Building structure safety monitoring and early warning method and system

Abstract

The invention provides a building structure safety monitoring and early warning method and system. The method comprises the steps of data preliminary collection, data real-time transmission, data preprocessing, feature value extraction, series-parallel connection early warning analysis and alarm sending. According to the method, real-time monitoring and early warning are carried out by adopting multiple logical judgment methods, the problem of single fixed value early warning is solved, and full-life-cycle safety guarantee is provided for the temporary supporting structure. According to the method, through a layered architecture of dual judgment of static set threshold preliminary screening, dynamic threshold fine judgment, isolated forest preliminary screening and One-Class SVM fine judgment, full-process automation from early risk identification to emergency response is realized. The core value of the system lies in that through data-driven accurate early warning, the structure safety risk is reduced, the operation and maintenance cost is reduced, and the reliability and safety of the system are improved. The method is suitable for key construction facility monitoring scenes with high requirements for real-time performance and accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of building safety early warning technology, and in particular to a building structure safety monitoring and early warning method and system. Background Art

[0002] Temporary support structures (such as scaffolding, formwork support systems, foundation pit supports, and temporary tunnel arches) are core facilities for ensuring construction safety in areas such as building engineering, bridge construction, and underground engineering. Their stability is directly related to personnel safety, project progress, and the safety of the surrounding environment. However, due to factors such as dynamic changes in construction loads, complex geological conditions, material degradation, and external environmental interference (such as wind, rain, and vibration), temporary support structures are prone to safety hazards such as excessive stress, displacement deformation, and local instability. Traditional monitoring methods are limited and mostly rely on manual inspection or simple monitoring equipment. They are unable to effectively monitor the real-time changes in stress conditions during construction, making it easy to overlook potential safety hazards and presenting significant shortcomings.

[0003] In recent years, with breakthroughs in technologies such as edge computing and machine learning, building an intelligent and highly reliable security monitoring system has become an urgent need in the industry. Summary of the Invention

[0004] The present invention proposes a method and system for real-time monitoring and early warning using multiple logical judgment methods, aiming to solve the above-mentioned technical bottlenecks and provide temporary support structures with full life cycle safety protection from "real-time perception" to "active prevention and control".

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions.

[0006] In one aspect, the present invention provides a building structure safety monitoring and early warning method, comprising the following steps: S1. Preliminary data collection: Based on the monitoring objectives and requirements, different types of sensors are arranged at appropriate locations for preliminary data collection. The arranged sensors include: axial force sensors, stress sensors, and inclination sensors. S2. Real-time data transmission: The sensor data is transmitted in real time to the monitoring database established in the computer. The present invention can connect the sensor to RS485, thereby transmitting the sensor data in real time to the monitoring database established in the computer. S3. Data preprocessing: The sensor data in the monitoring database is called and preprocessed, including data type conversion, dynamic wavelet filtering denoising, missing value processing, 3σ principle outlier processing and data standardization. S4. Extraction of eigenvalues: Calculate and extract eigenvalues from the preprocessed data. The eigenvalues include basic eigenvalues, mechanical parameters, and time series characteristics. The basic eigenvalues include real-time load values, cumulative load changes, and load change rates. The mechanical parameters include stress, strain, and safety factor. The time series characteristics include mean, variance, and trend. Real-time load value F t , cumulative load change ∆F=F t -F0 and load change rate ∆F / ∆t, where F0 is the initial load value and ∆t is the time change; Stress σ=F / A, strain ε=σ / E, safety factor S=F 设计 / F t , where A is the cross-sectional area, E is the elastic modulus; the real-time load value F t , F 设计 is the design load; The time series features are calculated by sliding the window (window size w) to calculate the mean μ w , variance σ 2 w , and extract the long-term trend term T through the Holt-Winters model t ; S5. Series-parallel warning analysis: The extracted feature values are compared with the set thresholds. Based on the comparison and analysis results, it is determined whether to issue a tiered warning: yellow warning, orange warning, or red warning. For feature values that meet the orange warning conditions, a dynamic threshold and an unsupervised machine learning model are used to further analyze whether they meet the red warning conditions. S6: Issue an alarm: When an abnormal situation is found in the analysis, the sound and light alarm device is triggered to issue an alarm of the corresponding warning level and the abnormal information is sent to relevant personnel through remote push.

[0007] Furthermore, in step S5, the extracted characteristic value is compared with the set threshold value, and whether to perform a tiered warning is determined based on the comparison and analysis results, specifically: The characteristic values for comparison and analysis with the set threshold value include 6 items: real-time load value, stress, strain and load change rate, stress change rate, and strain change rate; among them, the real-time load value F t , cumulative load change ∆F=F t -F0 and load change rate ∆F / ∆t, where F0 is the initial load value and ∆t is the time change; stress σ=F / A, strain ε=σ / E, stress change rate ∆σ / ∆t, and strain change rate ∆ε / ∆t.

[0008] When all six characteristic values are less than 80% of the set threshold, it is determined to be normal data, no warning is issued, and monitoring continues; When at least one of the six characteristic values is less than 90% and not less than 80% of the set threshold, a yellow warning is triggered, indicating a potential risk. For example, if the real-time load value, stress, or strain is ≥80% of the set threshold, or the respective change rates ∆F / ∆t, ∆σ / ∆t, and ∆ε / ∆t are ≥0.8, a yellow warning is triggered as long as at least one of the six characteristic values is less than 90% and not less than 80% of the set threshold, even if the other characteristic values are all less than 80% of the set threshold. When at least one of the six characteristic values is less than 100% of the set threshold and not less than 90% of the set threshold, an orange warning is triggered and a dynamic warning is activated. Further analysis is performed using a combination of dynamic thresholds and unsupervised machine learning models. When at least one of the six characteristic values is not less than 100% of the set threshold, a red alert is triggered. As long as at least one of the six characteristic values is not less than 100% of the set threshold, even if the other characteristic values are less than 80% of the set threshold, a red alert is triggered; Furthermore, the dynamic threshold is set as: F 动态阈值 =μ 24h +3σ 24h Real-time calculation of the mean load μ for the past 24 hours 24h and standard deviation σ 24h .

[0009] Furthermore, the unsupervised machine learning model uses a combination of isolation forest primary screening and one-class SVM precision judgment to classify and predict eigenvalue data by learning and training historical eigenvalue data. Isolation forest primary screening is used for mutation judgment to detect sudden changes in data, while one-class SVM precision judgment is used for trend judgment to analyze the long-term trend of eigenvalue data. The present invention introduces dual logic judgment of mutation and trend to improve the accuracy of early warning. The specific initial screening of the isolation forest is as follows: set n e stimators=100, contamination=0.05, calculate the anomaly score s(x)=2 -E(h(x)) / c(n) , filter out potential abnormal samples x with s(x)>0.8 潜在异常 ; Among them, n e Stimulators and contamination are the setting parameters in the isolation forest model, s(x) is the anomaly score, x is the feature matrix, E(h(x)) is the average path length of the samples, n is the number of samples, C(n)=2H(n-1)-2(n-1) / n, where H(n-1)=ln(n-1)+0.5772; One-Class SVM precise judgment is as follows: for potential abnormal samples x 潜在异常 After normalization, the RBF kernel is used to train the model and output the accurate judgment result f(x). If f(x)=-1, it is judged as abnormal and the real abnormal sample is output.

[0010] Furthermore, in step S5, a dynamic threshold and an unsupervised machine learning model are combined to further analyze whether the red warning condition is met, specifically: When the real-time load value is greater than the dynamic threshold, the load change rate calculated based on the real abnormal sample is greater than the preset safety threshold, and the real-time load value continues to be greater than the dynamic threshold for more than the preset time, a red warning is triggered, indicating that the current load deviates from the normal range.

[0011] When the real-time load value is greater than the dynamic threshold value F t ≥μ 24h +3σ 24h And dual judgment unit, mutation judgment: load change rate ∆F / ∆t>υ 安全阈值 ; Trend judgment: When the real-time load is continuously greater than the dynamic threshold for ≥5 minutes, that is, when the dynamic warning + dual judgment unit are established at the same time, a red warning is triggered, indicating that the current load deviates from the normal range.

[0012] Isolation Forest initially screens out potential anomalies and then uses actual anomalies to calculate the rate. One-Class SVM further verifies whether the data identified by Isolation Forest is an anomaly. Isolation Forest provides a quick screening method, while One-Class SVM provides a precise analysis.

[0013] In another aspect, the present invention provides a building structure safety monitoring and early warning system, comprising: Data preliminary acquisition module: used for preliminary data acquisition, including axial force sensor, stress sensor, and inclination sensor; Data real-time transmission module: used to transmit the initially collected data to the monitoring database established in the computer in real time; Data preprocessing module: used to preprocess the data in the monitoring database, including: data type conversion, dynamic wavelet filtering denoising, missing value processing, 3σ principle outlier processing and data standardization; Eigenvalue extraction module: used to calculate and extract eigenvalues from pre-processed data. The eigenvalues include basic eigenvalues, mechanical parameters, and time series characteristics. The basic eigenvalues include real-time load value, cumulative load change, and load change rate. The mechanical parameters include stress, strain, and safety factor. The time series characteristics include mean, variance, and trend. Series-parallel warning analysis module: used to compare and analyze characteristic values with set thresholds, and determine whether to issue tiered warnings based on the comparison and analysis results: yellow warning, orange warning, and red warning; dynamic thresholds and unsupervised machine learning models are used to further analyze characteristic values that enter orange warnings to determine whether they meet red warning conditions; Alarm module: used to trigger the sound and light alarm device to issue an alarm of the corresponding warning level and send abnormal information to relevant personnel through remote push.

[0014] The threshold set in the present invention can be understood as the initial judgment threshold, which is the first warning line. Its purpose is to determine whether to directly issue the highest level red warning or enter the dynamic threshold judgment according to the magnitude of the instantaneous change of the data during the data collection process by setting the threshold; setting the dynamic threshold is to check whether the data conforms to the data fluctuation; not directly setting the threshold as the dynamic threshold is conducive to reducing computing pressure. When a dynamic warning is needed, dynamic calculation is performed after the orange warning to improve detection efficiency.

[0015] This system automates the entire process from early risk identification to emergency response through a layered architecture combining static threshold screening with dynamic threshold refinement and machine learning enhancement (Isolation Forest screening with One-Class Support Vector Machine refinement). Its core value lies in providing precise, data-driven early warnings, reducing structural safety risks, lowering operation and maintenance costs, and improving system reliability and safety. It is suitable for monitoring critical construction facilities where real-time performance and accuracy are paramount. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a structural framework diagram of the monitoring and early warning system in the present invention; Figure 2 This is a flow chart of the monitoring and early warning method of the present invention; Figure 3 This is a series-parallel warning flow chart of the monitoring and early warning method in the present invention. DETAILED DESCRIPTION

[0017] Figure 1 This is a structural framework diagram of a building structure safety monitoring and early warning system provided by an embodiment of the present invention, the system comprising: Data preliminary acquisition module 100: used for preliminary data acquisition, including axial force sensor, stress sensor, and inclination sensor; Data real-time transmission module 200: used to transmit the initially collected data to the monitoring database established in the computer in real time; Data pre-processing module 300: used to pre-process the data in the monitoring database, including: data type conversion, dynamic wavelet filtering denoising, missing value processing, 3σ principle outlier processing and data standardization; Eigenvalue extraction module 400: used to calculate and extract eigenvalues from pre-processed data. The eigenvalues include basic eigenvalues, mechanical parameters, and time series characteristics. The basic eigenvalues include real-time load value, cumulative load change, and load change rate. The mechanical parameters include stress, strain, and safety factor. The time series characteristics include mean, variance, and trend. Series-parallel warning analysis module 500: used to compare and analyze characteristic values with set thresholds, and determine whether to issue a tiered warning based on the comparison and analysis results: yellow warning, orange warning, and red warning; dynamic thresholds and unsupervised machine learning models are used to further analyze characteristic values that enter orange warning conditions to determine whether they meet red warning conditions; Alarm module 600: used to trigger the sound and light alarm device to issue an alarm corresponding to the warning level and send abnormal information to relevant personnel through remote push.

[0018] Figure 2 An embodiment of the present invention provides a flow chart of a building structure safety monitoring and early warning method, wherein the method is: S1. Preliminary data collection: Based on the monitoring objectives and requirements, different types of sensors are arranged at appropriate locations for preliminary data collection. The arranged sensors include: axial force sensors, stress sensors, and inclination sensors. S2. Real-time data transmission: The sensor data is transmitted in real time to the monitoring database established in the computer. The present invention can connect the sensor to RS485, thereby transmitting the sensor data in real time to the monitoring database established in the computer. S3. Data preprocessing: The sensor data in the monitoring database is called and preprocessed, including data type conversion, dynamic wavelet filtering denoising, missing value processing, 3σ principle outlier processing and data standardization. S4. Extraction of eigenvalues: Calculate and extract eigenvalues from the preprocessed data. The eigenvalues include basic eigenvalues, mechanical parameters, and time series characteristics. The basic eigenvalues include real-time load values, cumulative load changes, and load change rates. The mechanical parameters include stress, strain, and safety factor. The time series characteristics include mean, variance, and trend. Real-time load value F t , cumulative load change ∆F=F t-F0 and load change rate ∆F / ∆t, where F0 is the initial load value and ∆t is the time change; Stress σ=F / A, strain ε=σ / E, safety factor S=F 设计 / F t , where A is the cross-sectional area, E is the elastic modulus; the real-time load value F t , F 设计 is the design load; The time series features are calculated by sliding the window (window size w) to calculate the mean μ w , variance σ 2 w , and extract the long-term trend term T through the Holt-Winters model t ; S5. Series-parallel warning analysis: The extracted feature values are compared with the set thresholds. Based on the comparison and analysis results, it is determined whether to issue a tiered warning: yellow warning, orange warning, or red warning. For feature values that meet the orange warning conditions, a dynamic threshold and an unsupervised machine learning model are used to further analyze whether they meet the red warning conditions. The characteristic values for comparison and analysis with the set threshold value include 6 items: real-time load value, stress, strain and load change rate, stress change rate, and strain change rate; among them, the real-time load value F t , cumulative load change ∆F=F t -F0 and load change rate ∆F / ∆t, where F0 is the initial load value and ∆t is the time change; stress σ=F / A, strain ε=σ / E, stress change rate ∆σ / ∆t, and strain change rate ∆ε / ∆t.

[0019] The dynamic threshold is set as: F 动态阈值 =μ 24h +3σ 24h Real-time calculation of the mean load μ for the past 24 hours 24h and standard deviation σ 24h .

[0020] The unsupervised machine learning model uses a combination of isolation forest primary screening and one-class support vector machine (SVM) precision judgment to classify and predict eigenvalue data by learning and training historical eigenvalue data. The isolation forest primary screening is used for mutation judgment to detect sudden changes in data, while the one-class support vector machine (SVM) precision judgment is used for trend judgment to analyze the long-term trend of eigenvalue data. The present invention introduces dual logic judgment of mutation and trend to improve the accuracy of early warning. The specific initial screening of the isolation forest is as follows: set n estimators=100, contamination=0.05, calculate the anomaly score s(x)=2 -E(h(x)) / c(n) , filter out potential abnormal samples x with s(x)>0.8 潜在异常 ; Among them, n e Stimulators and contamination are the setting parameters in the isolation forest model, s(x) is the anomaly score, x is the feature matrix, E(h(x)) is the average path length of the samples, n is the number of samples, C(n)=2H(n-1)-2(n-1) / n, where H(n-1)=ln(n-1)+0.5772; One-Class SVM precise judgment is as follows: for potential abnormal samples x 潜在异常 After normalization, the RBF kernel is used to train the model and output the accurate judgment result f(x). If f(x)=-1, it is judged as abnormal and the real abnormal sample is output.

[0021] Furthermore, in step S5, a dynamic threshold and an unsupervised machine learning model are combined to further analyze whether the red warning condition is met, specifically: When the real-time load value is greater than the dynamic threshold, the load change rate calculated based on the real abnormal sample is greater than the preset safety threshold, and the real-time load value is continuously greater than the dynamic threshold for more than the preset time, a red warning is triggered, indicating that the current load deviates from the normal range. t ≥μ 24h +3σ 24h And dual judgment unit, mutation judgment: load change rate ∆F / ∆t>υ 安全阈值 ; Trend judgment: When the real-time load is continuously greater than the dynamic threshold for ≥5 minutes, that is, when the dynamic warning + dual judgment unit are established at the same time, a red warning is triggered, indicating that the current load deviates from the normal range.

[0022] S6: Issue an alarm: When an abnormal situation is found in the analysis, the sound and light alarm device is triggered to issue an alarm of the corresponding warning level and the abnormal information is sent to relevant personnel through remote push.

[0023] Assume that a scaffold detects a sudden increase in load: Isolation Forest: A sudden increase in ΔF / Δt is detected, s(x)=0.85, and it is marked as a potential anomaly.

[0024] One-Class SVM: The load is found to be continuously higher than the dynamic threshold after the sudden increase, and f(x) = -1, confirming an anomaly.

[0025] System response: A red alert is triggered, indicating possible local instability.

[0026] This serial design avoids the time-consuming SVM model run on all data, while also reducing false alarms (such as temporary wind load interference) through two-stage filtering. Experiments have shown that this combination can improve warning accuracy by approximately 15-20% compared to a single model.

[0027] like Figure 3 As shown, it is a series-parallel warning flow chart of the safety monitoring and warning method of the temporary support structure of the building; When all six characteristic values are less than 80% of the set threshold, it is determined to be normal data, no warning is issued, and monitoring continues; When at least one of the six characteristic values is less than 90% and not less than 80% of the set threshold, a yellow warning is triggered, indicating a potential risk. For example, if the real-time load value, stress, or strain is ≥80% of the set threshold, or the respective change rates ∆F / ∆t, ∆σ / ∆t, and ∆ε / ∆t are ≥0.8, a yellow warning is triggered as long as at least one of the six characteristic values is less than 90% and not less than 80% of the set threshold, even if the other characteristic values are all less than 80% of the set threshold. When at least one of the six characteristic values is less than 100% of the set threshold and not less than 90% of the set threshold, an orange warning is triggered and a dynamic warning is activated. Further analysis is performed using a combination of dynamic thresholds and unsupervised machine learning models. A red alert is triggered when at least one of the six eigenvalues is not less than 100% of the set threshold. A red alert is triggered as long as at least one of the six eigenvalues is not less than 100% of the set threshold, even if the other eigenvalues are less than 80% of the set threshold.

Claims

1. A building structure safety monitoring and early warning method, characterized in that: The steps include: S1. Preliminary data collection: According to the monitoring objectives and monitoring requirements, different types of sensors are arranged at appropriate locations for the preliminary collection of sensor data; S2, real-time data transmission: the sensor data is transmitted to the monitoring database established in the computer in real time; S3. Data preprocessing: The sensor data in the monitoring database is called and preprocessed, including data type conversion, dynamic wavelet filtering denoising, missing value processing, 3σ principle outlier processing and data standardization; S4. Extraction of eigenvalues: Calculate and extract eigenvalues from the preprocessed data. The eigenvalues include basic eigenvalues, mechanical parameters, and time series characteristics. The basic eigenvalues include real-time load values, cumulative load changes, and load change rates. The mechanical parameters include stress, strain, and safety factor. The time series characteristics include mean, variance, and trend. S5. Series-parallel warning analysis: The extracted feature values are compared with the set thresholds. Based on the comparison and analysis results, it is determined whether to issue a tiered warning: yellow warning, orange warning, or red warning. For feature values that meet the orange warning conditions, a dynamic threshold and an unsupervised machine learning model are used to further analyze whether they meet the red warning conditions. S6: Issue an alarm: When an abnormal situation is found in the analysis, the sound and light alarm device is triggered to issue an alarm of the corresponding warning level and send the abnormal information to relevant personnel through remote push.

2. The building structure safety monitoring and early warning method according to claim 1, characterized in that: In step S5, the extracted characteristic value is compared with the set threshold value, and whether to perform a tiered warning is determined based on the comparison and analysis results. Specifically, The characteristic values compared with the set thresholds include 6 items: real-time load value, stress, strain and load change rate, stress change rate, and strain change rate; When all six characteristic values are less than 80% of the set threshold, it is determined to be normal data, no warning is issued, and monitoring continues; When at least one of the six characteristic values is less than 90% of the set threshold and not less than 80% of the set threshold, a yellow warning is triggered, indicating potential risks; When at least one of the six characteristic values is less than 100% of the set threshold and not less than 90% of the set threshold, an orange warning is triggered and a dynamic warning is activated. Further analysis is performed using a combination of dynamic thresholds and unsupervised machine learning models. When at least one of the six characteristic values is not less than 100% of the set threshold, a red alert is triggered.

3. The building structure safety monitoring and early warning method according to claim 2 is characterized in that: The dynamic threshold is set as: F 动态阈值 =μ 24h +3σ 24h Real-time calculation of the mean load μ for the past 24 hours 24h and standard deviation σ 24h .

4. The building structure safety monitoring and early warning method according to claim 3 is characterized in that: The unsupervised machine learning model uses a combination of isolation forest primary screening and one-class SVM precision judgment to classify and predict eigenvalue data by learning and training historical eigenvalue data. The isolation forest primary screening is used for mutation judgment and sudden changes in data, while the one-class SVM precision judgment is used for trend judgment and analysis of long-term changes in eigenvalue data. The specific initial screening of the isolation forest is as follows: set n e stimators=100, contamination=0.05, calculate the anomaly score s(x)=2 -E(h(x)) / c(n) , filter out potential abnormal samples x with s(x)>0.8 潜在异常 ; Among them, n e Stimulators and contamination are the setting parameters in the isolation forest model, s(x) is the anomaly score, x is the feature matrix, E(h(x)) is the average path length of the samples, n is the number of samples, C(n)=2H(n-1)-2(n-1) / n, where H(n-1)=ln(n-1)+0.5772; One-Class SVM precise judgment is as follows: for potential abnormal samples x 潜在异常 After normalization, the RBF kernel is used to train the model and output the accurate judgment result f(x). If f(x)=-1, it is judged as abnormal and the real abnormal sample is output.

5. The building structure safety monitoring and early warning method according to claim 4 is characterized in that: In step S5, a dynamic threshold and an unsupervised machine learning model are combined to further analyze whether the red warning condition is met, specifically: When the real-time load value is greater than the dynamic threshold, the load change rate calculated based on the real abnormal sample is greater than the preset safety threshold, and the real-time load value continues to be greater than the dynamic threshold for more than the preset time, a red warning is triggered, indicating that the current load deviates from the normal range.

6. A building structure safety monitoring and early warning system, characterized in that: include: Data preliminary acquisition module: used for preliminary data acquisition, including axial force sensor, stress sensor, and inclination sensor; Data real-time transmission module: used to transmit the initially collected data to the monitoring database established in the computer in real time; Data preprocessing module: used to preprocess the data in the monitoring database, including: data type conversion, dynamic wavelet filtering denoising, missing value processing, 3σ principle outlier processing and data standardization; Eigenvalue extraction module: used to calculate and extract eigenvalues from pre-processed data. The eigenvalues include basic eigenvalues, mechanical parameters, and time series characteristics. The basic eigenvalues include real-time load value, cumulative load change, and load change rate. The mechanical parameters include stress, strain, and safety factor. The time series characteristics include mean, variance, and trend. Series-parallel warning analysis module: used to compare and analyze characteristic values with set thresholds, and determine whether to issue tiered warnings based on the comparison and analysis results: yellow warning, orange warning, and red warning; dynamic thresholds and unsupervised machine learning models are used to further analyze characteristic values that enter orange warnings to determine whether they meet red warning conditions; Alarm module: used to trigger the sound and light alarm device to issue an alarm of the corresponding warning level and send abnormal information to relevant personnel through remote push.

Images