On-site detection method for bearing capacity of structural engineering

Abstract

The invention provides a structural engineering bearing capacity field detection method, and belongs to the technical field of structural engineering detection.The structural engineering bearing capacity field detection method comprises the steps that a multi-axial acceleration sensor and a strain gauge array are installed on the surface of a to-be-detected structural member, and an intelligent loading device is adopted to apply incremental loads; a mobile data acquisition terminal is utilized to acquire vibration response data and strain response data in real time, modal parameters are extracted based on a frequency domain analysis method to establish a load-modal parameter relation curve, the bearing capacity of a component is analyzed and calculated, a graph theory shortest path algorithm is adopted to optimize load transmission path analysis, and the load transmission path analysis is optimized. The bearing performance of the component is accurately evaluated through the structural response prediction model based on the Transform architecture, the technical integration of multi-dimensional detection data fusion, intelligent loading control, quantitative bearing capacity evaluation and accurate performance prediction is realized, and the technical problem of insufficient detection precision of the bearing capacity of the structural component is solved.

Description

Technical Field

[0001] This invention belongs to the field of structural engineering testing technology, and more specifically, relates to a method for on-site testing of the load-bearing capacity of structural engineering. Background Technology

[0002] Structural load-bearing capacity testing is a key technology for ensuring the safety of buildings and engineering structures. Traditional testing methods mainly rely on static loading tests, material sampling tests, and empirical assessments to determine the load-bearing capacity of structural components. These methods are widely used in bridge inspection, building safety assessment, and industrial equipment inspection, evaluating load-bearing capacity by applying predetermined loads and observing the structural response. However, traditional testing methods have significant drawbacks. Static loading tests can only obtain response data under specific load conditions and cannot comprehensively reflect the dynamic behavior characteristics of components under complex load histories. While material sampling tests can obtain basic material performance parameters, they are difficult to accurately reflect the overall load-bearing capacity and damage state of the component. Empirical assessment methods are highly subjective and lack quantitative analytical basis. In existing technologies, due to the lack of comprehensive analysis capabilities for multi-dimensional test data and precise quantitative assessment methods, traditional methods cannot accurately reflect the true load-bearing capacity and damage state of structural components, failing to meet the technical requirements of modern engineering for accurate structural safety assessment. In other words, existing technologies suffer from insufficient accuracy in testing the load-bearing capacity of structural components. Summary of the Invention

[0003] In view of this, the present invention provides a method for on-site testing of the load-bearing capacity of structural engineering, which can solve the technical problem of insufficient accuracy in testing the load-bearing capacity of structural components in the prior art.

[0004] This invention is implemented as follows: This invention provides a method for on-site testing of the load-bearing capacity of structural engineering components, comprising: installing multi-axial acceleration sensors on the surface of the structural member to be tested, with a density of 4 sensors per square meter and a sampling frequency of 1000Hz; simultaneously installing strain gauge arrays at key load-bearing locations of the member; and measuring the cross-sectional dimensions of the member using digital electronic calipers; applying an incremental load to the structural member to be tested using an intelligent loading device, the incremental load including static and dynamic loading, with the loading amplitude gradually increasing from 10% to 120% of the design load; and using a digital electronic rebound hammer to test the surface hardness of the member; and collecting real-time data of the structural member to be tested using a mobile data acquisition terminal. Vibration response and strain response data of structural members under different loads are collected. A mobile data acquisition terminal transmits these data to the on-site data processing system via a wireless data transmission module. The vibration response data is processed using frequency domain analysis to extract modal parameters of the structural member under test, including natural frequency, mode shape, and damping ratio. Load-modal parameter relationship curves are established. The load-bearing capacity of the structural member under test is analyzed and calculated using the material's elastoplastic damage constitutive equation, and the remaining load-bearing capacity assessment result is output. The load transfer path is optimized using the shortest path algorithm in graph theory. Finally, the subsequent load-bearing performance of the structural member under test is predicted using a structural response prediction model.

[0005] The structure of the structural response prediction model is a temporal prediction network based on the Transformer architecture. The structural response prediction model includes two main parts: an encoder and a decoder. The encoder is used to extract feature representations from historical detection data, and the decoder is used to generate prediction results of future load-bearing performance. The attention mechanism parameters in the structural response prediction model are dynamically adjusted according to the component type, load characteristics, and detection accuracy requirements.

[0006] The structural response prediction model outputs the assessment results of the load-bearing capacity degradation trend and remaining service life of the structural components under test, based on current detection data and historical performance data.

[0007] The digital electronic caliper is a high-precision digital caliper of model Mitutoyo CD-6”CSX, with a measurement accuracy of ±0.01mm and a measurement range of 0-150mm. It is equipped with Bluetooth data transmission function and is used to accurately measure the geometric parameters of the length, width, and thickness of components. The measured geometric parameters of the components are directly transmitted to the mobile data acquisition terminal for recording and subsequent analysis.

[0008] The intelligent loading device is an integrated testing platform that combines a hydraulic loading system, a load control system, and a data acquisition system. The intelligent loading device is equipped with high-precision load sensors and displacement sensors. The load sensor has an accuracy of ±0.5%, and the displacement sensor has an accuracy of ±0.005mm. The intelligent loading device has an adaptive load control function, which automatically adjusts the loading rate and loading mode according to the real-time response characteristics of the structural component to be tested.

[0009] The digital electronic rebound hammer is a concrete rebound hammer of model HT-225 with an impact energy of 2.207J. It is used to detect the rebound value of the surface material of the component. The rebound value is converted into a surface hardness value through the built-in data processing chip. The surface hardness value is used to evaluate the strength characteristics and uniformity of the component material.

[0010] The mobile data acquisition terminal has precise positioning, Bluetooth communication, information display, and Internet communication functions. The precise positioning function obtains the spatial coordinates of the mobile data acquisition terminal through a satellite positioning system with a positioning accuracy of centimeters. The spatial coordinates are used to determine the on-site location of the structural component to be inspected.

[0011] Specifically, the load-modal parameter relationship curve is a mathematical curve that describes the relationship between the load magnitude and the natural frequency, mode shape, and damping ratio.

[0012] The material elastic-plastic damage constitutive equation is used to describe the nonlinear behavior and damage evolution process of the structural component under load. The inputs of the material elastic-plastic damage constitutive equation include the component's geometric parameters, material elastic modulus, load history, and measured strain data. The outputs are the component's remaining bearing capacity assessment results and safety factor.

[0013] Specifically, the optimal load transfer path is the path with the least stress when the load is transferred inside the component under given load conditions.

[0014] Specifically, the steps of optimizing load transfer path analysis using the shortest path algorithm in graph theory involve discretizing the structural components to be tested into nodes and edges, and calculating the optimal load transfer path and stress distribution of key nodes under different load conditions.

[0015] Specifically, the multi-axial acceleration sensor is a sensor device capable of simultaneously measuring acceleration components in the X, Y, and Z directions.

[0016] Specifically, the strain gauge array is a sensor combination composed of multiple strain gauges arranged in a predetermined layout, used to measure the strain distribution at different locations of the component.

[0017] Specifically, the trend of load-bearing capacity degradation refers to the regular description of the change in the load-bearing capacity of the structural component under test over time.

[0018] The steps for establishing the training dataset for the structural response prediction model include collecting historical test data of different types of structural components. The historical test data includes component geometric parameters, material performance parameters, load history, vibration response data, and final bearing capacity. The raw data is preprocessed and standardized. The data is organized according to time series. The correspondence between input feature vectors and target output vectors is established. The dataset is divided into training set, validation set, and test set in a ratio of 8:1:1.

[0019] The training steps for the structural response prediction model include initializing model parameters, setting the learning rate to 0.001, the batch size to 32, the number of training epochs to 1000, using mean squared error as the loss function, using the Adam optimizer for parameter updates, employing an early stopping strategy to prevent overfitting during training, stopping training when the validation set loss does not decrease for 10 consecutive epochs, and evaluating the model performance on the test set after training is completed.

[0020] This invention integrates multiple detection devices, including multi-axial accelerometers, strain gauge arrays, and intelligent loading devices. Combined with frequency domain analysis, the material's elastoplastic damage constitutive equation, and a time-series prediction network based on the Transformer architecture, it establishes a multi-dimensional detection data fusion analysis system. This system can simultaneously acquire key information such as the vibration response, strain distribution, and modal parameters of structural members, overcoming the limitations of traditional methods in terms of single detection data and limited analytical dimensions. This invention employs real-time data acquisition and wireless transmission technology, combined with adaptive load control and an environmental compensation system, to achieve intelligent control of the detection process and real-time data processing and analysis. It quantitatively calculates the load-bearing capacity of structural members using the material's elastoplastic damage constitutive equation and optimizes load transfer path analysis using graph theory's shortest path algorithm, improving the accuracy and reliability of load-bearing capacity assessment and overcoming the limitations of traditional methods in terms of subjectivity and insufficient precision. This invention constructs a structural response prediction model based on historical and current detection data, achieving precise quantitative assessment of structural member load-bearing capacity, solving the technical problem of insufficient detection accuracy in traditional methods, and providing a scientific decision-making basis for structural safety management. In summary, this invention solves the technical problem of insufficient detection accuracy of structural member load-bearing capacity mentioned in the background art. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention.

[0022] Figure 2 This is a schematic diagram of the neural network structure of the structural response prediction model involved in the present invention.

[0023] Figure 3 This is a diagram of the data acquisition operation interface of the software involved in this invention.

[0024] Figure 4 This is a diagram of the data analysis interface of the software involved in this invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0026] like Figure 1 The diagram shown is a flowchart of a method for on-site testing of the load-bearing capacity of structural engineering provided by the present invention. This method includes the following steps:

[0027] S01. Install multi-axial acceleration sensors on the surface of the structural component to be tested. The multi-axial acceleration sensors are arranged at a density of 4 per square meter and the sampling frequency of the multi-axial acceleration sensors is 1000Hz. At the same time, install strain gauge arrays at the key load-bearing parts of the component and use digital electronic calipers to measure the cross-sectional dimensions of the component.

[0028] S02. An increasing load is applied to the structural component to be tested using an intelligent loading device. The increasing load includes static loading and dynamic loading. The loading amplitude gradually increases from 10% of the design load to 120%. A digital electronic rebound hammer is used to test the surface hardness of the component.

[0029] S03. The vibration response data and strain response data of the structural component under different loads are collected in real time by a mobile data acquisition terminal. The mobile data acquisition terminal sends the vibration response data and strain response data to the field data processing system through a wireless data transmission module.

[0030] S04. Process the vibration response data based on frequency domain analysis method, extract the modal parameters of the structural component to be tested, the modal parameters include natural frequency, mode shape and damping ratio, and establish load-modal parameter relationship curve;

[0031] S05. The load-bearing capacity of the structural component under test is analyzed and calculated using the material elastic-plastic damage constitutive equation. The input of the material elastic-plastic damage constitutive equation includes the component's geometric parameters, material elastic modulus, load history, and measured strain data. The output is the evaluation result of the component's remaining load-bearing capacity.

[0032] S06. The load transfer path analysis is optimized using the shortest path algorithm in graph theory. The structural component to be tested is discretized into nodes and edges, and the optimal load transfer path and stress distribution of key nodes under different load conditions are calculated.

[0033] S07. The subsequent load-bearing performance of the structural component to be tested is predicted by the structural response prediction model. The structural response prediction model outputs the load-bearing capacity degradation trend and remaining service life assessment results of the structural component to be tested based on the current test data and historical performance data.

[0034] The digital electronic caliper is a high-precision digital caliper, model Mitutoyo CD-6”CSX, with a measurement accuracy of ±0.01mm and a measurement range of 0-150mm. Equipped with Bluetooth data transmission, it is used to accurately measure the geometric dimensions of components, such as length, width, and thickness. The measured geometric parameters are directly transmitted to the mobile data acquisition terminal for recording and subsequent analysis. The digital electronic rebound hammer is a concrete rebound hammer, model HT-225, with an impact energy of 2.207J. It is used to detect the rebound value of the surface material of the component. The rebound value is converted into a surface hardness value by a built-in data processing chip. This surface hardness value is used to evaluate the strength characteristics and uniformity of the component material.

[0035] The intelligent loading device is an integrated testing platform combining a hydraulic loading system, a load control system, and a data acquisition system. It is equipped with high-precision load sensors and displacement sensors; the load sensor has an accuracy of ±0.5%, and the displacement sensor has an accuracy of ±0.005 mm. The intelligent loading device features adaptive load control, automatically adjusting the loading rate and loading mode based on the real-time response characteristics of the structural component under test. When an abnormal response is detected in the structural component, the loading rate is automatically reduced or loading is paused. The intelligent loading device integrates an environmental compensation system to eliminate the influence of temperature and humidity on the test results; the ambient temperature acquisition accuracy is ±0.1℃, and the ambient humidity acquisition accuracy is ±1%RH. To improve the reliability of the test data, the intelligent loading device employs multi-sensor fusion technology, identifying and eliminating abnormal data points by comparing and analyzing the measurement results of different types of sensors, ensuring the accuracy and consistency of the test results. The loading rate is adaptively adjusted based on the component's geometric parameters and the material's elastic modulus, and the loading mode is dynamically selected based on the load history and measured strain data.

[0036] The mobile data acquisition terminal has precise positioning, Bluetooth communication, information display, and internet communication functions. The precise positioning function obtains the spatial coordinates of the mobile data acquisition terminal through a satellite positioning system, with a positioning accuracy of centimeters. These spatial coordinates are used to determine the on-site location of the structural component to be inspected. The Bluetooth communication function receives measurement data transmitted by the digital electronic calipers and the digital electronic rebound hammer, with a communication distance of 10 meters and a data transmission rate of 2 Mbps. The mobile data acquisition terminal establishes a communication connection with the on-site data processing system through the wireless data transmission module. The wireless data transmission module supports both 4G and WiFi communication methods to ensure the stability and real-time performance of data transmission.

[0037] The material elastic-plastic damage constitutive equation is used to describe the nonlinear behavior and damage evolution process of the structural component under load. The inputs of the material elastic-plastic damage constitutive equation include the material elastic modulus, yield strength, ultimate strength and damage parameters. The output is the stress-strain relationship and damage state of the structural component under different load levels. The material elastic modulus is obtained from the detection results of the digital electronic rebound hammer and material test data. The yield strength and ultimate strength are obtained from the component design data and the results of field sampling tests. The damage parameters are calculated based on the strain response data and loading history.

[0038] The load-bearing capacity evaluation function is used to comprehensively analyze the load-bearing performance and safety status of the structural component under test. The input of the load-bearing capacity evaluation function includes measured modal parameters, strain response data, load history and material performance parameters. The output is the remaining load-bearing capacity evaluation result and safety factor of the component. The measured modal parameters are derived from the frequency domain analysis results of step S04. The material performance parameters include the material elastic modulus, yield strength and ultimate strength. The safety factor is used to evaluate the safety margin and service status of the structural component under test.

[0039] like Figure 2As shown, the structure of the structural response prediction model is a temporal prediction network based on the Transformer architecture. The model comprises two main parts: an encoder and a decoder. The encoder extracts feature representations from historical detection data, and the decoder generates predictions of future load-bearing capacity. The attention mechanism parameters in the structural response prediction model are dynamically adjusted according to the component type, load characteristics, and detection accuracy requirements. The steps for establishing the training dataset for the structural response prediction model include collecting historical detection data of different types of structural components. This historical detection data includes the component's geometric parameters, material performance parameters, load history, vibration response data, and final load-bearing capacity. The system's capacity is assessed by preprocessing and standardizing the raw data, organizing the data according to a time series, establishing a correspondence between input feature vectors and target output vectors, and dividing the dataset into training, validation, and test sets in an 8:1:1 ratio. The training steps for the structural response prediction model include initializing model parameters, setting the learning rate to 0.001, batch size to 32, training epochs to 1000, using mean squared error as the loss function, updating parameters using the Adam optimizer, employing an early stopping strategy to prevent overfitting during training, stopping training when the validation set loss does not decrease for 10 consecutive epochs, and evaluating model performance on the test set after training to ensure prediction accuracy meets engineering application requirements.

[0040] The attention weight adjustment function is used to adjust the attention mechanism parameters of the structural response prediction model. The function calculates weight adjustment coefficients based on component type, load characteristics, detection accuracy, and historical data quality. When the weight adjustment coefficient is between 0 and 0.3, a linear weight adjustment function is used to adjust the attention mechanism parameters. When the weight adjustment coefficient is between 0.3 and 0.7, an exponential weight adjustment function is used. When the weight adjustment coefficient is between 0.7 and 1.0, a logarithmic weight adjustment function is used. The component type is derived from engineering design data, the load characteristics are derived from load history analysis results, the detection accuracy is derived from sensor calibration data, and the historical data quality is determined through data integrity and consistency assessment.

[0041] The multi-axial acceleration sensor is a sensor device capable of simultaneously measuring acceleration components in the X, Y, and Z directions.

[0042] The strain gauge array is a sensor combination consisting of multiple strain gauges arranged in a predetermined layout, used to measure the strain distribution at different locations of the component.

[0043] The load-modal parameter relationship curve is a mathematical curve describing the relationship between the load magnitude and the natural frequency, mode shape, and damping ratio.

[0044] The optimal load transfer path is the path with the minimum stress when the load is transferred inside the component under a given load condition.

[0045] The load-bearing capacity degradation trend is a regular description of the change in the load-bearing capacity of the structural component under test over time.

[0046] Furthermore, it also includes a host computer, which is equipped with detection data analysis software to support the summarization and analysis of the collected data, such as... Figure 3-4 As shown.

[0047] The specific implementation methods of the above steps are described in detail below.

[0048] The specific implementation of step S01 is as follows: First, determine the sensor layout scheme. Calculate the coordinates of the layout points based on the geometry and dimensions of the structural component to be tested. Divide the component surface into several measurement areas using a grid-based layout principle, with each area controlled within 0.25 square meters to ensure 100% sensor coverage. Next, calibrate the multi-axial acceleration sensor. Use the gravity acceleration reference method to verify the sensor's triaxial accuracy, controlling the calibration error threshold within ±0.01g. Simultaneously, check the sensor's frequency response characteristics to ensure linearity meets requirements in the 1–500Hz frequency band. Then, install and fix the sensor using epoxy resin adhesive, firmly bonding it to the component surface for at least 24 hours. After installation, verify the coupling effect between the sensor and the component through a light tap test. The strain gauge array installation employs temperature compensation technology. First, clean the component surface to remove oil and oxide layers. After applying a primer, attach the strain gauges. The strain gauge arrangement follows the principal stress direction distribution principle, with longitudinal and transverse strain gauge groups set at the sections with maximum bending moment and maximum shear force, respectively. The digital electronic calipers operate on a multi-point averaging principle, measuring the length, width, and thickness of the same cross-section five times and averaging the results. Measurement accuracy is controlled within ±0.01mm. Measurement data is automatically transmitted to a mobile data acquisition terminal via Bluetooth for recording. The section with the maximum bending moment refers to the cross-section with the largest bending moment value along the length of the member under external loads. This section is typically located at the point of concentrated load application or at the mid-span of a uniformly distributed load. The section with the maximum shear force refers to the cross-section with the largest absolute value of shear force along the length of the member under external loads. This section is typically located near supports or at points of load abrupt change. These parameters are determined through structural mechanics analysis and calculation, providing a theoretical basis for the rational placement of sensors.

[0049] The specific implementation of step S02 involves establishing a control system for the intelligent loading device. A closed-loop feedback control principle is used to achieve precise load application. The control system automatically adjusts the hydraulic system output pressure according to a preset loading program, achieving a load control accuracy of ±0.5%. The loading scheme is based on the component's design load value. Static loading employs a graded loading method, with each load increment being 10% of the design load, and each load holding time not less than 5 minutes. Dynamic loading uses a sinusoidal excitation signal with an excitation frequency range of 1–100 Hz, and the excitation amplitude gradually increases. The digital electronic rebound hammer's testing follows statistical principles, uniformly selecting no fewer than 16 measuring points on the component surface. Each measuring point is measured three times consecutively, and the average value is taken. The coefficient of variation of the rebound value is controlled within 15%. An environmental compensation system monitors the temperature and humidity changes at the testing site in real time. When the temperature change exceeds ±2℃ or the humidity change exceeds ±5%, the system automatically performs environmental correction calculations to eliminate the influence of environmental factors on the test results. Multi-sensor fusion technology uses a weighted average algorithm to comprehensively process the measurement results of different sensors. The weighting coefficients are determined based on the sensor accuracy level and historical reliability data. Abnormal data identification adopts the three-standard-deviation criterion, and data points that exceed the normal range are automatically removed.

[0050] The specific implementation of step S03 involves establishing a communication network for the data acquisition system. The mobile data acquisition terminal acts as a data aggregation node, receiving signals from various sensors via both wired and wireless methods. Time synchronization technology ensures the synchronization of multi-channel data, with a time synchronization accuracy controlled within 1ms. The data preprocessing module filters and amplifies the raw signals, using a Butterworth low-pass filter to remove high-frequency noise, with a cutoff frequency set to 500Hz. The signal amplification factor is adaptively adjusted based on the sensor output characteristics. Data storage employs a circular buffer mechanism, automatically initiating the data upload program when storage capacity reaches 80%, ensuring no data loss. The wireless data transmission module uses dual-mode communication, prioritizing WiFi for higher transmission rates. When the WiFi signal is unstable, it automatically switches to a 4G network. The transmission protocol uses TCP to ensure data transmission reliability. Upon receiving data, the on-site data processing system immediately performs format verification and integrity checks. Abnormal data triggers a retransmission mechanism, and a data reception confirmation signal is automatically sent back to the mobile terminal.

[0051] The specific implementation of step S04 involves using a Fast Fourier Transform (FFT) algorithm to perform frequency domain transformation on the vibration response data, converting the time-domain signal into a frequency-domain signal to extract modal information. The transformation window length is set to 2048 sampling points, and the overlap rate is set to 50% to improve spectral resolution. Modal parameter identification employs a frequency domain decomposition method, identifying the structure's natural frequencies through the power spectral density function, with frequency identification accuracy controlled within ±0.1Hz. Mode shape extraction uses a singular value decomposition (SVD) algorithm, decomposing the frequency response function matrix to obtain mode shape vectors of each order. Mode shape normalization is performed using a mass normalization method. Damping ratio calculation uses the half-power point method, selecting two frequency points near the resonance peak where the power drops by 3dB, and calculating the damping ratio value through bandwidth calculation. The load-modal parameter relationship curve is established using a least-squares fitting method, performing polynomial fitting on the modal parameters under different load levels. The fitting order is determined based on the data distribution characteristics, and the fitting accuracy is evaluated through the correlation coefficient, requiring a correlation coefficient of not less than 0.95.

[0052] The specific implementation of step S05 involves establishing a material elastoplastic damage constitutive model. This model is based on the theory of continuous damage mechanics and uses the strain equivalence principle to describe the nonlinear behavior of the material. Input parameters include basic mechanical parameters such as component geometry, material elastic modulus, Poisson's ratio, yield strength, and ultimate strength, as well as measured load history and strain response data. The constitutive equations are solved using an incremental iteration method, discretizing the load history into several incremental steps. Within each incremental step, the Newton-Raphson iterative algorithm is used to solve the nonlinear equation system. The convergence criterion for iteration is that the relative error is less than [a certain value]. The damage variable is calculated based on the strain energy release rate theory. The degree of damage is determined by comparing the current strain state with the historical maximum strain state. The damage threshold is set to 80% of the material's yield strain. The load-bearing capacity calculation results include the component's current remaining load-bearing capacity, safety factor, and degree of damage. The safety factor is calculated using the limit state design method. When the safety factor is lower than 1.5, the load-bearing capacity is considered insufficient.

[0053] The specific implementation of step S06 involves using Dijkstra's algorithm in graph theory to establish a load transfer path optimization model. The structural member to be detected is discretized into a network of nodes and edges. Nodes represent the key cross-sectional locations of the member, and edges represent the load transfer paths between adjacent cross-sections. Node weights are determined based on cross-sectional geometry and material properties, while edge weights are calculated based on the distance and stiffness distribution between two nodes. The algorithm starts from the load application point and progressively searches for the shortest path to each node. The path length is represented by stress-level weighted distance, with higher stress resulting in higher weights. Key node identification uses centrality analysis to calculate the betweenness centrality and proximity centrality of each node. Nodes with high centrality values ​​are identified as key control points for load transfer. Stress distribution calculation is based on the finite element discretization principle, discretizing the continuous structure into a finite number of elements. Within each element, shape function interpolation is used to calculate the stress distribution. The element type is selected based on the member's geometry: beam elements are used for slender members, and plate elements are used for planar members.

[0054] The specific implementation of step S07 involves establishing a time-series prediction neural network model based on the Transformer architecture. This model uses an encoder-decoder structure to process historical detection data and predict future trends in load-bearing capacity. The encoder employs a multi-head self-attention mechanism to extract time-series features from historical data, with 8 attention heads and a hidden layer dimension of 512. The decoder generates prediction results for future time steps, with the prediction time span set to 10% of the remaining design service life of the component. Supervised learning is used for model training, with mean squared error as the loss function and the Adam algorithm as the optimizer. The initial learning rate is set to 0.001, and a cosine annealing scheduling strategy is used to dynamically adjust the learning rate. Training data preprocessing includes data standardization, missing value imputation, and outlier handling. Standardization uses the Z-score method, missing values ​​are imputed using linear interpolation, and outliers are identified and handled using the interquartile range method. Model validation employs time-series cross-validation, dividing the data into multiple training and test sets according to time order to verify the model's generalization ability. The prediction results include load-bearing capacity degradation curves, remaining service life probability distributions, and risk assessment levels. The risk levels are divided into three categories: low risk, medium risk, and high risk, each corresponding to different maintenance strategy recommendations.

[0055] The attention weight adjustment function is implemented based on a multi-factor comprehensive evaluation method. The component type factor is determined according to the structural form and stress characteristics, with a weight coefficient of 0.8 for beam components, 0.9 for column components, and 0.7 for slab components. The load characteristic factor is calculated based on the load type and variation law, with a weight coefficient of 0.6 for static load, 0.8 for dynamic load, and 1.0 for alternating load. The detection accuracy factor is determined based on sensor calibration results, with a weight coefficient of 1.0 for accuracy level A, 0.8 for level B, and 0.6 for level C. The historical data quality factor is determined through data integrity and consistency assessment, with a weight coefficient of 1.0 when the data integrity rate is higher than 95% and the consistency index is greater than 0.9. The weight adjustment coefficients are calculated using a weighted average method, with weights of 0.3, 0.3, 0.2, and 0.2 for each factor. Different adjustment functions are used depending on the numerical range of the weight adjustment coefficients. Linear functions are suitable for small adjustments, exponential functions are suitable for medium adjustments, and logarithmic functions are suitable for large adjustments, ensuring that the adjustment of the attention mechanism parameters is both accurate and stable.

[0056] It should be noted that the structural response prediction model adopts a time-series prediction network design based on the Transformer architecture. The overall structure consists of two core modules: an encoder and a decoder. It utilizes a multi-head self-attention mechanism and positional encoding technology to achieve deep learning and predictive analysis of the time-series data of the structural component's load-bearing performance. The encoder module is responsible for receiving and processing historical detection data, including multi-dimensional feature information such as component geometric parameters, material performance parameters, load history, vibration response data, and strain response data. Through a multi-layer neural network and attention mechanism, it extracts potential patterns and feature representations from the data, mapping the original high-dimensional time-series data into a low-dimensional feature vector space, providing an effective feature foundation for subsequent predictive analysis. The decoder module receives the feature representation output from the encoder, combines it with the current detection data and a preset prediction time window, and generates a predicted sequence of future load-bearing performance through a recurrent neural network structure and attention mechanism. The output includes the load-bearing capacity degradation trend and remaining service life assessment results.

[0057] The structural response prediction model employs a dynamic weight adjustment strategy for its attention mechanism. The attention weight adjustment function calculates weight adjustment coefficients based on factors such as component type, load characteristics, detection accuracy, and historical data quality. A linear weight adjustment function is used when the coefficient is between 0 and 0.3; an exponential weight adjustment function is used when it's between 0.3 and 0.7; and a logarithmic weight adjustment function is used when it's between 0.7 and 1.0. This piecewise weight adjustment mechanism allows the model to adaptively adjust the level of attention given to different features, improving the accuracy and robustness of the prediction results. The model also integrates a positional encoding layer and a multi-head attention layer. The positional encoding layer preserves the temporal order information of the time-series data, while the multi-head attention layer captures feature relationships at different levels and angles in the data through parallel computation of multiple attention heads, enhancing the model's ability to learn complex temporal patterns.

[0058] The process of establishing the training dataset for the structural response prediction model first involves collecting historical test data of different types of structural components from multiple engineering projects and laboratory tests. The data collection scope covers various component types, including steel structures, concrete structures, and composite material structures, ensuring the diversity and representativeness of the dataset. The collected raw data includes component geometric parameters such as length, width, thickness, and cross-sectional area; material performance parameters such as elastic modulus, yield strength, ultimate strength, and Poisson's ratio; load history data including dynamic information such as loading time series, load amplitude changes, and loading frequency; vibration response data including dynamic characteristics such as acceleration time history, frequency domain response, and modal parameters; strain response data including static characteristics such as strain time history, strain distribution, and damage indices; and the final load-bearing capacity test results and failure mode information.

[0059] The data preprocessing stage involves quality control and standardization of the collected raw data. First, data cleaning removes outliers and noise. Statistical analysis is used to identify and process missing data, and interpolation algorithms or deletion strategies are employed to handle incomplete data records. Then, features with different dimensions and numerical ranges are normalized to ensure that each feature is learned and compared on the same numerical scale. The data standardization process uses Z-score standardization, converting the value of each feature into a distribution with a mean of 0 and a standard deviation of 1, eliminating the impact of dimensional differences between different features on model training.

[0060] In the time-series data organization stage, the preprocessed data is rearranged and organized chronologically, establishing a correspondence between time windows and prediction windows. Time windows are used to input historical data for feature learning, while prediction windows are used to output future predictions for supervised learning. A sliding window technique is used to divide the continuous time-series data into multiple training samples. Each sample contains a fixed-length historical data as an input feature vector and the corresponding future data as a target output vector, establishing a mapping relationship between input and output. The dataset is ultimately divided into training, validation, and test sets in an 8:1:1 ratio. The training set is used for optimizing model parameters, the validation set for hyperparameter tuning and model selection, and the test set for objectively evaluating the final model performance, ensuring the model's generalization ability and practical application effectiveness.

[0061] The model training process follows a standard deep learning training workflow. First, model parameters, including the weight matrix and bias vector, are initialized. Key hyperparameters are set, including a learning rate of 0.001, a batch size of 32, and 1000 training epochs. Mean squared error (MSE) is chosen as the loss function to measure the difference between predicted and true values. The Adam optimizer is used for gradient descent and parameter updates. An early stopping strategy is implemented during training to prevent overfitting. Training automatically stops when the loss function on the validation set stops decreasing for 10 consecutive epochs, ensuring the model learns sufficiently on the training set while maintaining good generalization performance. After training, final performance is evaluated on an independent test set. Multiple evaluation metrics, including MSE, mean absolute error, and coefficient of determination, are used to comprehensively assess the model's prediction accuracy and stability, ensuring the model meets the accuracy and reliability requirements of practical engineering applications.

[0062] It should be noted that the sensor density formula is based on spatial sampling theory. By ensuring a density of four sensors per square meter, the spatial resolution requirements of structural vibration signals can be met. Compared with traditional empirical placement methods, this formulaic method ensures the scientific nature of the sensor placement and the completeness of the measurement data. The incremental load formula adopts a linear incremental strategy, gradually increasing from 10% of the design load to 120%, which can comprehensively reflect the response characteristics of the component under different load levels. Compared with single load level detection, this method can obtain richer structural performance information.

[0063] The multi-sensor fusion weighted average formula, based on information fusion theory, effectively improves the reliability and accuracy of measurement results by assigning appropriate weights to sensors with different accuracy levels. Compared to single-sensor measurements, this fusion method has advantages such as strong anti-interference capability and small measurement error. The three-standard-deviation criterion for anomaly data identification, based on normal distribution theory, effectively identifies and eliminates outlier data points during the measurement process, ensuring the accuracy of subsequent analysis and calculations. This method achieves an accuracy rate of 99.7%. The data standardization Z-score method eliminates differences in the dimensions and numerical ranges of different sensor outputs, providing a unified data foundation for multi-source data fusion and machine learning model training.

[0064] The Fast Fourier Transform (FFT) formula converts time-domain vibration signals into frequency-domain signals, effectively extracting the modal characteristics of structures. Its computational efficiency is far higher than traditional frequency-domain analysis methods, providing technical support for real-time online monitoring. The damping ratio calculation formula is based on the half-power point method. By identifying the frequency points on both sides of the resonance peak, it can accurately obtain the damping characteristics of the structure. Compared with other damping identification methods, this method has the advantages of simple calculation and high accuracy.

[0065] The elastoplastic damage constitutive equation incorporates damage variables into the stress-strain relationship, enabling it to describe the nonlinear behavior and damage evolution of materials under load. Compared to traditional linear elastic analysis methods, this equation can more accurately predict the degradation of a structure's load-bearing capacity. The damage variable evolution equation, based on continuous damage mechanics theory, describes the damage development law through an exponential function. This function form reflects the nonlinear characteristics and cumulative effects of damage development. The safety factor calculation formula, based on limit state design theory, quantitatively assesses the structure's safety status and service performance by comparing the remaining load-bearing capacity with the actual load effect.

[0066] The load transfer path weighting function comprehensively considers two key factors: physical distance and stress level. It quantifies the transfer efficiency of the path through a weighted combination. Compared to traditional methods that only consider geometric distance, this function can more accurately identify weak points in the structure. Dijkstra's algorithm uses dynamic programming in its distance update formula. By progressively optimizing the path length, it can efficiently find the global optimum. The time complexity of this algorithm is O(log n). Its computational efficiency is significantly better than that of exhaustive search methods.

[0067] The attention weight adjustment function comprehensively evaluates the model adjustment needs through multi-factor weighted averaging, and can adaptively adjust model parameters according to different detection conditions. Compared with prediction models with fixed parameters, this adjustment mechanism significantly improves the model's adaptability and prediction accuracy. The piecewise adjustment function is designed based on the characteristics of different adjustment magnitudes: linear functions are suitable for small adjustments to maintain stability, exponential functions are suitable for medium-amplitude adjustments to provide sufficient adjustment range, and logarithmic functions are suitable for large adjustments to avoid parameter abrupt changes. This piecewise processing strategy ensures the smoothness and effectiveness of parameter adjustment.

[0068] The key technical ideas of this invention are mainly reflected in the following aspects.

[0069] The first key technological approach is an intelligent detection system based on multi-sensor fusion. Traditional structural inspection methods typically rely on single-type sensors, making them susceptible to environmental interference and sensor inherent errors, resulting in unreliable test results. This invention employs a multi-axial acceleration sensor, strain gauge array, digital electronic calipers, and digital electronic rebound hammer, among other sensors, working collaboratively. By integrating different types of measurement data through multi-sensor fusion technology, it effectively improves detection accuracy and reliability. This approach is based on information fusion theory. By combining redundant and complementary measurements, it not only verifies the correctness of measurement results but also obtains comprehensive information that a single sensor cannot provide, thereby achieving a comprehensive and accurate assessment of the structural load-bearing capacity.

[0070] The second key technical approach is load transfer path optimization analysis based on graph theory algorithms. Traditional structural analysis methods mainly rely on finite element numerical calculations, which are relatively inefficient and fail to intuitively reflect the load transfer patterns within the structure. This invention innovatively introduces the shortest path algorithm from graph theory, discretizing the complex three-dimensional structure into a network model of nodes and edges. By finding the optimal load transfer path, it identifies weak points and critical control points in the structure. This method is not only computationally efficient but also clearly reveals the load transfer mechanism, providing important evidence for structural design optimization and safety assessment. Compared to traditional methods, this approach has advantages such as fast computation speed, clear physical meaning, and wide applicability.

[0071] The third key technological approach is an intelligent prediction model based on the Transformer architecture. Traditional structural performance prediction mainly relies on empirical formulas or simple statistical models, which have limited prediction accuracy and struggle to handle complex nonlinear relationships. This invention employs advanced deep learning technology to establish a time-series prediction neural network based on the Transformer architecture. This network can automatically learn the inherent patterns of structural performance changes from historical monitoring data, achieving accurate prediction of future load-bearing capacity degradation trends. This model possesses powerful feature extraction and long-sequence modeling capabilities, can handle complex dependencies in multi-dimensional time-series data, and exhibits significantly higher prediction accuracy than traditional methods.

[0072] The synergistic effect of these three key technological approaches has yielded significant technical benefits and advantages. The multi-sensor fusion system provides high-quality input data for graph theory algorithms, ensuring the accuracy of load transfer path analysis. The analysis results of the graph theory algorithm provide crucial structural feature information for the Transformer prediction model, improving its prediction accuracy. The prediction results of the Transformer model can then guide sensor placement optimization and detection scheme adjustments, forming a complete closed-loop optimization system. This synergistic effect gives the entire detection method comprehensive advantages such as high accuracy, high efficiency, and strong adaptability, enabling it to meet the on-site detection needs of complex engineering structures and providing scientific and reliable technical support for structural safety assessment and maintenance decisions.

[0073] Specifically, the principle of this invention is as follows: The core principle that enables this invention to solve the problems of existing technologies lies in establishing a technical framework for multi-dimensional detection data fusion analysis. By simultaneously acquiring the dynamic response and static strain information of components through multi-axial acceleration sensors and strain gauge arrays, more comprehensive and accurate structural behavior data is obtained than traditional single detection methods, providing a reliable data foundation for accurate assessment of load-bearing capacity. The adaptive control function of the intelligent loading device can dynamically adjust the loading parameters according to the real-time response characteristics of the components, ensuring the safety of the detection process and the validity of the data. The application of the environmental compensation system eliminates the interference of external factors on the detection results, improving the stability and consistency of detection accuracy.

[0074] The combined application of frequency domain analysis methods and the material's elastoplastic damage constitutive equations enables this invention to deeply analyze the load-bearing behavior of components from a mechanical mechanism perspective. Modal parameter extraction reveals the dynamic characteristics of the components, while the damage constitutive equations describe the nonlinear behavior and damage evolution process of materials under load. This physics-based analysis method is more scientific and accurate than traditional empirical assessments. The introduction of graph theory shortest path algorithms optimizes load transfer path analysis, enabling the identification of key stress regions within the components and providing more refined analytical results for load-bearing capacity assessment.

[0075] A time-series prediction network based on the Transformer architecture establishes a mathematical model of how a structural member's load-bearing capacity changes over time by learning regular features from historical inspection data. The application of an attention mechanism enables the model to automatically identify key factors affecting load-bearing capacity and dynamically adjust prediction weights, improving the accuracy and reliability of the prediction results. This deep learning-based prediction method combines the statistical regularities of a large amount of historical data with real-time information from current inspection data, enabling quantitative prediction of the future load-bearing performance of structural members and providing forward-looking technical support for structural safety management.

[0076] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0077] The specific implementation of step S01 involves determining the sensor layout coordinates based on the component's geometric characteristics and load distribution. The sensor layout density calculation formula is as follows:

[0078] ;

[0079] In the formula, Sensor density, expressed in units per square meter; This represents the total number of sensors arranged on the surface of the component. This represents the total surface area of ​​the component, expressed in square meters.

[0080] Sensor coordinate calculation uses a gridded layout method, the first The coordinates of each sensor are represented as follows:

[0081] ;

[0082] In the formula, For the first The spatial coordinates of each sensor; These are the three-dimensional coordinate components of the sensor, in meters; This is the row number where the sensor is located, and its value is a positive integer. The value is the column number where the sensor is located, and it is a positive integer. The grid is in direction and The directional spacing, in meters, is typically 0.5 meters.

[0083] The formula for calculating the calibration error of a multi-axis accelerometer is:

[0084] ;

[0085] In the formula, The relative error of sensor calibration is expressed as a percentage. The measured acceleration value is given by the sensor, in m / s². The reference value for gravitational acceleration is 9.8 m / s². Sensor calibration requirements. .

[0086] The specific implementation of step S02 is to establish an incremental load loading sequence, and the load increment formula is expressed as:

[0087] ;

[0088] In the formula, For the first Level load value, in Newtons; The reference value for the design load of the component is given in Newtons. This represents the loading level, with a value ranging from 1 to 12.

[0089] Multi-sensor fusion employs a weighted average algorithm, and the fusion result is calculated using the following formula:

[0090] ;

[0091] In the formula, The measurement results are obtained after multi-sensor fusion. For the first The measured values ​​of each sensor; For the first The weighting coefficients for each sensor range from 0 to 1, and ; The number of sensors participating in the fusion.

[0092] Outlier identification uses the three-standard-deviation criterion, and the judgment formula is as follows:

[0093] ;

[0094] In the formula, For the first One data point to be detected; The mean of the data sequence to be detected; denoted as the standard deviation of the data sequence to be tested.

[0095] The dynamic loading excitation signal is in the form of a sine wave, and the excitation force is expressed as:

[0096] ;

[0097] In the formula, for The dynamic stimulus of a moment, measured in Newtons; The excitation amplitude is expressed in Newtons. The excitation frequency is measured in Hz and ranges from 1 to 100 Hz. The initial phase angle is expressed in radians. This is a time variable, with the unit being seconds.

[0098] The statistical processing of rebound values ​​uses the arithmetic mean method, and the calculation formula is as follows:

[0099] ;

[0100] In the formula, This represents the average rebound value. The number of measurement points is required. ; For the first The rebound value at each measuring point.

[0101] The specific implementation of step S03 is to establish a data preprocessing system, and the data standardization adopts the Z-score method. The standardization formula is:

[0102] ;

[0103] In the formula, For the first Standardized values ​​of each data point, dimensionless; For the first One original data point; The mean of the original data sequence; denoted as the standard deviation of the original data sequence. The standardized data follows a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0104] The specific implementation of step S04 is to perform frequency domain analysis on the vibration signal using Fast Fourier Transform, and the transformation formula is expressed as follows:

[0105] ;

[0106] In the formula, For the frequency domain signal at the 1st... Complex values ​​at each frequency point; For the first The time-domain signal value of each sampling point; This represents the total number of sampling points, typically 2048. The imaginary unit satisfies ; The frequency point number, with a value range of 0 to... .

[0107] Natural frequency identification is achieved through power spectral density peak detection. The power spectral density is calculated using the following formula:

[0108] ;

[0109] In the formula, For frequency The power spectral density at that point is expressed in units of (m / s²)² / Hz. For frequency The Fourier transform result at the location; The duration of the signal, in seconds; The sampling frequency is 1000Hz.

[0110] The damping ratio is calculated using the half-power point method, and the formula is as follows:

[0111] ;

[0112] In the formula, The damping ratio is dimensionless and typically ranges from 0.01 to 0.05. The resonant frequency is expressed in Hz. This represents the frequency value corresponding to a 3dB power drop on both sides of the resonance peak, expressed in Hz.

[0113] The relationship between load and modal parameters is fitted using a polynomial, and the fitting function is expressed as:

[0114] ;

[0115] In the formula, For load The first under the action The natural frequency of the first order, in Hz; The applied load is expressed in Newtons. are the fitting coefficients, where The unit is Hz. The unit is Hz / N. The unit is Hz / N². The unit is Hz / N³; The fitting error term, in Hz, is required to... Hz; This represents the modal order, typically ranging from 1 to 5.

[0116] The specific implementation of step S05 is to establish a material elastic-plastic damage constitutive model, and the stress-strain relationship is expressed as:

[0117] ;

[0118] In the formula, Effective stress, in MPa; The damage variable takes a value from 0 to 1, where 0 represents no damage and 1 represents complete damage. This refers to the elastic modulus of the material, expressed in GPa. The strain is dimensionless.

[0119] The evolution equation of the damage variable is expressed as:

[0120] ;

[0121] In the formula, The damage initiation strain is dimensionless and typically taken as 80% of the yield strain. This is the largest response in history, dimensionless; For limiting strain, dimensionless.

[0122] The load-bearing capacity assessment function is expressed as follows:

[0123] ;

[0124] In the formula, Remaining load capacity, in Newtons; Initial load-bearing capacity, in Newtons; This is the material performance reduction factor, ranging from 0.8 to 1.0; This is the environmental impact factor, with a value ranging from 0.9 to 1.0.

[0125] The safety factor is calculated using the limit state design method, and the formula is as follows:

[0126] ;

[0127] In the formula, For the safety factor, dimensionless, the requirement is... ; The actual load effect is expressed in Newtons.

[0128] The specific implementation of step S06 is to use a graph theory algorithm to establish a load transfer path model, and the formula for calculating the distance weight between nodes is:

[0129] ;

[0130] In the formula, For nodes To the node The path weight is dimensionless. The physical distance between the two nodes, in meters; This represents the average stress along the path, in MPa. The allowable stress of the material is expressed in MPa. These are weighting coefficients, dimensionless, with typical values ​​of 0.6 and 0.4.

[0131] The shortest path search uses Dijkstra's algorithm, and the distance update formula is:

[0132] ;

[0133] In the formula, From the starting point to the node The shortest distance, dimensionless; From the starting point to the node The shortest distance, dimensionless; For nodes To the node The edge weights are dimensionless.

[0134] Key node identification uses betweenness centrality calculation, expressed by the formula:

[0135] ;

[0136] In the formula, For nodes The betweenness centrality of is dimensionless and takes values ​​from 0 to 1; For nodes To the node The total number of shortest paths; For the nodes The number of shortest paths.

[0137] The specific implementation of step S07 is to establish a prediction model based on deep learning, where the attention weight adjustment function is expressed as:

[0138] ;

[0139] In the formula, is the weighting adjustment coefficient, dimensionless, and its value ranges from 0 to 1; This is a component type factor, dimensionless, with a value of 0.8 for beam components, 0.9 for column components, and 0.7 for slab components; is the load characteristic factor, dimensionless, taken as 0.6 for static load, 0.8 for dynamic load, and 1.0 for alternating load; For the detection accuracy factor, which is dimensionless, we take 1.0 for Class A accuracy, 0.8 for Class B accuracy, and 0.6 for Class C accuracy. This is a historical data quality factor, dimensionless, with a value range of 0.6 to 1.0. The weight coefficients for each factor are dimensionless and take values ​​of 0.3, 0.3, 0.2, and 0.2, respectively.

[0140] Different adjustment functions are used based on the weighting adjustment coefficients. The linear adjustment function is expressed as follows:

[0141] ,when hour;

[0142] The exponential adjustment function is expressed as:

[0143] ,when hour;

[0144] The logarithmic adjustment function is expressed as:

[0145] ,when hour;

[0146] In the formula, The adjusted attention parameters are dimensionless. The attention parameters before adjustment are dimensionless. This is an adjustment coefficient, dimensionless, with typical values ​​of 0.1, 0.5, and 0.2.

[0147] It should be noted that the variables involved in this invention are explained in detail in Tables 1 and 2.

[0148] Table 1. Variable Explanation Table (Part 1)

[0149] Table 2. Variable Explanation Table (Part Two)

[0150] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting the load-carrying capacity of a structural engineering in situ, characterized by, The method comprises the following steps: A multi-axial acceleration sensor is installed on the surface of the structure member to be detected, the arrangement density of the multi-axial acceleration sensor is 4 per square meter, the sampling frequency of the multi-axial acceleration sensor is 1000 Hz, a strain gauge array is installed at a key part of the member to be detected, and a digital electronic caliper is used to measure the cross-sectional dimension of the member; an incremental load is applied to the structure member to be detected by an intelligent loading device, the incremental load comprises static loading and dynamic loading, the loading amplitude is gradually increased from 10% of the design load to 120%, and a digital electronic rebound instrument is used to detect the surface hardness of the member; vibration response data and strain response data of the structure member to be detected under different loads are collected in real time by a mobile data acquisition terminal, and the vibration response data and the strain response data are transmitted to a field data processing system through a wireless data transmission module; The vibration response data are processed based on a frequency domain analysis method, modal parameters of the structure member to be detected are extracted, the modal parameters comprise natural frequency, mode shape and damping ratio, and a load-modal parameter relationship curve is established; the bearing capacity of the structure member to be detected is analyzed and calculated by using an elastic-plastic damage constitutive equation of a material, and a residual bearing capacity evaluation result of the member is output.

2. The method of field testing the load bearing capacity of structural engineering according to claim 1, wherein, After the residual bearing capacity evaluation result of the member is output, a step of optimizing a load transfer path analysis by using a shortest path algorithm in graph theory and predicting subsequent bearing performance of the structure member to be detected by a structure response prediction model is further included.

3. The method of field testing the load bearing capacity of structural engineering according to claim 2, characterized in that, The structure response prediction model is a time series prediction network based on a Transformer architecture, the structure response prediction model comprises an encoder and a decoder, the encoder is used to extract feature representation of historical detection data, and the decoder is used to generate a prediction result of future bearing performance, and attention mechanism parameters in the structure response prediction model are dynamically adjusted according to a member type, a load feature and a detection accuracy requirement.

4. The method of field testing the load bearing capacity of a structural engineering according to claim 3, wherein, The digital electronic caliper has a measurement accuracy of ±0.01 mm, a measurement range of 0-150 mm and a Bluetooth data transmission function, and is used to accurately measure length, width and thickness geometric dimension parameters of the member.

5. The method of field testing the load bearing capacity of a structural engineering according to claim 4, wherein, The intelligent loading device is an integrated detection platform integrating a hydraulic loading system, a load control system and a data acquisition system, the intelligent loading device is provided with a high-precision load sensor and a displacement sensor, the load sensor has an accuracy of ±0.5%, and the displacement sensor has an accuracy of ±0.005 mm.

6. The method of field testing the load bearing capacity of a structural engineering according to claim 5, wherein, The digital electronic rebound instrument has an impact energy of 2.207 J, is used to detect a rebound value of a surface material of the member, and the rebound value is converted into a surface hardness value by a built-in data processing chip, and the surface hardness value is used to evaluate strength characteristics and uniformity of the material of the member.

7. The method of field testing the load bearing capacity of a structural engineering according to claim 6, wherein, The mobile data acquisition terminal has accurate positioning function, Bluetooth communication function, information display function and internet communication function, the accurate positioning function obtains spatial coordinates of the mobile data acquisition terminal through a satellite positioning system, the positioning accuracy is centimeter level, and the spatial coordinates are used to determine a field position of the structure member to be detected.

8. The method of field testing the load bearing capacity of a structural engineering according to claim 7, wherein, The load-mode parameter relationship curve, in particular, is a mathematical curve describing the change relationship between the load size and the natural frequency, mode shape and damping ratio.

9. The method of field testing the load bearing capacity of a structural engineering according to claim 8, wherein, The material elastic-plastic damage constitutive equation is used for describing the nonlinear behavior and damage evolution process of the structural member under the action of the load, and the input of the material elastic-plastic damage constitutive equation includes the geometric parameters of the member, the material elastic modulus, the load history and the measured strain data, and the output of the material elastic-plastic damage constitutive equation includes the residual bearing capacity evaluation result and the safety factor of the member.

10. The method of field testing the load bearing capacity of a structural engineering according to claim 9, wherein, The optimized load transmission path, in particular, is a path with the minimum stress when the load is transmitted in the member under the given load condition.

Images