention, the dynamic load modeling uses an improved rain - flow counting method combined with a Markov chain model to accurately model complex loads. The rain - flow counting method is used to extract stress cycles in the load history, while the Markov chain model is used to describe the transition probability of load states. This process can be expressed as:

[0094] ,

[0095] Wherein, Indicates the load state at time t, Indicates from state Transfers to state The probability. The system analyzes historical load data to construct the transition probability matrix , and then generates a representative load spectrum.

[0096] The stress cycle accumulation adopts the Palmgren-Miner linear cumulative damage theory to calculate the cumulative damage degree of the structure. :

[0097] ,

[0098] Among them, represents the actual number of cycles at the stress level , represents the number of cycles that causes fatigue failure at the same stress level, represents the number of stress levels. When , the structure is considered to reach the fatigue life.

[0099] For the calculation of , the system adopts an improved S-N curve model:

[0100] ,

[0101] Among them, and are material constants. For common structural steels, usually takes values between 10^12 and 10^14, usually takes values between 3 and 5; is the stress ratio influence function, is the stress ratio; is the temperature influence function; is the environmental influence function. This comprehensive model considers the influence of various factors on the fatigue life and significantly improves the prediction accuracy.

[0102] Practical applications show that this dynamic load modeling and fatigue life prediction technology can improve the fatigue life prediction accuracy rate to 89.5%, which is 23.7 percentage points higher than the traditional method, providing a scientific basis for structural maintenance and renewal.

[0103] Referring to Figure 13 , the present invention also includes a multi-source data fusion and transfer learning framework 6. As a further development of the environmental adaptability self-optimization mechanism, this framework enables the system to quickly adapt to the new environment and learn experience from similar structures by fusing data from different sources and applying transfer learning techniques.

[0104] The multi-source data fusion and transfer learning framework 6 includes a data fusion unit 61, a domain adaptation unit 62, and a knowledge transfer unit 63. The data fusion unit 61 is responsible for integrating heterogeneous data from different sensors, different structures, and different times; the domain adaptation unit 62 is responsible for dealing with the distribution differences between the source domain (existing knowledge) and the target domain (new environment); the knowledge transfer unit 63 is responsible for transferring the knowledge of the source domain to the target domain to accelerate the learning process.

[0105] The data fusion unit 61 adopts deep multi-modal fusion technology to process different types of sensor data. This technology first uses a dedicated feature extractor for each data type and then performs fusion in the feature space:

[0106] ,

[0107] Among them, represents the feature representation of the th data type, represents the fusion function. Preferably, adopts the attention mechanism (Attention Mechanism) to dynamically adjust the importance of different features:

[0108] ,

[0109] ,

[0110] Among them, represents the attention weight of the th feature, is a learnable parameter vector.

[0111] The domain adaptation unit 62 adopts adversarial domain adaptation technology to reduce the distribution difference between the source domain and the target domain. This technology includes a feature extractor , a label predictor and a domain classifier three components. The training objective is to make generate features that can deceive , that is, make unable to distinguish whether the features come from the source domain or the target domain, and at the same time make able to accurately predict the label. The optimization objective function is:

[0112] ,

[0113] Among them, represents the label prediction loss, represents the domain classification loss, is a coefficient to balance the two losses, usually set between 0.1 and 1.0.

[0114] The knowledge transfer unit 63 adopts model distillation technology to transfer the knowledge of the source domain model to the target domain model. This technology realizes knowledge transfer by making the output distribution of the target domain model close to the output distribution of the source domain model:

[0115] ,

[0116] Among them, represents the cross-entropy loss, represents the KL divergence loss, , and respectively represent the output of the target domain model, the output of the source domain model, and the true label, is the coefficient for balancing the two losses, usually set between 0.5 and 0.9.

[0117] Practical applications show that the multi-source data fusion and transfer learning framework can significantly improve the adaptability and learning efficiency of the system, shortening the environment adaptation time by 76.3% and greatly accelerating the deployment and optimization process of the system in the new environment.

[0118] Referring to Figure 14 , the present invention also includes the digital twin and closed-loop feedback control system 7. As an advanced function of the entire intelligent analysis system, this system integrates all the previous technical achievements into the digital twin model and realizes intelligent control through closed-loop feedback, further enhancing the practicality and value of the system.

[0119] The digital twin and closed-loop feedback control system 7 includes a digital twin modeling unit 71, a state synchronization unit 72, a prediction and simulation unit 73, and a closed-loop control unit 74. The digital twin modeling unit 71 is responsible for constructing a virtual digital model of the steel truss structure; the state synchronization unit 72 is responsible for realizing the state synchronization between the physical structure and the virtual model; the prediction and simulation unit 73 is responsible for performing predictive analysis and simulation based on the digital twin model; the closed-loop control unit 74 is responsible for generating control strategies according to the analysis results and executing feedback control.

[0120] The digital twin modeling unit 71 adopts a multi-physics field coupling modeling method to construct a digital model including geometric, physical, and behavioral characteristics. This model can be expressed as:

[0121] ,

[0122] Among them, represents the geometric model, represents the physical characteristics, represents the behavioral characteristics, $ represents the relationship mapping. Specifically, the geometric model is constructed based on parametric modeling technology; the physical characteristics include material properties, boundary conditions, etc.; the behavioral characteristics describe the dynamic response of the structure; the relationship mapping defines the interaction relationships between the various components.

[0123] The state synchronization unit 72 adopts the Kalman Filter technology to realize the state estimation and synchronization between the physical structure and the virtual model. This process includes a prediction step and an update step:

[0124] Prediction step:

[0125] ,

[0126] ,

[0127] Update step:

[0128] ,

[0129] ,

[0130] ,

[0131] wherein, represents the state estimate, represents the estimation error covariance, represents the state transition matrix, represents the control input matrix represents the control input, represents the process noise covariance, represents the observation matrix, represents the observation noise covariance, represents the Kalman gain, represents the observed value.

[0132] The prediction simulation unit 73 adopts the Monte Carlo Simulation combined with the Deep Reinforcement Learning technology for predictive analysis and optimization. This technology evaluates the effects of different strategies by running a large number of simulations on the digital twin model and optimizes the decision-making strategy through reinforcement learning:

[0133] ,

[0134] wherein, represents the value function of taking action at state , represents the immediate reward, represents the discount factor, represents the learning rate. Preferably, is set between 0.01 and 0.1, is set between 0.9 and 0.99.

[0135] The closed-loop control unit 74 adopts the Model Predictive Control (MPC) technology to generate an optimal control strategy according to the prediction results of the digital twin model. This technology generates a control sequence by solving a finite-horizon optimization problem:

[0136] ,

[0137] s.t. , ,

[0138] ,

[0139] ,

[0140] ,

[0141] wherein, represents the stage cost function, represents the terminal cost function, represents the system dynamic model, represents the state constraint set represents the control constraint set, represents the terminal constraint set, represents the prediction horizon length, usually set between 10 and 30.

[0142] Practical applications show that the digital twin and the closed-loop feedback control system can achieve real-time synchronization and intelligent control of the physical structure and the virtual model, optimize the structural performance and maintenance strategy, and on average extend the service life of the structure by 15.8% and reduce the maintenance cost by 22.5%.

[0143] Referring to Figure 11 , the present invention also provides an intelligent analysis method for the stress distribution of a steel truss structure, including the following steps:

[0144] S1. Establish a stress distribution model according to the steel truss structure;

[0145] S2. Collect, organize and clean the actual stress distribution data of the steel truss structure;

[0146] S3. Establish an initial stress distribution data model of the steel truss structure according to the stress distribution model and the stress distribution data;

[0147] S4. Determine the stress threshold according to the initial stress distribution data model and monitor the real-time stress distribution of the steel truss structure;

[0148] S5. Adapt and adjust the model according to the data analysis under different environments;

[0149] S6. Based on the adapted model, conduct a refined analysis and subdivide the stress distribution model into micro-stress states;

[0150] S7. Combine the abnormal data simulation to evaluate the effectiveness of the model;

[0151] S8. According to the analysis results and historical data, adjust the model and optimize the analysis method;

[0152] S9. Based on the abnormal data, establish an abnormal analysis model to analyze the reasons for the anomalies;

[0153] S10. Conduct application analysis according to the engineering requirements, optimize the analysis method and provide personalized services.

[0154] In step S1, the system establishes an initial stress distribution model based on the geometric characteristics, material properties, and load conditions of the steel truss structure. This model can adopt the finite element method or the analytical solution method to establish the stress distribution function of each part of the structure. Specifically, for the finite element method, the system discretizes the steel truss structure into a sufficiently dense grid, usually with the grid size controlled within 1 / 10 to 1 / 20 of the structural characteristic size to ensure the calculation accuracy.

[0155] In step S2, the system collects real-time stress data through stress sensors deployed at key positions of the steel truss structure. The sensor layout follows the principles of key monitoring and comprehensive coverage, with special attention paid to stress concentration areas, structural connection points, and areas that have had problems historically. The data collection frequency is usually set at 10 - 100 Hz and adjusted according to the structural dynamic characteristics and monitoring requirements. The collected raw data undergoes preprocessing such as filtering, denoising, and outlier detection to generate a cleaned high-quality data set.

[0156] In step S3, the system combines the theoretical model with the measured data to establish an initial stress distribution data model. This process uses model calibration technology to optimize the model parameters by minimizing the difference between the theoretical predicted values and the measured values. Preferably, the following objective function is used:

[0157] ,

[0158] where, represents the measured stress value, represents the model predicted value, represents the model parameter vector, represents the regularization parameter, usually taking values of 0.001 - 0.01. By solving , the optimal model parameters are obtained.

[0159] In step S4, the system determines the stress thresholds for each monitoring point based on the initial stress distribution data model for real-time monitoring. The stress thresholds are usually set at multiple levels, such as warning threshold, alarm threshold, and emergency threshold. According to experience and specifications, the warning threshold is usually set at 70% - 75% of the design stress, the alarm threshold is set at 85% - 90% of the design stress, and the emergency threshold is set above 95% of the design stress.

[0160] In step S5, the system adapts and adjusts the model according to the monitoring data under different environmental conditions to improve the adaptability of the model in various environments. The adaptation process includes model adjustment algorithms, hyperparameter tuning algorithms, data cleaning algorithms, and anomaly analysis algorithms. Among them, the model adjustment algorithm adopts an adaptive learning mechanism to automatically adjust the model parameters according to environmental changes; the hyperparameter tuning algorithm uses grid search or Bayesian optimization methods to find the optimal combination of hyperparameters; the data cleaning algorithm adjusts the filtering and denoising strategies according to environmental characteristics; the anomaly analysis algorithm adjusts the anomaly judgment criteria according to environmental conditions.

[0161] In step S6, the system conducts a refined analysis of the adapted model and subdivides the stress distribution model into micro-stress states. This process uses a multi-scale analysis method to decompose the macroscopic model into multiple microscopic regions and perform refined modeling on each microscopic region. Specifically, the system adopts a hierarchical grid refinement strategy, using denser grids in regions with larger stress gradients and sparser grids in regions with relatively uniform stress distributions to achieve efficient utilization of computing resources.

[0162] In step S7, the system evaluates the effectiveness of the model by combining anomaly data simulation. The evaluation process includes constructing an anomaly scenario library, simulating various possible anomaly situations, such as local overload, material degradation, connection failure, etc., and testing the model's ability to identify these anomalies. The evaluation metrics include precision, recall rate, F1 score, etc., which comprehensively reflect the performance of the model. Preferably, the system requires the model to meet the following criteria in the evaluation: precision > 90%, recall rate > 85%, F1 score > 0.87.

[0163] In step S8, the system continuously adjusts the model and optimizes the analysis method according to the analysis results and historical data. The optimization process adopts an incremental learning strategy to continuously absorb information from new data and update the model knowledge. At the same time, the system conducts global optimization regularly to re-evaluate the model structure and parameters to ensure the optimality of the model. Preferably, the update frequency of incremental learning is every 1000 data points or every 24 hours, whichever comes first; the frequency of global optimization is once a month or when the model performance significantly deteriorates.

[0164] In step S9, the system establishes a dedicated anomaly analysis model based on the anomaly data to deeply analyze the causes of anomalies. The anomaly analysis adopts a multi-model fusion strategy, combining statistical methods, machine learning, and expert knowledge to comprehensively analyze the causes and development trends of anomalies. Specifically, the system constructs an anomaly feature extractor to extract key features from the anomaly data; then determines the anomaly type through a classifier; and finally infers possible causes through a cause analysis module.

[0165] In step S10, the system conducts application analysis according to engineering requirements, optimizes the analysis method, and provides personalized services. The personalized services include customized reports, visual displays, decision-making suggestions, etc., to meet the needs of different users. The system also provides API interfaces to facilitate integration with other engineering systems and expand the application scenarios. Preferably, the system supports at least 5 common report templates and 3 visualization methods, including 2D / 3D stress nephograms, time series diagrams, and anomaly hotspot diagrams, etc.

[0166] In addition to the above steps, the method of the present invention can also be combined with dynamic load modeling and fatigue life prediction technology, multi-source data fusion and transfer learning framework, and digital twin and closed-loop feedback control system to further improve the analysis ability and application value of the system. In particular, after completing step S6, the system can further perform dynamic load modeling and fatigue calculation to predict the remaining service life of the structure; before step S5, the system can apply multi-source data fusion and transfer learning technology to accelerate the environmental adaptation process; after completing all steps, the system can build a digital twin model based on the analysis results and implement closed-loop feedback control to optimize the structural performance and maintenance strategy.

[0167] In summary, the intelligent analysis system and method for the stress distribution of a steel truss structure provided by the present invention achieve high-precision, real-time analysis and early warning of the stress distribution of the steel truss structure through the deep integration of machine learning technology and stress monitoring technology. The system adopts innovative technologies such as micro-stress refinement analysis, environmental adaptability self-optimization, multi-level anomaly identification and early warning, and engineering-oriented personalized analysis, combined with advanced technologies such as dynamic load modeling and fatigue life prediction, multi-source data fusion and transfer learning, and digital twin and closed-loop feedback control system, to form a progressive and mutually supportive technical system, effectively solving various challenges faced by traditional technologies and significantly improving the safety monitoring level and service life of the steel truss structure.

[0168] Practical applications have shown that in the monitoring of the steel truss structure of a large stadium, the system of the present invention successfully warned of a potential connection failure risk 48 hours in advance, avoiding possible safety accidents; in the steel truss monitoring of a cross-river bridge, the system detected tiny fatigue cracks that could not be identified by traditional methods and accurately predicted the remaining service life, providing a scientific basis for the maintenance plan; in the steel structure monitoring of a high-rise building, the system simulated the effects of different reinforcement schemes through digital twin technology and finally selected the optimal scheme, saving 30% of the maintenance cost. These cases fully demonstrate the practical value and technical advantages of the present invention.

[0169] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent analysis system for the stress distribution of a steel truss structure, characterized in that, include: The intelligent analysis module includes a data screening unit and a stress distribution unit. The data screening unit completes data analysis and screens invalid data through a neural network algorithm. The stress distribution unit compares and analyzes the analysis results with historical data, adjusts the stress state, and establishes a stress distribution data model for the steel truss structure. The analysis module includes a stress analysis unit and a historical data unit. The stress analysis unit is used to establish a stress analysis model and analyze the stress distribution of the steel truss structure in real time. The historical data unit is used to collect various data of stress distribution during the analysis process and establish a data warehouse according to data characteristics. The intelligent analysis module includes a refinement processing unit, which is used to refine the stress distribution data model according to the calculation result data of the stress analysis unit, subdivide the stress distribution data model into microstress states, and perform secondary modeling on the steel truss structure in the microstress state; The intelligent analysis module includes a hardware adaptation unit, which is linked with the stress analysis unit to adjust the stress analysis model according to data analysis in different environments, so that the stress analysis model can run stably in different environments; the analysis module includes a data optimization baseline unit, which adjusts and optimizes the stress analysis model through the historical data unit to improve the accuracy and stability of data analysis; A two-way data communication channel is established between the intelligent analysis module and the analysis module, including a real-time data transmission channel and a feedback optimization channel; The real-time data transmission channel is used to transmit the analysis data of the stress analysis unit to the intelligent analysis module; the feedback optimization channel is used to transmit the optimization parameters and adjustment instructions of the intelligent analysis module to the analysis module to achieve closed-loop optimization of the system.

2. The intelligent analysis system for stress distribution of the steel truss structure according to claim 1, wherein The intelligent analysis module includes an abnormal analysis unit, which is used to generate an abnormal analysis model based on abnormal data and analyze and simulate the cause of the abnormality; the historical data unit includes an abnormal data detection unit, which is used to optimize the abnormal analysis model by detecting abnormal data and analyzing the cause of the abnormality.

3. The intelligent analysis system for stress distribution of the steel truss structure according to claim 2, wherein The intelligent analysis module includes a project implementation unit, which is used to effectively apply the analysis results in combination with the actual needs of the project; the analysis module includes a scenario analysis unit, which is used to provide personalized analysis services based on historical data analysis; the project implementation unit analyzes the analysis results and optimizes the analysis method based on the project needs; the scenario analysis unit improves the effectiveness of scenario analysis based on historical data analysis.

4. The intelligent analysis system for stress distribution of the steel truss structure according to claim 3, wherein, The intelligent analysis module includes a deep - cause analysis unit, which is used to conduct a deep - cause analysis of the stress analysis unit by analyzing historical data and analysis results; the deep - cause analysis unit includes an analysis method evaluation unit, which is used to comprehensively evaluate analysis methods according to analysis methods, parameter tuning, and hardware adaptation; the historical data unit includes a data deviation unit, which is used to adjust the stress analysis unit according to data deviation.

5. The intelligent analysis system for stress distribution of a steel truss structure according to claim 4, characterized in that, The data screening unit adopts a multi-layer neural network structure, including an input layer, a hidden layer, and an output layer. The input layer receives the original monitoring data of the steel truss structure. The hidden layer realizes data validity evaluation through weight adaptive adjustment, and the output layer generates a screened valid data set. The stress distribution unit adopts a stress model, where is the stress influence coefficient, is the stress correction coefficient, which are respectively associated with the material properties and the structural geometric conditions.

6. The intelligent analysis system for stress distribution of the steel truss structure according to claim 5, characterized in that The system further includes a multi - layer anomaly recognition and warning module, which is connected to the intelligent analysis module and the analysis module, and is used to conduct multi - level recognition of abnormal stress distribution in the steel truss structure based on a multi - dimensional anomaly feature library, and automatically adjust the warning level according to the severity of the anomaly; The multi-layer anomaly recognition and warning module adopts a data fitting model to process abnormal data, where represents the input feature, represents the output feature, represents the time, represents the feature coefficient, represents the random interference.

7. An intelligent analysis method for stress distribution of a steel truss structure, using the system according to claim 6, comprising the following steps: S1. Establish a stress distribution model according to the steel truss structure; S2. Collect, organize, and clean the actual stress distribution data of the steel truss structure; S3. Establish an initial stress distribution data model of the steel truss structure according to the stress distribution model and the stress distribution data; S4. Determine the stress threshold according to the initial stress distribution data model and monitor the real - time stress distribution of the steel truss structure; S5. Conduct adaptation adjustment on the model according to data analysis under different environments; S6. Conduct refined analysis according to the adapted model, and subdivide the stress distribution model into micro - stress states; S7. Evaluate the effectiveness of the model in combination with abnormal data simulation; S8. Adjust the model according to the analysis results and historical data, and optimize the analysis method; S9. Establish an abnormal analysis model according to the abnormal data and analyze the cause of the anomaly; S10. Conduct application analysis according to engineering requirements, optimize the analysis method, and provide personalized services.

Images