A method, system and medium for monitoring safety risks of large-scale high-altitude steel structure corridors

Abstract

This application relates to a method, system, and medium for safety risk monitoring of large-scale steel structure corridors at high altitudes, belonging to the field of civil engineering structural health monitoring technology. The monitoring method includes: collecting raw monitoring data through a multi-source sensor group deployed at each node of the steel structure corridor; performing spatiotemporal alignment processing on the raw monitoring data to generate a sensor data matrix; performing dynamic noise suppression on the sensor data matrix based on an environmental noise transfer function model, outputting a clean signal matrix and damage characteristic frequency band identifiers; calculating thermal stress sensitivity factors, extracting multi-physics coupling features associated with damage characteristic frequency band identifiers from the clean signal matrix, fusing them to generate a high-dimensional damage characteristic tensor, and inputting it into a pre-constructed digital twin risk assessment model, outputting assessment results including risk level labels and damage location coordinates; and generating corresponding equipment control commands based on matching control strategies according to the risk level labels. This application can improve the scientificity and accuracy of risk decision-making.

Description

Technical Field

[0001] This application relates to the field of civil engineering structural health monitoring technology, and in particular to a method, system and medium for monitoring the safety risks of a large high-altitude steel structure connecting corridor. Background Technology

[0002] With the acceleration of urbanization and the continuous emergence of high-rise building complexes, large-span, high-altitude cantilevered steel structure corridors are widely used in commercial complexes, transportation hubs, and public buildings due to their excellent space utilization efficiency and aesthetics. However, these structures are exposed to complex external environments (such as wind loads, temperature changes, and vibration disturbances) for extended periods, making them susceptible to various potential threats such as fatigue damage, loose connections, and local buckling. If these threats are not detected and addressed in a timely manner, they may lead to serious safety accidents. Therefore, how to achieve comprehensive perception, accurate identification, and efficient response to these structures during their service life has become a research hotspot and technical challenge in the field of civil engineering structural safety monitoring.

[0003] Currently, common structural health monitoring systems typically rely on sensors (such as strain gauges or accelerometers) for local condition sensing, and use manual inspections or periodic checks to assist in judging the structural condition. However, when dealing with high-altitude, large-scale, multi-degree-of-freedom steel structure corridors, the actual monitoring signals from sensors are easily contaminated by environmental noise, and most systems lack in-depth feature mining capabilities after data acquisition, making it difficult to transform the raw signals into high-dimensional damage indicators that can be used for structural performance assessment, thus affecting the scientific validity and accuracy of subsequent risk decisions. Summary of the Invention

[0004] To improve the scientific nature and accuracy of risk decision-making, this application provides a method, system, and medium for monitoring the safety risks of large-scale steel structure connecting corridors at high altitudes.

[0005] Firstly, this application provides a method for monitoring the safety risks of large-scale steel structure connecting corridors at high altitudes, employing the following technical solution:

[0006] A method for monitoring safety risks of large high-altitude steel structure connecting corridors, the monitoring method comprising:

[0007] Raw monitoring data is collected by a multi-source sensor array deployed at each node of the steel structure corridor; the raw monitoring data includes strain signals, vibration signals, acoustic emission signals, and temperature signals.

[0008] The original monitoring data is spatiotemporally aligned to generate a time-synchronized sensor data matrix;

[0009] Based on a pre-trained environmental noise transfer function model, dynamic noise suppression is performed on the sensor data matrix, and the noise-reduced clean signal matrix and damage feature frequency band identifiers are output.

[0010] The thermal stress sensitivity factor is calculated based on the strain signal and temperature signal. Multi-physics coupling features associated with the damage feature frequency band identifier are extracted from the pure signal matrix and fused to generate a high-dimensional damage feature tensor.

[0011] The high-dimensional damage feature tensor is input into a pre-constructed digital twin risk assessment model. The digital twin risk assessment model refines the finite element mesh locally based on the feature values ​​and updates the material elastic modulus parameters, outputting an assessment result that includes a risk level label and damage location coordinates. The feature values ​​include the spatial strain gradient values ​​calculated based on the high-dimensional damage feature tensor.

[0012] The risk level label is used to match the control strategy and generate the corresponding equipment control command.

[0013] By adopting the above technical solutions, accurate identification and dynamic early warning of safety risks in large-scale steel structure corridors at high altitudes have been achieved, significantly improving the intelligence level and response efficiency of health monitoring for complex structures. This solution employs spatiotemporally aligned multi-source data fusion technology, combined with signal processing algorithms such as wavelet packet entropy, empirical mode decomposition, and matching pursuit, effectively solving the problems of severe environmental noise interference and weak, difficult-to-distinguish damage signals. By establishing a high-dimensional damage feature tensor containing multi-physics field characteristics such as strain-temperature coupling derivatives and frequency domain energy entropy, quantitative characterization of key damage mechanisms such as structural thermal stress effects and fatigue crack propagation has been achieved. A three-level decision tree risk assessment model based on digital twin technology can achieve real-time response while ensuring computational accuracy. Local mesh refinement and dynamic correction of material parameters ensure a high degree of consistency between simulation results and actual conditions. The hierarchical response mechanism ensures timely risk response while avoiding unnecessary resource waste. This application can effectively address the challenges posed by structural complexity, environmental interference, and uncertainty, providing a scientific and reliable monitoring method for the safe operation and maintenance of critical infrastructure such as large-scale steel structure corridors.

[0014] Optionally, the step of dynamically suppressing noise in the sensor data matrix based on a pre-trained environmental noise transfer function model and outputting a denoised clean signal matrix and damage feature frequency band identifiers includes:

[0015] Acquire a time-synchronized sensor data matrix; the sensor data matrix includes vibration signals, acoustic emission signals, wind speed, and temperature and humidity data;

[0016] Load the pre-trained environmental noise transfer function model;

[0017] Perform wavelet packet decomposition on the vibration signal and calculate the wavelet packet energy entropy value of each sub-band;

[0018] Based on the energy entropy value, sub-band components with entropy values ​​lower than a set threshold are selected and reconstructed into a primary noise-reduced signal;

[0019] Empirical mode decomposition is performed on the primary noise-reduced signal to extract intrinsic mode function components;

[0020] Based on the environmental noise transfer function model, the intrinsic mode function components related to wind speed are filtered out to generate a pure vibration signal;

[0021] A matching tracking algorithm is executed on the acoustic emission signal. When the temperature and humidity data exceed the preset temperature and humidity threshold, atoms in the preset rain noise frequency band are excluded, and feature atoms that match the crack waveform are extracted as crack feature waveform atoms.

[0022] By fusing the pure vibration signal with the crack characteristic waveform atoms, a pure signal matrix is ​​generated;

[0023] Calculate the effective frequency band boundary of the crack characteristic waveform atoms and output the damage characteristic frequency band boundary identifier.

[0024] By adopting the above technical solution, accurate processing of structural health monitoring signals under complex environmental conditions has been achieved. This solution not only considers the physical mechanism modeling of environmental interferences such as wind-induced noise, but also combines advanced signal processing techniques based on information theory and sparse representation. Through multi-dimensional feature extraction and noise suppression, it significantly improves the accuracy and reliability of structural damage identification, providing scientific and reliable technical support for the safety monitoring of large steel structure facilities.

[0025] Optionally, the steps for establishing the environmental noise transfer function model include:

[0026] Collect background noise data during historical periods of unstructured load;

[0027] Simultaneously record wind speed data as the independent variable;

[0028] Calculate the wind speed autopower spectrum P xx (f) and wind speed-noise cross power spectrum P xy (f);

[0029] The frequency domain transfer function is calculated using the cross-power spectral density, and the formula is: H(f) = P xy (f) / P xx (f).

[0030] By adopting the above technical solution, a frequency domain model that can accurately describe the wind speed-noise relationship was constructed, providing a reliable theoretical basis and calculation basis for subsequent dynamic noise suppression.

[0031] Optionally, the steps of calculating the thermal stress sensitivity factor based on the strain signal and temperature signal, extracting the multi-physics coupling features associated with the damage feature frequency band identifier in the pure signal matrix, and fusing them to generate a high-dimensional damage feature tensor include:

[0032] Receive a clean signal matrix and damage characteristic frequency band identifiers; the clean signal matrix includes clean vibration signals and crack characteristic waveform atoms;

[0033] Simultaneously acquire spatiotemporally aligned raw strain and temperature signals;

[0034] Based on the strain and temperature signals, the partial derivative of strain with respect to temperature is calculated within a preset time window to obtain the thermal stress sensitivity factor.

[0035] Extract the energy distribution of the crack feature waveform atoms in the time domain waveform, and calculate the crack energy entropy value within the damage feature frequency band identification range;

[0036] Perform frequency domain transformation on the pure vibration signal to extract the vibration frequency domain features within a preset characteristic frequency band;

[0037] The thermal stress sensitivity factor, crack energy entropy value and vibration frequency domain features are fused by a 1D-CNN neural network to output a fused high-dimensional damage feature tensor.

[0038] By adopting the above technical solutions, a complete processing chain from multi-physics field signal acquisition to high-dimensional feature fusion was constructed, realizing the refined identification and characterization of structural damage. By fully considering multi-dimensional information such as thermo-mechanical coupling effect, time-frequency localization characteristics and frequency domain dynamic characteristics, a damage identification method with engineering practical value was formed, improving the accuracy and reliability of the structural health monitoring system.

[0039] Optionally, the step of inputting the high-dimensional damage feature tensor into a pre-constructed digital twin risk assessment model, wherein the digital twin risk assessment model performs local finite element mesh refinement based on the feature values ​​and updates the material elastic modulus parameters, and outputs an assessment result including risk level labels and damage location coordinates, includes:

[0040] Receive the high-dimensional damage feature tensor, including thermal stress sensitivity factor, crack energy entropy value and vibration frequency domain features;

[0041] Calculate the spatial strain gradient value based on the aforementioned thermal stress sensitivity factor;

[0042] When the spatial strain gradient value exceeds a preset threshold, a local refinement operation of the finite element mesh is triggered in the pre-constructed digital twin risk assessment model;

[0043] Call the temperature-elastic modulus mapping table in the pre-stored material constitutive relation library, and dynamically update the elastic modulus parameters based on real-time temperature data;

[0044] Input the crack energy entropy value into the crack propagation rate model to calculate the critical damage index;

[0045] Perform finite element simulation on the refined mesh model after updating the elastic modulus parameters, and output the simulated stress field;

[0046] Based on preset physical rule thresholds, the thermal stress sensitivity factor and vibration frequency domain characteristics are compared to obtain the first-level decision result.

[0047] The vibration frequency domain features are input into a pre-trained LSTM time series model, and the residual between the predicted value and the actual value is calculated to obtain the secondary decision result.

[0048] The modal matching degree between the simulated stress field and the vibration frequency domain characteristics is calculated to obtain the three-level decision results;

[0049] The preset decision weight ratio is adjusted according to the critical damage index, and the first-level decision result, second-level decision result and third-level decision result are weighted and fused to generate a risk level label;

[0050] The peak coordinates in the simulated stress field are extracted as the damage location coordinates.

[0051] The assessment results are obtained by combining the risk level label and the coordinates of the damage location.

[0052] By adopting the above technical solutions, a digital twin risk assessment system with adaptability, dynamism and high precision was constructed. This model can not only achieve comprehensive perception and accurate positioning of structural damage status, but also dynamically adjust the assessment strategy according to different damage development stages, which significantly improves the timeliness and accuracy of risk warning.

[0053] Optionally, after the step of outputting assessment results including risk level labels and damage location coordinates, the following may also be included:

[0054] Extract the modal confidence factor deviation between the measured vibration frequency domain features at the damage location coordinates and the simulated stress field under the same spatial coordinates;

[0055] When the modal confidence factor deviation value exceeds the tolerance threshold for a preset number of consecutive times, the constitutive parameter inversion engine is triggered.

[0056] Based on the covariance matrix adaptive evolution algorithm, the temperature-elastic modulus mapping table in the material constitutive relation library is iteratively optimized.

[0057] The optimized temperature-elastic modulus mapping table is updated to the parameter storage area of ​​the digital twin risk assessment model, and a version iteration log is generated.

[0058] By adopting the above technical solutions, a complete closed-loop control system from model deviation detection to parameter adaptive optimization is constructed, realizing intelligent maintenance and dynamic updating of the digital twin risk assessment model. The system first quantifies the model's prediction accuracy through modal confidence factor deviation analysis. When persistent deviations are detected, a constitutive parameter inversion optimization mechanism based on the CMA-ES algorithm is initiated. Finally, version management achieves accurate updates of model parameters, ensuring that the digital twin system maintains long-term stable prediction performance in complex engineering environments, providing reliable technical support for structural health monitoring and fault early warning.

[0059] Secondly, this application provides a safety risk monitoring system for high-altitude large steel structure connecting corridors, which adopts the following technical solution:

[0060] A safety risk monitoring system for a large high-altitude steel structure connecting corridor, the monitoring system comprising:

[0061] The data acquisition module is used to collect raw monitoring data through a multi-source sensor group deployed at each node of the steel structure corridor; the raw monitoring data includes strain signals, vibration signals, acoustic emission signals and temperature signals;

[0062] The data processing module is used to perform spatiotemporal alignment processing on the raw monitoring data to generate a time-synchronized sensor data matrix;

[0063] The noise reduction module is used to perform dynamic noise suppression on the sensor data matrix based on a pre-trained environmental noise transfer function model, and output the noise-reduced clean signal matrix and damage feature frequency band identifiers.

[0064] The feature tensor generation module is used to calculate the thermal stress sensitivity factor based on the strain signal and temperature signal, extract the multi-physics coupling features associated with the damage feature frequency band identifier in the pure signal matrix, and fuse them to generate a high-dimensional damage feature tensor.

[0065] The risk assessment module is used to input the high-dimensional damage feature tensor into a pre-constructed digital twin risk assessment model. The digital twin risk assessment model performs local refinement of the finite element mesh and updates the material elastic modulus parameters based on the feature values, and outputs an assessment result containing risk level labels and damage location coordinates. The feature values ​​include spatial strain gradient values ​​calculated based on the high-dimensional damage feature tensor.

[0066] The equipment control module is used to match control strategies based on the risk level labels and generate corresponding equipment control commands.

[0067] Optionally, the monitoring system further includes:

[0068] The measured data acquisition module is used to acquire the measured vibration frequency domain characteristics at the coordinates of the damage location;

[0069] The deviation calculation module is used to calculate the modal confidence factor deviation between the measured vibration frequency domain characteristics and the simulated stress field under the same spatial coordinates.

[0070] The over-limit triggering module is used to trigger the constitutive parameter inversion engine when the modal confidence factor deviation value exceeds the tolerance threshold for a preset number of consecutive times.

[0071] The iterative optimization module is used to iteratively optimize the temperature-elastic modulus mapping table in the material constitutive relation library based on the covariance matrix adaptive evolution algorithm.

[0072] The iteration log generation module is used to update the optimized temperature-elastic modulus mapping table to the parameter storage area of ​​the digital twin risk assessment model and generate a version iteration log.

[0073] Thirdly, this application provides a computer-readable storage medium, which adopts the following technical solution:

[0074] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as in any of the methods in the first aspect.

[0075] In summary, this application has at least one of the following beneficial technical effects: This application can effectively address the challenges brought about by structural complexity, environmental interference and uncertainty, and provides a scientific and reliable monitoring method for the safe operation and maintenance of key infrastructure such as large steel structure corridors. Attached Figure Description

[0076] Figure 1 This is a schematic diagram of the first process of a safety risk monitoring method for a high-altitude large steel structure connecting corridor according to one embodiment of this application.

[0077] Figure 2 This is a schematic diagram of the second process of a safety risk monitoring method for a high-altitude large steel structure connecting corridor according to one embodiment of this application.

[0078] Figure 3 This is a schematic diagram of the third process of a safety risk monitoring method for a large high-altitude steel structure connecting corridor according to one embodiment of this application.

[0079] Figure 4 This is a schematic diagram of the fourth process of a safety risk monitoring method for a large high-altitude steel structure connecting corridor according to one embodiment of this application.

[0080] Figure 5This is a schematic diagram of the fifth process of a safety risk monitoring method for a high-altitude large steel structure connecting corridor according to one embodiment of this application.

[0081] Figure 6 This is a schematic diagram of the sixth process of a safety risk monitoring method for a large high-altitude steel structure connecting corridor according to one embodiment of this application. Detailed Implementation

[0082] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figures 1-6 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0083] This application discloses a method for monitoring the safety risks of a large high-altitude steel structure connecting corridor.

[0084] Reference Figure 1 A method for monitoring safety risks of large high-altitude steel structure connecting corridors, the monitoring method includes:

[0085] Step S101: Collect raw monitoring data by deploying a multi-source sensor group at each node of the steel structure corridor;

[0086] The raw monitoring data includes strain signals, vibration signals, acoustic emission signals, and temperature signals. During the data acquisition phase, raw monitoring data is obtained through a multi-source sensor array deployed at key nodes of the steel structure corridor (such as support points, connectors, and cantilever ends). These sensors include, but are not limited to, strain gauges, accelerometers, temperature and humidity sensors, wind speed sensors, and acoustic emission sensors.

[0087] Specifically, strain signals reflect the degree of structural deformation under external loads; vibration signals reflect the dynamic response characteristics of the structure under environmental excitation or human disturbance; and acoustic emission signals are used to capture the transient elastic wave energy released during the propagation of microcracks within the material. These three types of signals characterize the structural state from different dimensions, forming the basic data source for subsequent analysis. This multi-source heterogeneous data acquisition method not only improves the comprehensiveness of monitoring but also provides rich information support for subsequent data fusion and damage identification.

[0088] Step S102: Perform spatiotemporal alignment processing on the raw monitoring data to generate a time-synchronized sensor data matrix;

[0089] Since different types of sensors may have different sampling frequencies, transmission delays, and timestamp accuracies, failure to correct these will lead to time series misalignment and affect the accuracy of subsequent analysis.

[0090] Therefore, interpolation, resampling, and timestamp alignment can be used to unify the time reference of each sensor channel, and a unified data matrix structure can be constructed based on spatial location relationships. This process essentially involves building a four-dimensional data cube: a time axis, a sensor type axis, a spatial node axis, and a physical dimension axis. This matrix representation provides a structured input format for subsequent noise suppression and feature extraction, and also facilitates parallel computing and data flow scheduling in subsequent processing modules.

[0091] Step S103: Dynamic noise suppression is performed on the sensor data matrix based on the pre-trained environmental noise transfer function model, and the noise-reduced clean signal matrix and damage feature frequency band identifiers are output.

[0092] Among them, the environmental noise transfer function model is usually established through previous experiments or historical data analysis, and is used to describe the propagation path of environmental noise in a specific structure and its influence mechanism on the signals of various sensors.

[0093] Based on this, differentiated denoising strategies were adopted for different types of signals: For vibration signals, a hybrid denoising algorithm was formed by combining wavelet packet entropy analysis and empirical mode decomposition (EMD). Wavelet packet transform can decompose the signal into multiple frequency bands, and the main energy distribution region is identified by calculating the wavelet packet entropy of each sub-band, thereby retaining effective components and filtering out redundant noise; EMD adaptively decomposes non-stationary signals into several intrinsic mode functions (IMFs), and then further filters out the true vibration components through Hilbert transform. The combination of the two can enhance the signal-to-noise ratio while preserving the original vibration characteristics.

[0094] For acoustic emission signals, a matching pursuit algorithm is used to separate the characteristic frequency bands related to crack propagation. Matching pursuit is a sparse approximation method based on an atom library. It iteratively selects the optimal atom and matches it with the signal residual to gradually reconstruct the main components containing crack activity information and determine their corresponding set of frequency boundary values, i.e., the damage characteristic frequency band identifiers. This identifier serves as a key parameter for subsequent feature extraction, limiting the analysis range and improving detection sensitivity.

[0095] Step S104: Calculate the thermal stress sensitivity factor based on the strain signal and temperature signal, extract the multi-physics coupling features associated with the damage feature frequency band identifier in the pure signal matrix, and fuse them to generate a high-dimensional damage feature tensor.

[0096] Specifically, this stage aims to extract high-dimensional features closely related to structural damage from the pure signal matrix, forming a high-dimensional damage feature tensor that can be used for risk assessment.

[0097] First, the partial derivatives of the strain data with respect to temperature changes are calculated to obtain the thermal stress sensitivity factor. Steel structures experience significant thermal expansion and contraction under diurnal temperature variations or climate changes, which in turn induces additional stress. By solving the partial differential relationship between strain and temperature, the degree of influence of thermal stress on structural stability can be quantified, which is particularly useful for identifying problems such as local buckling or loosening of connections caused by excessive temperature gradients.

[0098] It should be noted that the strain signal was not subjected to dynamic noise suppression at the front end because its core interference is temperature drift rather than environmental noise, which is eliminated through subsequent physical model-driven coupled derivative calculations. Vibration and acoustic emission signals, due to frequency domain aliasing and transient volatility, require dedicated suppression algorithms. This layered processing strategy improves data processing efficiency.

[0099] Secondly, a frequency domain transformation is performed on the pure vibration signal to extract the vibration frequency domain features within a preset characteristic frequency band. At the same time, the crack energy entropy value (used to characterize the crack activity intensity) of the acoustic emission signal within the damage characteristic frequency band is extracted. The energy entropy reflects the uniformity of the energy distribution of the signal within a specified frequency band. If the energy in a certain frequency band is concentrated and continues to rise, it often indicates that the crack is expanding or fatigue damage is aggravated.

[0100] Finally, vibration frequency domain features, thermal stress sensitivity factors, and energy entropy values ​​can be fused using a 1D-CNN neural network. 1D-CNN has excellent temporal modeling capabilities, automatically extracting periodic patterns, resonance peak positions, and amplitude variation trends from vibration signals. These are then jointly modeled with thermal stress factors and energy entropy to form a high-dimensional nonlinear mapping relationship, ultimately outputting a high-dimensional damage feature tensor.

[0101] Step S105: Input the high-dimensional damage feature tensor into the pre-constructed digital twin risk assessment model. The digital twin risk assessment model refines the finite element mesh locally based on the feature values ​​and updates the material elastic modulus parameters, outputting the assessment results including risk level labels and damage location coordinates.

[0102] Among them, the eigenvalues ​​include the spatial strain gradient values ​​calculated based on the high-dimensional damage feature tensor;

[0103] In this embodiment, the digital twin risk assessment model integrates finite element simulation, real-time data updates, and a multi-level decision-making mechanism. When the strain gradient in the characteristic tensor exceeds a preset threshold, the system automatically triggers a local mesh refinement operation. By adjusting the finite element mesh size of the corresponding region to the 0.1mm level, the computational accuracy of that region is improved, allowing for a more accurate simulation of the potential damage evolution process. Simultaneously, considering the significant impact of temperature on material properties, the model also dynamically corrects the calculation formula parameters for the material's elastic modulus based on real-time temperature data, ensuring that the simulation results remain consistent with actual working conditions.

[0104] In terms of assessment logic, the model employs a three-level decision tree structure to determine the risk level: the first level is based on physical rule thresholds, such as whether indicators like strain and frequency drift exceed specified limits; the second level introduces an LSTM neural network to predict historical trends and analyzes the residual between the current state and the predicted value; if the residual continues to increase, it indicates that the structural state deviates from the normal trajectory; the third level assesses the consistency of the overall structural behavior by comparing the deviation between the current state and the simulation output of the digital twin risk assessment model. This multi-level, multi-dimensional assessment mechanism effectively improves the robustness and reliability of risk identification.

[0105] Step S106: Match the control strategy according to the risk level label and generate the corresponding equipment control command.

[0106] The design of the control strategy fully considers the differences in risk levels and the timeliness of emergency response.

[0107] For example, when the risk level is "Level 1," it means the structure is in an early abnormal state and has not yet reached a significant danger level. At this time, by increasing the sampling frequency of sensors around the damage location to 1kHz, the abnormal signals can be captured more precisely, providing more detailed information for subsequent diagnosis. When the risk level rises to "Level 2," it indicates that the damage is likely to develop further. At this time, a microscopic thermal imager re-inspection command containing the coordinates of the damage location is sent to the UAV dispatch system. The high-resolution imaging equipment on the UAV is used to visually inspect the suspicious area, which helps to confirm the damage morphology and degree of development. When the risk level reaches "Level 3," it means the structure is in a high-risk state, and the emergency plan must be activated immediately. This includes triggering the audible and visual alarms at the entrances and exits of the connecting corridor to warn personnel to evacuate, and simultaneously activating the flow control gates to prevent more people from entering the danger zone.

[0108] Understandably, this tiered response mechanism ensures both the timeliness of risk response and avoids unnecessary waste of resources.

[0109] The above-described embodiments achieve accurate identification and dynamic early warning of safety risks in high-altitude large steel structure corridors, significantly improving the intelligence level and response efficiency of health monitoring for complex structures. This scheme employs spatiotemporally aligned multi-source data fusion technology, combined with signal processing algorithms such as wavelet packet entropy, empirical mode decomposition, and matching pursuit, effectively solving the problems of severe environmental noise interference and weak, difficult-to-distinguish damage signals. By establishing a high-dimensional damage feature tensor containing multi-physics field characteristics such as strain-temperature coupling derivatives and frequency domain energy entropy, quantitative characterization of key damage mechanisms such as structural thermal stress effects and fatigue crack propagation is achieved. A three-level decision tree risk assessment model based on digital twin technology can achieve real-time response while ensuring computational accuracy, and ensures high consistency between simulation results and actual conditions through local mesh refinement and dynamic correction of material parameters. The hierarchical response mechanism ensures both timely risk response and avoids unnecessary resource waste. This application can effectively address the challenges posed by structural complexity, environmental interference, and uncertainty, providing a scientific and reliable monitoring method for the safe operation and maintenance of critical infrastructure such as large steel structure corridors.

[0110] Reference Figure 2 As one implementation of step S103, the step of dynamically suppressing noise in the sensor data matrix based on a pre-trained environmental noise transfer function model and outputting a denoised clean signal matrix and damage feature frequency band identifiers includes:

[0111] Step S201: Obtain the time-synchronized sensor data matrix; wherein, the sensor data matrix includes vibration signal, acoustic emission signal, wind speed and temperature and humidity data;

[0112] Time synchronization is a fundamental prerequisite for multiphysics coupling analysis. Due to differences in sampling frequencies and response characteristics among different sensors, a lack of precise time alignment will lead to deviations in subsequent feature extraction based on phase and causal relationships. In the sensor data matrix, vibration signals reflect the overall dynamic characteristics of the structure, acoustic emission signals can capture transient information about the initiation and propagation of microcracks within the material, while wind speed and temperature / humidity data provide necessary boundary conditions for environmental load modeling and noise source identification.

[0113] Step S202: Load the pre-trained environmental noise transfer function model;

[0114] Among them, the environmental noise transfer function model constructs the frequency domain transfer relationship with wind speed as the independent variable. This design is based on the generation mechanism of wind-induced noise in fluid mechanics, that is, the eddy shedding and turbulent pulsation generated when the wind flows over the surface of the structure will excite the structure to generate a vibration response with specific frequency components.

[0115] In some possible implementations, the environmental noise transfer function model is established by: collecting background noise samples during periods of no structural load; simultaneously recording wind speed data as an independent variable; and calculating the wind speed autopower spectrum P. xx (f) and wind speed-noise cross power spectrum P xy (f); The frequency domain transfer function is calculated using the cross-power spectral density: H(f) = P xy (f) / P xx (f).

[0116] Understandably, by collecting background noise samples and simultaneously recording wind speed data during periods without structural load, a wind speed-noise frequency domain mapping relationship was established. This modeling method fully utilizes the "quiet period" information of the structure in its non-operating state, avoiding interference from the actual response signal of the structure on noise modeling. The cross-power spectral density calculation method can effectively extract the linear influence relationship between wind speed changes and the noise spectrum distribution, providing a theoretical basis for subsequent adaptive noise suppression.

[0117] Step S203: Perform wavelet packet decomposition on the vibration signal and calculate the wavelet packet energy entropy value of each sub-band;

[0118] Among them, wavelet packet decomposition has a more flexible frequency band division capability compared with traditional wavelet decomposition, and can achieve uniform division of the signal spectrum, with each sub-band corresponding to a specific frequency range.

[0119] In some embodiments, energy entropy serves as an important indicator for measuring signal complexity and information content, and its calculation formula is E. i =-∑P i *log(P i ), where P i This represents the proportion of the energy of the i-th subband to the total energy.

[0120] Step S204: Filter sub-band components with entropy values ​​lower than a set threshold based on energy entropy values ​​and reconstruct them into a primary noise reduction signal;

[0121] Specifically, in structural health monitoring, environmental noise typically manifests as a broadband random signal with a relatively uniform energy distribution across its sub-bands, resulting in a high energy entropy value. In contrast, the characteristic signals generated by structural damage are often concentrated in specific frequency bands, exhibiting uneven energy distribution and relatively low entropy values. By setting a threshold to filter low-entropy sub-bands, frequency bands containing the true structural response information can be effectively identified. This information theory-based feature extraction method demonstrates good noise robustness.

[0122] Step S205: Perform empirical mode decomposition on the primary noise-reduced signal and extract the intrinsic mode function components;

[0123] The core of Empirical Mode Decomposition (EMD) lies in decomposing complex nonlinear and non-stationary signals into several Intrinsic Mode Function (IMF) components. Each IMF component satisfies two conditions: the number of extrema is equal to or differs from the number of zero-crossings by no more than 1; and the mean of the upper and lower envelopes is zero. This adaptive decomposition method can construct basis functions based on the time-frequency characteristics of the signal itself, avoiding the spectral leakage problem that may be caused by preset basis functions.

[0124] Step S206: Based on the environmental noise transfer function model, filter out the intrinsic mode function components related to wind speed to generate a pure vibration signal;

[0125] In this process, after obtaining the IMF components, the wind speed-related components are filtered out based on the pre-trained environmental noise transfer function model. This operation essentially integrates the time-frequency analysis results of empirical mode decomposition with the frequency domain modeling results of the transfer function. By identifying the main frequency components of each IMF component, the transfer function model is used to determine whether it belongs to the category of wind-induced noise, thereby achieving targeted noise suppression.

[0126] Step S207: Perform a matching tracking algorithm on the acoustic emission signal. When the temperature and humidity data exceed the preset temperature and humidity threshold, exclude atoms in the preset rain noise frequency band and extract feature atoms that match the crack waveform as crack feature waveform atoms.

[0127] Among them, the matching pursuit algorithm is based on the concept of an overcomplete dictionary and achieves sparse representation of the signal by iteratively selecting the atom that best matches the signal.

[0128] For example, when humidity > 90% and temperature > 5°C, it is determined to be a rainy environment. In this case, atoms in a preset rain noise frequency band (such as atoms in the 8-10kHz band) are removed from the overcomplete dictionary. This processing is based on the spectral characteristics of rain noise, where raindrops hitting sensors or structural surfaces will produce impact responses within a specific frequency range. At the same time, the atom matching calculation is focused on characteristic atoms that match the crack waveform (such as the 30-150kHz band). This frequency band is selected based on the propagation characteristics of elastic waves generated during crack propagation in metallic materials. The stress release at the crack tip will generate high-frequency stress waves, whose frequency range is typically between tens of kHz and hundreds of kHz.

[0129] Step S208: Fuse the pure vibration signal with the crack characteristic waveform atoms to generate a pure signal matrix;

[0130] In this process, feature signals obtained from different processing paths are integrated to construct a complete structural state representation matrix.

[0131] Step S209: Calculate the effective frequency band boundary of the crack characteristic waveform atoms and output the damage characteristic frequency band boundary identifier.

[0132] Among them, the center frequency f of the crack feature waveform atoms is extracted. c and bandwidth B w ; Calculate the effective frequency band boundary: [f c −αB w ,f c +αB w ], where α is the preset bandwidth expansion coefficient. This boundary calculation method takes into account the spectral diffusion characteristics of the actual signal. By adjusting the bandwidth expansion coefficient, it can ensure the integrity of feature extraction while avoiding noise mixing caused by excessively wide frequency bands.

[0133] The above implementation scheme achieves accurate processing of structural health monitoring signals under complex environmental conditions. This scheme not only considers the physical mechanism modeling of environmental interferences such as wind-induced noise, but also combines advanced signal processing techniques based on information theory and sparse representation. Through multi-dimensional feature extraction and noise suppression, it significantly improves the accuracy and reliability of structural damage identification, providing scientific and reliable technical support for the safety monitoring of large steel structure facilities.

[0134] Reference Figure 3 As one implementation of the environmental noise transfer function model, the steps for establishing the environmental noise transfer function model include:

[0135] Step S301: Collect background noise data during historical periods of unstructured load;

[0136] The unloaded period refers to the time window during which the structure is in a non-operating state or not bearing major loads. During this time, the effective response signal generated by the structure itself is so weak as to be negligible, and the signal collected by the sensors mainly consists of environmental noise. This data acquisition strategy embodies the core idea of ​​"signal-noise separation" in modern signal processing. By selecting an appropriate observation window, a clean noise sample is obtained, avoiding interference from the actual response signal of the structure on the extraction of noise features.

[0137] Step S302: Synchronously record wind speed data as an independent variable;

[0138] Wind speed, as a major driver of environmental noise, generates fluid dynamic phenomena such as eddies and turbulence through its interaction with structural surfaces, which are significant sources of environmental noise. By synchronously recording wind speed data, a temporal correspondence between environmental excitation and noise response was established. This synchronicity ensures the accuracy of causal relationships in subsequent transfer function calculations.

[0139] Step S303, calculate the wind speed autopower spectrum P xx (f) and wind speed-noise cross power spectrum P xy (f);

[0140] Among them, the self-power spectrum P xx(f) describes the energy distribution of the wind speed signal x(t) across its frequency components. Essentially, it is the squared modulus of the Fourier transform of the wind speed signal, reflecting the frequency domain characteristics of wind speed changes. The specific calculation formula is as follows: ;

[0141] In addition, the cross power spectrum P xy (f) describes the frequency domain correlation between the wind speed signal x(t) and the noise signal y(t), including the amplitude and phase relationships of the two signals at various frequency components. The calculation of the cross-power spectrum requires cross-correlation of the two signals followed by a Fourier transform. The result is a complex number; the real part reflects the relationship between the in-phase components of the two signals, and the imaginary part reflects the relationship between the quadrature components. The specific calculation formula is as follows: ;

[0142] In the above formula, the Fourier transforms of the wind speed signal x(t) and the noise signal y(t) are X(f) and Y(f), respectively. Let X(f) be the conjugate complex number, and T be the duration of signal observation.

[0143] In the embodiments of this application, these power spectrum calculations employ the classic Welch method or periodogram method, which improves the stability of spectrum estimation through piecewise averaging and effectively reduces the impact of random fluctuations on the results.

[0144] Step S304: Calculate the frequency domain transfer function using the cross-power spectral density. The calculation formula is: H(f) = P xy (f) / P xx (f).

[0145] The transfer function, mathematically speaking, describes the system's response characteristics to different frequency components. Its numerator, Pxy(f), contains the coupling information between the input and output signals, while the denominator, Pxx(f), serves a normalization function, eliminating the influence of variations in the input signal's intensity on the result. The magnitude of the transfer function reflects the system's amplification or attenuation characteristics for each frequency component, while the phase describes the time delay characteristics of each frequency component.

[0146] In the embodiments of this application, the transfer function essentially establishes a quantitative relationship between wind speed changes and the spectral distribution of environmental noise, providing accurate mathematical model support for subsequent adaptive noise suppression.

[0147] In the above implementation, a frequency domain model that can accurately describe the wind speed-noise relationship was constructed, providing a reliable theoretical basis and calculation basis for subsequent dynamic noise suppression.

[0148] Reference Figure 4As one implementation of step S104, the steps of calculating the thermal stress sensitivity factor based on the strain signal and temperature signal, extracting the multi-physics coupling features associated with the damage feature frequency band identifier in the pure signal matrix, and fusing them to generate a high-dimensional damage feature tensor include:

[0149] Step S401: Receive the clean signal matrix and the damage characteristic frequency band identifier;

[0150] The pure signal matrix comprises pure vibration signals and crack feature waveform atoms. The pure vibration signal refers to the true structural response signal obtained after prior noise suppression processing, while the crack feature waveform atoms are basis functions with specific time-frequency localization characteristics extracted based on matching pursuit algorithms or dictionary learning techniques, which can effectively characterize the transient features of structural damage.

[0151] Step S402: Simultaneously acquire the spatiotemporally aligned original strain signal and temperature signal;

[0152] The spatiotemporal alignment of strain and temperature signals requires that the physical locations of the two sensors be representative, typically using the same or adjacent measuring points to ensure that the collected physical quantities reflect the true state of the local area of ​​the structure. Time synchronization is achieved through hardware clock synchronization or software interpolation algorithms, with timestamp alignment accuracy usually required to be at the millisecond level or even higher to ensure the accuracy of subsequent differential calculations.

[0153] Step S403: Based on the strain signal and temperature signal, solve the partial derivative of strain with respect to temperature within a preset time window to obtain the thermal stress sensitivity factor.

[0154] The thermal stress sensitivity factor essentially reflects the sensitivity of a material or structure to the stress-temperature coupling relationship. Its physical meaning lies in describing the strain change caused by a unit temperature change, making it a key parameter in thermal stress analysis. The sliding time window method takes into account the time-varying characteristics of the structural response. The choice of window length requires a trade-off between time resolution and statistical stability, and is typically determined based on the characteristic period of the signal and the sampling frequency.

[0155] Specifically, the calculation of the thermal stress sensitivity factor includes: constructing a sliding time window with the temperature signal as the independent variable and the strain signal as the dependent variable; fitting the strain-temperature linear relationship using the least squares method, with the slope being the partial derivative value. This method obtains the optimal parameter estimate by minimizing the sum of squared residuals between the observed data and the theoretical model, exhibiting good numerical stability and statistical properties. The calculation of the slope, as an estimate of the partial derivative, involves matrix operations and linear algebra theory, specifically manifested in the solution process of the normal equation.

[0156] Step S404: Extract the energy distribution of the crack feature waveform atoms in the time domain waveform, and calculate the crack energy entropy value within the damage feature frequency band identification range;

[0157] The calculation of the energy entropy value of the crack characteristic waveform atoms includes: analyzing the time-domain envelope of the atomic waveform function gγ(t); integrating the signal energy E within the damage characteristic frequency band. i According to the entropy formula H=−∑(E) i / E total log2(E) i / E total Calculate the entropy value.

[0158] Specifically, the time-domain envelope analysis of crack characteristic waveform atoms employs the Hilbert transform technique. By converting the real signal into an analytic signal, instantaneous amplitude information is extracted. The envelope function effectively reflects the amplitude modulation characteristics of the signal, which is of great significance for identifying the transient impact response generated by cracks. The integral calculation of energy distribution is essentially a cumulative measure of signal power within a specific frequency band, reflecting the application of Passevar's theorem within a finite frequency band. Frequency domain integration quantifies the degree of energy concentration of the signal within the damage-sensitive frequency band. Entropy calculation uses the Shannon entropy formula to measure the uncertainty or complexity of the signal energy distribution. When crack development causes changes in signal characteristics, the entropy value of the energy distribution also changes accordingly, thus providing a quantitative indicator for damage identification.

[0159] Step S405: Perform frequency domain transformation on the pure vibration signal to extract the vibration frequency domain features within the preset characteristic frequency band;

[0160] Frequency domain transformation typically employs the Fast Fourier Transform (FFT) algorithm to convert time-domain signals into a frequency-domain representation. This transformation, based on Fourier analysis theory, decomposes complex time-domain waveforms into a superposition of different frequency components. Frequency domain feature extraction includes various forms such as amplitude spectrum, phase spectrum, and power spectral density. These features reflect the dynamic changes in the structure, providing a frequency-domain basis for damage identification.

[0161] In some embodiments, the preset characteristic frequency band is 0-200Hz, which is suitable for steel structure corridors with spans of 30-80m. The selection of the preset characteristic frequency band is based on structural dynamics theory and modal analysis results; the fundamental frequency of large steel structures is usually distributed within this range.

[0162] Step S406: The thermal stress sensitivity factor, crack energy entropy value and vibration frequency domain features are fused through a 1D-CNN neural network to output the fused high-dimensional damage feature tensor.

[0163] The 1D-CNN neural network architecture utilizes the temporal characteristics of one-dimensional signals, automatically extracting and fusing multi-dimensional features through a combination of convolutional layers, pooling layers, and fully connected layers. Thermal stress sensitivity factors provide thermo-mechanical coupling information, crack energy entropy reflects the time-frequency localization characteristics of damage, and vibration frequency domain features embody the overall dynamic characteristics of the structure. These three types of features complement each other physically, constituting a multi-dimensional damage characterization system. The final output high-dimensional damage feature tensor possesses rich semantic information and good discriminative ability, providing a high-quality feature representation for subsequent damage localization and quantitative assessment.

[0164] In some embodiments, the training process of the 1D-CNN neural network is based on the backpropagation algorithm and gradient descent optimization theory, and a nonlinear mapping relationship between input features and damage state is established through learning from a large number of samples.

[0165] In the above embodiments, a complete processing chain from multi-physics signal acquisition to high-dimensional feature fusion is constructed, realizing refined identification and characterization of structural damage. By fully considering multi-dimensional information such as thermo-mechanical coupling effect, time-frequency localization characteristics and frequency domain dynamic characteristics, a damage identification method with engineering practical value is formed, improving the accuracy and reliability of structural health monitoring system.

[0166] Reference Figure 5 As one implementation of step S105, the step of inputting a high-dimensional damage feature tensor into a pre-constructed digital twin risk assessment model, refining the finite element mesh locally based on the feature values ​​and updating the material elastic modulus parameters, and outputting an assessment result containing risk level labels and damage location coordinates includes:

[0167] Step S501: Receive the high-dimensional damage feature tensor, including thermal stress sensitivity factor, crack energy entropy value and vibration frequency domain features;

[0168] Step S502: Calculate the spatial strain gradient value based on the thermal stress sensitivity factor;

[0169] The calculation of the spatial strain gradient essentially quantifies the rate of change of the thermal stress sensitivity factor in the spatial domain. This process is based on partial differential equation theory and uses the finite difference method or finite element method to approximate the spatial derivative of the strain field. The physical meaning of the gradient value lies in describing the degree of drastic change of the strain field in space. High gradient regions usually correspond to stress concentrations or material discontinuities, and these regions are often high-risk areas for structural damage.

[0170] Step S503: When the spatial strain gradient value exceeds the preset threshold, a local refinement operation of the finite element mesh is triggered in the pre-constructed digital twin risk assessment model.

[0171] It should be noted that traditional finite element simulations often employ a uniform mesh generation method, making it difficult to balance global accuracy and computational efficiency. In this approach, the system dynamically adjusts the mesh density in certain key regions of the model by monitoring changes in the strain gradient in real time. This improves the simulation accuracy for high-risk areas without significantly increasing the overall computational burden.

[0172] This adaptive mesh refinement strategy based on physical field gradients is widely used in fields such as structural mechanics and fluid mechanics. Its advantage lies in its ability to significantly improve the ability to characterize local details while maintaining computational efficiency. It is especially suitable for dealing with damage evolution problems with strong non-uniformity and multi-scale characteristics.

[0173] Step S504: Call the temperature-elastic modulus mapping table in the pre-stored material constitutive relation library and dynamically update the elastic modulus parameters according to the real-time temperature data;

[0174] The elastic modulus of a material, as a crucial parameter describing its resistance to elastic deformation, changes significantly at high temperatures. Using material parameters from room temperature would distort the simulation results. Therefore, the system dynamically adjusts the material properties of corresponding elements in the model by accessing elastic modulus curves or functional relationships for different temperature conditions stored in a material database, combined with currently collected actual temperature data. This data-driven parameter update mechanism not only improves the model's accuracy but also enhances its adaptability to real-world operating conditions.

[0175] Step S505: Input the crack energy entropy value into the crack propagation rate model and calculate the critical damage index;

[0176] Specifically, the system utilizes fracture mechanics theory to establish a functional relationship between crack propagation rate and external loads and material properties. It then inversely calculates the crack development trend based on the current crack energy entropy value and accordingly calculates a critical damage index that comprehensively reflects the structural safety margin. This index can be considered a warning threshold before the structure enters a dangerous state; the closer its value is to 1, the closer the structure is to the failure threshold. By introducing this index, the system can rationally allocate the weights of different assessment sub-models in subsequent decision-making processes, achieving dynamic optimization of risk level assessment.

[0177] Step S506: Perform finite element simulation on the refined mesh model after updating the elastic modulus parameters, and output the simulated stress field;

[0178] Specifically, through the aforementioned adaptive mesh refinement and material parameter updates, the system constructs a more refined finite element model that closely resembles actual working conditions. Based on this, structural mechanics simulations can more accurately simulate the stress distribution of the structure under complex loads. The stress field obtained from the simulation not only includes the overall stress state of the structure but also reveals the location and intensity of local high-stress zones, providing crucial information for subsequent damage localization and risk assessment.

[0179] Step S507: Based on the preset physical rule threshold, the thermal stress sensitivity factor and vibration frequency domain characteristics are compared respectively to obtain the first-level decision result;

[0180] The system employs physical rules based on expert experience or experimental data to compare thresholds for two types of features. For example, when the thermal stress sensitivity factor exceeds a certain critical value, the system determines that the region may be at risk of thermal fatigue damage; when the frequency of a certain mode in the vibration frequency domain is below a preset lower limit, it may indicate a decrease in structural stiffness. This preliminary judgment method based on physical laws has strong interpretability and robustness, and can provide a basic basis for risk assessment even in the absence of a large number of training samples.

[0181] Step S508: Input the vibration frequency domain features into the pre-trained LSTM time series model, calculate the residual between the predicted value and the actual value, and obtain the secondary decision result;

[0182] Among them, LSTM (Long Short-Term Memory) networks, as a special type of recurrent neural network structure, excel at capturing long-term dependencies in time-series data. By pre-training this model on historical data, it can learn the evolution of structural responses over time. During operation, the system inputs the currently collected vibration frequency domain features into the model to obtain its predicted output of future states. By calculating the residual between the predicted and measured values, it determines whether the structure exhibits abnormal behavior. The larger the residual, the more severely the current state deviates from the normal pattern, and the higher the potential risk. This method compensates for the shortcomings of static threshold determination and enhances the ability to perceive dynamic evolution processes.

[0183] Step S509: Calculate the modal matching degree between the simulated stress field and the vibration frequency domain characteristics to obtain the three-level decision results;

[0184] The system employs modal analysis theory to compare the simulated stress field information with the measured vibration frequency domain characteristics, assessing the similarity between the two in terms of frequency, mode shape, and other aspects. A higher modal matching degree indicates that the simulation model more accurately reflects the true dynamic characteristics of the structure; conversely, a lower degree of modal matching may indicate model deviation or structural damage. This step, by introducing modal consistency analysis, enhances the objectivity and reliability of the evaluation results, avoiding the risk of misjudgment caused by a single feature source.

[0185] Step S510: Adjust the preset decision weight ratio according to the critical damage index, and perform weighted fusion of the first-level decision results, second-level decision results and third-level decision results to generate risk level labels;

[0186] Specifically, since different types of risk assessment methods have their own advantages and disadvantages, relying solely on a single method may lead to misjudgment. Therefore, the system dynamically adjusts the weight allocation of each sub-model based on the critical damage index output by the crack propagation rate model. For example, when the critical damage index is low, it indicates that the structure is still in the early stage of damage, and the first-level judgment based on physical rules is more reliable at this time, so it is given a higher weight; as the damage intensifies, the importance of the time series model and modal matching analysis increases, and the corresponding weights are adjusted accordingly. This weight adaptive mechanism based on the damage evolution state significantly improves the accuracy and adaptability of risk assessment.

[0187] For example, the weight ratio is dynamically allocated according to the critical damage index Dc. When Dc < 0.3, the weight ratio of the first-level, second-level, and third-level decision results is configured as 4:3:3; when 0.3 ≤ Dc < 0.7, the weight ratio is configured as 5:3:2; and when Dc ≥ 0.7, the weight ratio is configured as 6:2:2.

[0188] Step S511: Extract the peak coordinates in the simulated stress field as the damage location coordinates;

[0189] The system post-processes the simulation results to identify the maximum stress concentration points and maps them to a three-dimensional coordinate system, thereby accurately locating the specific location of potential damage. This method combines the high-resolution advantages of numerical simulation with the engineering requirements of damage localization, providing an intuitive spatial reference for subsequent maintenance decisions.

[0190] Step S512: Combine the risk level label and the coordinates of the damage location to obtain the assessment result.

[0191] In the above implementation, by constructing a digital twin risk assessment system with adaptability, dynamism and high precision, the model can not only achieve comprehensive perception and accurate positioning of structural damage status, but also dynamically adjust the assessment strategy according to different damage development stages, which significantly improves the timeliness and accuracy of risk warning.

[0192] In this embodiment, the digital twin risk assessment model adopts a hierarchical and progressive architecture, achieving multi-scale quantitative assessment of structural health status through deep integration of physical mechanisms and data-driven approaches. The core input of the model is a three-dimensional high-dimensional damage feature tensor, containing three key physical parameters: thermal stress sensitivity factor, crack energy entropy, and vibration frequency domain characteristics. In the initial processing stage, the system calculates the spatial strain gradient value based on the thermal stress sensitivity factor. This process quantifies the local rate of change of the strain field through a spatial difference algorithm. When the gradient value exceeds a preset threshold (e.g., 5.0 MPa / ℃·m⁻¹), the local mesh adaptive optimization mechanism of the digital twin risk assessment model is automatically triggered. This mechanism generates a spherical refined mesh domain in the potential damage area, with its radius topologically correlated with the distribution of nearby sensors. Micrometer-level refined cells are used within the refined domain, and a gradient mesh transition strategy is implemented in the transition zone. This significantly reduces the overall computational load while ensuring high-resolution simulation of stress concentration areas.

[0193] Simultaneously, the model synchronously calls the pre-stored material constitutive relation library and dynamically interpolates and updates the elastic modulus parameters based on real-time collected ambient temperature data. This process embeds temperature-modulus decay curves for 12 types of structural steel, accurately reflecting the nonlinear decay characteristics of material stiffness under high-temperature environments through a quadratic function model. After completing mesh topology optimization and material parameter updates, the system inputs the crack energy entropy value into the Paris crack propagation rate model and calculates the critical damage index (range 0-1) by inverting the crack evolution trajectory. This index, as a core indicator for quantifying structural safety margin, directly affects subsequent decision-making weight allocation strategies.

[0194] In the refined simulation layer, the model performs thermo-mechanical coupled finite element calculations, outputting a high-resolution three-dimensional stress field distribution. This stress field carries spatial coordinate information, and its peak coordinates are directly mapped to the damage location output. The decision layer employs a triple verification mechanism: the first-level decision compares the deviation between the thermal stress sensitivity factor and the fundamental vibration frequency based on physical rule thresholds; the second-level decision analyzes the temporal residuals of the vibration frequency domain characteristics through a pre-trained bidirectional LSTM network to capture dynamic characteristic anomalies; and the third-level decision calculates the confidence factors of the simulated stress field and the measured vibration modes to assess the dynamic consistency between the model and the entity. The three decision results are fused through a dynamic weight allocator, with weight configuration strictly following a segmentation rule guided by the critical damage index. The final fused result is input into the risk assessment engine, which generates three-level risk level labels based on pre-configured threshold intervals, and together with the damage spatial coordinates extracted from the stress field, constitutes the assessment result output.

[0195] In summary, the digital twin risk assessment model constructed in this application solves the three major technical bottlenecks of traditional methods, namely, the imbalance between simulation accuracy and efficiency, the neglect of temperature effects, and the misjudgment of damage stages, through a four-fold closed loop of strain gradient perception, dynamic parameter update, damage entropy feedback, and adaptive decision weight.

[0196] Reference Figure 6 As a further implementation of the risk monitoring method, after the step of outputting assessment results including risk level labels and damage location coordinates, the method further includes:

[0197] Step S601: Extract the modal confidence factor deviation between the measured vibration frequency domain characteristics and the simulated stress field at the damage location coordinates under the same spatial coordinates.

[0198] The modal confidence factor is a dimensionless index used to quantify the correlation between two modal vectors, obtained through normalization via the dot product of mode vectors. By calculating the deviation between the measured and simulated modal confidence factors at the damage location coordinates, the system can quantify the degree of difference between the model's prediction accuracy and the actual measurement results. This deviation reflects the credibility level of the digital twin risk assessment model under the current parameter configuration.

[0199] Step S602: When the modal confidence factor deviation value exceeds the tolerance threshold for a preset number of consecutive times, the constitutive parameter inversion engine is triggered.

[0200] In one embodiment of this application, when the calculated modal confidence factor deviation value exceeds a preset tolerance threshold three times consecutively, the system will trigger the constitutive parameter inversion engine. The setting of the tolerance threshold needs to comprehensively consider the accuracy limitations of the measurement system, environmental noise interference, and the actual needs of engineering applications, and is typically determined using statistical methods such as the 3σ criterion or empirical distributions based on historical data. The triggering mechanism of the constitutive parameter inversion engine embodies the core idea of ​​an adaptive control system: by monitoring the deviation between the system output and the expected value in real time, the parameter adjustment mechanism is automatically activated when the deviation exceeds an acceptable range. This design ensures that the digital twin risk assessment model can dynamically adapt to changes in actual working conditions and maintain the accuracy of model predictions.

[0201] Step S603: Based on the covariance matrix adaptive evolution algorithm, iteratively optimize the temperature-elastic modulus mapping table in the material constitutive relation library;

[0202] The core advantage of the covariance matrix adaptive evolutionary algorithm lies in its ability to adaptively adjust the covariance matrix of the search distribution, thereby achieving efficient global optimization in a high-dimensional parameter space. This algorithm generates candidate solutions by maintaining a multivariate Gaussian distribution. As the iteration process progresses, the mean of the distribution shifts towards a better solution region, and the covariance matrix is ​​adaptively updated based on the distribution information of successful individuals. This allows the search process to gradually converge to the vicinity of the optimal solution while maintaining its exploratory capabilities.

[0203] In the embodiments of this application, the temperature-elastic modulus mapping table is used as the optimization target. Its parameter space usually has high-dimensional nonlinear characteristics. Traditional gradient optimization methods are prone to getting trapped in local optima. However, the CMA-ES algorithm can effectively overcome this problem through population evolution mechanism and adaptive adjustment of covariance matrix, and achieve accurate identification of material constitutive parameters.

[0204] Step S604: Update the optimized temperature-elastic modulus mapping table to the parameter storage area of ​​the digital twin risk assessment model and generate a version iteration log.

[0205] Specifically, the temperature-elastic modulus mapping table, optimized using an adaptive evolutionary algorithm, is updated in the parameter storage area of ​​the digital twin risk assessment model, generating a corresponding version iteration log. This log records not only the timestamps and specific numerical changes of the parameters but also key statistical information from the optimization process, such as the objective function convergence curve, the number of iterations, and computation time. This information provides crucial data for subsequent model validation, parameter backtracking, and system performance analysis. Furthermore, this mechanism ensures that the digital twin risk assessment model can continuously learn and improve, maintaining its predictive accuracy and decision-making reliability in the face of complex and ever-changing engineering environments.

[0206] In the above implementation, a complete closed-loop control system is constructed from model deviation detection to parameter adaptive optimization, realizing intelligent maintenance and dynamic updating of the digital twin risk assessment model. The system first quantifies the model's prediction accuracy through modal confidence factor deviation analysis. When persistent deviations are detected, a constitutive parameter inversion optimization mechanism based on the CMA-ES algorithm is initiated. Finally, version management achieves accurate updates of model parameters, ensuring that the digital twin system maintains long-term stable prediction performance in complex engineering environments, providing reliable technical support for structural health monitoring and fault early warning.

[0207] This application also discloses a safety risk monitoring system for a large high-altitude steel structure connecting corridor.

[0208] A safety risk monitoring system for a large high-altitude steel structure connecting corridor, the monitoring system comprising:

[0209] The data acquisition module is used to collect raw monitoring data through a multi-source sensor array deployed at each node of the steel structure corridor; the raw monitoring data includes strain signals, vibration signals, acoustic emission signals, and temperature signals.

[0210] The data processing module is used to perform spatiotemporal alignment processing on the raw monitoring data to generate a time-synchronized sensor data matrix;

[0211] The noise reduction module is used to dynamically suppress noise in the sensor data matrix based on a pre-trained environmental noise transfer function model, and outputs a clean signal matrix after noise reduction and damage feature frequency band identification.

[0212] The feature tensor generation module is used to calculate the thermal stress sensitivity factor based on strain and temperature signals, extract multi-physics coupling features associated with damage feature frequency band identifiers in the pure signal matrix, and fuse them to generate a high-dimensional damage feature tensor.

[0213] The risk assessment module is used to input the high-dimensional damage feature tensor into a pre-built digital twin risk assessment model. The digital twin risk assessment model refines the finite element mesh locally based on the feature values ​​and updates the material elastic modulus parameters, outputting assessment results that include risk level labels and damage location coordinates. Among them, the feature values ​​include the spatial strain gradient values ​​calculated based on the high-dimensional damage feature tensor.

[0214] The equipment control module is used to match control strategies based on risk level labels and generate corresponding equipment control commands.

[0215] As a further implementation of the monitoring system, it also includes:

[0216] The measured data acquisition module is used to acquire the measured vibration frequency domain characteristics at the damage location coordinates.

[0217] The deviation calculation module is used to calculate the modal confidence factor deviation between the measured vibration frequency domain characteristics and the simulated stress field under the same spatial coordinates.

[0218] The over-limit trigger module is used to trigger the constitutive parameter inversion engine when the modal confidence factor deviation value exceeds the tolerance threshold for a preset number of consecutive times.

[0219] The iterative optimization module is used to iteratively optimize the temperature-elastic modulus mapping table in the material constitutive relation library based on the covariance matrix adaptive evolution algorithm.

[0220] The iteration log generation module is used to update the optimized temperature-elastic modulus mapping table to the parameter storage area of ​​the digital twin risk assessment model and generate a version iteration log.

[0221] The high-altitude large steel structure corridor safety risk monitoring system of this application embodiment can realize any of the above monitoring methods, and the specific working process of each module in the monitoring system can refer to the corresponding process in the above method embodiment.

[0222] In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the division of a certain module is merely a logical functional division, and in actual implementation there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0223] This application also discloses a computer-readable storage medium.

[0224] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above in any of the methods for monitoring safety risks of a large steel structure connecting corridor at high altitudes.

[0225] The computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device; the program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0226] It should be noted that the computer device and storage medium in the embodiments of this application are respectively electronic devices and storage media for applying the above-described method for monitoring safety risks of large-scale steel structure connecting corridors at high altitudes. Therefore, all embodiments of the above monitoring method are applicable to the computer device and storage medium, and can achieve the same or similar beneficial effects. For the computer device / storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple; relevant details can be found in the descriptions of the method embodiments.

[0227] In this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0228] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, disclosure, and appended claims in carrying out the claimed invention. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0229] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A method for monitoring the safety risks of a large high-altitude steel structure connecting corridor, characterized in that, The monitoring method includes: Raw monitoring data is collected by a multi-source sensor array deployed at each node of the steel structure corridor; the raw monitoring data includes strain signals, vibration signals, acoustic emission signals, and temperature signals. The original monitoring data is spatiotemporally aligned to generate a time-synchronized sensor data matrix; Based on a pre-trained environmental noise transfer function model, dynamic noise suppression is performed on the sensor data matrix, and the noise-reduced clean signal matrix and damage feature frequency band identifiers are output. The thermal stress sensitivity factor is calculated based on the strain signal and temperature signal. Multi-physics coupling features associated with the damage feature frequency band identifier are extracted from the pure signal matrix and fused to generate a high-dimensional damage feature tensor. The high-dimensional damage feature tensor is input into a pre-constructed digital twin risk assessment model. The digital twin risk assessment model refines the finite element mesh locally based on the feature values ​​and updates the material elastic modulus parameters, outputting an assessment result that includes a risk level label and damage location coordinates. The feature values ​​include the spatial strain gradient values ​​calculated based on the high-dimensional damage feature tensor. The risk level label is used to match the control strategy and generate the corresponding equipment control command.

2. The method for monitoring safety risks of a large high-altitude steel structure connecting corridor according to claim 1, characterized in that, The steps of dynamically suppressing noise in the sensor data matrix based on a pre-trained environmental noise transfer function model, and outputting a denoised clean signal matrix and damage feature frequency band identifiers, include: Acquire a time-synchronized sensor data matrix; the sensor data matrix includes vibration signals, acoustic emission signals, wind speed, and temperature and humidity data; Load the pre-trained environmental noise transfer function model; Perform wavelet packet decomposition on the vibration signal and calculate the wavelet packet energy entropy value of each sub-band; Based on the energy entropy value, sub-band components with entropy values ​​lower than a set threshold are selected and reconstructed into a primary noise-reduced signal; Empirical mode decomposition is performed on the primary noise-reduced signal to extract intrinsic mode function components; Based on the environmental noise transfer function model, the intrinsic mode function components related to wind speed are filtered out to generate a pure vibration signal; A matching tracking algorithm is executed on the acoustic emission signal. When the temperature and humidity data exceed the preset temperature and humidity threshold, atoms in the preset rain noise frequency band are excluded, and feature atoms that match the crack waveform are extracted as crack feature waveform atoms. By fusing the pure vibration signal with the crack characteristic waveform atoms, a pure signal matrix is ​​generated; Calculate the effective frequency band boundary of the crack characteristic waveform atoms and output the damage characteristic frequency band boundary identifier.

3. The method for monitoring safety risks of a large-scale steel structure connecting corridor at high altitudes according to claim 2, characterized in that, The steps for establishing the environmental noise transfer function model include: Collect background noise data during historical periods of unstructured load; Simultaneously record wind speed data as the independent variable; Calculate the wind speed autopower spectrum P xx (f) and wind speed-noise cross power spectrum P xy (f); The frequency domain transfer function is calculated using the cross-power spectral density, and the formula is: H(f) = P xy (f) / P xx (f).

4. The method for monitoring safety risks of a large high-altitude steel structure connecting corridor according to claim 1, characterized in that, The steps of calculating the thermal stress sensitivity factor based on the strain signal and temperature signal, extracting the multi-physics coupling features associated with the damage feature frequency band identifier in the pure signal matrix, and fusing them to generate a high-dimensional damage feature tensor include: Receive a clean signal matrix and damage characteristic frequency band identifiers; the clean signal matrix includes clean vibration signals and crack characteristic waveform atoms; Simultaneously acquire spatiotemporally aligned raw strain and temperature signals; Based on the strain and temperature signals, the partial derivative of strain with respect to temperature is calculated within a preset time window to obtain the thermal stress sensitivity factor. Extract the energy distribution of the crack feature waveform atoms in the time domain waveform, and calculate the crack energy entropy value within the damage feature frequency band identification range; Perform frequency domain transformation on the pure vibration signal to extract the vibration frequency domain features within a preset characteristic frequency band; The thermal stress sensitivity factor, crack energy entropy value and vibration frequency domain features are fused by a 1D-CNN neural network to output a fused high-dimensional damage feature tensor.

5. A method for monitoring safety risks of a large-scale steel structure connecting corridor at high altitudes according to any one of claims 1 to 4, characterized in that, The steps of inputting the high-dimensional damage feature tensor into a pre-constructed digital twin risk assessment model, wherein the digital twin risk assessment model performs local finite element mesh refinement based on the feature values ​​and updates the material elastic modulus parameters, and outputs an assessment result including risk level labels and damage location coordinates, include: Receive the high-dimensional damage feature tensor, including thermal stress sensitivity factor, crack energy entropy value and vibration frequency domain features; Calculate the spatial strain gradient value based on the aforementioned thermal stress sensitivity factor; When the spatial strain gradient value exceeds a preset threshold, a local refinement operation of the finite element mesh is triggered in the pre-constructed digital twin risk assessment model; Call the temperature-elastic modulus mapping table in the pre-stored material constitutive relation library, and dynamically update the elastic modulus parameters based on real-time temperature data; Input the crack energy entropy value into the crack propagation rate model to calculate the critical damage index; Perform finite element simulation on the refined mesh model after updating the elastic modulus parameters, and output the simulated stress field; Based on preset physical rule thresholds, the thermal stress sensitivity factor and vibration frequency domain characteristics are compared to obtain the first-level decision result. The vibration frequency domain features are input into a pre-trained LSTM time series model, and the residual between the predicted value and the actual value is calculated to obtain the secondary decision result. The modal matching degree between the simulated stress field and the vibration frequency domain characteristics is calculated to obtain the three-level decision results; The preset decision weight ratio is adjusted according to the critical damage index, and the first-level decision result, second-level decision result and third-level decision result are weighted and fused to generate a risk level label; The peak coordinates in the simulated stress field are extracted as the damage location coordinates. The assessment results are obtained by combining the risk level label and the coordinates of the damage location.

6. A method for monitoring safety risks of a large high-altitude steel structure connecting corridor according to claim 5, characterized in that, The step of outputting assessment results including risk level labels and damage location coordinates also includes: Extract the modal confidence factor deviation between the measured vibration frequency domain features at the damage location coordinates and the simulated stress field under the same spatial coordinates; When the modal confidence factor deviation value exceeds the tolerance threshold for a preset number of consecutive times, the constitutive parameter inversion engine is triggered. Based on the covariance matrix adaptive evolution algorithm, the temperature-elastic modulus mapping table in the material constitutive relation library is iteratively optimized. The optimized temperature-elastic modulus mapping table is updated to the parameter storage area of ​​the digital twin risk assessment model, and a version iteration log is generated.

7. A safety risk monitoring system for a large-scale steel structure connecting corridor at high altitude, characterized in that, The monitoring system includes: The data acquisition module is used to collect raw monitoring data through a multi-source sensor group deployed at each node of the steel structure corridor; the raw monitoring data includes strain signals, vibration signals, acoustic emission signals and temperature signals; The data processing module is used to perform spatiotemporal alignment processing on the raw monitoring data to generate a time-synchronized sensor data matrix; The noise reduction module is used to perform dynamic noise suppression on the sensor data matrix based on a pre-trained environmental noise transfer function model, and output the noise-reduced clean signal matrix and damage feature frequency band identifiers. The feature tensor generation module is used to calculate the thermal stress sensitivity factor based on the strain signal and temperature signal, extract the multi-physics coupling features associated with the damage feature frequency band identifier in the pure signal matrix, and fuse them to generate a high-dimensional damage feature tensor. The risk assessment module is used to input the high-dimensional damage feature tensor into a pre-constructed digital twin risk assessment model. The digital twin risk assessment model performs local refinement of the finite element mesh and updates the material elastic modulus parameters based on the feature values, and outputs an assessment result containing risk level labels and damage location coordinates. The feature values ​​include spatial strain gradient values ​​calculated based on the high-dimensional damage feature tensor. The equipment control module is used to match control strategies based on the risk level labels and generate corresponding equipment control commands.

8. A safety risk monitoring system for a large-scale steel structure connecting corridor at high altitudes according to claim 7, characterized in that, The monitoring system also includes: The measured data acquisition module is used to acquire the measured vibration frequency domain characteristics at the coordinates of the damage location; The deviation calculation module is used to calculate the modal confidence factor deviation between the measured vibration frequency domain characteristics and the simulated stress field under the same spatial coordinates. The over-limit triggering module is used to trigger the constitutive parameter inversion engine when the modal confidence factor deviation value exceeds the tolerance threshold for a preset number of consecutive times. The iterative optimization module is used to iteratively optimize the temperature-elastic modulus mapping table in the material constitutive relation library based on the covariance matrix adaptive evolution algorithm. The iteration log generation module is used to update the optimized temperature-elastic modulus mapping table to the parameter storage area of ​​the digital twin risk assessment model and generate a version iteration log.

9. A computer-readable storage medium, characterized in that: The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 6.

Images