Safety risk monitoring method for high-altitude large steel structure corridor

By laying multiple monitoring points and small model experiments in high-altitude large steel structure corridors, combining clustering algorithms and dual-branch neural networks, a risk index analysis model was established, which solved the problem of single detection methods and weak early warning capabilities, and realized Dynamic response analysis and risk warning to the corridor have been improved.

CN118941083APending Publication Date: 2024-11-12CHINA CONSTR EIGHTH BUREAU DEV & CONSTR CO LTD +1
View PDF 0 Cites 5 Cited by

Patent Information

Application Number
CN202410991259.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The prior art has problems such as single detection means and weak early warning capabilities in the safety risk monitoring of large steel structure corridors at high altitudes, making it difficult to fully grasp the dynamic response characteristics of the corridors and timely detect safety hazards.

Method used

Vibration sensors and deformation sensors are arranged at multiple representative monitoring points, and vibration experiments are carried out in combination with small scale model. The mapping relationship between vibration, deformation parameters and risk index of the corridor structure is established through clustering algorithm and dual-branch neural network. , calculate the risk index in real time and issue early warning signals.

Benefits of technology

A comprehensive dynamic response analysis of large corridor structures has been achieved, and a scientific risk assessment and early warning mechanism has been established, which can quickly and accurately assess safety risks and issue early warnings in a timely manner to reduce the risk of accidents.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118941083A_ABST
    Figure CN118941083A_ABST
Patent Text Reader

Abstract

The invention provides a safety risk monitoring method for a high-altitude large-scale steel structure corridor, and belongs to the technical field of high-altitude large-scale steel structure corridors, and the method comprises the steps: selecting a plurality of representative monitoring points, monitoring the vibration and deformation data of a corridor structure, building a small-scale model, and carrying out the experimental test, and feature intervals of different risk levels are divided by adopting a clustering analysis method. And then establishing a mapping relationship among the vibration and deformation parameters and the risk index of the corridor structure by using a machine learning method to obtain a corridor real-time risk index analysis model. And finally, inputting actual corridor monitoring data into the analysis model, comparing the actual corridor monitoring data with a preset risk threshold value, and sending out an early warning signal in time, thereby providing a basis for corridor safety monitoring and management. According to the method, measured data and simulation experiment data are fully utilized, the advanced data analysis technology is combined, the safety condition of the corridor structure can be effectively evaluated, and the technical problems that in the prior art, the detection means is single, and the early warning capacity is weak are solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of high-altitude large-scale steel structure corridors, and in particular, relates to a safety risk monitoring method for high-altitude large-scale steel structure corridors. Background Art

[0002] Large steel structure corridors are widely used in high-altitude scenes, such as airport terminals, internal corridors of complexes, rail transit stations, etc. These corridors generally have large spans and complex structures, and are subjected to complex dynamic loads, such as wind loads, earthquake loads, and human traffic loads. In addition, due to the influence of long-term use environment, they are prone to local stress concentration, overall deformation aggravation and other safety hazards. Therefore, how to effectively monitor and evaluate the safety risks of such large corridor structures has always been a key issue of concern to the industry.

[0003] At present, the industry mainly adopts the following common methods for safety risk monitoring of large steel structure corridors:

[0004] 1. Regular visual inspection method: Professional inspectors regularly conduct visual inspections on key parts of the corridor to check whether there are obvious deformations, cracks and other abnormal conditions. This method is simple and intuitive, but it is greatly affected by the subjective judgment of inspectors and it is difficult to fully and timely discover hidden safety hazards.

[0005] 2. Sensor monitoring method: Vibration sensors, strain sensors, etc. are deployed at key locations of the corridor to monitor the dynamic response characteristics of the structure in real time. By analyzing the sensor data, the safety status of the corridor can be preliminarily judged. This method can achieve continuous monitoring, but it requires a large number of sensors to be deployed, and the data analysis is also relatively complex, making it difficult to quickly and accurately assess risks.

[0006] The above methods have their own advantages and disadvantages, but generally they have the following common problems:

[0007] (1) The detection method is single and cannot fully grasp the dynamic response characteristics of the corridor. It is difficult to accurately evaluate the complex corridor structure by relying on only a small number of sensors or finite element analysis.

[0008] (2) Weak early warning capabilities make it difficult to detect safety hazards in a timely manner. Many methods can only detect problems through manual inspections or post-analysis, and are unable to monitor and warn the safety status of the corridor in real time. Summary of the invention

[0009] In view of this, the present invention provides a safety risk monitoring method for a high-altitude large-scale steel structure corridor, which can solve the technical problems of the prior art of single detection means and weak early warning capability.

[0010] The present invention is achieved in that:

[0011] A first aspect of the present invention provides a method for monitoring safety risks of a high-altitude large steel structure corridor, which comprises the following steps:

[0012] S10, selecting a plurality of representative monitoring points, using vibration sensors and deformation sensors to monitor the corridor structure, and obtaining vibration and deformation data;

[0013] S20, setting up the same monitoring points as those of the actual corridor on the pre-established small-scale corridor structure model, vibrating the model with different intensities using a vibration table, and collecting corresponding vibration and deformation data;

[0014] S30, using clustering algorithm to analyze the monitoring data of the actual corridor and the model, and dividing the characteristic intervals of different risk levels;

[0015] S40. Using machine learning methods, establish a mapping relationship between corridor structure vibration, deformation parameters and risk index, and obtain a corridor risk index analysis model;

[0016] S50, inputting the monitoring data of the actual corridor into the corridor risk index analysis model to calculate the real-time risk index of the corridor;

[0017] S60. Compare the real-time risk index of the corridor with the preset risk threshold. When the risk index exceeds the threshold, issue a warning signal in a timely manner.

[0018] Wherein, the step S10 specifically includes:

[0019] Select representative monitoring points in the corridor structure. These key monitoring points should cover the stress concentration points, potential weak points and other areas of the corridor. Vibration sensors and deformation sensors are deployed at these monitoring points. The vibration sensors are used to collect the vibration data of the corridor in the vertical, horizontal and torsional directions. The deformation sensors are used to measure the static deformation of the corridor under the action of force. These sensors are continuously monitored by the data acquisition system to obtain the vibration and deformation data representing the dynamic response characteristics of the corridor. The monitoring frequency is generally 1 Hz or higher.

[0020] The specific steps of step S20 include:

[0021] A small-scale model with similar geometric dimensions and structure to the actual corridor is established. Vibration sensors and deformation sensors are arranged on the small-scale model in the same positions and numbers as those of the actual corridor. Vibration loads of different intensities are applied to the small-scale model using a vibration table to simulate the dynamic response of the corridor under different external environments. While the model is subjected to vibration loads, the monitoring system collects data from the vibration sensors and deformation sensors in real time, including characteristics such as amplitude, frequency, phase, peak value, as well as amplitude and modal information of structural deformation. By analyzing these experimental data, the response characteristics of the small-scale model under different vibration conditions are established.

[0022] The specific implementation of step S30 is as follows:

[0023] Clustering algorithms are used to analyze the monitoring data of actual corridors and model experiments, such as K-means clustering or Gaussian mixture model clustering algorithms; these algorithms automatically divide the data into different cluster centers, each of which represents a typical vibration or deformation feature; by analyzing the statistical characteristics of these cluster centers, such as mean, standard deviation, extreme values, etc., the characteristic intervals of different risk levels are determined, such as dividing the results into three levels of "low risk", "medium risk" and "high risk", and giving the corresponding characteristic interval ranges.

[0024] The specific implementation of step S40 is as follows:

[0025] A dual-branch neural network structure is adopted as the corridor risk index analysis model, in which the micro-feature extraction branch uses multi-scale convolutional layers, spatial pyramid pooling layers and attention mechanisms to extract detailed features of local vibration and deformation of the corridor; the macro-feature extraction branch adopts deep convolutional networks, temporal convolutional networks and feature pyramids to capture the characteristics of the overall and long-term change trends of the corridor; in the feature fusion layer, the output features of the two branches are concatenated and fused through a fully connected layer to generate a comprehensive feature representation; the final output layer includes a fully connected layer and a Softmax classification layer, which maps the fused features to the risk index and risk level of the corridor.

[0026] The specific implementation of step S50 is as follows:

[0027] The actual corridor vibration and deformation monitoring data obtained from step S10 are preprocessed according to the input format of the model, such as normalization, supplementation of missing values, etc.; these data are input into the trained risk index analysis model, which will calculate the corresponding risk index value through forward propagation based on the input vibration and deformation characteristics, indicating the current risk level of the corridor; in order to ensure real-time performance, a sliding time window is set, and the monitoring data of the most recent period (such as 1 hour or 1 day) is input into the model for risk index calculation.

[0028] The specific implementation of step S60 is as follows:

[0029] Compare the real-time risk index of the corridor with the preset risk threshold. For example, divide the risk index into three levels: low risk (0-0.3), medium risk (0.3-0.7) and high risk (0.7-1.0), and set the corresponding risk thresholds to 0.3 and 0.7. When the risk index exceeds 0.7 (high risk threshold), it means that the corridor is facing a major safety hazard. The system should issue an early warning signal to notify the relevant responsible personnel to take necessary disposal actions, such as restricting use, strengthening and repairing, etc. When the risk index is in the medium risk range (0.3-0.7), it should also be taken seriously, and measures such as strengthening monitoring and checking for hidden dangers can be taken.

[0030] Among them, the corridor risk index analysis model adopts a dual-branch neural network structure, specifically including: a microscopic feature extraction branch for extracting the local vibration and deformation characteristics of the corridor structure and a macroscopic feature extraction branch for extracting the overall characteristics and long-term change trends of the corridor structure.

[0031] This dual-branch neural network structure can simultaneously capture the local vibration and deformation characteristics of the corridor structure, as well as the overall long-term change trend, thereby providing a more comprehensive and accurate risk assessment. By combining small-scale model experimental data and actual monitoring data, the model can effectively learn the characteristics of different risk levels and quickly calculate the real-time risk index of the corridor in practical applications.

[0032] Furthermore, it also includes a feature fusion layer that splices micro features and macro features together and performs feature fusion through several fully connected layers.

[0033] Furthermore, the microscopic feature extraction branch is used to extract local vibration and deformation characteristics of the corridor structure.

[0034] Furthermore, the microscopic feature extraction branch includes:

[0035] Multi-scale convolutional layer: Use convolution kernels of different sizes to process input data in parallel to capture vibration and deformation features of different scales;

[0036] Spatial pyramid pooling layer: divides the feature map into regions of different scales, performs maximum pooling in each region, and extracts multi-scale local features;

[0037] Attention mechanism: Channel attention and spatial attention are introduced to enable the corridor risk index analysis model to adaptively focus on important local vibration and deformation features.

[0038] Furthermore, the macro feature extraction branch is used to extract the overall characteristics and long-term change trends of the corridor structure.

[0039] Furthermore, the macro feature extraction branch includes:

[0040] Deep convolutional network: uses ResNet or LSTM as the backbone to extract long-term dependency features of time series data;

[0041] Temporal Convolutional Network: Use dilated convolution to expand the receptive field without increasing the number of parameters and capture contextual information over a longer time range;

[0042] Feature pyramid: Integrate features of different time scales to obtain both short-term and long-term structural change features.

[0043] The main differences between the micro feature extraction branch and the macro feature extraction branch:

[0044] 1. Different focus: The microscopic feature extraction branch focuses on capturing the local, subtle vibration and deformation characteristics of the corridor structure. These characteristics may include high-frequency vibrations at specific monitoring points, local stress concentrations, or small structural deformations; the macroscopic feature extraction branch focuses on the overall performance and long-term change trends of the corridor structure. This includes the low-frequency vibration modes of the overall structure, the overall deformation caused by the long-term settlement trend, etc.

[0045] 2. Network structure design: The micro-branch uses multi-scale convolution and spatial pyramid pooling, which is suitable for extracting local features of different scales. The attention mechanism helps the model focus on the most important local information; the macro-branch uses deep convolutional networks and temporal convolutional networks, which are more suitable for processing long sequence data and capturing long-term dependencies. The feature pyramid structure helps to integrate information of different time scales.

[0046] 3. Input data processing: The micro-branch may process more raw sensor data with high sampling rates, such as high-frequency vibration signals; the macro-branch may process downsampled or preprocessed data, such as daily average deformation data, monthly trends, etc.

[0047] 4. Time scale: The micro branch focuses on data changes in a short period of time, which may be a time window of seconds or minutes; the macro branch processes data with a longer time span, which may cover changes over days, weeks or longer.

[0048] Specifically:

[0049] 1. Frequency range:

[0050] Microscopic features:

[0051] High frequency vibration: usually above 1 Hz and may be as high as several hundred Hz or higher.

[0052] For example, a vibration of 101,000 Hz might be classified as a microscopic feature.

[0053] Macro features:

[0054] Low frequency vibration: usually below 1 Hz.

[0055] For example: a vibration of 0.11 Hz might be considered a macroscopic feature.

[0056] 2. Amplitude range:

[0057] Microscopic features:

[0058] Minor deformation: usually on the millimeter level or less.

[0059] For example: a deformation of 0.011 mm may be considered a microscopic feature.

[0060] Macro features:

[0061] Large-scale deformation: usually on the centimeter level or larger.

[0062] For example: a deformation of 110 cm might be classified as a macro feature.

[0063] 3. Time scale:

[0064] Microscopic features:

[0065] Short-term changes: from seconds to hours.

[0066] For example, data changes from 1 second to 1 hour may be considered as micro features.

[0067] Macro features:

[0068] Long-term trend: from day level to grade level.

[0069] For example, changes in data from 1 day to 1 year may be classified as macro features.

[0070] 4. Spatial scale:

[0071] Microscopic features:

[0072] Local features: data involving a single or a few monitoring points.

[0073] For example: data from a single sensor or 15 adjacent sensors may be considered as micro features.

[0074] Macro features:

[0075] Global features: data involving a large area or the entire structure.

[0076] For example, the combined data of more than 10 sensors or the deformation data of the entire corridor may be classified as macro features.

[0077] 5. Stress / strain range:

[0078] Microscopic features:

[0079] Localized stress concentrations: Stresses exceeding 50% of the material's yield strength may be seen as microscopic features.

[0080] For example: For a steel structure, if the yield strength is 235 MPa, then local stresses exceeding 117.5 MPa may be classified as microscopic features.

[0081] Macro features:

[0082] Overall stress distribution: Stresses below 50% of the material's yield strength may be considered macroscopic features.

[0083] For example: For the same steel structure, a stress distribution of 0.117.5MPa may be classified as a macroscopic feature.

[0084] 6. Deformation caused by temperature changes:

[0085] Microscopic features:

[0086] Local thermal deformation: deformation caused by temperature gradient in a small area.

[0087] For example, a local deformation caused by a temperature difference of 15°C may be seen as a microscopic feature.

[0088] Macro features:

[0089] Overall thermal deformation: overall deformation caused by large-scale temperature changes.

[0090] For example, global deformation caused by daily temperature differences or seasonal temperature changes (such as 1030°C) may be classified as macroscopic features.

[0091] The specific values ​​of these intervals may need to be adjusted according to the specific conditions of the corridor (such as size, material, use environment, etc.). In practical applications, it may be necessary to accurately define these intervals through data analysis and expert knowledge. In addition, some data may have both micro and macro characteristics. In this case, fuzzy sets or probabilistic methods can be considered for classification.

[0092] Furthermore, the feature fusion layer is used to splice the micro features and macro features together, perform feature fusion through multiple fully connected layers, and use residual connections to ensure that information can be effectively propagated in multiple fully connected layers.

[0093] How the feature fusion layer works:

[0094] 1. Feature stitching:

[0095] First, the output feature vectors of the micro-feature extraction branch and the macro-feature extraction branch are concatenated in the channel dimension. Assuming that the micro-feature dimension is [batch_size, m_channels] and the macro-feature dimension is [batch_size, n_channels], the concatenated feature dimension is [batch_size, m_channels + n_channels].

[0096] 2. Feature Fusion:

[0097] The concatenated features are further fused through a series of fully connected layers. The role of these layers is to learn the relationship and combination between micro and macro features.

[0098] Non-linear activation functions such as ReLU can be used to increase the expressiveness of the model.

[0099] 3. Residual Connection:

[0100] Residual connections are added between fully connected layers to ensure that the original micro and macro information can be directly passed to subsequent layers to prevent information loss.

[0101] 4. Attention Mechanism:

[0102] An attention mechanism can be introduced in the fusion layer to dynamically adjust the importance weights of different features. This helps the model flexibly focus on micro or macro features in different situations.

[0103] 5. Normalization:

[0104] Use batch normalization or layer normalization to stabilize feature distribution and speed up the training process.

[0105] 6. Dimensionality Reduction:

[0106] The last few fully connected layers can gradually reduce the feature dimensions and map the fused high-dimensional features to a low-dimensional space suitable for risk index prediction.

[0107] In this way, the feature fusion layer can effectively combine the advantages of micro and macro features to generate a comprehensive feature representation that contains both local detail information and global structural features, thus providing a powerful input for the final risk index prediction.

[0108] Wherein, the step of using a clustering algorithm to analyze the monitoring data of the actual corridor and the model adopts K-means clustering algorithm or Gaussian mixture model clustering.

[0109] The scale of the small model is selected to be between 1 / 10 and 1 / 50 of the actual corridor.

[0110] The vibration table can generate a sinusoidal vibration signal in the range of 0.1 Hz to 100 Hz, and the amplitude can be adjusted to the bearing limit of the corridor structure.

[0111] The number of clusters K of the K-means clustering algorithm can be set to between 3 and 10.

[0112] Among them, the low-risk characteristic interval is: amplitude less than 0.5mm and frequency less than 1Hz; the medium-risk characteristic interval is: amplitude between 0.5 and 2mm and frequency between 1 and 5Hz; the high-risk characteristic interval is: amplitude greater than 2mm and frequency greater than 5Hz.

[0113] Among them, the multi-scale convolution layer in the micro feature extraction branch uses three different sizes of convolution kernels: 1x1, 3x3 and 5x5 for parallel processing.

[0114] Among them, the deep convolutional network in the macro feature extraction branch uses ResNet-50 as the backbone network.

[0115] The feature fusion layer includes three fully connected layers, which have 256, 128 and 64 neurons respectively.

[0116] The length of the sliding time window is 1 hour.

[0117] Optionally, the early warning signal includes an audible and visual alarm and sending a text message to management personnel.

[0118] Optionally, the low risk threshold value 0.3 and the high risk threshold value 0.7 may be adjusted according to the specific use environment and material properties of the corridor.

[0119] Optionally, the measurement range of the vibration sensor is 0.01 mm / s to 100 mm / s, and the measurement frequency range is 0.1 Hz to 500 Hz.

[0120] Optionally, the measurement range of the deformation sensor is 0.01mm to 100mm, and the measurement accuracy is 0.001mm.

[0121] Optionally, the monitoring data is collected at a frequency between 1 Hz and 100 Hz, and the collection duration covers the entire life cycle of the corridor.

[0122] Compared with the prior art, the safety risk monitoring method for a high-altitude large steel structure corridor provided by the present invention has the following beneficial effects:

[0123] 1. Comprehensive monitoring methods can deeply analyze the dynamic response characteristics of the corridor

[0124] Vibration sensors and deformation sensors are used to comprehensively obtain vibration data and static deformation data of the corridor structure in multiple directions. Small model experiments are used to obtain more detailed dynamic response characteristics under vibration loads of different intensities to provide support for risk analysis. Through data analysis, we can simultaneously focus on the detailed changes in the local corridor and the overall long-term trend, fully reflecting the safety status of the structure.

[0125] 2. Scientific risk assessment system, which can quickly and accurately assess the safety risks of the corridor

[0126] Clustering algorithm is used to analyze monitoring data and divide characteristic intervals of different risk levels to establish a basis for subsequent risk assessment. A dual-branch neural network structure is designed to integrate microscopic vibration deformation characteristics and macroscopic structural characteristics to establish a mapping relationship between the vibration, deformation and risk index of the corridor. The risk index of the corridor can be calculated in real time and compared with the preset threshold to detect safety hazards in a timely manner.

[0127] 3. Strong early warning capability, measures can be taken in advance to reduce the risk of accidents

[0128] When the risk index exceeds the preset threshold, the system can promptly issue warning signals such as sound and light alarms and SMS notifications to remind relevant personnel to take necessary measures. Combined with historical data analysis and expert experience, the warning threshold can be adaptively adjusted to improve the sensitivity and pertinence of the warning. Multi-sensor fusion and dynamic risk assessment can be used to further improve the reliability and accuracy of the warning.

[0129] To sum up, the safety risk monitoring method of the present invention overcomes the limitations of the existing technology, realizes a comprehensive and in-depth dynamic response analysis of large corridor structures, establishes a scientific and systematic risk assessment and early warning mechanism, and solves the technical problems of the existing technology of single detection means and weak early warning capability. BRIEF DESCRIPTION OF THE DRAWINGS

[0130] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0131] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution 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.

[0132] like Figure 1 As shown, it is a flow chart of a safety risk monitoring method for a high-altitude large steel structure corridor provided by the present invention, and the method comprises the following steps:

[0133] S10, selecting a plurality of representative monitoring points, using vibration sensors and deformation sensors to monitor the corridor structure, and obtaining vibration and deformation data;

[0134] S20, setting up the same monitoring points as those of the actual corridor on the pre-established small-scale corridor structure model, vibrating the model with different intensities using a vibration table, and collecting corresponding vibration and deformation data;

[0135] S30, using clustering algorithm to analyze the monitoring data of the actual corridor and the model, and dividing the characteristic intervals of different risk levels;

[0136] S40. Using machine learning methods, establish a mapping relationship between corridor structure vibration, deformation parameters and risk index, and obtain a corridor risk index analysis model;

[0137] S50, inputting the monitoring data of the actual corridor into the corridor risk index analysis model to calculate the real-time risk index of the corridor;

[0138] S60. Compare the real-time risk index of the corridor with the preset risk threshold. When the risk index exceeds the threshold, issue a warning signal in a timely manner.

[0139] The specific implementation methods of the above steps are described in detail below:

[0140] Step S10: Select multiple representative monitoring points, use vibration sensors and deformation sensors to monitor the corridor structure, and obtain vibration and deformation data.

[0141] First, it is necessary to determine representative monitoring points in the corridor structure. These monitoring points should cover the key areas of the corridor, such as stress concentration points, potential weak points, etc. Finite element analysis or engineering experience can usually be used to determine these key monitoring points.

[0142] Next, vibration sensors and deformation sensors are installed at these key monitoring points. The vibration sensors are used to collect vibration data of the corridor structure in the vertical, horizontal and torsional directions, including characteristics such as amplitude and frequency. The deformation sensors are used to measure the static deformation of the corridor under force, including deflection, relative displacement, etc. The layout of the sensors should ensure the representativeness and accuracy of the data, such as avoiding installation in local stress concentration areas.

[0143] Finally, these sensors are continuously monitored by a data acquisition system to obtain vibration and deformation data. The monitoring frequency is generally 1 Hz or higher to fully capture the dynamic changes of the corridor structure. These real-time monitoring data will provide a basis for subsequent risk analysis.

[0144] Step S20: Arrange the same monitoring points as the actual corridor on the pre-established small-scale corridor structure model, use a vibration table to vibrate the model with different intensities, and collect corresponding vibration and deformation data

[0145] In order to establish the mapping relationship between the vibration, deformation and risk of the corridor structure, it is necessary to conduct experimental research with the help of a small-scale model. First, a small model with similar geometric dimensions and structure to the actual corridor should be established. Vibration sensors and deformation sensors with the same positions and numbers as those of the actual corridor should be arranged on this model.

[0146] Next, a vibration table is used to apply vibration loads of different intensities to this small model. The vibration table can generate vibration signals of various frequencies and amplitudes to simulate the dynamic response of the corridor under different external environments. By adjusting the parameters of the vibration table, the model can be subjected to vibration loads ranging from small to large.

[0147] When the model is subjected to vibration load, the monitoring system will collect data from vibration sensors and deformation sensors in real time. These data include characteristics such as amplitude, frequency, phase, peak value, as well as amplitude and modal information of structural deformation. By analyzing these data, the response characteristics of small models under different vibration conditions can be established.

[0148] These experimental data will provide support for the establishment of a corridor risk analysis model in the subsequent steps. By comparing the actual corridor monitoring data and the model experimental data, the corresponding relationship between the two can be found, laying the foundation for risk assessment.

[0149] Step S30: Use clustering algorithm to analyze the monitoring data of the actual corridor and the model, and divide the characteristic intervals of different risk levels

[0150] In order to establish a risk assessment model for corridor structures, it is first necessary to analyze and classify the monitoring data of actual corridors and model experiments. Clustering algorithms can be used to achieve this goal.

[0151] Clustering algorithm is an unsupervised learning method that can classify similar data samples into the same category. For the monitoring data of the corridor, clustering algorithm can help identify different vibration and deformation characteristic patterns and classify them into different risk levels.

[0152] Specifically, the K-means clustering algorithm or the Gaussian mixture model clustering algorithm can be used to analyze the monitoring data. These algorithms will automatically divide the data into K cluster centers, each of which represents a typical vibration or deformation feature. By analyzing the statistical characteristics of these cluster centers, such as the mean, standard deviation, extreme value, etc., the characteristic intervals of different risk levels can be determined.

[0153] For example, the clustering results can be divided into three levels: "low risk", "medium risk" and "high risk". The corresponding feature intervals can be: low risk - amplitude less than 0.5mm and frequency less than 1Hz; medium risk - amplitude between 0.5 and 2mm and frequency between 1 and 5Hz; high risk - amplitude greater than 2mm and frequency greater than 5Hz.

[0154] In this way, the corresponding relationship between the vibration and deformation characteristics of the corridor structure under different working conditions and the risk level can be established. These characteristic intervals will provide a basis for risk analysis in subsequent steps.

[0155] Step S40: Using machine learning methods, establish a mapping relationship between corridor structure vibration, deformation parameters and risk index, and obtain a corridor risk index analysis model

[0156] Based on the characteristic intervals of different risk levels obtained in step S30, a mapping relationship between the vibration and deformation parameters of the corridor structure and the risk index can be further established. A machine learning method can be used to achieve this goal.

[0157] Specifically, a dual-branch neural network structure can be used as a risk index analysis model. This network contains two parallel branches:

[0158] 1. Micro feature extraction branch:

[0159] -Multi-scale convolution layer: Use convolution kernels of different sizes (such as 1x1, 3x3, 5x5) to process input data in parallel to capture vibration and deformation features of different scales.

[0160] -Spatial Pyramid Pooling (SPP) layer: The feature map is divided into regions of different scales (such as 1x1, 2x2, 4x4), and maximum pooling is performed in each region to extract multi-scale local features.

[0161] -Attention mechanism: Channel attention and spatial attention are introduced to enable the model to adaptively focus on important local vibration and deformation features.

[0162] 2. Macro feature extraction branch:

[0163] -Deep convolutional network: Use deep networks such as ResNet or LSTM as the backbone to extract long-term dependency features of time series data.

[0164] -Temporal Convolutional Network (TCN): Use dilated convolution to expand the receptive field without increasing the number of parameters and capture contextual information over a longer time range.

[0165] -Feature pyramid: integrates features of different time scales to obtain short-term and long-term structural change features.

[0166] These two branches respectively extract the local vibration deformation characteristics (microscopic characteristics) and overall characteristics and long-term change trends (macroscopic characteristics) of the corridor structure.

[0167] In the feature fusion layer, the output features of the two branches are spliced ​​together and further fused through several fully connected layers. This can fully combine micro and macro features to generate a comprehensive feature representation.

[0168] Finally, a fully connected layer and a Softmax classification layer are used in the output layer to map the fused features to the risk index and risk level output of the corridor.

[0169] By pre-training the model experimental data obtained in step S20 and then applying it to the actual corridor monitoring data for fine-tuning, a more accurate corridor risk index analysis model can be obtained. This model can convert the vibration and deformation parameters of the corridor into a risk index, providing a basis for subsequent early warning.

[0170] Step S50: Input the actual corridor monitoring data into the risk index analysis model to calculate the real-time risk index of the corridor.

[0171] With the corridor risk index analysis model established in step S40, the real-time monitoring data of the actual corridor can be input into the model to calculate the real-time risk index of the corridor.

[0172] Specifically, the corridor vibration and deformation monitoring data obtained in step S10 are preprocessed according to the input format of the model, such as normalization, supplementation of missing values, etc. Then these data are input into the trained risk index analysis model.

[0173] Based on the input vibration and deformation characteristics, the model calculates the corresponding risk index value through forward propagation. This risk index is a continuous value that indicates the current risk level of the corridor. The higher the risk index, the greater the risk faced by the corridor.

[0174] It should be noted that in order to ensure real-time performance, a sliding time window should be set to input the monitoring data of the most recent period (such as 1 hour or 1 day) into the model for risk index calculation. This ensures that the risk index reflects the immediate status of the corridor without lagging due to the use of data from a longer period of time.

[0175] By continuously inputting real-time monitoring data into the model, the real-time risk index change trend throughout the life cycle of the corridor can be obtained, providing support for subsequent early warning and decision-making.

[0176] Step S60: Compare the real-time risk index of the corridor with the preset risk threshold. When the risk index exceeds the threshold, issue a warning signal in time.

[0177] After obtaining the real-time risk index of the corridor, it needs to be compared with the preset risk threshold to determine the current risk status.

[0178] The risk threshold can be determined based on the specific situation of the corridor, combined with engineering experience and safety standards. For example, the risk index can be divided into three levels: low risk (0-0.3), medium risk (0.3-0.7) and high risk (0.7-1.0). The corresponding risk thresholds can be set to 0.3 and 0.7.

[0179] When the real-time risk index of the corridor exceeds 0.7 (high risk threshold), it means that the corridor is facing a major safety hazard and timely measures need to be taken. At this time, the system should issue an early warning signal to notify relevant responsible personnel to take necessary disposal actions, such as restricting use, strengthening and repairing, etc.

[0180] Similarly, when the risk index is in the medium risk range (0.3-0.7), it should also be taken seriously, and measures such as strengthening monitoring and checking for hidden dangers can be taken. Only when the risk index is lower than 0.3 does it mean that the corridor is in a safe state and does not need special attention.

[0181] Through such risk threshold comparison and early warning mechanism, safety hazards of corridor structures can be discovered in a timely manner, providing support for decision makers and minimizing the risk of accidents.

[0182] Since the present invention involves some calculation processes, in order to better understand and implement the present invention, a more detailed description of the specific implementation of the present invention is given below in combination with specific formulas: Step S10: Select multiple representative monitoring points, use vibration sensors and deformation sensors to monitor the corridor structure, and obtain vibration and deformation data

[0183] In this step, it is necessary to determine representative monitoring points in the corridor structure. These monitoring points should cover key areas of the corridor, such as stress concentration points, potential weak points, etc. Finite element analysis or engineering experience can be used to determine these key monitoring points.

[0184] Assume that the corridor structure has N monitoring points in total, and use the set To represent these monitoring points. Each monitoring point p i Vibration sensors and deformation sensors are installed.

[0185] The vibration sensor is used to collect the vibration data of the corridor structure in the vertical, horizontal and torsional directions. To express the vibration condition at the tth monitoring point:

[0186]

[0187] Among them, v ix (t), v iy (t) and v iz (t) represents the components of vibration in the x, y and z directions respectively.

[0188] The deformation sensor is used to measure the static deformation of the corridor under the action of force. i (t) represents the deformation value at the i-th monitoring point.

[0189] Through the data acquisition system, the sampling frequency f s These sensors are continuously monitored to obtain vibration and deformation data representing the dynamic response characteristics of the corridor. Assuming the acquisition time is T, the vibration and deformation time series data at N monitoring points can be obtained:

[0190]

[0191]

[0192] These real-time monitoring data will provide a basis for subsequent risk analysis.

[0193] Step S20: Arrange the same monitoring points as the actual corridor on the pre-established small-scale corridor structure model, use a vibration table to vibrate the model with different intensities, and collect corresponding vibration and deformation data

[0194] In order to establish the mapping relationship between the vibration, deformation and risk of the corridor structure, it is necessary to conduct experimental research with the help of a small-scale model. Assuming that the scale of the small-scale model is one α of the actual corridor, the geometric dimensions, material properties, etc. of the model should be scaled according to the α ratio.

[0195] On this small model, the same position and number of vibration sensors and deformation sensors as those in the actual corridor are deployed. Assume that there are M monitoring points on the model, and use the set Similarly, the model vibration vector can be obtained and the deformation scalar v j (t) time series data:

[0196]

[0197]

[0198] Among them, T m Indicates the total duration of the model test.

[0199] Next, a vibration table is used to apply vibration loads of different intensities to this small model. The vibration table can generate sinusoidal vibration signals of various frequencies ω and amplitudes A to simulate the dynamic response of the corridor under different external environments. By adjusting the parameters of the vibration table, vibration loads ranging from small to large can be applied to the model.

[0200] Assume that the vibration signal of the vibration table can be expressed as:

[0201]

[0202] When the model is subjected to such vibration load, the monitoring system will collect data from vibration sensors and deformation sensors in real time. These data include characteristics such as amplitude, frequency, phase, peak value, as well as amplitude and modal information of structural deformation.

[0203] By analyzing these experimental data, the response characteristics of the small model under different vibration conditions can be established. These characteristics can be expressed by a set of statistical parameters, such as:

[0204]

[0205]

[0206] Among them, μ w and σ w Respectively represent the mean and standard deviation of the vibration response, A max and ω max Respectively represent the peak amplitude and frequency of vibration; μ v and σ v denote the mean and standard deviation of the deformation response, v max Indicates the peak value of deformation.

[0207] These experimental data will provide support for establishing a corridor risk analysis model in subsequent steps.

[0208] Step S30: Use clustering algorithm to analyze the monitoring data of the actual corridor and the model, and divide the characteristic intervals of different risk levels

[0209] In order to establish a risk assessment model for the corridor structure, it is necessary to analyze and classify the monitoring data of the actual corridor and the model experiment obtained in steps S10 and S20. A clustering algorithm can be used here to achieve this goal.

[0210] Specifically, the K-means clustering algorithm can be used to analyze the monitoring data. The goal of this algorithm is to divide N+M data samples into K clusters so that the sum of the squares of the distances from each sample to the center of the cluster to which it belongs is minimized. The cluster center μ k and the distance d from the sample to the center ik It can be expressed as follows:

[0211]

[0212]

[0213] in, represents the i-th sample (which can be a vibration vector Or the deformation scalar u i 、v j ), C k Represents the sample set contained in the kth cluster.

[0214] By optimizing the objective function, K cluster centers can be obtained. These cluster centers represent K typical vibration and deformation characteristic modes in the monitoring data.

[0215] Next, we analyze the statistical characteristics of these cluster centers, such as the mean μ k , standard deviation σ k , extreme values, etc., and divide them into different risk levels accordingly. For example, the clustering results can be divided into three levels: "low risk", "medium risk" and "high risk", and the corresponding feature intervals can be:

[0216] Low risk: μ k <μ L , σ k <σ L

[0217] Medium risk: μ L ≤μ k ≤μ H , σ L ≤σ k ≤σ H

[0218] High risk: μ k >μ H , σ k >σ H

[0219] Among them, μ L , σ L and μ H , σ H They are characteristic thresholds for low risk and high risk, respectively, and can be determined based on engineering experience and safety standards.

[0220] In this way, the corresponding relationship between the vibration and deformation characteristics of the corridor structure under different working conditions and the risk level can be established. These characteristic intervals will provide a basis for risk analysis in subsequent steps.

[0221] Step S40: Using machine learning methods, establish a mapping relationship between corridor structure vibration, deformation parameters and risk index, and obtain a corridor risk index analysis model

[0222] Based on the characteristic intervals of different risk levels obtained in step S30, a mapping relationship between the vibration and deformation parameters of the corridor structure and the risk index can be further established. Here, a dual-branch neural network structure can be used as a risk index analysis model.

[0223] The neural network contains two parallel branches: a micro-feature extraction branch and a macro-feature extraction branch.

[0224] The micro-feature extraction branch is used to extract the local vibration and deformation characteristics of the corridor structure. Specifically, it can be achieved using multi-scale convolutional layers and spatial pyramid pooling (SPP) layers:

[0225] Multi-scale convolutional layer:

[0226]

[0227] in, Indicates the convolution weights of different sizes of convolution kernels (1×1, 3×3, 5×5) used in the lth layer. is the bias, and σ(·) is the activation function (such as ReLU).

[0228] SPP layer:

[0229]

[0230] This layer divides the feature map into regions of different scales, performs maximum pooling in each region, and extracts multi-scale local features.

[0231] The macro feature extraction branch is used to extract the overall characteristics and long-term change trends of the corridor structure. This can be achieved using deep convolutional networks (such as ResNet) and temporal convolutional networks (TCN):

[0232] Deep Convolutional Networks:

[0233]

[0234]

[0235] TCN layer:

[0236]

[0237] TCN uses dilated convolution to expand the receptive field without increasing the number of parameters and capture contextual information over a longer time range.

[0238] In the feature fusion layer, micro features are and macroscopic characteristics Spliced ​​together, further fused through several fully connected layers:

[0239]

[0240]

[0241] Finally, a fully connected layer and a Softmax classification layer are used in the output layer to combine the fusion features The risk index and risk level output mapped to the corridor:

[0242]

[0243]

[0244] Among them, p i represents the probability of the ith risk level, and r is the final risk index.

[0245] By pre-training the model experimental data obtained in step S20 and then applying it to the actual corridor monitoring data for fine-tuning, a relatively accurate corridor risk index analysis model can be obtained.

[0246] Step S50: Input the monitoring data of the actual corridor into the risk index analysis model to calculate the real-time risk index of the corridor

[0247] With the corridor risk index analysis model established in step S40, the real-time monitoring data of the actual corridor obtained in step S10 can be input into the model to calculate the real-time risk index of the corridor.

[0248] Specifically, we first need to monitor the real-time data and Preprocess according to the model input format, such as normalization, missing value supplementation, etc. The preprocessed feature vector is

[0249] Then, Input it into the trained risk index analysis model and calculate the corresponding risk index value r through forward propagation:

[0250]

[0251] Among them, f θ Represents the risk index mapping function with model parameter θ.

[0252] It should be noted that in order to ensure real-time performance, a sliding time window should be set to input the monitoring data of the most recent period (such as 1 hour or 1 day) into the model for risk index calculation. This ensures that the risk index reflects the immediate status of the corridor, and will not fail to obtain results due to the use of a long period of time.

[0253] Step S60: Compare the real-time risk index of the corridor with the preset risk threshold. When the risk index exceeds the threshold, issue a warning signal in time.

[0254] After obtaining the real-time risk index r of the corridor, it needs to be compared with the preset risk threshold to determine the current risk status.

[0255] The risk threshold can be determined based on the specific situation of the corridor, combined with engineering experience and safety standards. For example, the risk index can be divided into three levels: low risk, medium risk and high risk, and the corresponding thresholds are r L and r H :

[0256] Low risk: 0 ≤ r <r L

[0257] Medium risk: r L ≤r <r H

[0258] High risk: r H ≤r≤1

[0259] When the real-time risk index r of the corridor exceeds the high risk threshold r H When the corridor is damaged, it means that the corridor is facing a major safety hazard and timely measures need to be taken. At this time, the system should issue an early warning signal to notify relevant responsible personnel to take necessary disposal actions, such as restricting use, strengthening and repairing, etc.

[0260] Warning signs can take the form of:

[0261] 1. Sound and light alarm: trigger the alarm lights and alarm speakers near the corridor to issue visual and auditory warnings.

[0262] 2. SMS notification: Send risk warning SMS to the preset corridor management personnel’s mobile phone, providing real-time risk index and recommended measures.

[0263] 3. Network push: Push risk warning information to the mobile devices of relevant departments or individuals through the network platform.

[0264] Similarly, when the risk index is in the medium risk range, it should also be taken seriously, and measures such as strengthening monitoring and regularly checking for hidden dangers can be taken. LOnly when the corridor is in a safe state does it mean that there is no need for special attention.

[0265] Through such risk threshold comparison and early warning mechanism, the safety hazards of the corridor structure can be discovered in time, support can be provided to decision makers, and the risk of accidents can be minimized.

[0266] In order to further improve the reliability and accuracy of early warning, some auxiliary means can also be considered in the early warning mechanism:

[0267] 1. Multi-sensor fusion: In addition to vibration and deformation data, other factors such as temperature, humidity, and environmental vibration can also be monitored to comprehensively analyze the safety status of the corridor.

[0268] 2. Historical data analysis: Combined with the long-term historical monitoring data of the corridor, analyze the long-term change trend of the risk index and predict possible risks in the future.

[0269] 3. Integration of expert knowledge: Invite experts in structural engineering, material mechanics, etc. to participate, and use their experience and judgment to provide support for risk assessment and early warning.

[0270] 4. Adaptive threshold: Dynamically adjust the risk threshold according to the actual use of the corridor to improve the pertinence and sensitivity of the early warning.

[0271] In general, this safety risk monitoring method for high-altitude large-scale steel structure corridors makes full use of the monitoring data of vibration sensors and deformation sensors, and establishes the mapping relationship between the vibration and deformation characteristics of the corridor structure and the risk index through small-scale model experiments, cluster analysis, deep learning and other technical means. This method can calculate the risk index of the corridor in real time, compare it with the preset threshold, and issue early warning signals in time, providing strong support for the safe operation and maintenance of the corridor.

[0272] The mathematical modeling and algorithm design of the whole method are comprehensive, taking into account the local vibration deformation characteristics and the long-term change trend of the corridor as a whole. A comprehensive risk assessment is finally obtained through feature fusion. At the same time, the experimental data of the small model is used for pre-training, and then applied to the actual monitoring data for fine-tuning, which can effectively learn the characteristics of different risk levels and improve the accuracy of the model in practical applications. In addition, auxiliary means such as expert knowledge and adaptive thresholds are introduced to further enhance the reliability and practicality of the method. In short, this is a comprehensive solution that integrates data monitoring, pattern recognition, and risk analysis, which is of great significance for improving the safety management of large corridor structures.

[0273] Specifically, the principle of the present invention is to make full use of the dynamic response data collected by vibration sensors and deformation sensors, and realize accurate identification and real-time warning of corridor structure safety risks through multi-level analysis and modeling. Specifically, it mainly includes the following technical innovations:

[0274] 1. Multi-source dynamic data acquisition: Select representative monitoring points and use vibration sensors and deformation sensors to conduct comprehensive monitoring of the corridor structure. These sensors can capture the vibration characteristics and static deformation of the corridor in multiple directions, providing basic data for subsequent risk analysis. At the same time, deploy the same monitoring points on the small model, use the vibration table to apply vibration loads of different intensities, and obtain the dynamic response data of the model under force. These test data provide support for establishing the mapping relationship between vibration, deformation and risk.

[0275] 2. Clustering-based feature interval division: An unsupervised clustering algorithm is used to analyze the monitoring data of the actual corridor and the model, and different vibration and deformation characteristic modes are automatically divided. Combined with the statistical characteristics of these characteristic modes, such as mean, standard deviation, extreme value, etc., characteristic intervals of different risk levels (such as low, medium and high risks) are determined. These characteristic intervals provide a basis for subsequent risk index analysis and can reflect the safety status of the corridor under different working conditions.

[0276] 3. Risk index modeling with dual-branch neural network: A dual-branch neural network structure is designed to simultaneously extract the local microscopic vibration deformation characteristics of the corridor and the overall macroscopic characteristics. The microscopic feature extraction branch uses multi-scale convolution and spatial pyramid pooling methods to capture the vibration and deformation information at the detail level. The macroscopic feature extraction branch uses deep convolutional networks and temporal convolutional networks to extract the dynamic response trend and long-term change characteristics of the corridor as a whole. Through the feature fusion layer, the microscopic and macroscopic features are organically combined to generate a comprehensive feature representation, which provides a powerful input for the final risk index prediction.

[0277] 4. Real-time risk warning mechanism: Input the real-time monitoring data of the actual corridor into the trained risk index analysis model to calculate the current risk index of the corridor. Compare this risk index with the preset safety threshold. Once the high risk threshold is exceeded, the system will promptly issue warning signals such as sound and light alarms and SMS notifications. At the same time, combined with historical data analysis and expert experience, the warning threshold is dynamically adjusted to improve the sensitivity and pertinence of the warning and reduce the probability of false alarms.

[0278] In general, the core logic of the method of the present invention is: through the comprehensive collection of multi-source dynamic data, cluster analysis is used to determine different risk characteristic intervals, and then the mapping relationship between vibration, deformation and risk index is established with the help of a double-branch neural network, and finally the real-time monitoring and early warning of the safety status of the corridor is realized. This method fully considers the complexity and dynamic characteristics of the corridor structure, can more accurately and timely discover safety hazards, provide reliable support for decision makers, and greatly improve the safety management level during the use of the corridor.

[0279] In order to better understand and implement the present invention, an embodiment of a specific application scenario of the present invention is given below: A modern high-speed railway station was newly built in a certain city, which includes a large steel structure corridor with a length of about 500 meters and a maximum span of 80 meters. This corridor adopts a box-shaped steel beam main truss structure, and the tower column is 60 meters high. It undertakes the important tasks of gathering and distributing the flow of people inside the station and connecting the station building. In order to ensure the safety of the corridor, the city's rail transit department decided to use the safety risk monitoring method proposed in the present invention for comprehensive detection and evaluation.

[0280] 1. Monitoring point layout and data collection

[0281] First, the rail transit department organized a team of structural engineers, material mechanics experts and inspection personnel to conduct on-site inspections of key parts of the corridor. Based on the results of finite element analysis and engineering experience, 30 representative monitoring points were identified, mainly distributed in the mid-span of the corridor, the base of the tower column, the intersection of beams and columns, and other areas prone to stress concentration or deformation.

[0282] At these monitoring points, the staff has successively deployed vibration sensors and deformation sensors. The vibration sensor uses a three-axis accelerometer, which can fully obtain the vibration response data of the corridor in the vertical, horizontal and torsional directions. The deformation sensor uses a displacement sensor to measure the static deformation of the corridor beam and column structure under force. To ensure the continuity and accuracy of the data, the sampling frequency of the sensor is set to 2Hz, which can fully capture the dynamic response characteristics of the corridor.

[0283] The data acquisition system records the monitoring data of these sensors in real time, forming the following two sets of time series data:

[0284] Vibration data

[0285] Deformation data

[0286] In which, each vibration vector It contains three components, representing the vibration response in the x, y and z directions. The monitoring time T is set to 1 week, which can preliminarily reflect the dynamic characteristics of the corridor within one week.

[0287] 2. Small model experiments and feature extraction

[0288] In order to further analyze the vibration and deformation characteristics of the corridor structure under different working conditions, the rail transit department also made a small model with a geometric size of 1 / 20 of the actual size. On this model, the same number and location of monitoring points as the actual corridor were set up, and a series of experiments were carried out on the model using a vibration table.

[0289] The vibration table can generate a sinusoidal vibration signal with an adjustable frequency between 0.1Hz and 10Hz, and a peak acceleration of up to 30m / s2. z , 3H z The model was subjected to three different vibration loads of different strengths at a frequency of 100 Hz and 5 Hz. The vibration vectors of each monitoring point on the model were collected in real time through the monitoring system. and deformation value v j (t), forming the following two sets of experimental data:

[0290] Vibration data

[0291] Deformation data

[0292] Among them, the monitoring time of the small model is T m It is set to 2 hours to fully record the dynamic response characteristics of the model under different vibration loads.

[0293] By analyzing the two sets of experimental data, the researchers extracted a series of statistical parameters that describe the characteristics of the small models in terms of vibration and deformation:

[0294] Vibration characteristics

[0295] Deformation characteristics

[0296] These characteristic parameters provide a basis for the subsequent establishment of a corridor risk analysis model.

[0297] 3. Determination of risk characteristic interval

[0298] With the monitoring data of the actual corridor and small-scale model experimental data The research team then used a clustering algorithm to analyze them and determine the characteristic intervals of different risk levels.

[0299] Specifically, we chose to use the K-means clustering algorithm to divide all monitoring data (45 samples in total) into three clusters. After iterative optimization, we finally obtained three cluster centers, representing the characteristics of the three risk levels of low, medium, and high:

[0300] Low risk cluster centers:

[0301]

[0302] Medium risk cluster center:

[0303]

[0304] High-risk cluster centers:

[0305]

[0306] Based on the characteristics of these cluster centers, the research team divided the risk levels into the following categories:

[0307] Low risk:

[0308] Medium risk:

[0309] High risk:

[0310] in, Represents the modulus of the vibration vector at the ith monitoring point. These characteristic intervals provide the basis for subsequent risk index modeling.

[0311] 4. Risk Index Modeling with Dual-branch Neural Network

[0312] With the above risk characteristic range, the research team then used a dual-branch neural network method to establish a mapping relationship between the dynamic response parameters of the corridor and the risk index.

[0313] This dual-branch network consists of two parallel feature extraction branches:

[0314] (1) Micro-feature extraction branch:

[0315] -Multi-scale convolution layer: Use convolution kernels of different sizes, such as 1x1, 3x3, and 5x53, to process the input vibration and deformation data in parallel and capture local features of different scales.

[0316] -Spatial Pyramid Pooling (SPP) layer: The feature map is divided into regions of 1x1, 2x2, and 4x43 scales, and maximum pooling is performed in each region to extract multi-scale local response features.

[0317] -Attention mechanism: Introduce channel attention and spatial attention to adaptively focus on the most important local vibration deformation features.

[0318] (2) Macro feature extraction branch:

[0319] -Deep convolutional network: ResNet-50 is used as the backbone network to extract long-term correlation features in time series data.

[0320] -Temporal Convolutional Network (TCN): Use dilated convolution to increase the receptive field and capture the overall dynamic response characteristics over a longer time range.

[0321] -Feature pyramid: integrates features of different time scales to obtain both short-term and long-term structural change information.

[0322] In the feature fusion layer, the research team spliced ​​the micro and macro features together and further fused them through three fully connected layers. This can fully combine local details and overall trends to generate a comprehensive feature representation.

[0323] Finally, a fully connected layer and a Softmax classification layer are set in the output layer to map the fused features to a risk index in the range of 0 to 1, and divide it into three risk levels: low, medium, and high.

[0324] The training process of the entire neural network is as follows:

[0325] 1) First, pre-training is performed using small model experimental data to learn the preliminary mapping relationship between vibration deformation characteristics and risk index.

[0326] 2) Then, the pre-trained model is applied to the actual corridor monitoring data for fine-tuning to further optimize the model parameters.

[0327] 3) The cross entropy loss function and Adam optimizer are used for training, and L2 regularization and dropout are introduced to prevent overfitting.

[0328] After sufficient training, the prediction accuracy of the risk index analysis model reached more than 90%, and it can convert the dynamic response characteristics of the corridor into a risk index more accurately.

[0329] 5. Real-time risk monitoring and early warning

[0330] With the above risk analysis model, the research team can use the real-time monitoring data of the actual corridor and To ensure real-time performance, a sliding time window of 1 hour is set, and the data of the last hour is fed into the model every 10 minutes to calculate the risk index.

[0331] According to the model output, we can see that the risk index of the corridor remained at around 0.2 most of the time, which was in a low-risk state. However, in some periods, the risk index had a short peak, reaching a maximum of 0.85, entering the high-risk area.

[0332] According to the risk thresholds set previously, the low risk range is 0-0.3, the medium risk range is 0.3-0.7, and the high risk range is greater than 0.7. When the risk index exceeds 0.7, it means that the corridor is facing a major safety hazard and immediate measures need to be taken. As a result, the system automatically issued an audible and visual alarm and sent a warning text message to the relevant management department to inform them of the current high-risk situation of the corridor.

Claims

1. A safety risk monitoring method for a high-altitude large steel structure corridor, characterized in that: The following steps are involved: S10, selecting a plurality of representative monitoring points, using vibration sensors and deformation sensors to monitor the corridor structure, and obtaining vibration and deformation data; S20, setting up the same monitoring points as those of the actual corridor on the pre-established small-scale corridor structure model, vibrating the model with different intensities using a vibration table, and collecting corresponding vibration and deformation data; S30, using clustering algorithm to analyze the monitoring data of the actual corridor and the model, and dividing the characteristic intervals of different risk levels; S40. Using machine learning methods, establish a mapping relationship between corridor structure vibration, deformation parameters and risk index, and obtain a corridor risk index analysis model; S50, inputting the monitoring data of the actual corridor into the corridor risk index analysis model to calculate the real-time risk index of the corridor; S60. Compare the real-time risk index of the corridor with the preset risk threshold. When the risk index exceeds the threshold, issue a warning signal in a timely manner.

2. A safety risk monitoring method for a high-altitude large steel structure corridor according to claim 1, characterized in that: The corridor risk index analysis model adopts a dual-branch neural network structure, specifically including: a microscopic feature extraction branch for extracting local vibration and deformation characteristics of the corridor structure and a macroscopic feature extraction branch for extracting the overall characteristics and long-term change trends of the corridor structure.

3. A safety risk monitoring method for a high-altitude large steel structure corridor according to claim 2, characterized in that: It also includes a feature fusion layer that concatenates micro features and macro features and performs feature fusion through several fully connected layers.

4. A safety risk monitoring method for a high-altitude large steel structure corridor according to claim 3, characterized in that: The microscopic feature extraction branch is used to extract the local vibration and deformation characteristics of the corridor structure.

5. A safety risk monitoring method for a high-altitude large steel structure corridor according to claim 4, characterized in that: The microscopic feature extraction branch includes: Multi-scale convolutional layer: Use convolution kernels of different sizes to process input data in parallel to capture vibration and deformation features of different scales; Spatial pyramid pooling layer: divides the feature map into regions of different scales, performs maximum pooling in each region, and extracts multi-scale local features; Attention mechanism: Channel attention and spatial attention are introduced to enable the corridor risk index analysis model to adaptively focus on important local vibration and deformation features.

6. A safety risk monitoring method for a high-altitude large steel structure corridor according to claim 5, characterized in that: The macro feature extraction branch is used to extract the overall features and long-term variation trends of the corridor structure.

7. A safety risk monitoring method for a high-altitude large steel structure corridor according to claim 6, characterized in that: The macro feature extraction branch includes: Deep convolutional network: uses ResNet or LSTM as the backbone to extract long-term dependency features of time series data; Temporal Convolutional Network: Use dilated convolution to expand the receptive field without increasing the number of parameters and capture contextual information over a longer time range; Feature pyramid: Integrate features of different time scales to obtain both short-term and long-term structural change features.

8. The safety risk monitoring method for a high-altitude large steel structure corridor according to claim 7 is characterized in that: The feature fusion layer is used to splice micro features and macro features together, perform feature fusion through multiple fully connected layers, and use residual connections to ensure that information can be effectively propagated in multiple fully connected layers.

9. The safety risk monitoring method for a high-altitude large steel structure corridor according to claim 1 is characterized in that: The step of using a clustering algorithm to analyze the monitoring data of the actual corridor and the model uses a K-means clustering algorithm or a Gaussian mixture model clustering.

10. The safety risk monitoring method for a high-altitude large steel structure corridor according to claim 1 is characterized in that: The scale of the small model is selected to be between 1 / 10 and 1 / 50 of the actual corridor.

Citation Information

Cited By

  • Safety monitoring method and device for steel structure

    CN119124520A

  • High and large formwork construction safety monitoring system

    CN119130765A

  • Large cantilever structure construction safety risk assessment system

    CN119180507A

  • Deformed steel structure deformation assessment method based on monitoring point data association

    CN119692208A

  • Intelligent analysis system and method for stress distribution of steel truss structure

    CN119880226A