A method for evaluating the performance of L-shaped corrugated steel web connections
By performing performance characteristics sequence data analysis and clustering of the connecting nodes of L-shaped corrugated steel webs, combining residual and trend analysis, abnormal nodes are identified, and the problem of difficulty in evaluating the connection performance of L-shaped corrugated steel webs in the existing technology is solved. Efficient and accurate connection performance evaluation and failure prediction are achieved.
Patent Information
- Application Number
- CN202411197380.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-29
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-08-29
AI Technical Summary
The existing evaluation methods are difficult to comprehensively evaluate the overall connection performance of L-shaped corrugated steel webs under complex stress conditions, and cannot effectively capture the changes in the performance of the connection nodes with load changes and environmental erosion, which affects the safety and service life of the structure.
By obtaining the connection performance detection data of multiple connected nodes, extracting performance feature sequence data, performing node correlation analysis and clustering, building performance parameter status curves, predicting connection performance, and identifying abnormal nodes based on residual and trend analysis, using long and short-term Memory network builds a connection performance analysis model to achieve efficient and accurate connection performance evaluation.
Significantly reduce the risk of failure, ensure the safe and stable operation of the structure, accurately evaluate the connection status, capture changes in the performance of the connection node, and improve evaluation accuracy and efficiency.
Smart Images

Figure CN118690217B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of data processing, and in particular to a method for evaluating the connection performance of an L-shaped corrugated steel web. Background Art
[0002] As modern buildings and bridge projects have increasingly higher requirements for structural strength, stability and durability, steel structures are increasingly being used in large-scale projects. As a new type of structural component, L-shaped corrugated steel webs are gradually being widely used in bridges and high-rise buildings due to their advantages such as light weight, good seismic resistance and easy installation. However, with the increase in service life and the influence of external environmental factors, the connection performance of L-shaped corrugated steel webs may deteriorate, affecting the safety and service life of the overall structure.
[0003] Existing evaluation methods mostly focus on the analysis of a single node or local area, which makes it difficult to comprehensively evaluate the overall connection performance of the L-shaped corrugated steel web under complex stress conditions. With the changes in load and environmental erosion during use, the performance of different connection nodes in the L-shaped corrugated steel web will gradually change. How to effectively capture these changes and evaluate the connection performance of the L-shaped corrugated steel web has become an urgent problem to be solved in the engineering field. Summary of the invention
[0004] In order to solve the above technical problems, the present invention proposes an L-shaped corrugated steel web connection performance evaluation method for efficiently and accurately evaluating and predicting the connection performance of the L-shaped corrugated steel web structure.
[0005] To achieve the above object, the present invention provides a method for evaluating the connection performance of an L-shaped corrugated steel web, comprising:
[0006] Acquire connection performance test data of multiple connection nodes in the L-shaped corrugated steel web to be analyzed, and acquire multiple groups of reference benchmark data according to specification parameters of the L-shaped corrugated steel web to be analyzed;
[0007] Extracting performance characteristic sequence data of each connection node from multiple groups of reference benchmark data, performing node association analysis on the multiple groups of performance characteristic sequence data, extracting node association characteristics of multiple connection nodes, clustering the multiple connection nodes based on the node association characteristics and the performance characteristic sequence data to obtain multiple clustering sets, specifically including calculating correlation coefficients between the multiple connection nodes, performing a first clustering based on the correlation coefficients between the multiple connection nodes, extracting association influence characteristics between the multiple connection nodes and performing a second clustering;
[0008] Extracting a performance parameter state curve of each cluster set according to the performance characteristic sequence data of the connection nodes, and predicting the connection performance prediction data of each connection node based on the performance parameter state curve;
[0009] The abnormal nodes in the L-shaped corrugated steel web to be analyzed are determined according to the connection performance prediction data and the connection performance test data. The connection performance test data of the abnormal nodes are input into the connection performance analysis model to obtain the connection performance evaluation results of the L-shaped corrugated steel web to be analyzed.
[0010] Preferably, extracting node association features of a plurality of connection nodes, and clustering the plurality of connection nodes based on the node association features and the performance feature sequence data, comprises:
[0011] Determine multiple performance feature vectors of each connection node according to the performance feature sequence data, generate a performance feature matrix of each connection node according to the multiple performance feature vectors, calculate the Frobenius norm between any two connection nodes, determine the distance parameter for clustering the multiple connection nodes based on the Frobenius norm and construct a distance parameter matrix;
[0012] Using the distance parameter matrix as input data, multiple connection nodes are clustered for the first time based on the K-Means clustering algorithm to obtain multiple reference clusters, and multiple target connection nodes in each reference cluster are determined;
[0013] The topological relationship data between multiple connection nodes in the L-shaped corrugated steel web to be analyzed are obtained, a multivariate linear regression model is established based on the topological relationship data and the performance characteristic sequence data, and the correlation influence characteristics of multiple target connection nodes are extracted based on the multivariate linear regression model;
[0014] The associated influence features of each target connection node are combined into the corresponding performance feature vector to obtain the target associated features, and the multiple target connection nodes are clustered for the second time according to the target associated features to obtain multiple cluster sets.
[0015] Preferably, extracting a performance parameter state curve of each cluster set according to the performance characteristic sequence data of the connection nodes, and predicting connection performance prediction data of each connection node based on the performance parameter state curve, comprises:
[0016] For any cluster set, based on the performance characteristic sequence data of each connection node in the cluster set, a historical data reference set corresponding to each connection performance parameter in the cluster set is constructed, and the data in the historical data reference set corresponding to each connection performance parameter is fitted by cubic spline interpolation to obtain the performance parameter state curve of the cluster set;
[0017] For multiple connection nodes that do not belong to any cluster set, a cubic spline interpolation fitting is performed based on the performance characteristic sequence data of each connection node to obtain a performance parameter state curve corresponding to each connection node in the multiple connection nodes that do not belong to any cluster set;
[0018] The fitting value of each connection performance parameter of each connection node on the performance parameter state curve at the current time is determined according to the performance parameter state curve, and the connection performance prediction data of the connection node is determined according to the fitting value of each connection performance parameter.
[0019] Preferably, determining the abnormal nodes in the L-shaped corrugated steel web to be analyzed according to the connection performance prediction data and the connection performance detection data includes:
[0020] The residual corresponding to each connection performance parameter of any connection node is calculated based on the connection performance prediction data and the connection performance detection data, and the residual feature vector of each connection node is constructed. The residual parameter of each connection node is calculated based on the residual feature vector.
[0021] Determine multiple residual time series vectors of each connection node according to the performance characteristic sequence data of the connection node and the corresponding performance parameter state curve;
[0022] Perform linear regression analysis on multiple residual time series vectors of each connection node, extract the residual change rate of each residual time series vector, and calculate multiple trend parameters of each connection node based on the residual change rate;
[0023] The abnormal factor of each connection node is calculated according to the residual parameter and the trend parameter, and multiple abnormal nodes in the L-shaped corrugated steel web to be analyzed are determined according to the preset abnormal threshold.
[0024] Preferably, determining the abnormal nodes in the L-shaped corrugated steel web to be analyzed according to the connection performance prediction data and the connection performance detection data also includes:
[0025] The two norms of the residual eigenvector of each connection node are calculated as the residual parameter of the corresponding connection node, and the trend parameter is calculated based on the residual change rate using the following formula:
[0026] ;
[0027] In the formula, is the trend parameter, is the residual change rate, To adjust the trend;
[0028] After determining the residual parameters and trend parameters, the abnormal factor is calculated based on the following formula:
[0029] ;
[0030] In the formula, is the abnormal factor, is the residual parameter, The first node of the connection trend parameters, is the total number of trend parameters, , are the weight factors of residual parameters and trend parameters respectively;
[0031] After calculating the abnormal factor of each connection node, multiple connection nodes whose abnormal factors are greater than a preset abnormal threshold are screened out and recorded as abnormal nodes in the L-shaped corrugated steel web to be analyzed.
[0032] Preferably, the connection performance analysis model further includes:
[0033] According to the specification parameters of the L-shaped corrugated steel web to be analyzed, multiple groups of historical test data are obtained, and multiple groups of historical connection performance test data and status labels of each group of historical connection performance test data are extracted from the historical test data. A training data set is constructed according to the multiple groups of historical connection performance test data, and a status label is added to each group of historical connection performance test data. A connection performance analysis model is obtained by training based on the training data set.
[0034] Preferably, a connection performance analysis model is constructed based on a long short-term memory network.
[0035] Preferably, a second clustering is performed on multiple target connection nodes based on the AGNES algorithm.
[0036] The present invention has the following beneficial effects:
[0037] The present invention constructs an efficient and accurate L-shaped corrugated steel web connection performance evaluation method through systematic data processing, multi-level cluster analysis of data, performance prediction based on historical data, and anomaly detection combined with residuals and trends. The method can well capture the gradual changes in the performance of different connection nodes in the L-shaped corrugated steel web under load changes and environmental erosion during use, accurately evaluate the connection status of the L-shaped corrugated steel web, significantly reduce the risk of failure, and ensure the safe and stable operation of the L-shaped corrugated steel web structure. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 A schematic flow chart of a method for evaluating the connection performance of an L-shaped corrugated steel web provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0039] In order to make those skilled in the art better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0040] The embodiment of the present invention describes a method for evaluating the connection performance of an L-shaped corrugated steel web, which is used to evaluate and analyze the connection performance of an L-shaped corrugated steel web. Figure 1 , a method for evaluating the connection performance of an L-shaped corrugated steel web, specifically comprising the following steps:
[0041] S1. Obtain connection performance test data of multiple connection nodes in the L-shaped corrugated steel web to be analyzed, and obtain multiple groups of reference benchmark data according to specification parameters of the L-shaped corrugated steel web to be analyzed.
[0042] In this embodiment, connection performance test data are obtained at multiple connection nodes in the L-shaped corrugated steel web to be analyzed, including but not limited to stress, strain, displacement and other related parameters for characterizing the connection performance of the L-shaped corrugated steel web at different connection nodes, reflecting the actual working state of each connection node under different environmental and load conditions. In order to comprehensively analyze the connection performance of the L-shaped corrugated steel web to be analyzed, multiple groups of reference benchmark data are also obtained according to the specification parameters (such as geometric dimensions, material properties, design loads, design structures, etc.) of the L-shaped corrugated steel web to be analyzed. These reference benchmark data can be obtained from standard specifications or historical data to provide data reference for analyzing and judging the connection performance of different connection nodes in the subsequent process.
[0043] S2. Extract performance feature sequence data of each connection node from multiple groups of reference benchmark data, perform node association analysis on the multiple groups of performance feature sequence data, extract node association features of multiple connection nodes, cluster the multiple connection nodes based on the node association features and the performance feature sequence data, and obtain multiple cluster sets.
[0044] In this embodiment, after obtaining multiple sets of reference benchmark data, performance characteristic sequence data of each connection node are extracted therefrom. Specifically, the historical data involved in the reference benchmark data includes performance test data on the L-shaped corrugated steel web in different periods. These data are analyzed and sorted out, and multiple sets of data on the changes in different connection performance parameters over time are extracted based on the change data during use, and the performance characteristic sequence data of the connection node are constructed.
[0045] L-shaped corrugated steel webs are usually connected to other structural members through multiple connection nodes, and the performance of these nodes directly affects the safety and stability of the overall structure. The mechanical behaviors of various connection nodes in complex structures may affect each other. For example, excessive deformation of one node may increase stress concentration at adjacent nodes. Therefore, analyzing the connection performance of multiple nodes helps to understand the synergistic effect between nodes. Node correlation analysis is performed on multiple sets of performance characteristic sequence data in order to analyze the correlation between different connection nodes and quantify the similarity and dependence of multiple connection nodes in performance.
[0046] In this embodiment, the nodes are first clustered based on the calculated correlation coefficients between the multiple connected nodes. The purpose of this step is to classify nodes with similar performance into the same category based on the similarity of the nodes for further analysis. The correlation influence characteristics between the multiple connected nodes, such as stress transfer coefficient, deformation coordination and other parameters, are extracted from the preliminary clustering results. Based on the extracted correlation influence characteristics, the multiple connected nodes are clustered for the second time. The purpose of the second clustering is to more finely identify the node groups with deep associations so as to more accurately evaluate the overall connection performance of the nodes in subsequent analysis.
[0047] S3. Extract the performance parameter state curve of each cluster set according to the performance characteristic sequence data of the connection nodes, and predict the connection performance prediction data of each connection node based on the performance parameter state curve.
[0048] In this embodiment, after clustering is completed, the performance parameter state curve of each cluster set is extracted based on the performance feature sequence data corresponding to multiple connection nodes in the cluster set. These state curves reflect the change trend of node performance parameters over time in a healthy state, providing a standardized benchmark curve. The future connection performance of each connection node is predicted through these performance parameter state curves to help identify the possible performance changes of the node in the future, and provide a basis for predicting abnormal connection performance of different nodes.
[0049] S4. Determine the abnormal nodes in the L-shaped corrugated steel web to be analyzed based on the connection performance prediction data and the connection performance test data, input the connection performance test data of the abnormal nodes into the connection performance analysis model, and obtain the connection performance evaluation result of the L-shaped corrugated steel web to be analyzed.
[0050] In this embodiment, after obtaining the connection performance prediction data, it is compared and analyzed with the actually detected connection performance data to identify those abnormal nodes that deviate greatly from the predicted values. Abnormal nodes refer to those nodes whose actual detection data deviate significantly from the expected performance curve, and these nodes may have potential connection problems or failure risks. For the identified abnormal nodes, the corresponding connection performance detection data is input into the connection performance analysis model to perform an in-depth analysis of the input data, comprehensively consider historical data and current data, and finally output the connection performance evaluation results of the L-shaped corrugated steel web to be analyzed, such as determining whether multiple connection nodes have abnormal connection performance under the current connection performance detection data, so as to provide maintenance suggestions for the L-shaped corrugated steel web to be analyzed, and provide data support for subsequent structural maintenance and management.
[0051] In step S2, node association features of multiple connection nodes are extracted, and multiple connection nodes are clustered based on the node association features and performance feature sequence data, specifically including:
[0052] S21. Determine multiple performance feature vectors of each connection node according to the performance feature sequence data, generate a performance feature matrix of each connection node according to the multiple performance feature vectors, calculate the Frobenius norm between any two connection nodes, use the Frobenius norm as the correlation coefficient between the connection nodes, determine the distance parameter for clustering the multiple connection nodes based on the Frobenius norm and construct a distance parameter matrix.
[0053] In this embodiment, each performance characteristic vector corresponds to the value of a specific performance parameter at multiple time points, reflecting the characteristics of the performance parameter changing over time. After determining multiple performance characteristic vectors, these characteristic vectors are combined to generate a performance characteristic matrix for each connection node. Exemplarily, each row of the performance characteristic matrix represents a time point, and each column represents a performance parameter. After constructing the performance characteristic matrix of each connection node, the Frobenius norm between any two connection nodes is calculated to analyze and compare the similarity of the connection nodes in performance. The Frobenius norm can quantify the distance between the two matrices, that is, measure their overall similarity. Based on the calculated Frobenius norm, the distance parameter for clustering multiple connection nodes is further determined and a distance parameter matrix is constructed. Each element in the matrix represents the distance between any two connection nodes. These distance parameters will be used as input data for subsequent clustering analysis.
[0054] S22. Using the distance parameter matrix as input data, performing a first clustering on multiple connection nodes based on the K-Means clustering algorithm to obtain multiple reference clusters, and determining multiple target connection nodes in each reference cluster.
[0055] In this embodiment, the constructed distance parameter matrix is used as input data, and the K-Means clustering algorithm is used to perform the first clustering of multiple connection nodes. Through the K-Means algorithm, multiple connection nodes can be divided into multiple reference clusters according to the distance parameter matrix. The nodes in each cluster have high similarity in performance characteristics, indicating that they may have similar mechanical behaviors or working conditions in the structure. After completing the first clustering, multiple target connection nodes in each reference cluster are further determined. The target connection nodes refer to nodes that show significant characteristics in the cluster analysis. They represent the main characteristics of their respective clusters and will be the focus of subsequent analysis. Optionally, for multiple reference clusters, the target connection nodes can be screened out based on the distance threshold, for example, based on the distance threshold, some connection nodes in each reference cluster that are closer to the cluster center are screened out, thereby determining multiple target connection nodes in each reference cluster.
[0056] S23. Obtain topological relationship data between multiple connection nodes in the L-shaped corrugated steel web to be analyzed, establish a multivariate linear regression model based on the topological relationship data and the performance characteristic sequence data, and extract the correlation influence characteristics of multiple target connection nodes based on the multivariate linear regression model.
[0057] In this embodiment, in order to more accurately analyze and evaluate the performance of the target connection node, the topological relationship data between multiple connection nodes in the L-shaped corrugated steel web to be analyzed is obtained. The topological relationship data refers to the physical position relationship of the node in the steel web structure, the connection mode of the adjacent nodes, and the mechanical coupling situation. These data can reflect the physical connection and mutual influence between the nodes, and are an important basis for building a multivariate linear regression model. A multivariate linear regression model is established based on the topological relationship data and the previous performance characteristic sequence data, which can be used to analyze the relationship between a dependent variable and multiple independent variables to extract the associated influence characteristics of multiple target connection nodes. For example, the target connection node to be analyzed is used as an independent variable, and the surrounding connection nodes are used as dependent variables. The associated influence characteristics of each target connection node are extracted based on the multivariate linear regression model. The associated influence characteristics reflect the key mechanical behaviors such as interaction, stress transfer, and deformation coordination between nodes, wherein the associated influence characteristics can be characteristic parameters obtained based on the analysis of the multivariate linear regression model, such as the stress transfer coefficient and deformation coordination coefficient of each target connection node determined based on the multivariate linear regression model.
[0058] S24. Combining the associated influence features of each target connection node into the corresponding performance feature vector to obtain a target associated feature, and performing a second clustering on the multiple target connection nodes according to the target associated feature to obtain multiple cluster sets.
[0059] In this embodiment, the target association feature is a comprehensive feature vector, which includes the original performance data of the node and the association influence features extracted from the multivariate linear regression model, and can more accurately describe the overall performance status of each target node and its association with the surrounding nodes. A second clustering analysis is performed on multiple target connection nodes based on the target association feature. Specifically, a hierarchical clustering algorithm can be used to re-cluster multiple target connection nodes, such as re-clustering multiple target connection nodes based on the AGNES algorithm, which is not specifically limited in this embodiment. The purpose of the second clustering is to more accurately divide and identify node groups after considering the association influence between nodes, thereby obtaining multiple cluster sets, each of which represents a node group with high similarity in performance and node association.
[0060] It is worth noting that the goal of the initial clustering is to cluster nodes based on their performance parameters (such as stress, strain, etc.) and identify groups of nodes that perform similarly on these parameters. The nodes in these groups have commonalities in the initial analysis phase, that is, they show similar change patterns on certain performance indicators. As for the second clustering process, after extracting the associated influence characteristics such as stress transfer coefficient and deformation coordination coefficient, it is no longer limited to the individuals in the initial clustering group, but analyzes the mechanical coupling relationship between each key node and its surrounding nodes based on the physical topological relationship. Finally, the group obtained by clustering based on the target associated characteristics shows similarities in the mechanical coupling characteristics and the performance characteristics of the initial clustering, even if the actual physical positions of different nodes in the group are quite different, which means that the mutual influence between these nodes and the coupled nodes around them is relatively similar. The result of the second clustering not only retains the similarity of performance parameters in the initial clustering, but also introduces mechanical coupling characteristics based on topological relationships, thereby more comprehensively characterizing the complex relationship between nodes. The nodes in the group are similar in the way they interact with the coupled nodes around them, which may be understood as their mechanical performance in the structure may have similar risks or characteristics. This consistency helps to identify similar risk patterns or potential problems in structural health monitoring.
[0061] In step S3, a performance parameter state curve of each cluster set is extracted according to the performance characteristic sequence data of the connection nodes, and connection performance prediction data of each connection node is predicted based on the performance parameter state curve, specifically including:
[0062] According to the performance characteristic sequence data of each connection node in the clustering set, a historical data reference set corresponding to each connection performance parameter in the clustering set is constructed, and cubic spline interpolation fitting is performed on the data in the historical data reference set corresponding to each connection performance parameter to obtain the performance parameter state curve of the clustering set.
[0063] In this embodiment, any cluster set is taken as an example for explanation. The performance characteristic sequence data of the connection node includes relevant data of different performance parameters at multiple time points. By combining the data of the same performance parameter at different time points, a historical data reference set corresponding to each connection performance parameter in the cluster set is constructed. The historical data reference set reflects the historical performance of all connection nodes in the cluster set in various performance parameters, which can provide a data basis for fitting the performance parameter state curve. For the fitting of the connection performance parameters, in this embodiment, cubic spline interpolation fitting is performed on the data in the historical data reference set corresponding to each connection performance parameter to obtain a fitting curve corresponding to each connection performance parameter, which can describe the change trend of the performance characteristics of multiple connection nodes in the cluster set over time. The performance parameter state curve includes fitting curves corresponding to multiple connection performance parameters, which can effectively reflect the performance change law of the connection node in a healthy state.
[0064] For multiple connection nodes that do not belong to any cluster set, their performance characteristics are analyzed separately to extract their performance parameter state curves, specifically, cubic spline interpolation fitting is performed based on the performance characteristic sequence data of each connection node to obtain the performance parameter state curve corresponding to each connection node in the multiple connection nodes that do not belong to any cluster set;
[0065] After determining the performance parameter status curve of each connection node, determine the fitting value of each connection performance parameter of each connection node on the performance parameter status curve at the current time according to the performance parameter status curve, and determine the connection performance prediction data of the connection node according to the fitting value of each connection performance parameter.
[0066] In this embodiment, the fitting value of each connection performance parameter at the current time point is determined based on the performance parameter status curve of each connection node. The performance parameter status curve describes the changing trend of the performance parameters at different times. Therefore, these curves can be used to infer the expected value of each connection performance parameter at the current time point. The multiple expected values obtained based on the trend prediction of historical data can provide a benchmark for the performance evaluation at the current moment. These expected values are integrated to generate connection performance prediction data for each connection node. The connection performance prediction data comprehensively summarizes the expected performance of all key performance parameters at the current moment.
[0067] In step S4, in order to accurately identify the abnormal conditions of the connection nodes in the L-shaped corrugated steel web, the connection performance prediction data and the actual connection performance test data are combined to determine the abnormal nodes by calculating the residuals, analyzing the trends, etc. The abnormal nodes in the L-shaped corrugated steel web to be analyzed are determined based on the connection performance prediction data and the connection performance test data, specifically including:
[0068] The residual corresponding to each connection performance parameter of any connection node is calculated based on the connection performance prediction data and the connection performance detection data, the residual feature vector of each connection node is constructed, and the residual parameter of each connection node is calculated based on the residual feature vector.
[0069] In this embodiment, for any connection node, the residual of the connection node on each connection performance parameter is calculated based on the connection performance prediction data and the connection performance detection data. The residual reflects the deviation between the prediction data and the actual detection data. The residual values of the connection node on all performance parameters are combined together to construct a residual feature vector. The residual feature vector describes the deviation of the node on all performance parameters. The second norm of the residual feature vector is used as the residual parameter of the connection node. The residual parameter combines the multi-dimensional residuals into a comprehensive metric to provide a quantitative assessment of the overall deviation of the node. The residual parameter represents the comprehensive residual size of a node on all monitoring parameters. The larger the norm, the greater the overall deviation of the node from the healthy state on these parameters.
[0070] According to the performance characteristic sequence data of the connection node and the corresponding performance parameter state curve, multiple residual time series vectors of each connection node are determined, linear regression analysis is performed on the multiple residual time series vectors of each connection node, the residual change rate of each residual time series vector is extracted, and multiple trend parameters of each connection node are calculated according to the residual change rate.
[0071] In this embodiment, the residuals of each connection node at multiple time points are calculated based on the performance characteristic sequence data of the connection node and the corresponding performance parameter state curve, and a residual time series vector corresponding to each connection performance parameter is constructed. The residual time series vector reflects the change of the residual of a certain connection performance parameter of the node over time. Linear regression analysis is performed on the residual time series vector of each connection node, and the residual change rate is extracted. The residual change rate is the slope of the linear regression, which indicates the change trend of the residual in the time dimension. If the residual change rate is positive, it means that the deviation of the node is gradually increasing, and if it is negative, it means that the deviation is decreasing. According to the residual change rate, multiple trend parameters of each connection node are calculated, and the following formula is used for calculation:
[0072] ;
[0073] In the formula, is the trend parameter, is the residual change rate, To adjust the trend. When it is large, the trend parameter will be close to 0, indicating that the deviation of the node has changed significantly and there may be potential anomalies; when the residual change rate When it is close to 0, the trend parameter The value is close to 1, indicating that the node deviation does not change much and the state is relatively stable. The trend parameter can effectively distinguish nodes with drastic residual changes and nodes with stable changes, ensuring the sensitivity and accuracy of anomaly detection.
[0074] After determining the residual parameters and trend parameters, the abnormal factor is calculated based on the following formula:
[0075] ;
[0076] In the formula, is the abnormal factor, is the residual parameter, The first node of the connection trend parameters, is the total number of trend parameters, , are the weight factors of the residual parameter and trend parameter respectively. The calculation of the abnormal factor comprehensively considers the size of the residual and the significance of the trend, and can accurately reflect the overall health status of the node.
[0077] After the abnormal factor of each connection node is calculated according to the residual parameter and the trend parameter, multiple abnormal nodes in the L-shaped corrugated steel web to be analyzed are determined according to the preset abnormal threshold. Specifically, multiple connection nodes with abnormal factors greater than the preset abnormal threshold are screened out and recorded as abnormal nodes in the L-shaped corrugated steel web to be analyzed, so as to effectively screen out the connection nodes with potential risks in the L-shaped corrugated steel web.
[0078] In step S4, for the connection performance analysis model, it also includes:
[0079] A connection performance analysis model is constructed based on the long short-term memory network, and a training data set is constructed based on multiple groups of historical detection data, and the training is obtained based on the training data set.
[0080] In this embodiment, multiple groups of historical detection data are obtained according to the specification parameters of the L-shaped corrugated steel web to be analyzed, and multiple groups of historical connection performance detection data are extracted from the historical detection data. These data include performance parameters of different connection nodes, such as stress, strain and other parameters, which can reflect the working status of the connection nodes at different historical moments. And obtain the status label of each group of historical connection performance detection data. The status label may include: normal state: the connection node behaves normally without abnormal phenomena; abnormal state: the connection node exhibits abnormal characteristics, such as stress exceeding the safety range; fault state: the connection node has shown fault characteristics in the historical detection data and needs maintenance or replacement. The specific status label can be reasonably set based on expert experience, and this embodiment does not specifically limit it. By adding a status label to each group of historical connection performance detection data, multiple groups of historical connection performance detection data are converted to obtain a training data set, providing a data basis for the training of the connection performance analysis model.
[0081] In this embodiment, the basic structure of the connection performance analysis model is designed as follows: input layer: receiving the connection performance data of the L-shaped corrugated steel web, including key performance parameters at multiple time points; LSTM layer: used to capture the long-term dependencies in the time series data and extract the implicit state features of the connection nodes at different times; fully connected layer: the output of the LSTM layer is further processed by the fully connected layer, and the complex time series features are mapped to the target output space; output layer: the output layer maps the final output of the network to the prediction result of the connection performance, such as the state label corresponding to the output input data. The connection performance analysis model after training can be used for actual connection performance analysis. For the L-shaped corrugated steel web to be analyzed, the real-time detection data is input into the LSTM model, and the model will output the health status prediction results of different connection nodes, that is, the state label corresponding to the connection node. Based on the state label of the connection node output by the connection performance analysis model, the health status of the connection node can be monitored in real time. After predicting and analyzing that a certain connection node may be abnormal, an early warning can be issued in time, indicating that further inspection and maintenance are required, providing technical support for abnormal detection of L-shaped corrugated steel webs.
[0082] The embodiment of the present invention provides a systematic and intelligent method for evaluating the connection performance of L-shaped corrugated steel webs, which improves the accuracy and efficiency of evaluating the connection performance of L-shaped corrugated steel webs. Specifically, by collecting and processing the connection performance detection data of multiple connection nodes in the L-shaped corrugated steel web to be analyzed and multiple groups of reference benchmark data, the node association analysis is performed through the two steps of primary clustering and secondary clustering, and the node groups with similar performance are first identified. On the basis of considering the node performance parameters, the mechanical coupling characteristics based on the topological relationship are introduced, and the complex mutual influence relationship between the nodes is deeply analyzed, which provides a more detailed foundation for subsequent performance prediction and anomaly detection; after the clustering analysis is completed, the performance parameter state curve of each node is constructed by performing cubic spline interpolation fitting on the connection nodes in the cluster set, which is used to predict the performance parameters of the connection nodes, thereby improving The predictability and accuracy of structural health monitoring are improved; after predicting the connection performance of each node, the residual between the connection performance prediction data and the actual detection data is calculated, and the abnormal factor of each node is calculated by combining the trend analysis of the residual time series. The dual analysis method of residual and trend can comprehensively and accurately identify potential abnormal nodes, avoiding misjudgment or missed detection that may be caused by simple threshold method, and improving the sensitivity and reliability of anomaly detection; further introduce a connection performance analysis model based on long short-term memory network (LSTM) to further analyze potential abnormal nodes, evaluate the health status of nodes, and improve the accuracy of L-type corrugated steel web connection performance evaluation.
[0083] The above is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.
Claims
1. A method for evaluating the connection performance of an L-shaped corrugated steel web, characterized in that: include: Acquire connection performance test data of multiple connection nodes in the L-shaped corrugated steel web to be analyzed, and acquire multiple groups of reference benchmark data according to specification parameters of the L-shaped corrugated steel web to be analyzed; Extracting performance characteristic sequence data of each connection node from multiple groups of reference benchmark data, performing node association analysis on the multiple groups of performance characteristic sequence data, extracting node association characteristics of multiple connection nodes, clustering the multiple connection nodes based on the node association characteristics and the performance characteristic sequence data to obtain multiple clustering sets, specifically including calculating correlation coefficients between the multiple connection nodes, performing a first clustering based on the correlation coefficients between the multiple connection nodes, extracting association influence characteristics between the multiple connection nodes and performing a second clustering; Extracting a performance parameter state curve of each cluster set according to the performance characteristic sequence data of the connection nodes, and predicting the connection performance prediction data of each connection node based on the performance parameter state curve; The abnormal nodes in the L-shaped corrugated steel web to be analyzed are determined according to the connection performance prediction data and the connection performance test data. The connection performance test data of the abnormal nodes are input into the connection performance analysis model to obtain the connection performance evaluation results of the L-shaped corrugated steel web to be analyzed.
2. The method for evaluating the connection performance of an L-shaped corrugated steel web according to claim 1, characterized in that: Node association features of multiple connected nodes are extracted, and multiple connected nodes are clustered based on the node association features and performance feature sequence data, including: Determine multiple performance feature vectors of each connection node according to the performance feature sequence data, generate a performance feature matrix of each connection node according to the multiple performance feature vectors, calculate the Frobenius norm between any two connection nodes, determine the distance parameter for clustering the multiple connection nodes based on the Frobenius norm and construct a distance parameter matrix; Using the distance parameter matrix as input data, multiple connection nodes are clustered for the first time based on the K-Means clustering algorithm to obtain multiple reference clusters, and multiple target connection nodes in each reference cluster are determined; The topological relationship data between multiple connection nodes in the L-shaped corrugated steel web to be analyzed are obtained, a multivariate linear regression model is established based on the topological relationship data and the performance characteristic sequence data, and the correlation influence characteristics of multiple target connection nodes are extracted based on the multivariate linear regression model; The associated influence features of each target connection node are combined into the corresponding performance feature vector to obtain the target associated features, and the multiple target connection nodes are clustered for the second time according to the target associated features to obtain multiple cluster sets.
3. The method for evaluating the connection performance of an L-shaped corrugated steel web according to claim 1, characterized in that: A performance parameter state curve of each cluster set is extracted according to the performance characteristic sequence data of the connection nodes, and connection performance prediction data of each connection node is predicted based on the performance parameter state curve, including: For any cluster set, based on the performance characteristic sequence data of each connection node in the cluster set, a historical data reference set corresponding to each connection performance parameter in the cluster set is constructed, and the data in the historical data reference set corresponding to each connection performance parameter is fitted by cubic spline interpolation to obtain the performance parameter state curve of the cluster set; For multiple connection nodes that do not belong to any cluster set, a cubic spline interpolation fitting is performed based on the performance characteristic sequence data of each connection node to obtain a performance parameter state curve corresponding to each connection node in the multiple connection nodes that do not belong to any cluster set; The fitting value of each connection performance parameter of each connection node on the performance parameter state curve at the current time is determined according to the performance parameter state curve, and the connection performance prediction data of the connection node is determined according to the fitting value of each connection performance parameter.
4. The method for evaluating the connection performance of an L-shaped corrugated steel web according to claim 3, characterized in that: According to the connection performance prediction data and connection performance test data, the abnormal nodes in the L-shaped corrugated steel web to be analyzed are determined, including: The residual corresponding to each connection performance parameter of any connection node is calculated based on the connection performance prediction data and the connection performance detection data, and the residual feature vector of each connection node is constructed. The residual parameter of each connection node is calculated based on the residual feature vector. Determine multiple residual time series vectors of each connection node according to the performance characteristic sequence data of the connection node and the corresponding performance parameter state curve; Perform linear regression analysis on multiple residual time series vectors of each connection node, extract the residual change rate of each residual time series vector, and calculate multiple trend parameters of each connection node based on the residual change rate; The abnormal factor of each connection node is calculated according to the residual parameter and the trend parameter, and multiple abnormal nodes in the L-shaped corrugated steel web to be analyzed are determined according to the preset abnormal threshold.
5. The method for evaluating the connection performance of an L-shaped corrugated steel web according to claim 4, characterized in that: Determine the abnormal nodes in the L-shaped corrugated steel web to be analyzed based on the connection performance prediction data and the connection performance test data, and also include: The two norms of the residual eigenvector of each connection node are calculated as the residual parameter of the corresponding connection node, and the trend parameter is calculated based on the residual change rate using the following formula: ; In the formula, is the trend parameter, is the residual change rate, To adjust the trend; After determining the residual parameters and trend parameters, the abnormal factor is calculated based on the following formula: ; In the formula, is the abnormal factor, is the residual parameter, The first node of the connection trend parameters, is the total number of trend parameters, , are the weight factors of residual parameters and trend parameters respectively; After calculating the abnormal factor of each connection node, multiple connection nodes whose abnormal factors are greater than a preset abnormal threshold are screened out and recorded as abnormal nodes in the L-shaped corrugated steel web to be analyzed.
6. The method for evaluating the connection performance of an L-shaped corrugated steel web according to claim 1, characterized in that: For the connection performance analysis model, also includes: According to the specification parameters of the L-shaped corrugated steel web to be analyzed, multiple groups of historical test data are obtained, and multiple groups of historical connection performance test data and status labels of each group of historical connection performance test data are extracted from the historical test data. A training data set is constructed according to the multiple groups of historical connection performance test data, and a status label is added to each group of historical connection performance test data. A connection performance analysis model is obtained by training based on the training data set.
7. The method for evaluating the connection performance of an L-shaped corrugated steel web according to claim 6, characterized in that: Construct a connection performance analysis model based on long short-term memory network.
8. The method for evaluating the connection performance of an L-shaped corrugated steel web according to claim 2, characterized in that: A second clustering of multiple target connection nodes is performed based on the AGNES algorithm.
Citation Information
Patent Citations
Bridge health monitoring system and method based on big data
CN114282398A
Abnormality identification and mode distinguishing method and system for monitoring data of power transformer
CN116861354A
Foundation pit supporting performance evaluation method and system based on data analysis
CN118133672A
Tensile damage online characterization method based on multi-fractal and unsupervised clustering
CN118296415A
Device state monitoring method based on multi-index cluster analysis
WO2022252505A1
