Method for predicting and evaluating service life of steel structure building

Abstract

The invention provides a method for predicting and evaluating the service life of a steel structure building, and the method comprises the steps: achieving the real-time collection, preprocessing and spatial mapping of structure and environment data through the deployment of a multi-type sensor network, fusing the structure state and environment variable time sequence characteristics, and mining a key degradation factor through principal component analysis and an attention mechanism; on this basis, an environment driving factor is used as a time-varying parameter to be embedded into a physical basic degradation model, Bayesian reasoning and an online learning mechanism are combined, model parameters are dynamically adjusted and optimized, and confidence interval evaluation of a structure degradation path and residual life is output; according to the method, the accuracy and generalization ability of steel structure degradation mechanism understanding and life prediction can be improved, and the adaptability in a complex environment is enhanced.

Description

Technical Field

[0001] This invention relates to the field of steel structure health monitoring and life prediction technology, and in particular to a method for predicting and assessing the service life of steel structure buildings. Background Technology

[0002] In the current field of steel structure building health monitoring and life prediction technology, mainstream technologies generally adopt life prediction methods based on structural performance monitoring and historical degradation data. Common solutions include deploying stress, strain, displacement, and crack propagation sensors to achieve real-time monitoring of the physical state of the steel structure, and combining this with periodic inspection and maintenance records to assess the structural health status and remaining life. Meanwhile, some cutting-edge systems are beginning to introduce data-driven degradation models, such as applying statistical regression, machine learning algorithms, or crack growth models based on physical mechanisms (e.g., the Weibull degradation model, Paris formula, etc.), in order to improve the accuracy and automation of life prediction. In practical engineering applications, the industry has gradually promoted technologies such as multi-source data fusion, structural state modeling, and life risk assessment, emphasizing monitoring accuracy and assessment efficiency. However, in complex service environments, the degradation process of steel structures is not only affected by their own material properties and mechanical loads, but also highly dependent on external environmental variables such as temperature, humidity, wind speed, load fluctuations, and corrosion rates. Existing mainstream degradation models still have significant shortcomings in modeling the dynamic influence of environmental parameters. Most models tend to treat environmental factors as static correction terms or simple weighted parameters, failing to achieve a synchronous correlation between dynamic changes in environmental variables and structural degradation paths. This leads to errors in model predictions that deviate from reality under rapid environmental changes or multiple operating conditions. Currently, most representative structural life prediction technologies are applicable to standard operating conditions with relatively simple environmental loads or limited parameter variations. For example, crack propagation formulas based on physical mechanisms only consider load cycles and material properties, neglecting the time-varying nature of environmental parameters such as temperature, humidity, and corrosion. While data-driven machine learning models can capture certain correlations, they exhibit insufficient generalization ability under conditions of drastic fluctuations in environmental variables or novel operating conditions. Furthermore, although there are comprehensive evaluation systems based on multi-source health monitoring data in the industry, environment-structure coupling modeling still relies mainly on static grouping or empirical weights, failing to dynamically track the real-time modulation effect of environmental variables on the degradation rate, and exhibiting limited physical interpretability and trend-following capabilities. Summary of the Invention

[0003] In order to solve the above-mentioned technical problems, the present invention provides a method for predicting and evaluating the service life of steel structure buildings.

[0004] The technical solution of this invention is implemented as follows: A method for predicting and assessing the service life of steel structure buildings, comprising: S1: Collect multi-source heterogeneous data of steel structure buildings during different service cycles. The multi-source heterogeneous data includes structural state parameters, environmental parameters and historical maintenance records. Add timestamps and collection location tags to each type of data. S2: Normalize and imput missing values ​​of the collected structural state parameters to eliminate data anomalies caused by sensor drift or communication interruption, thereby improving the stability of subsequent feature extraction. S3: Based on the time series characteristics of environmental parameters, extract environmental driving factors such as temperature and humidity fluctuation rate, load change rate and corrosion rate, and construct a time-varying feature matrix of environmental variables; S4: Multimodal feature fusion is performed on the time-varying feature matrix of structural state parameters and environmental variables. A combined approach of principal component analysis and attention mechanism is used to extract key degradation features in order to reduce redundant information and enhance the expressive power of the model. S5: Based on the physical degradation mechanism, a basic degradation equation is established, and the extracted environmental driving factors are embedded into the degradation equation as time-varying parameters to construct an environment-driven dynamic degradation model. S6: The dynamic degradation model is trained and optimized using historical service data, and the uncertainty of the model parameters is estimated using Bayesian inference method to improve the generalization ability of the model under complex working conditions. S7: Input real-time monitoring data into the optimized dynamic degradation model, dynamically update the model parameters to adapt to the evolution of structural performance, and generate structural degradation path predictions that take into account the influence of environmental variables. S8: Based on the degradation path prediction results output by the model, combined with the structural safety threshold, determine whether the current service status meets the design life requirements, and generate a confidence interval report of the remaining service life. S9: If it is determined that there is a significant deviation between the current combination of environmental variables and the distribution of historical training data, the incremental update mechanism of the model is triggered to improve the model's adaptability to new working conditions.

[0005] The present invention provides a method for predicting and evaluating the service life of steel structure buildings, which has the following beneficial effects: (1) This invention innovatively uses multi-dimensional environmental factors such as temperature and humidity fluctuation rate, load change rate, and corrosion rate as time-varying driving parameters. Through physical mechanisms and dynamic system theory, these factors are dynamically embedded into the degradation differential equation, realizing the real-time dynamic response of the structural degradation path to the actual service environment. Compared with traditional degradation models that rely solely on static or simplified environmental correction coefficients, this invention significantly improves physical interpretability and trend tracking capabilities, effectively reflecting the phased acceleration effect of environmental abrupt changes on the degradation rate under complex working conditions. (2) This invention utilizes a real-time sensor network to synchronously collect structural state information such as stress, strain, displacement, and crack propagation, as well as environmental parameters such as temperature, humidity, wind speed, and corrosion rate. It also strictly aligns timestamps and spatial location information, combining historical maintenance records to form a high-dimensional data input set oriented towards the entire service life. Multimodal data is used to collaboratively extract key degradation features through principal component analysis and attention mechanisms, reducing redundant information and dimensional interference, thus enabling the model to have stronger generalization capabilities for various service environments. Compared with conventional models based on a single sensor or data dimension, this invention exhibits higher robustness and stability in cross-condition and cross-regional life prediction scenarios. (3) This invention employs Bayesian inference and Markov chain Monte Carlo (MCMC) techniques to comprehensively model the uncertainty of the core parameters of the dynamic degradation model and optimize the posterior distribution, thereby enabling the model output to provide a lifetime assessment with a confidence interval. This expands the assessment of the remaining lifetime of a structure from a single numerical value to a probability density prediction, providing a complete risk warning and decision-making basis for actual operation and maintenance. Furthermore, the lifetime prediction error is significantly reduced in complex environments, the remaining lifetime confidence interval is significantly narrowed, and the reliability is significantly improved. (4) This invention integrates incremental learning and online model update mechanisms to monitor the distribution of environmental variables in real-time monitoring data. When the current environmental state is found to deviate significantly from the historical training data, local parameters are automatically updated quickly to ensure that the model still has high-precision lifetime prediction capability in unknown or extreme environments (such as extreme humidity, sudden corrosion, unconventional load fluctuations, etc.). In typical abrupt change environments, the model's adaptive adjustment can significantly reduce the prediction error, and the model's transfer and expansion performance is particularly outstanding. Attached Figure Description

[0006] Figure 1 This is a flowchart of a method for predicting and evaluating the service life of steel structure buildings according to the present invention; Figure 2 This is a sub-flowchart of a method for predicting and evaluating the service life of steel structure buildings according to the present invention. Figure 3 This is another sub-flowchart of the method for predicting and evaluating the service life of steel structure buildings according to the present invention. Detailed Implementation

[0007] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0008] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0009] When used herein, the singular forms of “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising / including” or “having,” etc., specify the presence of the stated features, wholes, steps, operations, components, parts, or combinations thereof, but do not preclude the possibility of the presence or addition of one or more other features, wholes, steps, operations, components, parts, or combinations thereof. Meanwhile, the term “and / or” as used in this specification includes any and all combinations of the associated listed items.

[0010] Please see Figures 1-3 As shown, a method for predicting and assessing the service life of steel structure buildings includes: S1: Collect multi-source heterogeneous data of steel structure buildings during different service cycles. The multi-source heterogeneous data includes structural state parameters, environmental parameters and historical maintenance records. Add timestamps and collection location tags to each type of data. S2: Normalize and imput missing values ​​of the collected structural state parameters to eliminate data anomalies caused by sensor drift or communication interruption, thereby improving the stability of subsequent feature extraction. S3: Based on the time series characteristics of environmental parameters, extract environmental driving factors such as temperature and humidity fluctuation rate, load change rate and corrosion rate, and construct a time-varying feature matrix of environmental variables; S4: Multimodal feature fusion is performed on the time-varying feature matrix of structural state parameters and environmental variables. A combined approach of principal component analysis and attention mechanism is used to extract key degradation features in order to reduce redundant information and enhance the expressive power of the model. S5: Based on the physical degradation mechanism, a basic degradation equation is established, and the extracted environmental driving factors are embedded into the degradation equation as time-varying parameters to construct an environment-driven dynamic degradation model. S6: The dynamic degradation model is trained and optimized using historical service data, and the uncertainty of the model parameters is estimated using Bayesian inference method to improve the generalization ability of the model under complex working conditions. S7: Input real-time monitoring data into the optimized dynamic degradation model, dynamically update the model parameters to adapt to the evolution of structural performance, and generate structural degradation path predictions that take into account the influence of environmental variables. S8: Based on the degradation path prediction results output by the model, combined with the structural safety threshold, determine whether the current service status meets the design life requirements, and generate a confidence interval report of the remaining service life. S9: If it is determined that there is a significant deviation between the current combination of environmental variables and the distribution of historical training data, the incremental update mechanism of the model is triggered to improve the model's adaptability to new working conditions.

[0011] Step S1: Collect multi-source heterogeneous data of the steel structure building during different service cycles. This multi-source heterogeneous data includes structural state parameters, environmental parameters, and historical maintenance records. A timestamp and collection location label are added to each type of data. Specifically, this includes: S1.1: Deploy a multi-type sensor network to collect real-time structural state parameters such as stress, strain, displacement, deformation, and crack propagation rate of key structural nodes in steel structure buildings, so as to obtain physical quantity data reflecting the degradation of structural performance. The input conditions include the location of key structural nodes in steel structure buildings and the requirements for collecting their mechanical response characteristic parameters, involving deployment schemes for various types of sensor networks such as stress-strain sensors, displacement sensors, and crack propagation monitoring equipment; A distributed stress-strain sensor network (parameters: range 1-500MPa, sampling frequency 1-1000Hz) is used to measure the stress at a specified beam-column joint. and strain Real-time data collection; Furthermore, by using a laser displacement sensor array (parameters: resolution 0.001-0.1 mm, measurement frequency 1-500 Hz), displacement deformation measurement of key span areas is achieved, and displacement response time-series data is obtained. ; Furthermore, the energy release rate at the crack tip is achieved using an acoustic emission crack propagation monitoring device (threshold frequency range of 20-400kHz). Dynamic acquisition and generation of crack length increments. sequence; Furthermore, stress-strain data is synchronized via a time synchronization protocol. Displacement data With crack increment data Alignment is performed using a unified timestamp to form a comprehensive acquisition matrix of structural state parameters. ; By using sensor calibration and data synchronization processing, the physical quantity acquisition results of the previous step are transformed into a raw dataset of structural states that can be directly used for degradation analysis, achieving the technical effect of high-precision, multi-dimensional structural performance degradation input. For example, in a span Stress and strain sensors are deployed on a steel structure bridge. One, range MPa, sampling frequency Hz; Laser displacement sensors are deployed in the middle of the main span. One, resolution mm; Acoustic emission crack monitoring probes were installed at three welding nodes. One, frequency range kHz. During operation, the stress collected at a certain moment is... MPa, strain is × The displacement is mm, crack length increment is mm. This data can be aligned using a uniform timestamp to build... The matrix was used as input to the degradation model for subsequent analysis, verifying that the sensor network deployment scheme could... Multi-source data synchronization is achieved within millisecond-level time deviations, ensuring the accuracy and completeness required for degradation path modeling; S1.2: Periodically collect service environment parameters based on temperature and humidity sensors, wind speed and pressure sensors, and corrosion rate monitoring devices to construct a time-series dataset characterizing environmental changes; Input conditions include raw data collected from temperature and humidity sensors, wind speed and pressure sensors, and corrosion rate monitoring devices located at different key locations in the steel structure building, covering temperature... ,humidity Wind speed Wind pressure and corrosion rate Multidimensional time series data; Periodic acquisition algorithm is used (parameter: acquisition period) s, sampling accuracy temperature ± ℃, humidity ± %RH, wind speed ± m / s, wind pressure ± (Pa), to realize the function of timed acquisition of environmental parameters; Furthermore, through the time synchronization protocol (parameter: NTP time synchronization precision) (ms), to unify the timestamps of collected data on temperature, humidity, wind speed, wind pressure, and corrosion rate, and to obtain the original time-series matrix of environmental parameters. ; Furthermore, a sliding window mean filtering algorithm is adopted (window length...) (Points) to suppress noise in the original time series matrix and generate a smoothed environmental data sequence; Furthermore, a volatility calculation method is used to calculate the temperature volatility index, thereby quantifying the intensity of temperature changes and obtaining volatility parameters. Furthermore, a differential calculation method is used to extract the wind speed change rate characteristics and generate a wind speed dynamic response sequence; Through the above algorithm chain, the original synchronized environmental data from the previous step is transformed into a time-series dataset containing temperature and humidity fluctuation rate, wind speed change rate and corrosion rate, thereby achieving high-precision time-series feature output that characterizes the dynamic changes of the service environment. For example, temperature and humidity sensors are installed on the north and south sides of a coastal steel structure factory building. One, accuracy ± ℃, ± %RH; Wind speed and pressure sensors installed on the roof One, accuracy ± m / s, ± Pa; Corrosion rate monitoring devices are installed at the bottom steel beam anti-corrosion coating. One, precision mm / a. The periodic acquisition period is set to... s, NTP time synchronization accuracy The temperature sequence within a certain window is [25.1, 25.3, 25.4, 25.2, 25.5]℃. Calculate the mean. = ℃, volatility ≈ The environmental temporal feature matrix was obtained after time synchronization and filtering. The wind speed sequence was [4.2, 4.3, 4.6, 4.4, 4.5] m / s, the difference rate sequence was [0.1, 0.3, -0.2, 0.1] m / s, and the measured corrosion rate was [0.021, 0.019, 0.020, 0.022] mm / a. It accurately reflects the correlation between temperature and humidity fluctuations and corrosion rate under the influence of monsoons on the north and south sides of the factory, providing low-noise and high-real-time basic data for the construction of environmental driving factors in subsequent degradation models; S1.3: Call the historical maintenance database to extract structural intervention information such as maintenance records, inspection reports, and repair timestamps generated during the maintenance, repair, and inspection of the steel structure building, so as to form a historical maintenance record dataset; S1.4: Add a unified format timestamp tag to the collected structural state parameters, environmental parameters and historical maintenance records to achieve time alignment and service cycle mapping of multi-source data; S1.5: Attach a structural spatial location label to each type of collected data, and annotate the sensor deployment location and corresponding component number based on the structural topology map to achieve spatial data positioning and structural mapping.

[0012] Step S2: Normalize and imput missing values ​​in the collected structural state parameters to eliminate data anomalies caused by sensor drift or communication interruption, thereby improving the stability of subsequent feature extraction. Specifically, this includes: S2.1: Perform data cleaning operations on the collected structural state parameters, and use the sliding window mean filtering algorithm to identify and remove outliers to obtain the original structural state data sequence after noise reduction; The structural state parameter matrix acquired via step S1 Perform data cleaning operations, with input conditions including stress. ,strain Displacement and crack increment Multidimensional time-series records; A sliding window mean filtering algorithm is used (parameter: window length). Point, window slide step (Points) to achieve local statistical smoothing of each physical quantity sequence; Furthermore, by calculating the mean of the measurements within the window... with standard deviation Outliers are identified using the three-standard-deviation principle, and the detection formula is as follows:

[0013] in For the first in the window Each sample value; Furthermore, for the identified abnormal samples, the window mean replacement method is used to remove them and replace them with the mean value of the current window to correct the abnormal data points; Furthermore, boundary smoothing is performed on the anomaly-corrected sequence, using a weighted average filter (with the weight sequence centered on the weights). Adjacent weights ), to reduce the step change at the data correction point; Through the above anomaly identification and smoothing processes, the original structural state parameter matrix is... The original data sequence of the structural state is transformed into a denoised data sequence, which improves the stability of the data and the effectiveness of subsequent feature extraction. For example, in a span In stress monitoring of a steel bridge with a diameter of m, the stress sequence of a certain beam segment was collected as [352, 355, 497, 358, 354, 351, 349] MPa. The window length was used... Calculate the mean = MPa, standard deviation = MPa. For the third value MPa performs anomaly checks: MPa, greater than MPa, identified as an outlier and replaced with the mean. The replaced sequence is [352, 355, 388, 358, 354, 351, 349] MPa. After weighted average filtering, the sequence becomes [353.25, 364.25, 370.25, 365.25, 353.25] MPa. This sequence shows a 35% reduction in noise power in the frequency domain analysis, verifying that denoising improves data stability and laying the foundation for subsequent S2.2 normalization processing. S2.2: Based on the original data sequence of the denoised structural state, the Z-score normalization method is used to normalize the multi-dimensional structural state parameters to eliminate the influence of the difference in the dimensions of different sensors on the subsequent feature extraction, and obtain the standardized structural state data under the unified dimension. S2.3: To address the missing values ​​in the standardized structural state data, a time-series-based linear interpolation method is used to fill in the missing timestamps to obtain complete and continuous structural state time series data. S2.4: Based on complete and continuous structural state time series data, a sliding window segmentation mechanism is used to extract time-domain feature segments to construct a set of structural state feature segments to support subsequent feature engineering processing; S2.5: Perform variance stability tests on each segment in the set of structural state feature segments, use the sliding window standard deviation analysis method to identify data drift points, and correct local data offsets through a dynamic threshold adjustment mechanism to obtain a structural state feature dataset with enhanced stability.

[0014] Step S3: Based on the time-series characteristics of environmental parameters, extract environmental driving factors such as temperature and humidity fluctuation rate, load change rate, and corrosion rate, and construct a time-varying feature matrix of environmental variables. For example... Figure 2 As shown, it specifically includes: S3.1: Perform sliding window segmentation on the temperature and humidity sensor data in the environmental parameters to obtain the local change trend of the time series and obtain the segmented feature vector of the temperature and humidity time series for subsequent volatility calculation; S3.2: Based on the segmented feature vector of temperature and humidity time series within the sliding window, the standard deviation and range analysis methods are used to calculate the temperature and humidity fluctuation index in order to quantify the potential impact of environmental temperature and humidity changes on the structural corrosion process. Based on the segmented feature vectors of temperature and humidity time series within a sliding window, standard deviation and range analysis methods are used (parameter: window length). Sampling points, step size (Sampling points) to achieve quantitative calculation of local temperature and humidity fluctuations; Furthermore, by calculating the standard deviation, the dispersion of temperature and humidity within each window is measured, yielding the variance component of the temperature and humidity fluctuation rate, as shown in the formula:

[0015] in The current window standard deviation, For the first in the window A temperature or humidity sample, The mean within the window. This represents the number of samples in the window. Furthermore, by calculating the range, the maximum and minimum differences in temperature and humidity within the same window are calculated to obtain the amplitude components of the temperature and humidity fluctuation rate, as shown in the formula:

[0016] in This represents the range of the current window. This refers to the maximum temperature or humidity value inside the window. This represents the minimum temperature or humidity value inside the window. Furthermore, by normalizing the standard deviation and range, a unified dimensional expression for fluctuation indicators of different scale windows is achieved, thereby obtaining a comparable temperature and humidity fluctuation rate index matrix, the formula of which is:

[0017] in Normalized volatility; Furthermore, the volatility index matrix of each window is aligned with the environmental data collection time according to the timestamp to form a time series curve of temperature and humidity volatility, which is used for subsequent dynamic modeling of the corrosion process. By using the above-mentioned standard deviation and range joint analysis method, the piecewise eigenvector of the sliding window is transformed into a volatility index that reflects the intensity of temperature and humidity changes, thereby achieving a quantitative characterization of the impact of environmental temperature and humidity changes on structural corrosion. For example, in the environmental monitoring of a steel structure factory building in a coastal area, the temperature sensor sampling frequency is... Hz, the humidity sensor sampling frequency is Hz, using window length Point, step size Calculate the standard deviation and range for a given temperature window: [25.2, 25.8, 26.0, 25.7, 26.4, 27.1, 26.9, 26.3, 26.6, 27.0]℃. Calculate the window mean. = ℃, standard deviation = ℃, range = ℃, normalized volatility = The humidity window data is [68.1, 68.4, 68.9, 69.2, 69.0, 69.7, 70.1, 70.3, 70.0, 69.8]%RH, with the window mean... = %RH, standard deviation = %RH, range = %RH, normalized volatility = In this scenario, after a year of monitoring and model training verification, the corrosion rate corresponding to the high volatility segment increased by 23%, proving that this volatility index is effective in predicting the corrosion process. S3.3: Perform differential processing on the structural load monitoring data to extract the load change rate characteristics, so as to reflect the degree of influence of the dynamic changes of external load on the fatigue damage accumulation rate of steel structure; The structural load time series matrix acquired via a load sensor network The first-order difference method is used (parameter: difference order). Sampling interval for s), to achieve instantaneous progressive calculation of load changes at adjacent sampling points; Furthermore, by using the differential calculation formula, the quantitative extraction of the load change per unit time is achieved, and the original sequence data of the load change rate is obtained; Furthermore, a sliding window mean filter is applied to the difference results (parameter: window length). Sampling points, step size Sampling points are used to reduce sudden noise interference and highlight trend change characteristics to obtain a smoothed load change rate sequence; Furthermore, through standardized calculations, the load change rate data is transformed into a normalized index with zero mean and unit variance, ensuring that the load change rate under different working conditions is comparable under a unified dimension. Furthermore, a peak locking and duration identification algorithm is performed on the normalized load change rate sequence (the peak threshold is selected as the standard deviation). (Multiple times), to screen out the key locations and durations of rapidly changing load events, in order to characterize the intensity of the impact of dynamic external load impact on the fatigue damage accumulation rate; Through the above differential processing, smoothing filtering, normalization and peak locking, the load time series matrix is ​​transformed into a load change rate index with time-series characteristics, realizing the effective embedding of external load dynamic change information in fatigue damage prediction. For example, in a span In the crane operation load monitoring of the m-shaped steel structure factory building, the sensor sampling frequency is [missing information]. The original load sequence was [102.5, 106.8, 115.4, 118.0, 117.2, 130.5, 128.9] kN. The first-order finite difference formula was used to calculate... The rate of change at point 2 = kN / s, the rate of change at point 3 kN / s, and so on, to obtain the original rate of change sequence [4.3, 8.6, 2.6, -0.8, 13.3, -1.6] kN / s. This is followed by a sliding window of length... Step length Mean filtering smooths the rate of change sequence to [4.18, 4.18, 4.28, 3.7, 5.86, 4.38] kN / s. After standardization, the first value is normalized to... = The generated normalized sequence identified a peak event with a duration of 1 second at the 5th point in the peak locking. This peak corresponds to the rapid load change caused by the instantaneous acceleration of the crane. The fatigue damage model calculation verified that the contribution of this event to the fatigue accumulation rate was increased by 18%, indicating that this load change rate index can effectively reflect the impact of external load dynamics on structural degradation. S3.4: Based on the corrosion rate model, linear combination and nonlinear mapping of temperature and humidity fluctuation rate and load change rate are performed to generate comprehensive corrosion rate characteristic parameters, which are used to characterize the coupling relationship between environmental factors and structural degradation rate. Based on the temperature and humidity fluctuation index matrix obtained through steps S3.2 and S3.3 Normalized sequence of load change rate A linear combination method using the corrosion rate model (parameter: linear weight coefficient) is employed. , (This can be determined by historical regression analysis), enabling the estimation of the basic contribution of the linear superposition of environmental factors to the corrosion rate; Furthermore, through the linear combination formula:

[0018] in This is the baseline value for the linear combination corrosion rate. This is the weighting coefficient for temperature and humidity fluctuation rate. As the load change rate weighting coefficient, linear rate synthesis based on two types of environmental driving factors is realized; Furthermore, to address the nonlinear acceleration effect of the corrosion process, an exponential nonlinear mapping method is employed (parameter: nonlinear acceleration coefficient). Curvature correction coefficient This enables the nonlinear transformation of environmentally driven combined indicators into the actual corrosion rate response; The formula for nonlinear mapping is expressed as:

[0019] in The comprehensive corrosion rate characteristic parameters after nonlinear mapping are used to express the acceleration effect of corrosion rate under high fluctuation environment; Furthermore, on Perform normalization (normalization range [0,1]) to ensure the comparability of comprehensive corrosion rates under different working conditions and time periods; Furthermore, a sliding window mean filter is used (parameter: window length). Sampling points, step size Sampling points) Smoothing and normalization The sequence is used to reduce the impact of instantaneous spikes on the model input, making the corrosion rate characteristic curve more trend-like; Through the above linear combination, nonlinear mapping and normalization smoothing, the synergistic effect of temperature and humidity fluctuation rate and load change rate is transformed into a comprehensive corrosion rate characteristic parameter, realizing the quantitative and stable expression of the coupling relationship between environmental factors and structural degradation rate. For example, in the monitoring of a coastal steel structure factory building, the temperature and humidity fluctuation rate... = Load change rate = Linear weighting coefficients = , = Calculate the linear combination value according to the formula. Nonlinear mapping parameters are taken as follows: = , = Mapping value = Normalizing to the range [0,1] yields... , through window length After mean smoothing, the mean of the comprehensive corrosion rate curve during the high-fluctuation period in the one-year monitoring data increased by about 19%. Model validation showed that the prediction error of the annual corrosion depth corresponding to this feature parameter was reduced to ±5%, proving the ability of this feature to characterize the environmentally driven degradation path. S3.5: Align the temperature and humidity fluctuation rate, load change rate and comprehensive corrosion rate characteristic parameters according to the timestamp to construct a time-varying characteristic matrix of environmental variables, which will serve as input data for subsequent dynamic degradation modeling and multimodal feature fusion.

[0020] Step S4: Multimodal feature fusion is performed on the time-varying feature matrix of structural state parameters and environmental variables. A combined principal component analysis and attention mechanism method is used to extract key degradation features to reduce redundant information and enhance the model's expressive power. For example... Figure 3 As shown, it specifically includes: S4.1: Perform principal component analysis to reduce the dimensionality of the structural state parameters to obtain a low-dimensional structural degradation feature vector, thereby reducing the feature dimension and removing linear correlations in the structural state data. For the collected and preprocessed structural state parameter matrix, principal component analysis (PCA) was used (parameters: covariance matrix calculation based on zero-mean data, Jacobi iteration as the eigenvalue decomposition algorithm, and the number of principal components retained based on a cumulative contribution rate of ≥95%) to achieve linear dimensionality reduction of high-dimensional structural state data; Furthermore, by using a centralized processing method (parameter: mean calculation is based on the average value of each dimension of the entire sample), the data is aligned with the zero mean in each dimension, providing a stable input for covariance matrix calculation; Furthermore, a formula for constructing the covariance matrix is ​​used to ensure that the linear correlation between the dimensions of each parameter is reflected; Furthermore, based on the eigenvalue decomposition method, the covariance matrix is ​​solved for eigenvalues ​​and eigenvectors to obtain a set of principal component loading vectors sorted by eigenvalue size; Furthermore, the cumulative contribution rate is calculated using the following formula:

[0021] in For the front The cumulative contribution rate of each principal component For the first 1 eigenvalue, The entire eigenvalue set is used to determine the number of principal components and perform dimensionality reduction mapping; Furthermore, by multiplying the principal component loading matrix with the original centered data matrix, a set of low-dimensional structural degradation feature vectors is generated, thereby realizing a non-redundant representation of structural state data in a low-dimensional space. By using the PCA dimensionality reduction method described above, the original structural state parameters are transformed into low-dimensional structural degradation feature vectors, achieving the expected technical effect of reducing feature dimensionality, removing linear correlation, and retaining the main degradation information. For example, in a steel structure bridge monitoring scenario, the structural state parameters include three dimensions: stress, strain, and displacement, with a sampling frequency of [missing information]. Hz, number of samples Point. After mean centering, the mean stress is MPa, mean displacement is mm, average strain value Construct the covariance matrix with values ​​of [[12.4, 2.1, 0.003], [2.1, 0.55, 0.0009], [0.003, 0.0009, 1.2e-6]]. The eigenvalue decomposition result is λ = [13.1, 0.9, 1.5e-6]. Calculate the cumulative contribution rate. ≈ The first two principal components are retained, and the dimensionality is reduced to a 2-dimensional degenerate feature vector. The 2-dimensional features output after dimensionality reduction reduce the model training time by 22% and the prediction error from ±8% to ±4% in subsequent degenerate modeling and multimodal fusion, demonstrating the effectiveness of this dimensionality reduction process in improving generalization performance and feature representation efficiency. S4.2: Based on the time-varying feature matrix of environmental variables, a sliding window is used to extract the statistical features of the time series, including the mean, variance and trend term, in order to construct an environmental degradation feature vector and improve the dynamic response capability of environmental variables to degradation paths. Based on the time-varying feature matrix of environmental variables, a sliding window is used to extract time series statistical features (parameter: window length). Sampling points, step size Sampling points) enable the capture of statistical characteristics of local environmental data segments; Furthermore, by using the mean calculation method, the central tendency of environmental variables within the window is measured, and the mean characteristic sequence is obtained. ; Furthermore, by using variance calculation methods, the dispersion of environmental variables within the window is measured, and the variance feature sequence is obtained. ; Furthermore, a linear regression trend term calculation method is used to quantify the trend of parameter changes over time within the window, and to obtain the trend characteristic sequence. ; Furthermore, normalization processing (parameter: Z-score standardization) is performed on the mean, variance, and trend feature sequences to achieve comparability of different environmental variables under a unified scale, and to obtain a normalized environmental degradation feature matrix. By extracting and normalizing the mean, variance, and trend as described above, the time-varying feature matrix of environmental variables is transformed into an environmental degradation feature vector, thereby enabling a quantitative characterization of the dynamic response capability of environmental factors to degradation paths. For example, in the monitoring of a coastal steel structure bridge, the time-varying characteristic matrix of environmental variables includes temperature, humidity, and corrosion rate, with a sampling frequency of [missing information]. Hz, window length Point, step size Point. Given a temperature window with data of [23.5, 23.6, 23.8, 24.0, 24.2, 24.4, 24.6, 24.7, 24.8, 25.0, 25.1, 25.3]℃, calculate the mean. = = ℃, variance calculation = ≈ ℃, trend term calculation ≈ ℃ / s. Z-score normalization was performed on the above features, yielding normalized values ​​of 0.02, 0.05, and 0.03, which were then incorporated into the environmental degradation feature matrix. In the corrosion degradation model validation, the dynamic response characteristics of this matrix improved the prediction accuracy by 14% during periods of sudden high temperature and humidity events, demonstrating the effectiveness of this environmental feature extraction method in modeling degradation paths. S4.3: Concatenate the structural degradation feature vector and the environmental degradation feature vector into a tensor to generate a multimodal fusion feature tensor, thereby forming a joint structure-environment representation space to support the implementation of subsequent feature enhancement mechanisms; The set of low-dimensional structural degradation feature vectors obtained by step S4.1 and the set of environmental degradation feature vectors output by step S4.2 are used to achieve joint encoding of the two types of feature vectors in homogeneous space format by tensor concatenation method (parameters: concatenation dimension is set to the last dimension of the feature, concatenation mode is concat non-overlapping concatenation). Furthermore, through the tensor splicing formula:

[0022] in It is a set of low-dimensional structural degenerative feature vectors. It is a set of feature vectors of environmental degradation. This is a multimodal fusion feature tensor that enables the non-loss combination of structural and environmental information in a unified feature space; Furthermore, a dimension rearrangement algorithm is performed on the concatenated multimodal fusion feature tensor (parameters: prioritize structural feature dimensions, in the order of time, space, and magnitude; arrange environmental feature dimensions according to the physical quantity category) to achieve semantic consistency and storage layout optimization within the feature tensor; Furthermore, a normalization method (parameter: Z-score standardization based on the full feature dimension) is adopted to achieve a unified dimensional representation of features from different sources on a numerical scale, and a standardized fused feature tensor is obtained. ; Furthermore, on Perform shape checks and missing value detection (methods: NaN detection and calculation of the proportion of non-zero elements in the sparse matrix) to ensure the integrity and computability of the data input to subsequent feature enhancement mechanisms; Through the above tensor splicing, dimension rearrangement, normalization and integrity detection processing, structural degradation features and environmental degradation features are mathematically formed into a unified structure-environment joint representation space, realizing the expected technical effect of direct calling and efficient enhancement of multimodal degradation features in subsequent attention mechanisms; For example, in a monitoring scenario of a coastal steel structure bridge, the structural degradation feature vector obtained after dimensionality reduction using S4.1 is a two-dimensional matrix. Size is × The environmental degradation feature vector constructed by S4.2 Size is × The concat concatenation mode is used to fuse feature tensors. Size is × The data are arranged in order of structural dimension (principal components of stress, principal components of displacement) and environmental dimension (mean temperature, variance of humidity, corrosion tendency). After Z-score standardization, the original vector of a fused sample [0.12, -0.08, 24.25, 0.61, 0.16] is transformed into [-0.02, -0.05, 0.02, 0.05, 0.03], ensuring the comparability of features with different dimensions. Missing value detection shows no NaN elements, and the non-zero proportion of the sparse matrix is ​​100%. In subsequent attention mechanism processing, the multimodal consistency of this fused feature tensor improves the stability of key degenerate feature weight allocation by 17%, verifying the effectiveness of this tensor concatenation method in structure-environment joint modeling. S4.4: Apply a multi-head self-attention mechanism to the multimodal fusion feature tensor to calculate the correlation weights between each feature dimension in order to generate attention-weighted feature vectors and enhance the expressive power of key degenerate features. S4.5: Based on attention-weighted feature vectors, a feature selection algorithm is executed to filter out a subset of features that are highly sensitive to the degradation process, and the key degradation feature matrix is ​​output for use as input to the subsequent dynamic degradation model construction module.

[0023] Step S5: Establish a basic degradation equation based on the physical degradation mechanism, and embed the extracted environmental driving factors as time-varying parameters into the degradation equation to construct an environment-driven dynamic degradation model. Specifically, this includes: S5.1: Based on structural mechanics and material degradation mechanisms, the basic degradation behavior characteristics of typical steel structures during service are obtained. The basic degradation equation is constructed using the Weibull degradation model or the Paris crack propagation model to form a degradation evolution framework with a physical basis. S5.2: The environmental driving factors extracted in S3, including time-varying environmental variables such as temperature and humidity fluctuation rate, load change rate and corrosion rate, are normalized and time-aligned to obtain a set of environmental driving time series parameters that evolve synchronously with the structural degradation state. For time-varying environmental variables such as temperature and humidity fluctuation rate, load change rate, and corrosion rate extracted from S3, the Z-score standardization method (parameters: mean and standard deviation are calculated based on full-time samples) is used to achieve numerical scale unification, so as to eliminate modeling bias caused by differences in the dimensions and numerical ranges of different physical quantities. Furthermore, by using a timestamp alignment algorithm (parameters: the reference time series is set to the structural state sampling time series, and the interpolation method is linear interpolation), the environmental variables are strictly synchronized with the structural degradation state in the time dimension. Furthermore, a dynamic resampling method (parameters: resampling frequency set to 1 Hz, consistent with structural state sampling, and constant value extension selected for boundary processing) is adopted to reconstruct the environmental variable time series according to a uniform sampling rate, thereby obtaining an environmental driving data sequence of equal length and frequency. Furthermore, based on the formula:

[0024] in These are the original time-varying environmental variable sequence values. The mean of the sequence. To calculate the standard deviation of the sequence, normalized sequence values ​​are calculated to obtain a standardized set of environmental driving factor sequences with zero mean and unit variance. ; Furthermore, a time-sliding window smoothing process (window length 12 points, step size 1 point) is performed on the normalized sequence set to smooth out short-term fluctuations and obtain a more stable set of environment-driven time series parameters. Through the above normalization, time alignment, resampling and smoothing processes, the original environmental variables are transformed into a set of environmental-driven time-series parameters that are strictly synchronized with the structural degradation state and have uniform dimensions, so as to realize the accurate temporal embedding and dynamic response characterization of environmental factors in degradation modeling. For example, in a monitoring scenario of a coastal steel structure factory building, the temperature and humidity fluctuation rate sequence has a length of 1000 points and a mean of [missing value]. The standard deviation is The load change rate sequence has a length of 980 points and a mean of [missing value]. kN / s, standard deviation is The corrosion rate sequence has a length of 995 points and a mean of [missing value]. mm / d, standard deviation is The three types of sequences are aligned to a uniform length of 1000 points using linear interpolation, and normalized values, such as the corrosion rate at a certain point, are calculated according to the formula described above. Normalization = After sliding window smoothing, the standard deviation of local fluctuations in the environment-driven time series parameter set decreased to 0.015, and the noise impact during modeling was reduced by approximately 38%. In the subsequent S5.3 stage of constructing the environment-structure coupling equations, the fitting accuracy of the degradation path was improved by 12%. S5.3: Based on dynamic system theory, the processed environment-driven time-series parameters are embedded as external excitation terms into the basic degradation equation to construct an environment-structure coupled dynamic degradation differential equation model, so as to realize the real-time modulation effect of environmental variables on the degradation path. The input consists of the set of environment-driven timing parameters obtained by standardization, time alignment and smoothing in step S5.2, and the mathematical structure of the basic degradation equation established in step S5.1. A dynamic system modeling method (parameters: system type is set as nonlinear time-varying, state variable is degradation damage amount D(t), external excitation is environmental driving factor sequence, and solution framework is differential equation structure) is adopted to realize the real-time modulation of degradation state by environmental influence as dynamic external input; Furthermore, through an external excitation coupling strategy (parameters: the coupling form is set to a hybrid mode of multiplicative coupling and additive coupling, and the weight coefficients are determined based on the sensitivity analysis of the influence of environmental factors on the degradation rate), the direct effect of environmental driving factors on the rate of state change in the basic degradation equation is realized, and the coupled equation structure is obtained. Furthermore, parameter normalization and time-domain synchronization ensure that the time indices of each environmental driving term in the equation are consistent, and a dynamic degenerate differential equation model that meets the input conditions is generated. Furthermore, by performing stability analysis on the coupled equations (method: Lyapunov function construction and sign derivative test), unstable coupled parameter combinations are screened to ensure the physical rationality and numerical stability of the model during long-term integration. By using the above dynamic system coupling processing method, the environmental driving timing parameters obtained in the previous step are embedded into the basic degradation equation in the form of strictly synchronized external excitation, so as to realize the real-time modulation effect of environmental conditions on the structural degradation path and improve the dynamic response capability and physical interpretability of the degradation model. For example, in a monitoring scenario of a coastal steel structure bridge, the Paris crack propagation model is selected as the foundation degradation equation. ,in This represents the stress intensity factor amplitude. and For material constants. Environmental driving parameters include the average temperature and humidity fluctuation rate E1(t) of 0.12, the average load change rate E2(t) of 2.5 kN / s, and the average corrosion rate E3(t) of 0.0025 mm / d, with coupling coefficients set as α1 = 0.08, α2 = 0.15, and α3 = 0.21, respectively. The coupling equations are constructed as follows: The stability of the system was tested using the Lyapunov function V(D)=D^2, and numerical verification showed that the system remained stable within an environmental fluctuation range of ±15%. The output environmental-structure coupling equation accurately reflects the sudden increase in crack propagation rate under high humidity stress fluctuations, with a prediction accuracy improved by approximately 13% compared to the uncoupled model, thus verifying the effectiveness of this dynamic coupling modeling. S5.4: The Runge-Kutta numerical integration method is used to discretize and solve the constructed dynamic degradation differential equation, generating a numerical solution sequence of the structural degradation state over time, so as to obtain the degradation path response curve under different combinations of environmental variables. S5.5: Based on the key degradation features extracted in S4, parameter identifiability analysis is performed on the constructed dynamic degradation model, and the least squares optimization algorithm is used to identify and correct the environmental coupling coefficient in the model, so as to improve the prediction accuracy and stability of the model under actual working conditions.

[0025] Step S6: The dynamic degradation model is trained and optimized using historical service data. Bayesian inference is used to estimate the uncertainty of the model parameters to improve the model's generalization ability under complex operating conditions. Specifically, this includes: S6.1: Perform time-series alignment and operating condition grouping on the structural degradation samples in the historical service data to eliminate the initial state deviation caused by differences in service cycles and obtain a standardized historical degradation trajectory dataset. S6.2: Based on the mathematical structure of the dynamic degradation model, perform parameter sensitivity analysis on the standardized historical degradation trajectory dataset to identify key parameters that significantly affect the model output, so as to guide the selection of key variables for subsequent uncertainty modeling; Based on the mathematical structure of the dynamic degradation model, the input for performing parameter sensitivity analysis on the standardized historical degradation trajectory dataset is the time series of structural degradation state and the synchronization sequence of environmental driving factors after processing in step S6.1. The local perturbation method (parameters: perturbation amplitude set to ±5% of the nominal parameter value, perturbation step size of 0.5%) is used to make small adjustments to the model parameters one by one, and the degradation path deviation of the model output is calculated to achieve preliminary quantification of the model output response to each parameter. Furthermore, using the Sobol method for global sensitivity analysis (parameters: 10,000 samples, uniform sampling distribution), the total sensitivity index and first-order sensitivity index of each parameter are calculated to decompose the contribution of different parameters to the variance of the degradation path and obtain the sensitivity ranking results. Furthermore, based on the variance attribution results, a parameter sensitivity matrix is ​​constructed. ,in For parameter index, For output response feature indices, matrix elements represent the first... The parameter for the first Sensitivity values ​​of each output feature are used to map key parameters to specific structural performance indicators. Furthermore, a normalization process is applied (parameter: normalized to [0,1] based on the maximum value of each column) to obtain the parameter importance score matrix, and then based on a preset threshold. A subset of highly sensitive parameters is selected to form the key variable set for subsequent uncertainty modeling; Through sensitivity analysis, variance attribution, and parameter screening, the complex parameter space is transformed into a set of key parameters that significantly affect the model output, thereby improving the computational efficiency of subsequent Bayesian inference steps and focusing uncertainty modeling. For example, in a coastal steel bridge scenario, the historical degradation trajectory dataset includes a crack length time series (sampling frequency 1 Hz, length 2000 points) and three types of environmental driving factor sequences (mean temperature and humidity fluctuation rate 0.12, mean load change rate 2.5 kN / s, and mean corrosion rate 0.0025 mm / d). Parameters are coupled using the basic degradation equation. =0.08、 =0.15、 =0.21 is the nominal value. The change in crack propagation rate is calculated based on a ±5% amplitude disturbance, and the preliminary response degree ranking is obtained as follows: > > Global analysis using the Sobol method yielded total sensitivity indices of 0.72, 0.64, and 0.58, with corresponding first-order indices of 0.69, 0.61, and 0.55. A parameter sensitivity matrix was then constructed. After normalization, the results were filtered out using a threshold of 0.65. and This serves as a key parameter set. Verification has shown that this set can reduce sampling computation by approximately 33% in subsequent Bayesian inference while maintaining prediction accuracy within ±2%, effectively improving the model's generalization ability in complex environments. S6.3: The Markov Chain Monte Carlo (MCMC) sampling method is used to perform Bayesian inference on key parameters. Based on the likelihood function and prior distribution, a posterior probability model is established to quantify the uncertainty distribution of model parameters under different service conditions. S6.4: Perform maximum a posteriori estimation (MAP) and uncertainty interval extraction on the posterior distribution of parameters obtained by Bayesian inference to generate a statistically significant set of optimized parameters to support the enhancement of the model's robustness under different combinations of environmental variables; S6.5: Based on the parameter optimization set, the dynamic degradation model is trained and iterated multiple times. The generalization performance of the model under different environmental driving factors is evaluated by combining the cross-validation mechanism to obtain the optimal model parameter configuration and the corresponding uncertainty confidence interval.

[0026] Step S7: Input real-time monitoring data into the optimized dynamic degradation model, dynamically update the model parameters to adapt to structural performance evolution, and generate a structural degradation path prediction that considers the influence of environmental variables. Specifically, this includes: S7.1: Acquire real-time structural state data and environmental parameter data from a distributed sensor network. The structural state data includes displacement, strain, and vibration frequency, and the environmental parameters include temperature and humidity, load fluctuation, and corrosion rate, in order to construct a complete real-time input dataset. S7.2: Based on the key degradation feature templates extracted in S4, feature mapping and normalization are performed on the real-time input dataset obtained in S7.1 to obtain a feature space representation consistent with the training data, thereby improving the consistency of model input and prediction stability. S7.3: Input the standardized feature vector output from S7.2 into the environment-driven dynamic degradation model constructed in S5, and perform forward propagation calculation based on the current parameter configuration of the model to generate the initial degradation path prediction value; The standardized feature vector obtained by step S7.2 is input into the environment-driven dynamic degradation model constructed in S5. The forward propagation calculation method (parameters: integration step size Δt=0.1d, number of iterations N=500) is used to realize the prediction calculation of the model under the current parameter configuration. Furthermore, by using matrix multiplication and a temporal iterative update mechanism (parameters: coefficient matrix size m×n, sparsity <0.1), the coupled calculation of structural state and environment driving terms is realized to obtain the instantaneous dynamic response sequence of the degradation path; Furthermore, based on the numerical integration method of the Runge-Kutta fourth-order scheme (parameters: time interval T=100d, initial conditions from the latest real-time input), the degenerate differential equation is discretized and solved, generating a high-precision calculation sequence of the degenerate state over time. ; Furthermore, the degradation prediction value is calculated using the feature response mapping function. ; Furthermore, noise suppression is performed on the discrete prediction sequence using a smoothing filtering algorithm (parameters: window length w=5, weight type is Gaussian distribution) to obtain smoothed initial degradation path prediction values; Through the above iterative solution and smoothing process, the real-time input data after standardization mapping is transformed into the initial degradation path prediction curve, so as to realize the real-time evolution calculation of the structural state under the influence of environmental variables. For example, in a real-time monitoring scenario of a coastal steel structure bridge, the input feature vector includes displacement of 0.45 mm, strain of 450 με, vibration frequency of 12.5 Hz, temperature and humidity fluctuation rate of 0.15, load change rate of 3.0 kN / s, and corrosion rate of 0.003 mm / d. Under the conditions of Δt = 0.1d and N = 500, the Runge-Kutta fourth-order integral generates... The value of the sequence at T=50d is 0.85 (standardized dimension of degradation). After Gaussian smoothing, the variance is reduced to 0.65 times the original value. The mean absolute error of the prediction curve with historical validation data under short-term operating conditions remains within ±1.8%, realizing dynamic response capture and high-precision degradation trend tracking of the impact of complex environment. S7.4: Based on the parameter uncertainty distribution estimated by the Bayesian inference method in S6, the probability density model of the initial degradation path prediction value output by S7.3 is performed to generate a degradation path probability distribution function with confidence assessment. S7.5: Based on the degradation path probability distribution function output by S7.4, and combined with the environmental data distribution deviation threshold preset in S9, determine whether the current service environment is within the coverage of the model training data, and generate a decision signal on whether the model needs to trigger incremental updates. S7.6: If S7.5 does not trigger the incremental update mechanism, the model parameters will be dynamically adjusted based on the latest prediction results. An online learning mechanism will be used to perform lightweight parameter fine-tuning on the degraded model to enhance the model's adaptability to the current service status and prediction accuracy. S7.7: Outputs the degradation path prediction results updated in S7.6, including the trend curve of the degradation state over time and its confidence interval, providing input for the structural life assessment in S8.

[0027] Step S8: Based on the degradation path prediction results output by the model, and combined with the structural safety threshold, determine whether the current service status meets the design life requirements, and generate a confidence interval report for the remaining service life. Specifically, this includes: S8.1: Perform time-series alignment and normalization on the degradation path prediction results output by the model to eliminate evaluation bias caused by differences in sampling frequency or inconsistent model output scale. S8.2: Based on the normalized degradation path data, the sliding window method is used to calculate the time series characteristics of the structural degradation rate in order to capture the local trend changes in the degradation process; S8.3: Using the structural safety threshold as the evaluation benchmark, perform threshold comparison processing on the degradation rate time series to identify whether the current service status is close to or exceeds the design life critical point; S8.4: Based on the parameter uncertainty distribution output by the Bayesian inference module, the Monte Carlo simulation method is used to generate multiple sets of degradation path simulation samples to construct a probability distribution model of the remaining useful life; S8.5: Estimate the confidence interval of the probability distribution model of the remaining useful life and generate a report on the upper and lower limits of the remaining useful life at a 95% confidence level to provide a quantitative basis for operation and maintenance decisions; S8.6: Visualize and integrate the confidence interval report with the structural service status assessment results, and output a comprehensive assessment report including degradation trend, safety status and remaining life for operation and maintenance personnel to refer to.

[0028] Step S9: If it is determined that the current combination of environmental variables deviates significantly from the distribution of historical training data, an incremental model update mechanism is triggered to improve the model's adaptability to new working conditions. Specifically, this includes: S9.1: Perform statistical distribution modeling on the combination of environmental variables collected in real time, and use the kernel density estimation method to calculate the probability density distribution of the current environmental state in the multidimensional feature space to obtain the distribution characteristics of the current combination of environmental variables; S9.2: Based on the multidimensional feature space of the historical training dataset, the K-means clustering algorithm is used to divide the typical working condition areas of environmental variables, and an environmental feature label is generated for each cluster center to establish a benchmark reference system for the distribution of environmental variables. S9.3: Calculate the similarity between the probability density distribution of the current combination of environmental variables and the distribution characteristics of the historical cluster centers, and use the Bach distance to measure the distribution difference between the current environment and historical conditions in order to quantify the degree of deviation of the current environment; S9.4: When the Bach distance is greater than the preset threshold, it is determined that there is a significant deviation between the current combination of environmental variables and the distribution of historical training data, an environmental anomaly event flag is generated, and a model incremental update request signal is triggered to start the model online learning process; S9.5: Based on newly acquired environmental-structural response data samples, an online sequential extreme learning machine algorithm is used to locally update the parameters of the dynamic degradation model in order to integrate new working condition information and improve the model's adaptability to the current environmental conditions. S9.6: After the incremental model update is completed, perform cross-validation tests on the updated model, calculate the mean squared error index of the predicted output under the new environmental conditions, in order to evaluate the model update effect and decide whether to retain the updated model parameters.

[0029] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0030] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and rules of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting and evaluating the service life of a steel structure building, characterized by, The method comprises the following steps: S1: Collecting multi-source heterogeneous data of steel structure buildings in different service periods, and adding time stamps and collection location labels to each type of data; S2: Preprocessing the collected structure state parameters by normalization and missing value filling; S3: Extracting environmental driving factors based on the time series characteristics of environmental parameters, and constructing an environmental variable time-varying feature matrix; S4: Fusing the structure state parameters and the environmental variable time-varying feature matrix to extract key degradation features; S5: Establishing a basic degradation equation based on the physical degradation mechanism, and embedding the extracted environmental driving factors as time-varying parameters into the degradation equation to construct a dynamic degradation model; S6: Training and optimizing the dynamic degradation model, and estimating the uncertainty of the model parameters; S7: Inputting real-time monitoring data into the optimized dynamic degradation model, dynamically updating the model parameters, and generating a structure degradation path prediction; S8: Based on the degradation path prediction result, combining the structure safety threshold to judge whether the current service state meets the design life requirement, and generating a confidence interval report of the remaining service life; S9: If it is judged that the current environmental variable combination is significantly deviated from the historical training data distribution, the model incremental update mechanism is triggered.

2. The method of claim 1, wherein the steel structure building is a building. The step S1 specifically comprises: Deploying multiple types of sensor networks to collect real-time structure state parameters of key structure nodes of steel structure buildings, and obtaining physical quantity data reflecting structure performance degradation; Periodically collecting environmental parameters based on temperature and humidity sensors, wind speed and pressure sensors, and corrosion rate monitoring devices to construct a time series data set representing environmental changes; Calling a historical operation and maintenance database to extract structure intervention information generated during previous maintenance, repair, and detection of steel structure buildings to form a historical maintenance record data set; Adding uniform format time stamp labels to the structure state parameters, the environmental parameters, and the historical maintenance records; Adding a structure spatial position label to each type of collected data, and labeling the sensor deployment location and corresponding component number based on the structure topology graph.

3. The method of claim 2, wherein the steel structure building is a building having a steel frame structure. In the step S1, the structure state parameters are collected by a distributed multi-type sensor network, including stress and strain sensors with a range of 1-500MPa and a sampling frequency of 1-1000Hz, displacement sensors with a resolution of 0.001-0.1mm and a measurement frequency of 1-500Hz, and acoustic emission crack propagation monitoring devices with a detection frequency range of 20-400kHz, and the multi-source data is aligned through a high-precision time synchronization protocol.

4. The method of claim 1, wherein the steel structure building is a building. The step S2 specifically comprises: Performing data cleaning operations on the collected structure state parameters, identifying and removing abnormal points using a sliding window mean filtering algorithm to obtain denoised structure state original data sequences; Based on the denoised structure state original data sequences, normalizing multi-dimensional structure state parameters using a standardization method to obtain structure state standardized data; For the missing values in the structure state standardized data, a linear interpolation method based on time series is used to fill in the missing data at the timestamp position, and a complete and continuous structure state time series data is obtained; Based on the complete and continuous structure state time series data, a sliding window segmentation mechanism is used to extract time domain feature segments, and a structure state feature segment set is constructed; The variance stability test is performed on each segment in the structure state feature segment set, the data drift point is identified by using the sliding window standard deviation analysis method, and the local data offset is corrected by the dynamic threshold adjustment mechanism, and the stability enhanced structure state feature data set is obtained.

5. The method of claim 1, wherein the steel structure building is a building. The step S3 specifically includes: The temperature and humidity sensor data in the environmental parameters are processed by sliding window segmentation to obtain the local change trend of the time series, and the temperature and humidity time series segmentation feature vector is obtained; Based on the temperature and humidity time series segmentation feature vector, the standard deviation and range analysis method is used to calculate the temperature and humidity fluctuation rate index; The load change rate feature is extracted by performing difference processing on the structure load monitoring data; Based on the corrosion rate model, the temperature and humidity fluctuation rate and the load change rate are linearly combined and nonlinearly mapped to generate a comprehensive corrosion rate feature parameter; The temperature and humidity fluctuation rate index, the load change rate feature and the comprehensive corrosion rate feature parameter are aligned according to the timestamp to construct an environmental variable time-varying feature matrix.

6. The method of claim 5, wherein the steel structure building life prediction evaluation method is characterized by, In the step S3, the temperature and humidity fluctuation rate is extracted by joint normalization of sliding window standard deviation and range, the load change rate is extracted by first-order difference and sliding mean filtering, and based on linear combination and exponential nonlinear mapping of corrosion rate, a comprehensive corrosion rate feature is formed, and all time-varying features are aligned according to a unified time.

7. The method of claim 1, wherein the steel structure building is a building. The step S4 specifically includes: Perform principal component analysis dimension reduction processing on the structure state parameters to obtain a structure degradation feature vector; Based on the environmental variable time-varying feature matrix, the statistical features of the time series are extracted by using a sliding window to construct an environmental degradation feature vector; The structure degradation feature vector and the environmental degradation feature vector are spliced to generate a multi-modal fusion feature tensor; The multi-modal fusion feature tensor is applied to the multi-head self-attention mechanism to calculate the correlation weight between each feature dimension, and an attention weighted feature vector is generated; Based on the attention weighted feature vector, a feature selection algorithm is executed to filter out a feature subset with high sensitivity to the degradation process, and a key degradation feature matrix is output.

8. The method for predicting and evaluating the service life of a steel structure building according to claim 7, characterized in that, In the S4 step, the principal component analysis method is used to linearly reduce the multi-dimensional structure parameters, and the cumulative contribution rate is preferably not less than 95%, the environmental variable statistical features include mean, variance and trend, the multi-modal feature fusion is performed by tensor splicing and standardization, and the multi-head self-attention mechanism and the feature selection algorithm are applied to the fusion feature to filter out the features with high sensitivity to the degradation process.

9. The method of claim 1, wherein the steel structure building is a building. The step S5 specifically includes: Based on the structure mechanics and material degradation mechanism, the basic degradation behavior characteristics of a typical steel structure in the service process are obtained, and a basic degradation equation is constructed; The extracted environmental driving factors are normalized and time-aligned to obtain a set of environmental driving time series parameters that evolve synchronously with the structural degradation state; Based on dynamic system theory, the environmental driving time series parameters are embedded into the basic degradation equation as external excitation terms to construct a dynamic degradation differential equation model coupled with the environment and structure; The constructed dynamic degradation differential equation is discretized and solved by the Runge-Kutta numerical integration method to generate a sequence of numerical solutions of the structural degradation state evolution over time, and obtain the degradation path response curves under different combinations of environmental variables; Based on the extracted key degradation characteristics, the parameter identifiability of the constructed dynamic degradation model is analyzed, and the environmental coupling coefficients in the model are identified and corrected by using the least squares optimization algorithm.

10. The method of claim 9, wherein the steel structure building is a building. In the step S5, the Weibull or Paris crack propagation model is selected for the basic degradation equation, the normalized and time-aligned environmental driving parameters are coupled with the basic equation as external excitation terms through multiplicative coupling and additive coupling, and the numerical stability of the coupled equation is verified by the Lyapunov method. The differential equation is discretized by the Runge-Kutta numerical integration.

Images