Intelligent bridge reinforcement decision generation method, device and equipment based on damage identification

Abstract

The invention relates to the technical field of data processing, and discloses an intelligent bridge reinforcement decision generation method, device and equipment based on damage identification. The method comprises the following steps: carrying out bridge damage identification and evaluation on a bridge state data set to obtain a bridge damage characteristic spectrum; performing intelligent decision analysis on the bridge damage characteristic spectrum, generating a preliminary reinforcement scheme, and obtaining a candidate reinforcement strategy set; performing parametric modeling and multi-scale simulation on the candidate reinforcement strategy set, and performing coupling calculation evaluation on a reinforcement effect to obtain an optimized reinforcement scheme; virtual construction planning is conducted on the optimized reinforcement scheme, a construction process model is generated through an intelligent scheduling algorithm, and a target construction scheme is obtained; and monitoring and feeding back the target construction scheme in real time, and dynamically optimizing the construction process through an adaptive control algorithm to obtain a target intelligent reinforcement implementation scheme. The intelligent bridge reinforcement decision generation efficiency and accuracy based on damage identification are improved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a method, device and equipment for generating intelligent bridge reinforcement decisions based on damage identification. Background Art

[0002] Existing bridge reinforcement decision-making methods mainly rely on manual experience and traditional structural analysis techniques. Engineers evaluate the structural condition of bridges through regular inspections and non-destructive testing, and then develop reinforcement plans based on static analysis and finite element simulation. These methods usually include visual inspection, ultrasonic testing, strain measurement and other technologies, as well as structural analysis using commercial finite element software. During the reinforcement design stage, engineers will select appropriate reinforcement materials and methods based on specification requirements and experience, such as external prestressing, carbon fiber reinforcement, concrete repair, etc.

[0003] However, these traditional methods have some limitations. First, manual inspection is easily affected by subjective factors, making it difficult to detect potential structural problems in a timely manner. Second, conventional structural analysis methods find it difficult to fully consider the dynamic behavior and long-term performance evolution of bridges. Furthermore, the traditional reinforcement decision-making process lacks systematicity and intelligence, making it difficult to find the optimal balance between multiple objectives (such as safety, economy, and durability). In addition, existing construction planning and monitoring methods are often static and difficult to adapt to complex site conditions and uncertainties. These deficiencies limit the efficiency and accuracy of reinforcement decisions, which may affect the overall effect and resource utilization of the reinforcement project. Summary of the invention

[0004] The present application provides a method, device and equipment for generating intelligent bridge reinforcement decisions based on damage identification, which are used to improve the efficiency and accuracy of generating intelligent bridge reinforcement decisions based on damage identification.

[0005] In a first aspect, the present application provides an intelligent bridge reinforcement decision-making method based on damage identification, and the intelligent bridge reinforcement decision-making method based on damage identification includes: collecting multi-source data on the bridge structure, preprocessing the collected data through a multimodal data fusion algorithm, and obtaining a bridge status data set; identifying and evaluating bridge damage on the bridge status data set through a multi-branch neural network model to obtain a bridge damage feature map; performing intelligent decision-making analysis on the bridge damage feature map to generate a preliminary reinforcement plan and obtain a candidate reinforcement strategy set; performing parameterized modeling and multi-scale simulation on the candidate reinforcement strategy set, and performing coupled calculation evaluation on the reinforcement effect to obtain an optimized reinforcement plan; performing virtual construction planning on the optimized reinforcement plan, generating a construction process model through an intelligent scheduling algorithm, and obtaining a target construction plan; real-time monitoring and feedback on the target construction plan, and dynamically optimizing the construction process through an adaptive control algorithm to obtain a target intelligent reinforcement implementation plan.

[0006] In a second aspect, the present application provides an intelligent bridge reinforcement decision-making device based on damage identification, and the intelligent bridge reinforcement decision-making device based on damage identification includes:

[0007] The acquisition module is used to collect multi-source data of the bridge structure and pre-process the collected data through a multi-modal data fusion algorithm to obtain a bridge status data set;

[0008] An evaluation module, used for identifying and evaluating bridge damage on the bridge status data set through a multi-branch neural network model to obtain a bridge damage feature map;

[0009] An analysis module is used to perform intelligent decision-making analysis on the bridge damage characteristic map, generate a preliminary reinforcement plan, and obtain a candidate reinforcement strategy set;

[0010] A simulation module is used to perform parameterized modeling and multi-scale simulation on the candidate reinforcement strategy set, and to perform coupled calculation evaluation on the reinforcement effect to obtain an optimized reinforcement plan;

[0011] A planning module is used to perform virtual construction planning for the optimized reinforcement scheme, generate a construction process model through an intelligent scheduling algorithm, and obtain a target construction scheme;

[0012] The feedback module is used to monitor and provide feedback on the target construction plan in real time, and dynamically optimize the construction process through an adaptive control algorithm to obtain a target intelligent reinforcement implementation plan.

[0013] The third aspect of the present application provides a computer device, wherein the memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor and the memory communicate via a bus, and when the machine-readable instructions are executed by the processor, the steps of the above-mentioned intelligent bridge reinforcement decision-making method based on damage identification are performed.

[0014] In the technical solution provided by the present application, a comprehensive and high-quality bridge status data set is obtained by collecting multi-source data of the bridge structure and preprocessing it using a multimodal data fusion algorithm, which provides a reliable data basis for subsequent analysis and effectively improves the accuracy and comprehensiveness of the bridge status assessment. Secondly, a multi-branch neural network model is used to identify and evaluate the damage of the bridge status data set, and a bridge damage feature map is generated, which greatly improves the accuracy and efficiency of damage identification, can timely discover potential structural problems, and provides a key basis for reinforcement decisions. Furthermore, an intelligent decision-making analysis is performed on the bridge damage feature map, a preliminary reinforcement plan is generated, and a candidate reinforcement strategy set is obtained, which realizes the automation and intelligence of the reinforcement plan generation, and greatly reduces the subjectivity and uncertainty of manual decision-making. Then, by performing parametric modeling and multi-scale simulation on the candidate reinforcement strategy set, and performing coupled calculation evaluation, an optimized reinforcement plan is obtained. This process fully considers multiple factors such as reinforcement effect, cost, and construction difficulty, ensuring the scientificity and feasibility of the reinforcement plan. In addition, virtual construction planning is performed on the optimized reinforcement plan, and a construction process model is generated through an intelligent scheduling algorithm to obtain the target construction plan, which significantly improves the efficiency and rationality of construction planning and helps reduce resource waste and time delays during the construction process. Finally, the target construction plan is monitored and fed back in real time, and the construction process is dynamically optimized through an adaptive control algorithm to obtain the target intelligent reinforcement implementation plan. This closed-loop control mechanism greatly enhances the adaptability and robustness of the reinforcement implementation process, and can effectively deal with various uncertainties and emergencies during the construction process. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0016] Figure 1 A schematic diagram of an embodiment of a method for generating intelligent bridge reinforcement decisions based on damage identification in an embodiment of the present application;

[0017] Figure 2 A schematic diagram of an embodiment of an intelligent bridge reinforcement decision-making device based on damage identification in an embodiment of the present application;

[0018] Figure 3 It is a schematic diagram of the structure of a computer device in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The embodiments of the present application provide a method, device and equipment for making intelligent bridge reinforcement decisions based on damage identification. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0020] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the intelligent bridge reinforcement decision generation method based on damage identification includes:

[0021] Step S101, collecting multi-source data on the bridge structure, and pre-processing the collected data through a multimodal data fusion algorithm to obtain a bridge status data set;

[0022] Step S102, performing bridge damage identification and evaluation on the bridge status data set through a multi-branch neural network model to obtain a bridge damage feature map;

[0023] Step S103, performing intelligent decision analysis on the bridge damage characteristic map, generating a preliminary reinforcement plan, and obtaining a candidate reinforcement strategy set;

[0024] Step S104: perform parameterized modeling and multi-scale simulation on the candidate reinforcement strategy set, and perform coupled calculation evaluation on the reinforcement effect to obtain an optimized reinforcement plan;

[0025] Step S105: Perform virtual construction planning on the optimized reinforcement scheme, generate a construction process model through an intelligent scheduling algorithm, and obtain a target construction scheme;

[0026] Step S106: Real-time monitoring and feedback of the target construction plan are performed, and the construction process is dynamically optimized through an adaptive control algorithm to obtain a target intelligent reinforcement implementation plan.

[0027] It is understandable that the execution subject of the present application may be an intelligent bridge reinforcement decision-making device based on damage identification, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0028] Specifically, multi-source data of the bridge structure is first collected, including vibration signals, strain data and displacement data. For vibration signals, wavelet transform is used to eliminate noise, and then fast Fourier transform is used to obtain frequency domain features. Wavelet transform can effectively remove high-frequency noise and retain important features of the signal. Strain data is subjected to median filtering to remove outliers, and then principal component analysis is performed to extract main features. Median filtering can effectively remove sudden noise, while principal component analysis can reduce data dimensions and retain key information. Displacement data is smoothed by Kalman filtering, and then time series analysis is performed to obtain displacement trend features. Kalman filtering can effectively estimate the state of the dynamic system and reduce the influence of measurement noise. These processed data are multi-modally fused by tensor decomposition algorithm to form a comprehensive bridge status data set. A multi-branch neural network model is used to identify and evaluate damage of the bridge status data set. First, time window division is performed, and the data is segmented by sliding window algorithm, and then Fourier transform is performed on each time series feature subset to obtain frequency domain feature matrix. These frequency domain feature matrices are processed by convolutional neural network to extract local features, and global feature vectors are obtained by global average pooling. The long short-term memory network is used to capture long-term temporal features, and the self-attention mechanism further optimizes the feature representation. Finally, the probability distribution of damage types is obtained through the fully connected layer and softmax classification, and the final bridge damage feature map is obtained through threshold filtering and spatial mapping. Based on the obtained bridge damage feature map, intelligent decision analysis is performed to generate a preliminary reinforcement plan. First, the damage feature map is processed using a graph convolutional network to extract the topological relationship of the damage, and the damage representation vector is obtained through graph embedding. K-means clustering analysis is performed on these vectors to group similar damages, and the severity is evaluated to obtain the reinforcement priority ranking. Then, the applicable reinforcement technology is selected through knowledge graph matching to form a preliminary reinforcement plan. After these plans are checked for constraints, the decision path of the plan is obtained through multi-objective optimization and decision tree analysis. Finally, the robustness of the plan is evaluated through Monte Carlo simulation to obtain a set of candidate reinforcement strategies.

[0029] The candidate reinforcement strategy set is parametrically modeled and multi-scale simulated. First, the geometric model of the reinforcement component is generated using the non-uniform rational B-spline technology, and the finite element model is obtained by meshing. Then, the material properties are assigned, the boundary conditions are set, and the computationally ready model is obtained. The macro-scale simulation is performed by the nonlinear finite element method to calculate the overall structural response. The critical area is simulated at the mesoscopic scale to simulate the internal damage evolution of the material and obtain the microscopic crack propagation path. The critical stress intensity factor is obtained by fracture mechanics analysis. Finally, the reliability evaluation is performed to obtain the reliability index of the reinforcement scheme, and the optimized reinforcement scheme is obtained by comprehensive sorting. Virtual construction planning is performed for the optimized reinforcement scheme. First, the process is decomposed and the priority is determined by the hierarchical analysis method to obtain the process dependency diagram and the key process sequence. The resource demand analysis and resource balance optimization of the key process sequence are performed to obtain the preliminary resource allocation plan. The construction progress is evaluated by discrete event simulation technology to generate progress adjustment suggestions. Multiple scheduling schemes are generated using heuristic algorithms, and the optimal scheduling scheme is obtained through multi-criteria evaluation. Finally, risk assessment is performed to obtain risk response strategies, which are integrated into the construction process to form a target construction plan.

[0030] Finally, the target construction plan is monitored and fed back in real time. The sensor layout is planned through the optimal coverage algorithm, the data collection frequency is set, and a real-time monitoring plan is formed. The collected data is preprocessed by edge computing technology for anomaly detection. The detected abnormal events are analyzed by Bayesian network reasoning to obtain a construction deviation report and adjustment priority list. A set of candidate adjustment plans is generated through a reinforcement learning algorithm, and the optimal adjustment plan is obtained through simulation verification. Finally, a specific construction adjustment instruction set is generated and integrated into the original construction plan to obtain the target intelligent reinforcement implementation plan.

[0031] For example, suppose a reinforced concrete bridge is damaged and reinforced. First, the vibration signal of the bridge is collected by the vibration sensor. After wavelet transform denoising, the vibration amplitude is reduced by 20%. The main frequency components obtained after fast Fourier transform are 2.5Hz, 5.8Hz and 9.3Hz. After the strain data is median filtered, the outliers are reduced by 15%, and the principal component analysis retains 95% of the information. After the displacement data is smoothed by Kalman filter, the noise variance is reduced by 30%. After multimodal data fusion, a 50×100×3 tensor data set is obtained. After the multi-branch neural network model analysis, it is identified that the main beam has microcracks (confidence 0.92) and the support has slight corrosion (confidence 0.87). Based on these damage characteristics, the intelligent decision-making system generates three candidate reinforcement strategies: carbon fiber reinforcement, external prestressing and concrete repair. Through multi-scale simulation, the reliability index of the carbon fiber reinforcement scheme is the highest, reaching 0.95. The virtual construction planning decomposed the entire reinforcement process into 15 steps, with 7 steps included in the critical path and an estimated construction period of 45 days. During the real-time monitoring process, it was detected that the construction progress fell behind by 2% on the third day. The system automatically adjusted the resource allocation of subsequent steps, and the actual construction period was finally in line with the plan, and the reinforcement effect achieved the expected goal.

[0032] In the embodiment of the present invention, by collecting multi-source data of the bridge structure and preprocessing it using a multimodal data fusion algorithm, a comprehensive and high-quality bridge status data set is obtained, which provides a reliable data basis for subsequent analysis and effectively improves the accuracy and comprehensiveness of the bridge status assessment. Secondly, a multi-branch neural network model is used to identify and evaluate the damage of the bridge status data set, and a bridge damage feature map is generated, which greatly improves the accuracy and efficiency of damage identification, can timely discover potential structural problems, and provides a key basis for reinforcement decisions. Furthermore, an intelligent decision-making analysis is performed on the bridge damage feature map, a preliminary reinforcement plan is generated, and a candidate reinforcement strategy set is obtained, which realizes the automation and intelligence of the reinforcement plan generation, and greatly reduces the subjectivity and uncertainty of manual decision-making. Then, by performing parametric modeling and multi-scale simulation on the candidate reinforcement strategy set, and performing coupled calculation evaluation, an optimized reinforcement plan is obtained. This process fully considers multiple factors such as reinforcement effect, cost, and construction difficulty, ensuring the scientificity and feasibility of the reinforcement plan. In addition, virtual construction planning is performed on the optimized reinforcement plan, and a construction process model is generated through an intelligent scheduling algorithm to obtain the target construction plan, which significantly improves the efficiency and rationality of construction planning and helps reduce resource waste and time delays during the construction process. Finally, the target construction plan is monitored and fed back in real time, and the construction process is dynamically optimized through an adaptive control algorithm to obtain the target intelligent reinforcement implementation plan. This closed-loop control mechanism greatly enhances the adaptability and robustness of the reinforcement implementation process, and can effectively deal with various uncertainties and emergencies during the construction process.

[0033] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0034] (1) Collect vibration signals of the bridge structure, eliminate noise from the vibration signals through wavelet transform to obtain noise-reduced vibration data, and perform fast Fourier transform on the noise-reduced vibration data to obtain frequency domain features;

[0035] (2) Strain data of the bridge structure is collected, outliers are removed from the strain data by median filtering to obtain filtered strain data, and principal component analysis is performed on the filtered strain data to obtain the strain principal components;

[0036] (3) Collecting displacement data of the bridge structure, smoothing the displacement data through Kalman filtering to obtain smoothed displacement data, and performing time series analysis on the smoothed displacement data to obtain displacement trend characteristics;

[0037] (4) performing data standardization on frequency domain features, strain principal components, and displacement trend features to obtain standardized feature data, and performing multimodal fusion on the standardized feature data through a tensor decomposition algorithm to obtain a fused feature tensor;

[0038] (5) Perform dimensionality reduction processing on the fused feature tensor to obtain the reduced-dimensional feature vector, and organize the reduced-dimensional feature vector into a time series to obtain the bridge status dataset.

[0039] Specifically, vibration signals of bridge structures are collected, and vibration data are collected at key locations of bridges using high-precision acceleration sensors. The collected original vibration signals usually contain various noises, so noise elimination is required. Wavelet transform is an effective signal denoising method, which decomposes the signal into wavelet coefficients of different scales, removes the noise coefficients through threshold processing, and then reconstructs the signal to obtain the denoised vibration data. Fast Fourier transform is performed on the denoised vibration data to convert the time domain signal into frequency domain features, thereby obtaining important information such as the natural frequency and mode of the bridge structure. Strain data of bridge structures are collected, and fiber Bragg grating strain sensors are deployed at the main stress-bearing parts of the bridge to collect strain data. Since strain data may have outliers, outliers need to be removed through median filtering. Median filtering is a nonlinear filtering method that replaces the value of the center point with the median of the data in the sliding window, which can effectively remove sudden noise. After filtering, smoother and more reliable strain data are obtained. The filtered strain data is subjected to principal component analysis. The covariance matrix of the data is calculated, the eigenvalues ​​and eigenvectors are solved, and the main eigenvectors are selected as principal components, thereby reducing the data dimension and extracting the most representative strain features.

[0040] At the same time, the displacement data of the bridge structure is collected, and the displacement of the key parts of the bridge is measured using a laser displacement sensor or a GNSS system. The displacement data usually contains measurement noise and random errors, and needs to be smoothed by Kalman filtering. Kalman filtering is a recursive estimation algorithm that continuously optimizes the state estimation through two steps of prediction and update to obtain smoothed displacement data. Time series analysis is performed on the smoothed displacement data, including trend analysis, periodicity analysis, and autocorrelation analysis, to obtain displacement trend characteristics and reflect the long-term deformation trend of the bridge. In order to comprehensively utilize different types of data, it is necessary to standardize the frequency domain features, strain principal components, and displacement trend features. Standardization can eliminate the dimensional differences between different features. The Z-score standardization method is usually used to convert each feature into a standard normal distribution with a mean of 0 and a variance of 1. The standardized feature data is multimodally fused through a tensor decomposition algorithm. Tensor decomposition is a multidimensional data processing technology that decomposes high-dimensional data into the product of low-dimensional factors. Commonly used methods include Tucker decomposition and CP decomposition. Through tensor decomposition, the correlation between different modal data can be captured to obtain a fused feature tensor.

[0041] Finally, the fused feature tensor is subjected to dimensionality reduction. Common methods include principal component analysis (PCA) or t-SNE (t-distributed random neighbor embedding). Dimensionality reduction can further reduce the data dimension and extract the most important feature information. The feature vectors after dimensionality reduction are organized in chronological order to form a time series, and finally a bridge status data set reflecting the overall status of the bridge is obtained. For example, suppose a steel box girder bridge with a span of 200 meters is monitored for health. Three accelerometers are arranged at the 1 / 4 span, 1 / 2 span and 3 / 4 span of the bridge, with a sampling frequency of 100Hz, and vibration data is collected continuously for 24 hours. The original data contains vibration caused by environmental noise and traffic load. Through wavelet transform denoising, db4 wavelet is selected, the signal is decomposed into 5 layers, and the signal is reconstructed after soft threshold processing, and the noise amplitude is reduced by about 40%. The denoised data is subjected to fast Fourier transform, and the first three natural frequencies of the bridge are identified to be 0.52Hz, 1.18Hz and 2.35Hz respectively.

[0042] 20 fiber Bragg grating strain sensors were arranged on the bottom plate of the bridge main beam with a sampling frequency of 10Hz. There were some sudden outliers in the original strain data. After using the median filter with a window size of 5, the number of outliers was reduced by 95%. The filtered strain data was subjected to principal component analysis, and the first three principal components were selected, with a cumulative explained variance of 92%. A GNSS receiver was installed on the top of the bridge tower with a sampling frequency of 1Hz to measure the three-dimensional displacement of the bridge tower. There were centimeter-level random errors in the original displacement data. After Kalman filtering, the standard deviation of the displacement data was reduced from 3.2cm to 0.8cm. Time series analysis of the smoothed displacement data revealed that the bridge tower had a small periodic swing with a period of about 1 hour and an amplitude within the range of ±5mm.

[0043] The processed frequency domain features (9 features, 3 measurement points × 3rd order frequency), strain principal components (3 features), and displacement trend features (6 features, displacement and velocity in the XYZ direction) are combined into an 18-dimensional feature vector. The 18 features are Z-score standardized, and then a 18×24×3 third-order tensor is constructed (18 features, 24 hours, 3 data types). The tensor is decomposed using Tucker decomposition to obtain a 5×6×2 core tensor and three factor matrices. Finally, the core tensor is flattened and reorganized in chronological order to obtain a 60×24 bridge status dataset, with one state vector per hour, containing 60 fused features. This dataset comprehensively reflects the changes in the health status of the bridge within 24 hours, providing a reliable data basis for subsequent damage identification and reinforcement decisions.

[0044] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0045] (1) The bridge status data set is divided into time windows, and the data is segmented using a sliding window algorithm to obtain a time series feature subset. The time series feature subset is then Fourier transformed to obtain a frequency domain feature matrix.

[0046] (2) Perform convolutional neural network processing on the frequency domain feature matrix, extract local features through multi-layer convolution and pooling operations to obtain a convolution feature map, and perform global average pooling on the convolution feature map to obtain a global feature vector;

[0047] (3) The global feature vector is processed by a long short-term memory network, and the long-term features are captured by temporal dependency modeling to obtain a temporal feature vector. The temporal feature vector is processed by a self-attention mechanism to obtain an attention-weighted feature.

[0048] (4) Performing full connection layer processing on the attention weighted features, performing feature mapping through a multi-layer perceptron to obtain a high-dimensional feature representation, and performing softmax classification on the high-dimensional feature representation to obtain the probability distribution of the damage type;

[0049] (5) Threshold filtering is performed on the probability distribution of damage types. High-confidence predictions are screened by setting a confidence threshold to obtain damage type determination results. The damage type determination results are spatially mapped to obtain a bridge damage characteristic map.

[0050] Specifically, the bridge status data set is divided into time windows, and the sliding window algorithm is used to process the data in segments. The sliding window algorithm sets a fixed-size time window, such as 1 hour, and slides along the time axis with a certain step size, such as 10 minutes, to divide the continuous time series data into multiple overlapping time series feature subsets. This processing method can capture the local time characteristics of the data while maintaining the continuity of the data. Fourier transform is performed on each time series feature subset to convert the time domain signal into a frequency domain representation to obtain a frequency domain feature matrix. Fourier transform can reveal the frequency composition of the signal and help identify the vibration characteristics and potential abnormal patterns of the bridge structure. Convolutional neural network (CNN) is used to process the frequency domain feature matrix. CNN extracts local features through multi-layer convolution and pooling operations, in which the convolution operation uses different convolution kernels to filter the input data and extract features of different scales and types; the pooling operation reduces the data dimension by downsampling to improve computational efficiency. This process obtains a convolution feature map, and then the convolution feature map is globally averaged pooled, and the average value of each feature map is used as the global representation of the feature, thereby obtaining a global feature vector. Global average pooling can effectively reduce the number of parameters while retaining important feature information.

[0051] The global feature vector is then input into a long short-term memory network (LSTM) for processing. LSTM is a special recurrent neural network that can effectively process long sequence data and capture long-term dependencies. LSTM controls the flow of information through a gating mechanism, including input gates, forget gates, and output gates, so that it can learn long-term temporal features in the data. The temporal feature vector obtained after LSTM processing is then processed by a self-attention mechanism. The self-attention mechanism allows the model to focus on different parts of the input sequence and weight them according to their relevance to the current moment, thereby obtaining attention-weighted features. This mechanism can capture long-range dependencies in the sequence and improve the model's sensitivity to important information. The attention-weighted features are processed by a fully connected layer, which is actually a multi-layer perceptron (MLP) that maps and transforms features through nonlinear activation functions to obtain high-dimensional feature representations. This step can further extract abstract features and enhance the expressiveness of the model. The high-dimensional feature representation is finally processed by a softmax classifier to obtain a probability distribution of damage types. The softmax function converts high-dimensional features into probability values ​​for each damage type, and the sum of the probability values ​​is 1, reflecting the model's confidence in predicting different damage types.

[0052] Finally, the probability distribution of damage types is filtered by threshold, and a confidence threshold (such as 0.8) is set to screen high-confidence prediction results. Only predictions with probability values ​​exceeding the threshold are considered reliable damage type determination results. These determination results are then mapped to the actual structural position of the bridge through spatial mapping, thereby obtaining an intuitive bridge damage feature map.

[0053] For example, suppose a steel box girder bridge with a length of 500 meters is used for damage identification. The bridge status dataset contains 7 days of monitoring data from 200 sensor points with a sampling frequency of 100Hz. First, the data is segmented using a sliding window with a length of 1 hour and a step size of 30 minutes to obtain 336 time series feature subsets. Fourier transform is performed on each subset, and the frequency components in the range of 0-50Hz are selected to obtain a frequency domain feature matrix of size 200×500 (200 sensor points, 500 frequency components). These frequency domain feature matrices are input into a CNN with 4 convolutional layers, each layer using 64 3×3 convolution kernels, followedby ReLU activation function and 2×2 maximum pooling. After CNN processing, a 12×12×64 feature map is obtained, and then a 64-dimensional global feature vector is obtained by global average pooling.

[0054] The global feature vector is input to a two-layer LSTM network with 128 hidden units in each layer. The 128-dimensional time series feature vector output by the LSTM passes through a multi-head self-attention layer with 8 attention heads to obtain a 512-dimensional attention-weighted feature. The attention-weighted feature is processed through a four-layer fully connected network with hidden layer sizes of 256, 128, 64, and 32, respectively, using the ReLU activation function. The output of the last layer is subjected to softmax classification, assuming that 6 predefined damage types (such as cracks, corrosion, fatigue, deformation, support damage, and prestress loss) are pre-defined.

[0055] The model outputs the probability distribution of 6 damage types, for example [0.03, 0.02, 0.89, 0.01, 0.04, 0.01]. The confidence threshold is set to 0.85, so only the third damage type (fatigue) with a probability of 0.89 is judged as a reliable result. Finally, the fatigue damage judgment result is mapped to the bridge structure diagram, marking the specific location where fatigue damage may occur, such as a certain welded joint area in the middle span of the main beam. In this way, the model successfully identified the fatigue damage of the bridge and accurately located the damage location. This result provides timely early warning information for bridge managers, enabling them to take targeted reinforcement measures, such as local reinforcement of fatigue damage areas or replacement of high fatigue strength materials, thereby effectively extending the service life of the bridge and ensuring the safe operation of the bridge.

[0056] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0057] (1) The bridge damage feature map is processed by graph convolutional network, and the topological relationship of the damage is obtained by spatial feature extraction to obtain a damage association map. The damage association map is then embedded into a graph to obtain a damage representation vector.

[0058] (2) Perform cluster analysis on the damage representation vectors, group similar damages using the K-means algorithm to obtain a set of damage categories, and evaluate the severity of the damage category set to obtain a reinforcement priority ranking;

[0059] (3) Perform knowledge graph matching on the reinforcement priority ranking, select the applicable reinforcement technology through semantic similarity calculation, obtain the preliminary reinforcement plan, and check the constraints of the preliminary reinforcement plan to obtain a set of feasible reinforcement plans;

[0060] (4) Perform multi-objective optimization on the feasible reinforcement scheme set, balance the reinforcement effect and cost through genetic algorithm, obtain the Pareto optimal solution set, and perform decision tree analysis on the Pareto optimal solution set to obtain the scheme decision path;

[0061] (5) Monte Carlo simulation is performed on the decision path of the scheme, and the robustness of the scheme is evaluated through random sampling to obtain the scheme reliability index. The scheme reliability index is then sorted and screened to obtain a set of candidate reinforcement strategies.

[0062] Specifically, the bridge damage feature map is processed by graph convolutional network (GCN), and the topological relationship of damage is obtained by spatial feature extraction. GCN regards the bridge structure as a graph, in which nodes represent structural elements and edges represent the connection relationship between elements. GCN updates the features of the central node by aggregating the information of neighboring nodes, thereby capturing the spatial distribution and mutual influence of damage in the structure. This process obtains a damage association graph, which reflects the relationship between different damage points. Subsequently, the damage association graph is embedded in the graph to compress the high-dimensional graph structure information into a low-dimensional space to obtain a damage representation vector. Graph embedding technologies such as Node2Vec or GraphSAGE can be used in this step, which can retain the structural information and node characteristics of the graph. Cluster analysis is performed on the damage representation vector, and the K-means algorithm is used to group similar damages. The K-means algorithm divides the data points into K clusters through iterative optimization, so that each data point belongs to the nearest cluster center. This process obtains a set of damage categories and classifies similar damage patterns. Then, the severity of the damage category set is evaluated, considering factors such as damage degree, location importance, development trend, etc., and the comprehensive score is used to obtain the reinforcement priority ranking.

[0063] The knowledge graph is matched to the reinforcement priority, and the applicable reinforcement technology is selected through semantic similarity calculation. The knowledge graph contains information such as various reinforcement technologies, applicable conditions, and effect evaluation. Semantic similarity calculation can use methods such as cosine similarity or Word2Vec to match the damage characteristics with the description of the reinforcement technology to obtain a preliminary reinforcement plan. Subsequently, the constraints of the preliminary reinforcement plan are checked, and factors such as structural safety, construction feasibility, and economy are considered to screen out a set of feasible reinforcement plans that meet all constraints. Multi-objective optimization is performed on the feasible reinforcement plan set, and the reinforcement effect and cost are balanced through genetic algorithms. Genetic algorithms simulate the biological evolution process and continuously optimize the solution space through operations such as selection, crossover, and mutation. The optimization objectives include multiple dimensions such as improved structural safety, extended service life, and construction cost. This process obtains a Pareto optimal solution set, that is, a non-dominated solution set that achieves a balance between multiple objectives. A decision tree analysis is performed on the Pareto optimal solution set to construct a hierarchical decision structure, consider the optimal choice under different conditions, and obtain the decision path of the plan.

[0064] Finally, a Monte Carlo simulation is performed on the decision path of the solution, and the robustness of the solution is evaluated through random sampling. The Monte Carlo method simulates uncertainty through a large number of random samplings to evaluate the performance of the solution under different circumstances. This process obtains the solution reliability index, which reflects the stability of the solution under various possible circumstances. The solution reliability index is sorted and screened, and finally a set of candidate reinforcement strategies is obtained, providing decision makers with multiple high-quality alternatives.

[0065] For example, suppose a cross-sea bridge is subjected to damage identification and reinforcement decision making. The bridge damage feature map contains 50 damage points, each with 10 feature dimensions. Through 3 layers of GCN processing, 64 hidden units in each layer, a 50×64 damage association map is obtained. Node2Vec is used for graph embedding to map each damage point to a 32-dimensional vector space. K-means clustering (K=5) classifies the damage into 5 categories: cracks, corrosion, fatigue, deformation, and bearing damage. The severity assessment considers the degree of damage (40% weight), location importance (30% weight), and development trend (30% weight), with a score range of 0-100. The knowledge graph contains 20 reinforcement technologies, and the semantic similarity is calculated by Word2Vec to select the top-3 applicable technologies for each type of damage. The constraint check considers a budget cap of 50 million yuan and a construction period of no more than 6 months, and screens out 10 feasible reinforcement schemes. The multi-objective optimization sets the population size to 100, iterates 500 generations, and obtains 30 Pareto optimal solutions. Decision tree analysis considers three key nodes: budget, construction period, and traffic impact, and constructs a decision path.

[0066] The Monte Carlo simulation was performed with 10,000 random samplings, taking into account uncertainties such as material price fluctuations (±10%), construction delays (0-30 days), and environmental factors (±5% effect). The reliability index calculation was based on the performance within a 90% confidence interval. Finally, five candidate reinforcement strategies were selected, including carbon fiber reinforcement of the main beam (reliability 0.92), external prestressed reinforcement (reliability 0.89), and cathodic protection (reliability 0.95).

[0067] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0068] (1) Parametric modeling is performed on the candidate reinforcement strategy set, and the geometric model of the reinforcement component is generated by the non-uniform rational B-spline technology to obtain the reinforcement component model set, and the reinforcement component model set is meshed to obtain a finite element model;

[0069] (2) Assign material properties to the finite element model, select appropriate material parameters through the constitutive relationship database, obtain a complete structural model, and set boundary conditions for the complete structural model to obtain a computationally ready model;

[0070] (3) Perform macro-scale simulation on the computationally ready model, calculate the overall structural response using the nonlinear finite element method, obtain macro-mechanical performance indicators, and perform local stress analysis on the macro-mechanical performance indicators to obtain stress distribution in key areas;

[0071] (4) Perform mesoscopic simulation on the stress distribution in the key area, simulate the internal damage evolution of the material by discrete element method, obtain the micro crack propagation path, and perform fracture mechanics analysis on the micro crack propagation path to obtain the critical stress intensity factor;

[0072] (5) The reliability of the critical stress intensity factor is evaluated, and the structural failure probability is calculated by the first-order and second-order moment methods to obtain the reliability index of the reinforcement scheme. The reliability index of the reinforcement scheme is comprehensively ranked to obtain the optimized reinforcement scheme.

[0073] Specifically, the candidate reinforcement strategy set is parametrically modeled, and the geometric model of the reinforcement component is generated by the non-uniform rational B-spline (NURBS) technology. NURBS is a mathematical method for accurately expressing complex surfaces, which defines the geometric shape by control points, node vectors and weights. This technology can accurately describe the shapes of various reinforcement components, such as carbon fiber plates, steel plates, etc. The generated reinforcement component model set contains the geometric information of all possible reinforcement schemes. Subsequently, the reinforcement component model set is meshed to obtain a finite element model. During the meshing process, the adaptive meshing technology is used to use a finer mesh in the stress concentration area to improve the calculation accuracy. The material properties are assigned to the finite element model. Through the pre-established constitutive relationship database, parameters suitable for various materials, such as elastic modulus, Poisson's ratio, yield strength, etc. are selected. The constitutive relationship database contains the mechanical properties data of common reinforcement materials, such as high-strength concrete, carbon fiber, steel, etc. After the material property assignment is completed, a complete structural model is obtained. Then, the boundary conditions of the complete structural model are set, including support constraints, load conditions, etc., to obtain a calculation-ready model.

[0074] The computationally ready model is simulated at macroscale, and the overall structural response is calculated using the nonlinear finite element method. Nonlinear analysis takes into account factors such as the nonlinear behavior of the material and large geometric deformation, and can more accurately simulate the actual structural behavior. By solving the nonlinear equilibrium equation, the displacement field, stress field and other information of the structure are obtained, and then the macroscopic mechanical performance indicators such as bearing capacity and deformation are calculated. Local stress analysis is performed on the macroscopic mechanical performance indicators to identify the stress concentration area and obtain the stress distribution in the key area. The stress distribution in the key area is simulated at mesoscale, and the discrete element method (DEM) is used to simulate the internal damage evolution of the material. DEM regards the material as composed of discrete particles and describes the material behavior by simulating the interaction between particles. This method can simulate the microscopic processes such as crack initiation and propagation, and obtain the microscopic crack propagation path. Fracture mechanics analysis is performed on the microscopic crack propagation path, the stress intensity factor is calculated, and the crack propagation rate is predicted by methods such as Paris's law, and finally the critical stress intensity factor is obtained.

[0075] Finally, the reliability of the critical stress intensity factor is evaluated, and the first-order second-order moment method is used to calculate the probability of structural failure. The first-order second-order moment method is a commonly used structural reliability analysis method, which estimates the failure probability by calculating the mean and variance of the limit state function. This process obtains the reliability index of the reinforcement scheme, which reflects the safety and reliability of the reinforcement scheme. The reliability index of the reinforcement scheme is comprehensively ranked, considering multiple factors such as reliability, economy, and construction difficulty, and finally the optimized reinforcement scheme is obtained.

[0076] For example, suppose a prestressed concrete continuous beam bridge is to be reinforced. Candidate reinforcement strategies include external carbon fiber plate bonding, external prestressing and steel plate reinforcement. First, NURBS technology is used to model the carbon fiber plate model using a 20×20 control point grid to accurately describe its curved surface shape at the bottom of the beam. The external prestressed steel strand uses a cubic NURBS curve to describe its spatial trajectory, which contains 15 control points. When meshing, a grid size of 0.1m is used at the edge of the carbon fiber plate and the prestressed anchorage area, and 0.5m is used in other areas, with a total number of about 500,000 units. When assigning material properties, the elastic modulus of 230GPa and the ultimate strength of 3500MPa for the carbon fiber plate are selected from the constitutive relationship database; the elastic modulus of 195GPa and the yield strength of 1860MPa for the prestressed steel strand. The boundary condition settings include fixed constraints on the piers, temperature loads (±25℃), vehicle loads (standard vehicle model), etc.

[0077] The macro-scale nonlinear analysis uses the Newton-Raphson iteration method, and the convergence criteria are set as displacement error less than 0.1mm and force balance error less than 1kN. The analysis results show that the carbon fiber plate reinforcement scheme reduces the mid-span deflection by 12mm and the tensile stress at the bottom of the main beam by 15MPa. The local stress analysis found that there is stress concentration at the end of the carbon fiber plate, and the maximum shear stress reaches 3.2MPa. DEM analysis is performed on the stress concentration area, simulating a 1cm 3 The volume contains 100,000 discrete element particles. The simulation results show that microcracks appear at the interface, with a maximum crack length of 0.5 mm. Fracture mechanics analysis calculates the critical stress intensity factor KIC to be 0.8 MPa·m^(1 / 2).

[0078] The reliability evaluation was performed using the first-order second-order moment method, taking into account uncertainties such as material strength and load, and the reliability index β of the reinforced structure was calculated to be 4.2, corresponding to a failure probability of 1.33×10^-5. Taking reliability, cost and construction difficulty into consideration, the carbon fiber plate external reinforcement solution was ultimately determined to be the best choice, which is expected to extend the service life of the bridge by 20 years.

[0079] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0080] (1) Decompose the process of the optimized reinforcement scheme, determine the process priority through the hierarchical analysis method, obtain the process dependency graph, and perform critical path analysis on the process dependency graph to obtain the key process sequence;

[0081] (2) Analyze the resource requirements of the key process sequence, optimize resource allocation through resource balancing algorithm, obtain a preliminary resource allocation plan, and perform conflict detection on the preliminary resource allocation plan to obtain a feasible resource plan;

[0082] (3) Conduct construction simulation for feasible resource solutions, evaluate the construction progress through discrete event simulation technology, obtain construction progress forecast data, and conduct deviation analysis on the construction progress forecast data to obtain progress adjustment suggestions;

[0083] (4) Intelligent scheduling optimization is performed on the schedule adjustment suggestions. Multiple scheduling schemes are generated through heuristic algorithms to obtain a candidate scheduling set. The candidate scheduling set is evaluated by multiple criteria to obtain the optimal scheduling scheme.

[0084] (5) Conduct risk assessment on the optimal scheduling plan, analyze potential risk factors through Monte Carlo simulation, obtain risk response strategies, and integrate risk response strategies into the construction process to obtain the target construction plan.

[0085] Specifically, the optimized reinforcement scheme is decomposed into processes, and the process priority is determined by the analytic hierarchy process (AHP). The AHP method decomposes complex problems into a hierarchical structure, establishes a judgment matrix through expert scoring, calculates the eigenvector to obtain the weight of each process, and thus determines the priority. This process obtains a process dependency graph, which reflects the logical relationship between the processes. Subsequently, the process dependency graph is subjected to critical path analysis, and the forward and backward traversal algorithms are used to calculate the earliest start time, the latest start time, and the total time difference of each process, and identify the key process sequence with the greatest impact on the total construction period. The resource demand analysis of the key process sequence is carried out, and the resource allocation is optimized by the resource balancing algorithm. The resource balancing algorithm considers resource constraints such as manpower, equipment, and materials, and adjusts the start time of non-critical processes to make the resource usage curve tend to be stable, avoiding excessive concentration or idleness of resources. This process obtains a preliminary resource allocation plan. Then, the preliminary resource allocation plan is subjected to conflict detection to check resource conflicts in space and time, such as multiple processes using the same equipment at the same time or overlapping work surfaces. By adjusting the process sequence or resource allocation, the conflict is resolved and a feasible resource plan is obtained.

[0086] The construction simulation of feasible resource solutions is carried out, and the discrete event simulation technology is used to evaluate the construction progress. Discrete event simulation abstracts the construction process into a series of discrete events. By simulating the occurrence and mutual influence of these events, the dynamic changes of the entire construction process are predicted. The simulation process takes into account random factors, such as weather influence and equipment failure, to obtain construction progress forecast data. Deviation analysis is carried out on the construction progress forecast data, the difference between the predicted progress and the planned progress is compared, potential delay risks are identified, and progress adjustment suggestions are obtained. The progress adjustment suggestions are intelligently optimized and multiple scheduling solutions are generated through heuristic algorithms. Heuristic algorithms such as genetic algorithms or ant colony algorithms search for approximate optimal solutions in the solution space by simulating the principles of biological evolution or swarm intelligence. This process generates multiple candidate scheduling solutions and forms a candidate scheduling set. The candidate scheduling set is evaluated by multiple criteria, considering multiple objectives such as construction period, cost, quality, and safety, and adopts multi-criteria decision-making methods such as TOPSIS method (topic ideal solution ranking method) to comprehensively evaluate the advantages and disadvantages of each solution and obtain the optimal scheduling solution.

[0087] Finally, the optimal scheduling plan is risk assessed and potential risk factors are analyzed through Monte Carlo simulation. The Monte Carlo method simulates the impact of various uncertain factors on the project, such as construction delays, cost overruns, quality issues, etc., through a large number of random sampling. Based on the simulation results, the probability and impact of the risk are calculated to obtain the risk response strategy. These strategies include risk avoidance, risk transfer, risk mitigation and risk acceptance. Integrate the risk response strategy into the construction process, adjust the construction plan and resource allocation, and finally obtain the target construction plan.

[0088] For example, suppose a reinforced concrete arch bridge is reinforced. The optimized reinforcement plan includes carbon fiber reinforcement of the main arch rib and replacement of the bearing. The process decomposition results in 20 main processes, including traffic diversion, scaffolding, surface treatment, carbon fiber paving, maintenance, and bearing replacement. Through the AHP method analysis, it is found that the weight of carbon fiber paving is 0.25, the weight of bearing replacement is 0.2, and the weights of other processes are distributed between 0.02 and 0.1. Critical path analysis identifies 7 key processes, and the total construction period is expected to be 60 days. Resource demand analysis shows that the carbon fiber paving process requires 10 professional workers and 2 special equipment; bearing replacement requires 5 workers and 1 large crane. After resource balancing, non-critical processes such as traffic diversion are started 5 days in advance to avoid human resource peaks exceeding 20 people. Conflict detection found that there was a one-day overlap in the use of cranes, and this conflict was resolved by adjusting the order of bearing replacement.

[0089] Discrete event simulation takes into account the impact of rainy days with a probability of 30%. After 1,000 simulations, the average construction period is 63.5 days and the standard deviation is 2.1 days. Deviation analysis shows that the maintenance time of carbon fiber may be insufficient, and it is recommended to add 1 day of buffer time. Intelligent scheduling optimization uses genetic algorithm, with a population size of 100 and 500 iterations to obtain 20 candidate solutions. Multi-criteria evaluation considers construction period (weight 0.4), cost (weight 0.3), quality (weight 0.2) and safety (weight 0.1). The optimal solution is expected to have a construction period of 62 days, a cost increase of 2%, but a 5% increase in quality and safety scores. Risk assessment identified weather risks (30% probability of impact, average delay of 2 days) and material supply risks (10% probability of impact, average delay of 3 days) as the main risks through 10,000 Monte Carlo simulations. The response strategies formulated include increasing the construction period buffer by 15% and establishing a backup material supply channel. The final target construction plan adjusts the construction period to 65 days and adds a material advance inspection procedure before the key process.

[0090] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0091] (1) Plan the sensor layout for the target construction plan, determine the location of the monitoring points through the optimal coverage algorithm, obtain the monitoring network topology, and set the data collection frequency for the monitoring network topology to obtain a real-time monitoring plan;

[0092] (2) Collect data for the real-time monitoring solution, pre-process the collected data through edge computing technology to obtain an effective monitoring data stream, and perform anomaly detection on the effective monitoring data stream to obtain abnormal events in the construction process;

[0093] (3) Diagnose the causes of abnormal events in the construction process, analyze the root causes of abnormalities through Bayesian network reasoning, obtain a construction deviation report, and conduct an impact assessment on the construction deviation report to obtain an adjustment priority list;

[0094] (4) Generate a plan for the adjustment priority list, formulate an adjustment strategy through a reinforcement learning algorithm, obtain a set of candidate adjustment plans, and perform simulation verification on the set of candidate adjustment plans to obtain the optimal adjustment plan;

[0095] (5) Generate execution instructions for the optimal adjustment plan, convert it into specific operation instructions through semantic parsing technology, obtain a construction adjustment instruction set, and generate a target intelligent reinforcement implementation plan based on the construction adjustment instruction set.

[0096] Specifically, the sensor layout is planned for the target construction plan, and the location of the monitoring point is determined by the optimal coverage algorithm. The optimal coverage algorithm is based on a greedy strategy, and iteratively selects the sensor location that can maximize the coverage area until the preset coverage requirement is met. This process obtains the monitoring network topology, which reflects the spatial distribution and connection relationship of the sensors. Subsequently, the data acquisition frequency is set for the monitoring network topology, taking into account the characteristics and monitoring targets of different types of sensors. For example, strain sensors may require a higher sampling frequency (such as 100Hz), while displacement sensors may have a lower sampling frequency (such as 1Hz). In this way, a comprehensive real-time monitoring solution is obtained. Data is collected for the real-time monitoring solution, and edge computing technology is used to pre-process the collected data. Edge computing delegates some data processing tasks to edge nodes close to the data source, reducing the amount of data transmission and improving the response speed. Preprocessing includes data cleaning, outlier detection, feature extraction, etc., to obtain an effective monitoring data stream. Anomaly detection is performed on the effective monitoring data stream, and algorithms such as isolation forest or local anomaly factor (LOF) are used to identify data points that are significantly different from the normal mode, thereby obtaining abnormal events in the construction process.

[0097] Cause diagnosis of abnormal events in the construction process is performed, and the root causes of abnormalities are analyzed through Bayesian network reasoning. Bayesian network is a probabilistic graphical model that can represent conditional dependencies between variables. By constructing a Bayesian network containing nodes such as construction process, environmental factors, and equipment status, and setting a conditional probability table based on historical data and expert knowledge, the most likely cause of the abnormal event can be inferred. This process results in a construction deviation report that describes in detail the nature, location, and possible causes of the abnormality. An impact assessment is performed on the construction deviation report, and an adjustment priority list is obtained by considering the severity, development trend, and potential risks of the deviation. Schemes are generated for the adjustment priority list, and an adjustment strategy is formulated using a reinforcement learning algorithm. Reinforcement learning learns the optimal strategy through the interaction between the agent and the environment, and is suitable for dealing with complex decision-making problems. Specifically, a deep Q network (DQN) or a policy gradient method can be used to take the current state (including construction progress, resource allocation, abnormal conditions, etc.) as input and output the optimal adjustment action. This process results in a set of candidate adjustment schemes, which contains multiple possible adjustment strategies. The candidate adjustment scheme set is simulated and verified, and the effect and feasibility of each scheme are evaluated using discrete event simulation or system dynamics models, and the optimal adjustment scheme is finally obtained.

[0098] Finally, the execution instructions for the optimal adjustment plan are generated, and the high-level adjustment strategy is converted into specific operation instructions through semantic parsing technology. Based on natural language processing and domain knowledge graphs, semantic parsing technology can understand the semantic content of the adjustment plan and generate detailed instructions that comply with construction specifications and operation procedures. This process obtains a construction adjustment instruction set, which includes specific process adjustments, resource reallocation, quality control measures, etc. According to the construction adjustment instruction set, combined with the original construction plan, an updated target intelligent reinforcement implementation plan is generated, realizing dynamic optimization and intelligent control of the construction process.

[0099] For example, suppose the main beam of a cross-sea bridge is reinforced with carbon fiber. First, 50 strain sensors and 20 accelerometers are arranged on the 1,000-meter-long main beam through the optimal coverage algorithm, with a coverage rate of 95%. The sampling frequency of the strain sensor is set to 100Hz and the accelerometer is 200Hz. During the construction process, the edge computing node reduces the noise and compression of the original data, reducing the data volume by 60%. The anomaly detection algorithm found an abnormal event in which the strain value suddenly increased by 20% in the middle section of the bridge. Bayesian network analysis showed that there was an 80% probability that this anomaly was caused by bubbles in the carbon fiber laying process. The impact assessment ranked this issue as the highest priority and needed to be dealt with immediately. The reinforcement learning algorithm generated three candidate adjustment plans based on historical data and current status: 1) local rework and re-laying; 2) increase vacuum pressure; 3) extend the curing time. Simulation verification shows that the success rate of plan 1 is the highest, reaching 95%. The semantic parsing system converts the optimal solution into specific instructions: 1) stop the current process; 2) define the rework range (pile number K15+200 to K15+220); 3) remove defective carbon fiber; 4) re-process the surface; 5) re-lay the carbon fiber according to the standard process; 6) increase the curing time by 2 hours. These instructions are integrated into the original construction plan to form a new intelligent reinforcement implementation plan. Through this real-time monitoring and dynamic adjustment, potential quality problems are discovered and solved in a timely manner, ensuring the overall effect of the reinforcement project.

[0100] The above describes the intelligent bridge reinforcement decision-making method based on damage identification in the embodiment of the present application. The following describes the intelligent bridge reinforcement decision-making device based on damage identification in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of an intelligent bridge reinforcement decision-making device based on damage identification includes:

[0101] The acquisition module 201 is used to collect multi-source data of the bridge structure and pre-process the collected data through a multi-modal data fusion algorithm to obtain a bridge status data set;

[0102] An evaluation module 202 is used to identify and evaluate bridge damage on the bridge status data set through a multi-branch neural network model to obtain a bridge damage feature map;

[0103] The analysis module 203 is used to perform intelligent decision analysis on the bridge damage characteristic map, generate a preliminary reinforcement plan, and obtain a candidate reinforcement strategy set;

[0104] The simulation module 204 is used to perform parameterized modeling and multi-scale simulation on the candidate reinforcement strategy set, and to perform coupled calculation evaluation on the reinforcement effect to obtain an optimized reinforcement scheme;

[0105] The planning module 205 is used to perform virtual construction planning on the optimized reinforcement scheme, generate a construction process model through an intelligent scheduling algorithm, and obtain a target construction scheme;

[0106] The feedback module 206 is used to monitor and provide feedback on the target construction plan in real time, and dynamically optimize the construction process through an adaptive control algorithm to obtain a target intelligent reinforcement implementation plan.

[0107] Through the collaboration of the above components, a comprehensive and high-quality bridge status data set is obtained by collecting multi-source data of the bridge structure and preprocessing it using a multimodal data fusion algorithm, which provides a reliable data basis for subsequent analysis and effectively improves the accuracy and comprehensiveness of bridge status assessment. Secondly, a multi-branch neural network model is used to identify and evaluate the damage of the bridge status data set, and a bridge damage feature map is generated, which greatly improves the accuracy and efficiency of damage identification, can timely discover potential structural problems, and provides a key basis for reinforcement decisions. Furthermore, intelligent decision-making analysis is performed on the bridge damage feature map to generate a preliminary reinforcement plan and obtain a candidate reinforcement strategy set, which realizes the automation and intelligence of reinforcement plan generation and greatly reduces the subjectivity and uncertainty of manual decision-making. Then, by performing parametric modeling and multi-scale simulation on the candidate reinforcement strategy set, and performing coupled calculation evaluation, an optimized reinforcement plan is obtained. This process fully considers multiple factors such as reinforcement effect, cost, and construction difficulty, ensuring the scientificity and feasibility of the reinforcement plan. In addition, virtual construction planning is performed on the optimized reinforcement plan, and a construction process model is generated through an intelligent scheduling algorithm to obtain the target construction plan, which significantly improves the efficiency and rationality of construction planning and helps reduce resource waste and time delays during the construction process. Finally, the target construction plan is monitored and fed back in real time, and the construction process is dynamically optimized through an adaptive control algorithm to obtain the target intelligent reinforcement implementation plan. This closed-loop control mechanism greatly enhances the adaptability and robustness of the reinforcement implementation process, and can effectively deal with various uncertainties and emergencies during the construction process.

[0108] Based on the same technical concept, the embodiment of the present application also provides an electronic device. Figure 3 As shown, it is a schematic diagram of the structure of the electronic device 300 provided in the embodiment of the present application, including a processor 301, a memory 302, and a bus 303. Among them, the memory 302 is used to store execution instructions, including a memory 3021 and an external memory 3022; the memory 3021 here is also called an internal memory, which is used to temporarily store the operation data in the processor 301, and the data exchanged with the external memory 3022 such as a hard disk. The processor 301 exchanges data with the external memory 3022 through the memory 3021. When the electronic device 300 is running, the processor 301 communicates with the memory 302 through the bus 303.

[0109] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0110] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0111] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent bridge reinforcement decision-making method based on damage identification, characterized in that: The intelligent bridge reinforcement decision-making method based on damage identification includes: Collect multi-source data of bridge structures, pre-process the collected data through multi-modal data fusion algorithm, and obtain the bridge status data set; Using a multi-branch neural network model to identify and evaluate bridge damage on the bridge status data set, and obtain a bridge damage feature map; Perform intelligent decision-making analysis on the bridge damage characteristic map, generate a preliminary reinforcement plan, and obtain a candidate reinforcement strategy set; Performing parameterized modeling and multi-scale simulation on the candidate reinforcement strategy set, and performing coupled calculation evaluation on the reinforcement effect to obtain an optimized reinforcement scheme; Performing virtual construction planning on the optimized reinforcement scheme, generating a construction process model through an intelligent scheduling algorithm, and obtaining a target construction scheme; The target construction plan is monitored and fed back in real time, and the construction process is dynamically optimized through an adaptive control algorithm to obtain a target intelligent reinforcement implementation plan.

2. The intelligent bridge reinforcement decision-making method based on damage identification according to claim 1 is characterized in that: The multi-source data collection of the bridge structure is performed, and the collected data is preprocessed by a multi-modal data fusion algorithm to obtain a bridge status data set, including: The vibration signal of the bridge structure is collected, and the noise of the vibration signal is eliminated by wavelet transform to obtain the noise-reduced vibration data, and the noise-reduced vibration data is subjected to fast Fourier transform to obtain the frequency domain characteristics; Strain data of the bridge structure is collected, outliers are removed from the strain data by median filtering to obtain filtered strain data, and principal component analysis is performed on the filtered strain data to obtain the strain principal component; The displacement data of the bridge structure is collected, and the displacement data is smoothed by Kalman filtering to obtain the smoothed displacement data. The smoothed displacement data is then analyzed in time series to obtain the displacement trend characteristics. The frequency domain features, strain principal components and displacement trend features are standardized to obtain standardized feature data, and the standardized feature data are multi-modally fused through the tensor decomposition algorithm to obtain a fused feature tensor; The fused feature tensor is subjected to dimensionality reduction to obtain the reduced feature vector, which is then organized into a time series to obtain the bridge status dataset.

3. The intelligent bridge reinforcement decision-making method based on damage identification according to claim 1 is characterized in that: The bridge damage identification and evaluation is performed on the bridge status data set by a multi-branch neural network model to obtain a bridge damage feature map, including: The bridge status data set is divided into time windows, and the data is segmented using a sliding window algorithm to obtain a time series feature subset. The time series feature subset is then Fourier transformed to obtain a frequency domain feature matrix. The frequency domain feature matrix is ​​processed by convolutional neural network, local features are extracted through multi-layer convolution and pooling operations to obtain a convolution feature map, and the convolution feature map is globally averaged pooled to obtain a global feature vector; The global feature vector is processed by the long short-term memory network, and the long-term features are captured by temporal dependency modeling to obtain the temporal feature vector. The temporal feature vector is processed by the self-attention mechanism to obtain the attention-weighted feature. The attention weighted features are processed by a fully connected layer, and feature mapping is performed through a multi-layer perceptron to obtain a high-dimensional feature representation. The high-dimensional feature representation is then classified by softmax to obtain the probability distribution of the damage type. The probability distribution of damage types is filtered by threshold, and high-confidence predictions are screened by setting confidence thresholds to obtain damage type determination results. The damage type determination results are spatially mapped to obtain a bridge damage feature map.

4. The intelligent bridge reinforcement decision-making method based on damage identification according to claim 1 is characterized in that: The intelligent decision-making analysis is performed on the bridge damage characteristic map to generate a preliminary reinforcement plan and obtain a candidate reinforcement strategy set, including: The bridge damage feature map is processed by graph convolutional network, and the topological relationship of the damage is obtained by spatial feature extraction to obtain a damage association map. The damage association map is then embedded into a graph to obtain a damage representation vector. Perform cluster analysis on the damage characterization vectors, group similar damages using the K-means algorithm to obtain a set of damage categories, and evaluate the severity of the damage category set to obtain a reinforcement priority ranking; Perform knowledge graph matching on the reinforcement priority ranking, select the applicable reinforcement technology through semantic similarity calculation, obtain the preliminary reinforcement plan, and check the constraints of the preliminary reinforcement plan to obtain a set of feasible reinforcement plans; Perform multi-objective optimization on the feasible reinforcement scheme set, balance the reinforcement effect and cost through genetic algorithm, obtain the Pareto optimal solution set, and perform decision tree analysis on the Pareto optimal solution set to obtain the scheme decision path; The Monte Carlo simulation is performed on the decision path of the scheme, and the robustness of the scheme is evaluated through random sampling to obtain the scheme reliability index. The scheme reliability index is then sorted and screened to obtain a set of candidate reinforcement strategies.

5. The intelligent bridge reinforcement decision-making method based on damage identification according to claim 1 is characterized in that: The intelligent decision-making analysis is performed on the bridge damage characteristic map to generate a preliminary reinforcement plan and obtain a candidate reinforcement strategy set, including: Performing parameterized modeling on the candidate reinforcement strategy set, generating a geometric model of the reinforcement component by using a non-uniform rational B-spline technique to obtain a reinforcement component model set, and performing meshing on the reinforcement component model set to obtain a finite element model; Assigning material properties to the finite element model, selecting appropriate material parameters through a constitutive relationship database to obtain a complete structural model, and setting boundary conditions for the complete structural model to obtain a calculation-ready model; Performing macro-scale simulation on the computationally ready model, calculating the overall structural response by nonlinear finite element method to obtain macro-mechanical performance indicators, and performing local stress analysis on the macro-mechanical performance indicators to obtain stress distribution in key areas; Performing mesoscopic simulation on the stress distribution in the key area, simulating the internal damage evolution of the material by discrete element method to obtain the micro crack propagation path, and performing fracture mechanics analysis on the micro crack propagation path to obtain the critical stress intensity factor; The reliability of the critical stress intensity factor is evaluated, the structural failure probability is calculated by the first-order and second-order moment method, the reliability index of the reinforcement scheme is obtained, and the reliability index of the reinforcement scheme is comprehensively ranked to obtain the optimized reinforcement scheme.

6. The intelligent bridge reinforcement decision-making method based on damage identification according to claim 1 is characterized in that: The virtual construction planning of the optimized reinforcement scheme is performed, and a construction process model is generated through an intelligent scheduling algorithm to obtain a target construction scheme, including: Decomposing the optimized reinforcement scheme into processes, determining the process priority through the hierarchical analysis method, obtaining a process dependency graph, and performing a critical path analysis on the process dependency graph to obtain a critical process sequence; Performing resource demand analysis on the key process sequence, optimizing resource allocation through a resource balancing algorithm to obtain a preliminary resource allocation plan, and performing conflict detection on the preliminary resource allocation plan to obtain a feasible resource plan; Conduct construction simulation on the feasible resource scheme, evaluate the construction progress through discrete event simulation technology, obtain construction progress forecast data, and conduct deviation analysis on the construction progress forecast data to obtain progress adjustment suggestions; Performing intelligent scheduling optimization on the schedule adjustment suggestion, generating multiple scheduling schemes through a heuristic algorithm to obtain a candidate scheduling set, and performing multi-criteria evaluation on the candidate scheduling set to obtain an optimal scheduling scheme; A risk assessment is performed on the optimal scheduling plan, potential risk factors are analyzed through Monte Carlo simulation, a risk response strategy is obtained, and the risk response strategy is integrated into the construction process to obtain a target construction plan.

7. The intelligent bridge reinforcement decision-making method based on damage identification according to claim 1 is characterized in that: The target construction plan is monitored and fed back in real time, and the construction process is dynamically optimized through an adaptive control algorithm to obtain a target intelligent reinforcement implementation plan, including: Plan the sensor layout for the target construction plan, determine the location of the monitoring points through the optimal coverage algorithm, obtain the monitoring network topology, and set the data collection frequency for the monitoring network topology to obtain a real-time monitoring plan; Collect data for the real-time monitoring solution, pre-process the collected data through edge computing technology to obtain an effective monitoring data stream, and perform anomaly detection on the effective monitoring data stream to obtain abnormal events in the construction process; Diagnose the causes of abnormal events in the construction process, analyze the root causes of the abnormalities through Bayesian network reasoning, obtain a construction deviation report, and conduct an impact assessment on the construction deviation report to obtain an adjustment priority list; Generating a scheme for the adjustment priority list, formulating an adjustment strategy through a reinforcement learning algorithm, obtaining a set of candidate adjustment schemes, and performing simulation verification on the set of candidate adjustment schemes to obtain an optimal adjustment scheme; An execution instruction is generated for the optimal adjustment plan, which is converted into specific operation instructions through semantic parsing technology to obtain a construction adjustment instruction set, and a target intelligent reinforcement implementation plan is generated based on the construction adjustment instruction set.

8. An intelligent bridge reinforcement decision-making device based on damage identification, used to implement the intelligent bridge reinforcement decision-making method based on damage identification as described in any one of claims 1 to 7, characterized in that: The intelligent bridge reinforcement decision-making device based on damage identification includes: The acquisition module is used to collect multi-source data of the bridge structure and pre-process the collected data through a multi-modal data fusion algorithm to obtain a bridge status data set; An evaluation module, used for identifying and evaluating bridge damage on the bridge status data set through a multi-branch neural network model to obtain a bridge damage feature map; An analysis module is used to perform intelligent decision-making analysis on the bridge damage characteristic map, generate a preliminary reinforcement plan, and obtain a candidate reinforcement strategy set; A simulation module is used to perform parameterized modeling and multi-scale simulation on the candidate reinforcement strategy set, and to perform coupled calculation evaluation on the reinforcement effect to obtain an optimized reinforcement plan; A planning module is used to perform virtual construction planning for the optimized reinforcement scheme, generate a construction process model through an intelligent scheduling algorithm, and obtain a target construction scheme; The feedback module is used to monitor and provide feedback on the target construction plan in real time, and dynamically optimize the construction process through an adaptive control algorithm to obtain a target intelligent reinforcement implementation plan.

9. A computer device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor and the memory communicate via the bus, and when the machine-readable instructions are executed by the processor, the intelligent bridge reinforcement decision-making method based on damage identification as described in any one of claims 1 to 7 is executed.

Images