A steel structure health monitoring and early warning method, system, terminal and medium

Abstract

The application provides a steel structure health monitoring and early warning method, system, terminal and medium, and belongs to the technical field of structure health monitoring. The method comprises the following steps: collecting acceleration, strain, inclination, temperature and displacement data; using a pre-fitted temperature-strain causal influence function to separate the temperature effect from the original strain through counterfactual reasoning, and outputting a real mechanical strain value; extracting a vibration main frequency, a vibration amplitude, an inclination change rate and a strain change rate to construct a health state feature vector, and calculating a health index through a data-driven model; constructing a pattern feature vector from the health state feature vector, a displacement change rate, a temperature change rate, a strain direction reversal frequency and a strain mutation amplitude, matching the pattern feature vector with a pre-defined structure damage pattern library, and outputting an early warning and a corresponding pattern type. The application can effectively separate the temperature effect, accurately identify the precursors of different damage patterns of the steel structure and suppress the false reports of environmental interference.

Description

Technical Field

[0001] This invention relates to the field of steel structure health monitoring technology, specifically to a method, system, terminal, and medium for steel structure health monitoring and early warning. Background Technology

[0002] Steel structures are widely used in engineering fields such as large-span stadiums, high-rise buildings, bridges, and industrial plants due to their advantages such as high strength, light weight, and good plasticity and toughness. However, during service, steel structures are susceptible to damage such as local component yielding and buckling, node connection degradation, and overall instability due to factors such as fatigue loads, environmental erosion, and accidental impacts.

[0003] Currently, steel structure health monitoring primarily utilizes sensors such as strain gauges, accelerometers, inclinometers, and displacement gauges to collect structural response data. The structural condition is assessed by analyzing changes in physical quantities such as strain, vibration, and deformation. In strain monitoring, a common temperature compensation method is based on statistical regression fitting to establish an empirical formula between temperature and strain. This involves collecting historical temperature and strain data, fitting the correlation between the two, and then subtracting the temperature term from the original strain based on the current temperature value. However, statistical regression fits the correlation between variables rather than causality. When temperature changes and structural deformation occur simultaneously, the regression model cannot distinguish the contributions of both, easily attributing the true mechanical strain incorrectly to temperature and deducting it. Furthermore, strain gauges suffer from zero-point drift and sensitivity variations during long-term monitoring, further affecting the reliability of the measurement data. In structural condition assessment, existing methods often employ an independent evaluation model for individual indicators, such as comparing peak strain, maximum displacement, and vibration amplitude with their respective preset thresholds. This approach fails to comprehensively characterize the overall health of the structure and cannot distinguish between different types of anomalies, such as localized component damage and node connection degradation. In terms of early warning decision-making, existing methods generally adopt a fixed threshold over-limit triggering mechanism, that is, any monitoring parameter exceeding the corresponding threshold will trigger an alarm. This lacks the ability to identify environmental interference, is prone to false alarms and missed alarms, and has poor early warning reliability. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides a method, system, terminal, and medium for steel structure health monitoring and early warning, which effectively improves the accuracy of steel structure strain monitoring, the reliability of health assessment, and the accuracy of early warning, and can identify abnormal mode types.

[0005] In a first aspect, the technical solution of the present invention provides a method for health monitoring and early warning of steel structures, comprising the following steps: Acceleration, strain, tilt angle, temperature and displacement data are collected. The displacement change rate is calculated based on the collected displacement data, the temperature change rate is calculated based on the collected temperature data, and the strain direction reversal frequency and strain change amplitude are obtained based on the strain data. Based on the pre-fitted causal influence function of temperature on strain, and the real-time acquired temperature and strain data, the temperature effect is separated from the original strain through counterfactual reasoning, and the true mechanical strain value is output. Vibration dominant frequency and vibration amplitude features are extracted based on the collected acceleration data, the tilt angle change rate is calculated based on the collected tilt angle data, and the strain change rate is calculated based on the actual mechanical strain value to obtain a health state feature vector; using the health state feature vector as input, a health index is calculated through a data-driven model. The health status feature vector, displacement change rate, temperature change rate, strain direction reversal frequency, and strain mutation amplitude are used to construct a mode feature vector. The mode feature vector is then matched with predefined structural failure modes, including component failure modes, node failure modes, overall instability modes, and disturbance modes. If the matching result is a component failure mode, node failure mode, or overall instability mode and the health index is lower than the warning threshold, a warning and the corresponding mode type are output.

[0006] Secondly, the technical solution of the present invention provides a steel structure health monitoring and early warning system, comprising: The data acquisition module is used to collect acceleration, strain, tilt angle, temperature and displacement data, and calculate the displacement change rate based on the collected displacement data and the temperature change rate based on the collected temperature data. The causal inference compensation module is used to separate the temperature effect from the original strain through counterfactual reasoning based on the pre-fitted causal influence function of temperature on strain, as well as the real-time acquired temperature and strain data, and output the true mechanical strain value. The health status assessment module is used to extract the vibration dominant frequency and vibration amplitude features based on the collected acceleration data, calculate the tilt angle change rate based on the collected tilt angle data, and calculate the strain change rate based on the actual mechanical strain value to obtain a health status feature vector; using the health status feature vector as input, the health index is calculated through a data-driven model. The early warning decision module is used to construct a mode feature vector from the health status feature vector, displacement change rate, temperature change rate, strain direction reversal frequency, and strain mutation amplitude. The mode feature vector is then matched with predefined structural failure modes, which include component failure modes, node failure modes, overall instability modes, and disturbance modes. If the matching result is a component failure mode, node failure mode, or overall instability mode and the health index is lower than the early warning threshold, an early warning and the corresponding mode type are output.

[0007] Thirdly, the technical solution of the present invention provides a terminal, including: The memory is used to store the steel structure health monitoring and early warning program; A processor is configured to implement the steps of the steel structure health monitoring and early warning method as described above when executing the steel structure health monitoring and early warning program.

[0008] Fourthly, the present invention provides a computer-readable storage medium storing a steel structure health monitoring and early warning program, wherein the steel structure health monitoring and early warning program, when executed by a processor, implements the steps of the steel structure health monitoring and early warning method as described in any of the above claims.

[0009] As can be seen from the above technical solutions, this application has the following advantages: By constructing a causal influence function of temperature on strain through causal inference combined with counterfactual reasoning, the interference of exogenous confounding variables such as wind speed and solar radiation is effectively removed, and the pseudo-strain caused by temperature drift is separated from the true mechanical strain of the structure. This reduces the strain monitoring error in complex field environments and can provide reliable basic monitoring data for structural health assessment, thereby improving the reliability of early warning. By integrating the dominant vibration frequency, vibration amplitude, tilt angle change rate, and strain change rate corresponding to the true mechanical strain to construct a multi-dimensional health state feature vector, and realizing multi-feature nonlinear fusion assessment through a data-driven model, the overall health state of the structure can be characterized more completely, and a quantitative evaluation of the structural health state can be achieved. By constructing a standardized structural failure mode library that includes component failure modes, node failure modes, overall instability modes, and disturbance modes, and matching and identifying the mode feature vector with the mode template, and combining it with the health index threshold for dual judgment, it can effectively distinguish between the true precursors of structural instability and false anomalies caused by environmental disturbances. At the same time, it can determine the failure mode type to which the anomaly belongs, reducing the probability of false alarms and missed alarms in early warning. Attached Figure Description

[0010] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of a method for monitoring and early warning of the health of steel structures provided in an embodiment of the present invention.

[0012] Figure 2 This is a schematic block diagram of a steel structure health monitoring and early warning system provided in an embodiment of the present invention.

[0013] Figure 3 This is a schematic diagram of the structure of a terminal provided in an embodiment of the present invention. Detailed Implementation

[0014] To make the purpose, features, and advantages of this application more apparent and understandable, specific embodiments and accompanying drawings will be used to clearly and completely describe the technical solution protected by this application. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] Unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this application and in the specification of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0016] Figure 1 This is a schematic flowchart of a steel structure health monitoring and early warning method provided in an embodiment of the present invention. Figure 1 The implementing entity can be a steel structure health monitoring and early warning system. The steel structure health monitoring and early warning method provided in this embodiment of the invention is executed by computer equipment; correspondingly, the steel structure health monitoring and early warning system runs on the computer equipment. Depending on different needs, the order of the steps in this flowchart can be changed, and some steps can be omitted.

[0017] like Figure 1 As shown, the method includes the following steps.

[0018] S1 collects acceleration, strain, tilt angle, temperature and displacement data, and calculates the displacement change rate based on the collected displacement data, the temperature change rate based on the collected temperature data, and obtains the strain direction reversal frequency and strain change amplitude based on the strain data.

[0019] S2, based on the pre-fitted causal influence function of temperature on strain, and the real-time acquired temperature and strain data, separates the temperature effect from the original strain through counterfactual reasoning and outputs the true mechanical strain value.

[0020] S3. Based on the collected acceleration data, the vibration dominant frequency and vibration amplitude features are extracted. Based on the collected tilt angle data, the tilt angle change rate is calculated. Based on the actual mechanical strain value, the strain change rate is calculated to obtain the health state feature vector. Using the health state feature vector as input, the health index is calculated through the data-driven model.

[0021] S4. Construct a mode feature vector from the health status feature vector, displacement change rate, temperature change rate, strain direction reversal frequency, and strain mutation amplitude. Match the mode feature vector with predefined structural failure modes, including component failure mode, node failure mode, overall instability mode, and disturbance mode. If the matching result is a component failure mode, node failure mode, or overall instability mode and the health index is lower than the warning threshold, then output a warning and the corresponding mode type.

[0022] As a refinement and extension of the specific implementation of the above embodiments, in order to fully explain the specific implementation process of this embodiment, the following will provide possible embodiments to describe the specific implementation of the above steps in a non-limiting manner.

[0023] In some alternative implementations, step S1 specifically includes the following steps S101 and S102.

[0024] S101: Collect the tilt angle, strain, and displacement data of the structure at the first sampling frequency, and calculate the rate of change of each parameter.

[0025] The first sampling frequency is a low-frequency sampling frequency, which results in lower power consumption and is suitable for long-term continuous monitoring. Specifically, only tilt angle, strain, and displacement data are collected at the first sampling frequency for rate of change monitoring and trigger judgment. Acceleration data is collected after switching to the second sampling frequency to further reduce power consumption.

[0026] The rates of change of the collected tilt angle, strain, and displacement data were calculated. The method for calculating the rates of change was as follows: using a sliding window approach, least-squares linear fitting was performed on the data points within each sliding window, and the slope of the fitted line was taken as the rate of change of the parameters within that window.

[0027] In addition, the strain direction reversal frequency and strain abrupt change amplitude were calculated for subsequent mode matching.

[0028] For the strain direction reversal frequency, obtain the true mechanical strain value sequence, and detect the case where the strain product of two adjacent sampling points is negative, i.e., satisfying... Each time interval is recorded as one reversal. The total number of reversals per unit time is counted, and this number is divided by the unit time length to obtain the strain direction reversal frequency. This characteristic reflects the frequency of strain direction changes in a component. When local buckling occurs, strain alternates between tension and compression, and the reversal frequency increases significantly.

[0029] For the magnitude of strain abrupt change, obtain the sequence of actual mechanical strain values ​​and calculate the absolute value of the strain difference between adjacent sampling points. Take the maximum absolute value of the difference within a unit of time. This represents the magnitude of strain abrupt change. This characteristic reflects whether there are large jumps in the strain sequence of a component. When local buckling occurs, the sudden instability of the cross-section will produce strain jumps.

[0030] S102, when the rate of change of any parameter continuously exceeds the corresponding preset silent threshold, the sampling rate is switched to a second sampling frequency higher than the first sampling frequency, and acceleration, strain, tilt angle, temperature and displacement data are collected at the second sampling frequency.

[0031] The calculated rates of change of tilt angle, strain, and displacement are compared with their respective preset silent thresholds. It should be noted that the silent thresholds are much lower than the conventional warning thresholds because they are used to capture the continuous trend of parameter changes, rather than absolute value exceeding limits.

[0032] The trigger condition for switching is that the rate of change of any parameter continuously exceeds the corresponding silent threshold for a preset duration. For example, the trigger condition is met when the rate of change of tilt angle is greater than the tilt angle silent threshold for three consecutive sliding windows, or the rate of change of strain is greater than the strain silent threshold for three consecutive sliding windows, or the rate of change of displacement is greater than the displacement silent threshold for three consecutive sliding windows.

[0033] When the trigger condition is met, the sampling frequency is switched from the first sampling frequency to the second sampling frequency. The second sampling frequency is a high-frequency sampling frequency, and acceleration, strain, tilt angle, temperature and displacement data are collected at the second sampling frequency for subsequent feature extraction and health assessment.

[0034] In some alternative implementations, step S2 is used to separate the temperature effect from the original strain by using a pre-fitted causal effect function of temperature on strain, and output the true mechanical strain value.

[0035] Strain gauges suffer from temperature drift in field monitoring; that is, temperature changes cause the strain gauge to produce thermal output (pseudo-strain), which is superimposed on the true mechanical strain, resulting in measurement errors. A common method to address this is regression fitting to subtract the temperature term. ,in, This is the strain value after temperature compensation, i.e., the strain after deducting the effect of temperature. The raw strain value is directly acquired by the strain sensor. The temperature compensation coefficient represents the change in strain caused by a unit change in temperature, and is determined by the material's coefficient of thermal expansion. This represents the change in current temperature relative to a reference temperature. However, when temperature change occurs simultaneously with actual mechanical strain, such as in a frost heave scenario where temperature drop and soil expansion occur simultaneously, the regression fitting method may incorrectly attribute some of the actual mechanical strain to temperature and subtract it, leading to distorted measurement results.

[0036] This embodiment uses regression fitting to process statistical correlation, while the relationship between real mechanical strain and temperature is essentially a causal relationship. Therefore, a causal inference method is adopted to identify the causal structure between temperature and strain by constructing a causal graph model, and then use counterfactual reasoning to separate the temperature effect, thereby improving the accuracy of the obtained real mechanical strain value.

[0037] To achieve causal inference compensation, a causal influence function is pre-fitted. This process is completed offline before or in the early stages of equipment deployment and includes the following sub-steps S201 to S203.

[0038] S201, acquire historical monitoring data, which includes time-synchronized temperature sequences, strain sequences, and exogenous variable sequences, wherein the exogenous variables are selected from wind speed, solar radiation, or a combination of both.

[0039] The selection of exogenous variables must meet the following conditions: they must have a strong correlation with temperature, such as wind speed and solar radiation directly affecting temperature; and they must not have a direct causal relationship with structural deformation, meaning that the exogenous variable can only indirectly affect strain by influencing temperature, and does not directly act on strain. Exogenous variables that meet the above conditions can be used as instrumental variables to identify the causal effect between temperature and strain.

[0040] S202, Apply conditional independence test to the historical monitoring data to determine the direction of causal relationship between variables and generate a directed acyclic graph describing the causal structure between variables. In this graph, a directed edge pointing to the temperature node is constructed with the exogenous variable as the root node; a directed edge pointing to the strain node is constructed with the temperature node as the intermediate node; and a directed edge pointing to the strain node is constructed with the unknown real mechanical strain as the latent variable.

[0041] The principle of conditional independence testing is to determine whether two variables are independent given a third variable. If the two variables remain correlated after controlling for other variables, a direct causal relationship exists; if they become independent, the correlation is a spurious correlation caused by other variables.

[0042] Based on the results of the conditional independence test, a directed acyclic graph is constructed, including: 1) Construct directed edges pointing to temperature nodes, with the exogenous variable as the root node. This structure reflects the physical fact that wind speed and solar radiation affect temperature, but are not affected by temperature.

[0043] 2) Using the temperature node as the intermediate node, construct directed edges pointing to the strain node. This structure reflects the direct effect of temperature on the strain gauge, that is, temperature changes cause the strain gauge to generate heat output.

[0044] 3) Using the unknown real mechanical strain as a hidden variable, a directed edge pointing to the strain node is constructed. This structure reflects the direct influence of the real mechanical strain (caused by structural stress or deformation) on the strain gauge output.

[0045] Through the above directed acyclic diagram, the original strain is decomposed into a component caused by temperature and a component caused by actual mechanical strain.

[0046] S203, based on the directed acyclic graph, the backdoor criterion is used to identify the causal effect of temperature on the causal relationship processing, and historical data is used to fit the causal influence function of temperature on strain. This function is used to predict the strain component caused by temperature alone based on the current temperature value.

[0047] The backdoor criterion is used in causal inference to eliminate confounding bias. In a directed acyclic graph, a confounding variable is a variable that simultaneously affects both the causal variable (temperature) and the outcome variable (strain). If confounding variables are not controlled, the correlation between temperature and strain will contain spurious components.

[0048] The backdoor criterion includes: identifying the set of parent nodes of temperature nodes as the adjustment set for the backdoor path, and eliminating spurious correlations caused by confounding variables by stratifying or weighting the adjustment set in the causal effect estimation.

[0049] Specifically, the set of parent nodes for temperature nodes is first identified as the adjustment set for the backdoor path. In a directed acyclic graph, the parent nodes of temperature nodes include exogenous variables such as wind speed and solar radiation. These variables affect both temperature and strain and are confounding variables that need to be controlled. Then, in the causal effect estimation, the adjustment set is stratified or weighted. The stratification method simply involves dividing the variables in the adjustment set into several layers according to their values, calculating the partial correlation coefficient between temperature and strain in each layer, and then averaging by layer to obtain the causal effect after removing the influence of confounding variables. The weighting method involves calculating the propensity score of each variable in the adjustment set, using the propensity score as the sample weight, constructing a pseudo-population, and fitting the relationship between temperature and strain in the pseudo-population.

[0050] By using the backdoor criterion to estimate the causal effect, spurious correlations caused by confounding variables are eliminated, and a pure causal effect of temperature on strain is obtained.

[0051] Finally, a causal function f(T) representing the effect of temperature on strain is fitted using historical data. This function describes the strain component caused solely by temperature, excluding the influence of actual mechanical strain. Fitting methods can include linear regression, polynomial regression, or Gaussian process regression. For example, linear regression can be used for materials with a constant coefficient of thermal expansion. For materials whose coefficient of thermal expansion varies with temperature, second-order polynomial regression can be used. .

[0052] In some alternative implementations, separating the temperature effect from the original strain through counterfactual reasoning and outputting the true mechanical strain value specifically includes the following steps S204 to S206, which are performed during real-time monitoring.

[0053] The idea behind counterfactual reasoning is to construct a hypothetical counterfactual scenario, namely "what would have happened if the circumstances had been different", and compare the actual observed results with the predicted results under the counterfactual scenario, thereby separating the contributions of different causes to the outcome.

[0054] In this embodiment, the actual observed original strain It consists of two parts: the strain component caused by temperature. and the actual mechanical strain caused by structural stress or deformation However, these two parts are not directly distinguishable in the observational data. Counterfactual reasoning achieves separation by answering the following question: If the current temperature is not the actual observed value... Instead, it is another reference temperature. So what is the strain component caused by temperature? By subtracting the temperature contribution under the counterfactual scenario from the original strain observed in reality, the remaining part is the true mechanical strain.

[0055] Since a causal graphical model describing the causal relationship between temperature and strain has been constructed offline, and a causal influence function f(T) of temperature on strain has been fitted, representing "the strain component caused solely by temperature, excluding interference from actual mechanical strain," the counterfactual temperature... Substituting f(T) into the equation allows us to predict the temperature contribution in counterfactual scenarios.

[0056] S204, set a counterfactual condition, which assumes that the current temperature value is equal to the reference temperature value, where the reference temperature value is the ambient temperature during device calibration, the historical average temperature for the same period, or a preset benchmark temperature.

[0057] S205 uses the counterfactual temperature value as the independent variable to input the causal influence function, and calculates the strain component corresponding to that temperature.

[0058] Counterfactual temperature value The pre-fitted causal influence function f(T) is used as the independent variable to calculate the strain component corresponding to that temperature, which is expressed as:

[0059] The calculation result represents the strain component caused solely by temperature under counterfactual conditions. Since the causal influence function f(T) was obtained offline after eliminating interference from actual mechanical strain using a backdoor criterion, therefore... It represents the strain contribution caused purely by temperature and does not include mechanical strain components.

[0060] S206, subtract the strain component predicted under the counterfactual conditions from the actual observed original strain value to obtain the true mechanical strain value.

[0061] The actual observed raw strain values Subtract the temperature strain component predicted under counterfactual conditions The actual mechanical strain value is obtained, expressed as:

[0062] It is the total strain actually output by the sensor, including both temperature and mechanical contributions. The temperature contribution that "should exist" under counterfactual scenarios is the sum of the two. The remaining part is the strain beyond the temperature contribution, which is caused by the actual stress or deformation of the structure.

[0063] In some optional implementations, step S3 first extracts key features reflecting the structural health status from the collected multi-source data. Based on the collected acceleration data, a fast Fourier transform is used to convert the time-domain acceleration signal into a frequency-domain signal, and the dominant vibration frequency and vibration amplitude features are extracted from the frequency-domain signal. The dominant vibration frequency refers to the frequency corresponding to the maximum amplitude in the frequency domain, representing the natural frequency of the structure. The dominant vibration frequency reflects the structural stiffness, and the vibration amplitude reflects the structural vibration intensity. Based on the collected tilt angle data, the tilt angle change rate is calculated using a sliding window least squares linear fitting method. Based on the actual mechanical strain value output from the causal inference compensation step, the strain change rate is calculated using the same sliding window least squares linear fitting method as the tilt angle change rate. The extracted dominant vibration frequency, vibration amplitude, tilt angle change rate, and strain change rate are combined to form a health status feature vector.

[0064] Step S3 is used to map the health status feature vector into a quantitative index representing the structural health level through a data-driven model. This step is executed in real time at the edge, with the health status feature vector as input and the health index as output. Specifically, it includes the following sub-steps S301 to S304.

[0065] In structural health monitoring, vibration characteristics (dominant frequency, amplitude) and quasi-static deformation characteristics (rate of change of tilt angle, rate of change of strain) reflect the structural state from different physical dimensions. In an optional implementation, a lightweight neural network is used as the data-driven model, leveraging the nonlinear fitting capability of the neural network to automatically learn the mapping relationship between characteristics and health status. Simultaneously, the network structure is designed to be lightweight, including a 4-node input layer, 8-node and 4-node hidden layers, and a 1-node output layer, with few parameters, making it suitable for real-time operation on edge computing devices.

[0066] S301, Obtain the health status feature vector, which includes the dominant vibration frequency, vibration amplitude, tilt angle change rate, and strain change rate.

[0067] S302, normalize the features of each dimension in the health status feature vector and map their numerical range to a preset interval.

[0068] S303, the normalized health status feature vector is input into a pre-trained lightweight neural network, which includes an input layer, at least one hidden layer and an output layer. Each neuron in the network is fully connected to the next neuron in the next layer, and a nonlinear activation function is used to introduce a nonlinear transformation.

[0069] S304, obtain the output layer result of the neural network, which is a value between 0 and 1, and serves as a health index.

[0070] A lightweight neural network is pre-trained using historical monitoring data, including feature vector samples from various health states, with each sample labeled with a real health tag. A binary cross-entropy loss function is employed, along with the Adam optimizer. Training stops when the loss function value on the validation set no longer decreases for 10 consecutive training epochs.

[0071] The dominant vibration frequency is the natural frequency of the structure, reflecting the overall stiffness of the structure. The greater the stiffness, the higher the frequency. When the steel structure experiences component section loss, node connection degradation, or overall stiffness decreases, the frequency decreases. The vibration amplitude is the structural vibration intensity, reflecting the external excitation or inherent instability of the structure. When the steel structure experiences component cracking, bolt loosening, or weld damage, the vibration response intensifies, and the amplitude increases. The tilt angle change rate is the structural tilt rate, reflecting whether the structure is undergoing continuous deformation. When the steel structure foundation experiences uneven settlement, component buckling, or overall tilting, the tilt angle changes continuously. The strain change rate is the rate of change of structural stress, reflecting whether the structural stress is accumulating rapidly. When the steel structure component enters the yielding stage, the section experiences local buckling, or stress concentration occurs in the node area, the strain increases rapidly.

[0072] In some optional implementations, the structural failure mode matching process in the early warning decision step S4 is used to compare the current monitoring data with predefined structural failure modes in the structural failure mode library to identify the structural failure type to which the current state belongs. Specifically, it includes the following steps S401 to S403.

[0073] The occurrence of steel structure failure is usually accompanied by the coordinated changes of multiple physical quantities, and different failure types have different characteristic combinations. For example, component failure (including local buckling) is characterized by a rapid increase in strain, frequent reversals in strain direction, and abrupt strain jumps, but the overall structure has not yet tilted significantly; node failure is characterized by significant changes in vibration characteristics (decreased dominant frequency, increased amplitude) and strain concentration in the node area; overall instability is characterized by a continuous increase in tilt angle and displacement, and the entire structure begins to tilt and move; environmental disturbances are characterized by only strain changes without changes in tilt angle and displacement. By identifying the degree of matching between the current characteristic combination and the characteristic templates of each failure mode, it is possible to determine which mode the current state belongs to, thus providing a basis for early warning decisions.

[0074] It should be noted that the data-driven model outputs a continuous health index, while the structural damage pattern matching outputs a discrete pattern type. The health index reflects the degree of health, and the pattern type reflects the type of abnormality.

[0075] S401, Read the predefined structural failure modes. The structural failure modes include at least component failure modes, node failure modes, overall instability modes, and disturbance modes. Each mode corresponds to a set of mode feature vector templates composed of multiple features and their matching conditions, including tilt angle change rate, displacement change rate, strain change rate, vibration dominant frequency, vibration amplitude, temperature change rate, strain direction reversal frequency, strain mutation amplitude, etc.

[0076] The collapse mode corresponds to the precursor characteristics of overall sliding. Its physical mechanism is as follows: the slope undergoes overall displacement along the sliding surface, resulting in continuous changes in the dip angle and displacement. However, since the sliding body maintains its integrity, the local strain changes are relatively small, and the vibration characteristics do not change significantly.

[0077] The feature vector template for the landslide mode consists of the following four features and their matching conditions: the first feature is the rate of change of dip angle, with the matching condition that the rate of change of dip angle is positive and continuously increasing; the second feature is the rate of change of displacement, with the matching condition that the rate of change of displacement is positive and continuously increasing; the third feature is the rate of change of strain, with the matching condition that the rate of change of strain is positive but the change amplitude is less than a preset threshold; and the fourth feature is the amplitude of the decrease in the dominant vibration frequency, with the matching condition that the amplitude of the decrease in the dominant vibration frequency is less than a first frequency threshold. The matching condition for the landslide mode is that the current feature vector simultaneously satisfies the matching conditions corresponding to the above four features.

[0078] The instability mode corresponds to the precursor characteristics of local failure. Its physical mechanism is as follows: cracks, yielding or loosening occur in local parts of the structure, leading to a rapid increase in strain, a decrease in stiffness and an increase in vibration, but it has not yet developed into overall sliding, so the changes in tilt angle and displacement are not obvious.

[0079] The feature vector template for the instability mode consists of the following five features and their matching conditions: The first feature is the strain rate of change, with the matching condition that the strain rate of change is positive and continuously increasing; the second feature is the magnitude of the decrease in the dominant vibration frequency, with the matching condition that the magnitude of the decrease in the dominant vibration frequency is greater than a second frequency threshold; the third feature is the vibration amplitude, with the matching condition that the vibration amplitude increases beyond the amplitude threshold; the fourth feature is the rate of change of tilt angle, with the matching condition that the rate of change of tilt angle is zero or less than the tilt angle threshold; and the fifth feature is the rate of change of displacement, with the matching condition that the rate of change of displacement is zero or less than the displacement threshold. The matching condition for the instability mode is that the current feature vector simultaneously satisfies the matching conditions corresponding to the above five features.

[0080] The disturbance mode corresponds to the false anomaly caused by environmental factors. Its physical mechanism is: drastic temperature changes cause the strain gauge to generate heat output (false strain), but the structure itself does not undergo real deformation.

[0081] The feature vector template for the interference mode consists of the following four features and their matching conditions: the first feature is the strain rate of change, with the matching condition that the strain rate of change is positive; the second feature is the tilt rate of change, with the matching condition that the tilt rate of change is zero or less than a tilt threshold; the third feature is the displacement rate of change, with the matching condition that the displacement rate of change is zero or less than a displacement threshold; and the fourth feature is the temperature rate of change, with the matching condition that the temperature rate of change is greater than a temperature threshold. The matching condition for the interference mode is that the current feature vector simultaneously satisfies the matching conditions corresponding to the above four features.

[0082] The component failure modes correspond to the precursory characteristics of reduced load-bearing capacity in steel structure components (steel beams, steel columns, braces, connecting plates, etc.) due to insufficient strength, fatigue cracking, section loss, or local buckling. The mode feature vector template consists of the following six features and their matching conditions: The first feature is the strain rate of change, with a matching condition of a positive value and continuous increase; the second feature is the strain direction reversal frequency, with a matching condition of being greater than a preset reversal frequency threshold; the third feature is the strain abrupt change amplitude, with a matching condition of being greater than a preset abrupt change amplitude threshold; the fourth feature is the amplitude of the decrease in the dominant vibration frequency, with a matching condition of being less than a first frequency threshold; the fifth feature is the vibration amplitude, with a matching condition of increasing beyond an amplitude threshold; and the sixth feature is the tilt angle rate of change, with a matching condition of being zero or less than a tilt angle threshold. The matching condition for the component failure mode is that the current feature vector simultaneously satisfies the matching conditions corresponding to the above six features.

[0083] The node failure mode corresponds to the precursory characteristics of bolt loosening, weld cracking, or gusset plate buckling at steel structure nodes (beam-column nodes, support nodes, splice nodes, weld connection zones, etc.). Its mode feature vector template consists of the following five features and their matching conditions: the first feature is the strain rate of change, with the matching condition being a positive value that continuously increases; the second feature is the magnitude of the decrease in the dominant vibration frequency, with the matching condition being greater than a second frequency threshold; the third feature is the vibration amplitude, with the matching condition being an increase exceeding the amplitude threshold; the fourth feature is the rate of change of tilt angle, with the matching condition being zero or less than the tilt angle threshold; and the fifth feature is the rate of change of displacement, with the matching condition being zero or less than the displacement threshold. The matching condition for the node failure mode is that the current feature vector simultaneously satisfies the matching conditions corresponding to the above five features.

[0084] The overall instability mode corresponds to the precursory characteristics of a steel structure losing its overall stability and bearing capacity. Its physical mechanism involves the degradation of the overall structural stiffness to a critical level, leading to overall collapse or overturning under its own weight or external loads. The mode feature vector template consists of the following five features and their matching conditions: the first feature is the rate of change of tilt angle, with a matching condition of a positive value and continuous increase; the second feature is the rate of change of displacement, with a matching condition of a positive value and continuous increase; the third feature is the rate of change of strain, with a matching condition of a positive value but a change amplitude less than a preset threshold; the fourth feature is the magnitude of the decrease in the dominant vibration frequency, with a matching condition of being less than a first frequency threshold; and the fifth feature is the vibration amplitude, with a matching condition of increasing beyond the amplitude threshold. The matching condition for the overall instability mode is that the current feature vector simultaneously satisfies the matching conditions corresponding to the above five features.

[0085] The interference mode corresponds to spurious anomalies caused by environmental factors. Its physical mechanism includes drastic temperature changes leading to thermal output (pseudo-strain) from strain gauges, and wind-induced vibration or electromagnetic interference causing abnormal vibration data, but the structure itself does not undergo actual deformation or damage. Its mode feature vector template consists of the following four features and their matching conditions: the first feature is the strain rate of change, with a positive matching condition; the second feature is the tilt rate of change, with a matching condition of zero or less than the tilt threshold; the third feature is the displacement rate of change, with a matching condition of zero or less than the displacement threshold; and the fourth feature is the temperature rate of change, with a matching condition of greater than the temperature threshold. The matching condition for the interference mode is that the current feature vector simultaneously satisfies the matching conditions corresponding to the above four features.

[0086] The pattern library is stored in a parameterized form, including the feature dimensions, matching conditions, and threshold parameters for each pattern.

[0087] S402, calculate the matching degree between the pattern feature vector and each pattern feature vector template. This matching degree is obtained by comparing the feature values ​​of each dimension in the pattern feature vector with the matching conditions of the corresponding template and counting the ratio of the number of features that meet the conditions to the total number of features.

[0088] The pattern feature vector includes four dimensions of the health state feature vector and six dimensions in total, namely the rate of change of tilt angle, the rate of change of displacement, the rate of change of strain, the dominant vibration frequency, the vibration amplitude, the rate of change of temperature, the frequency of strain direction reversal, and the amplitude of strain abrupt change.

[0089] The matching degree is calculated using a scoring method based on the proportion of features that satisfy the criteria. For each pattern, the feature values ​​of each dimension in the pattern feature vector are checked one by one to see if they satisfy the matching conditions of the corresponding template. The number of features that satisfy the conditions is counted, and the matching degree is calculated using the formula: Matching degree = Number of features that satisfy the conditions / Total number of features in the pattern template.

[0090] S403 uses the pattern corresponding to the maximum matching degree as the matching result.

[0091] In some optional implementations, the warning output process in step S4 is used to comprehensively determine whether to output a warning and what type of warning to output based on the pattern matching results and the health index.

[0092] Judging solely based on health indices may lead to false alarms due to sensor malfunctions or environmental interference. Judging solely based on pattern matching results may result in misjudgments due to ambiguities in the matching degree calculation. Therefore, a dual-condition approach is adopted: an alert is only issued when the pattern matching result indicates a potential disaster and the health index is below the warning threshold, ensuring the accuracy and reliability of the warning.

[0093] The foregoing has described in detail an embodiment of a method for monitoring and early warning of steel structure health. Based on the steel structure health monitoring and early warning method described in the above embodiment, this invention also provides a steel structure health monitoring and early warning system corresponding to the method.

[0094] Figure 2 This is a schematic block diagram of a steel structure health monitoring and early warning system provided in an embodiment of the present invention. In this embodiment, the steel structure health monitoring and early warning system can be divided into multiple functional modules according to the functions it performs. A module, as referred to in this invention, is a series of computer program segments that can be executed by at least one processor and perform a fixed function, and is stored in memory.

[0095] The data acquisition module is used to collect acceleration, strain, tilt angle, temperature and displacement data, and calculate the displacement change rate based on the collected displacement data, and calculate the temperature change rate based on the collected temperature data.

[0096] The causal inference compensation module is used to separate the temperature effect from the original strain through counterfactual reasoning based on the pre-fitted causal influence function of temperature on strain, as well as the real-time acquired temperature and strain data, and output the true mechanical strain value.

[0097] The health status assessment module is used to extract the vibration dominant frequency and vibration amplitude features based on the collected acceleration data, calculate the tilt angle change rate based on the collected tilt angle data, and calculate the strain change rate based on the actual mechanical strain value to obtain a health status feature vector; using the health status feature vector as input, a data-driven model is used to calculate the health index.

[0098] The early warning decision module is used to construct a pattern feature vector from the health status feature vector, displacement change rate, and temperature change rate, and match the pattern feature vector with predefined structural failure modes, including collapse mode, instability mode, and disturbance mode. If the matching result is a collapse mode or instability mode and the health index is lower than the early warning threshold, an early warning and the corresponding mode type are output.

[0099] The steel structure health monitoring and early warning system of this embodiment is used to implement the aforementioned steel structure health monitoring and early warning method. Therefore, the specific implementation of this system can be found in the embodiment section of the steel structure health monitoring and early warning method above. Thus, the specific implementation can be referred to the description of the corresponding embodiments, and will not be elaborated here.

[0100] Furthermore, since the steel structure health monitoring and early warning system in this embodiment is used to implement the aforementioned steel structure health monitoring and early warning method, its function corresponds to the function of the above method, and will not be repeated here.

[0101] Figure 3 This is a schematic diagram of a terminal 300 provided in an embodiment of the present invention, including: a processor 310, a memory 320, and a communication unit 330. The processor 310 is used to implement the process steps of the above-described steel structure health monitoring and early warning method embodiment when implementing the steel structure health monitoring and early warning program stored in the memory 320.

[0102] This invention also provides a computer storage medium, which may be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc. The computer storage medium stores a steel structure health monitoring and early warning program. When the steel structure health monitoring and early warning program is executed by a processor, it implements the process steps of the above-described steel structure health monitoring and early warning method embodiments.

[0103] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for health monitoring and early warning of steel structures, characterized in that, Includes the following steps: Acceleration, strain, tilt angle, temperature and displacement data are collected. The displacement change rate is calculated based on the collected displacement data, the temperature change rate is calculated based on the collected temperature data, and the strain direction reversal frequency and strain change amplitude are obtained based on the strain data. Based on the pre-fitted causal influence function of temperature on strain, and the real-time acquired temperature and strain data, the temperature effect is separated from the original strain through counterfactual reasoning, and the true mechanical strain value is output. Vibration dominant frequency and vibration amplitude features are extracted based on the collected acceleration data, the tilt angle change rate is calculated based on the collected tilt angle data, and the strain change rate is calculated based on the actual mechanical strain value to obtain a health state feature vector; using the health state feature vector as input, a health index is calculated through a data-driven model. The health status feature vector, displacement change rate, temperature change rate, strain direction reversal frequency, and strain mutation amplitude are used to construct a mode feature vector. The mode feature vector is then matched with predefined structural failure modes, including component failure modes, node failure modes, overall instability modes, and disturbance modes. If the matching result is a component failure mode, node failure mode, or overall instability mode and the health index is lower than the warning threshold, a warning and the corresponding mode type are output.

2. The method for health monitoring and early warning of steel structures according to claim 1, characterized in that, Data on acceleration, strain, tilt angle, temperature, and displacement are collected, specifically including: The tilt angle, strain, and displacement data of the structure are collected at the first sampling frequency, and the rate of change of each parameter is calculated. When the rate of change of any parameter continuously exceeds the corresponding preset silent threshold, the sampling rate is switched to a second sampling frequency higher than the first sampling frequency, and acceleration, strain, tilt angle, temperature, and displacement data are collected at the second sampling frequency.

3. The method for health monitoring and early warning of steel structures according to claim 1, characterized in that, The causal effect function of temperature on strain is prefitted through the following steps: Acquire historical monitoring data, which includes time-synchronized temperature sequences, strain sequences, and exogenous variable sequences, wherein the exogenous variables are selected from wind speed, solar radiation, or a combination of both. Conditional independence tests are applied to the historical monitoring data to determine the direction of causal relationships between variables, and a directed acyclic graph describing the causal structure between variables is generated. In this graph, an exogenous variable is used as the root node to construct a directed edge pointing to the temperature node; a temperature node is used as the intermediate node to construct a directed edge pointing to the strain node; and an unknown real mechanical strain is used as the latent variable to construct a directed edge pointing to the strain node. Based on the directed acyclic graph, the backdoor criterion is used to identify the causal effect of temperature on the causal relationship. Historical data is used to fit a causal influence function of temperature on strain, which is used to predict the strain component caused by temperature alone based on the current temperature value.

4. The method for health monitoring and early warning of steel structures according to claim 3, characterized in that, The backdoor criterion includes: identifying the set of parent nodes of the temperature node as the adjustment set for the backdoor path, and eliminating spurious correlations caused by confounding variables by stratifying or weighting the adjustment set in the causal effect estimation.

5. The method for health monitoring and early warning of steel structures according to claim 3, characterized in that, By using counterfactual reasoning to isolate the temperature effect from the original strain, the true mechanical strain value is output, specifically including: Set a counterfactual condition, which assumes that the current temperature value is equal to the reference temperature value, where the reference temperature value is the ambient temperature during equipment calibration, the historical average temperature for the same period, or a preset baseline temperature. By inputting the counterfactual temperature value as an independent variable into the causal influence function, the strain component corresponding to that temperature is calculated. The true mechanical strain value is obtained by subtracting the strain component predicted under the counterfactual conditions from the actual observed original strain value.

6. The method for health monitoring and early warning of steel structures according to claim 1, characterized in that, Using a health status feature vector as input, a data-driven model calculates a health index, specifically including: Obtain the health state feature vector, which includes the dominant vibration frequency, vibration amplitude, tilt angle change rate, and strain change rate; Normalize the features of each dimension in the health status feature vector and map their numerical range to a preset interval; The normalized health status feature vector is input into a pre-trained lightweight neural network, which includes an input layer, at least one hidden layer and an output layer. Each neuron in the network is fully connected to the next neuron in the network, and a non-linear activation function is used to introduce a non-linear transformation. Obtain the output layer result of the neural network, which is a value between 0 and 1, as a health index.

7. The method for health monitoring and early warning of steel structures according to claim 1, characterized in that, Matching the pattern feature vector with predefined structural damage patterns, specifically including: Read the predefined structural failure modes, which include at least component failure modes, node failure modes, overall instability modes and disturbance modes. Each mode corresponds to a set of mode feature vector templates composed of multiple features and their matching conditions, including tilt angle change rate, displacement change rate, strain change rate, vibration dominant frequency, vibration amplitude, temperature change rate, strain direction reversal frequency, strain mutation amplitude, etc. The matching degree between the pattern feature vector and each pattern feature vector template is calculated. This matching degree is obtained by comparing the feature values ​​of each dimension in the pattern feature vector with the matching conditions of the corresponding template and counting the ratio of the number of features that meet the conditions to the total number of features. The pattern corresponding to the maximum matching score is taken as the matching result.

8. A steel structure health monitoring and early warning system, characterized in that, include: The data acquisition module is used to collect acceleration, strain, tilt angle, temperature and displacement data, and calculate the displacement change rate based on the collected displacement data, calculate the temperature change rate based on the collected temperature data, and obtain the strain direction reversal frequency and strain change amplitude based on the strain data. The causal inference compensation module is used to separate the temperature effect from the original strain through counterfactual reasoning based on the pre-fitted causal influence function of temperature on strain, as well as the real-time acquired temperature and strain data, and output the true mechanical strain value. The health status assessment module is used to extract the vibration dominant frequency and vibration amplitude features based on the collected acceleration data, calculate the tilt angle change rate based on the collected tilt angle data, and calculate the strain change rate based on the actual mechanical strain value to obtain a health status feature vector; using the health status feature vector as input, the health index is calculated through a data-driven model. The early warning decision module is used to construct a mode feature vector from the health status feature vector, displacement change rate, temperature change rate, strain direction reversal frequency, and strain mutation amplitude. The mode feature vector is then matched with predefined structural failure modes, including component failure modes, node failure modes, overall instability modes, and disturbance modes. If the matching result is a component failure mode, node failure mode, or overall instability mode and the health index is lower than the early warning threshold, an early warning and the corresponding mode type are output.

9. A terminal, characterized in that, include: The memory is used to store the steel structure health monitoring and early warning program; A processor is configured to implement the steps of the steel structure health monitoring and early warning method as described in any one of claims 1 to 7 when executing the steel structure health monitoring and early warning program.

10. A computer-readable storage medium, characterized in that, The readable storage medium stores a steel structure health monitoring and early warning program, which, when executed by a processor, implements the steps of the steel structure health monitoring and early warning method as described in any one of claims 1 to 7.

Images