Highway bridge safety state assessment method and system

Abstract

The invention discloses a safety state assessment method and system for a highway bridge, and relates to the technical field of highway bridges, and the method comprises the steps: carrying out the distributed deployment of multi-type sensing equipment, collecting the multi-dimensional data, such as structural stress and deflection, and carrying out the abnormal value elimination and missing value filling preprocessing, thereby obtaining a standardized data set; statistical and trend features are extracted, and feature weights are calculated by adopting a combined weighting model of an improved AHP combined entropy weight method; constructing an improved LSTM and ELM fusion evaluation model introducing an attention mechanism, and inputting a feature vector and a weight to obtain an initial evaluation result; and in combination with historical disease data and an environmental influence coefficient correction result, dividing four security levels and outputting a report. The matching system comprises a data acquisition module, a preprocessing module, a feature extraction and weight distribution module, a fusion evaluation module and the like, a remote early warning function is additionally arranged, 5G + edge calculation and cloud edge collaborative storage are adopted, the evaluation accuracy rate reaches 96.67%, the system adapts to complex working conditions, and accurate and reliable technical support is provided for bridge safety management.

Description

Technical Field

[0001] This invention relates to the field of highway bridge technology, specifically to a method and system for assessing the safety status of highway bridges. Background Technology

[0002] In the field of highway bridge engineering, safety status assessment is a core component in ensuring bridge operational safety and extending their service life. With the continuous increase in traffic volume, rising load levels, and long-term environmental erosion, bridge structures are prone to problems such as stress accumulation, deflection deformation, and material aging. Accurate assessment of their safety status has become a key focus of the industry. Existing technologies for bridge safety assessment primarily revolve around data acquisition and model analysis. Common technical approaches include traditional assessment methods based on single sensor data and intelligent assessment methods combining machine learning models. Machine learning models are represented by Long Short-Term Memory Networks (LSTM) and Extreme Learning Machines (ELM), and weighting methods often employ the Analytic Hierarchy Process (AHP). By extracting features from bridge structural status data and inputting them into the model, safety level determination is achieved, providing data support for bridge maintenance decisions.

[0003] However, existing technologies have significant shortcomings in data processing and feature utilization. On the one hand, data collection dimensions are limited, with most schemes focusing only on core mechanical parameters such as structural stress and deflection, neglecting key influencing factors such as environmental temperature and humidity, and vehicle load fluctuations. Furthermore, data preprocessing methods are simplistic, lacking precision in outlier removal and missing value imputation, resulting in low quality of standardized data. On the other hand, feature extraction lacks systematicity, focusing only on statistical features while ignoring trend feature mining, making it difficult to comprehensively reflect the long-term evolution of bridge structures. Additionally, the traditional AHP weighting method relies on expert subjective judgment to construct the judgment matrix, which is prone to consistency bias, and the objectivity and rationality of weight allocation are insufficient, directly affecting the credibility of the evaluation results.

[0004] Existing assessment models and result correction mechanisms also have shortcomings. In model construction, while a single LSTM model possesses time-series data processing capabilities, its convergence speed is slow and it lacks sufficient focus on key features. Conversely, a single ELM model has limited generalization capabilities and struggles to adapt to the complex and ever-changing operational scenarios of bridges. The lack of effective integration between the two leads to a trade-off between model assessment accuracy and efficiency. Regarding result correction, most solutions fail to fully consider the reference value of historical bridge damage data and lack a scientific mechanism for quantifying environmental impacts. This makes it difficult to effectively avoid interference from factors such as historical damage accumulation, temperature and humidity corrosion, and extreme weather on the assessment results. Consequently, the assessment results deviate from the actual safety status of the bridge, failing to meet the needs of refined and precise safety management. Furthermore, misjudgments may even lead to untimely or excessive maintenance.

[0005] Therefore, a safety status assessment method and system for highway bridges is proposed to address the above problems. Summary of the Invention

[0006] In view of this, the technical problem to be solved by the present invention is to propose a method and system for assessing the safety status of highway bridges, so as to solve the problems in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method and system for assessing the safety status of highway bridges, comprising the following steps: S1. Multi-source data acquisition: Sensing devices are deployed at key structural parts of highway bridges to collect multi-dimensional state data during bridge operation. The multi-dimensional state data includes structural stress data σ, deflection data w, vibration frequency data f, environmental temperature and humidity data T / H, and vehicle load data P, where i is the acquisition time number, i=1,2,...,n, and n is the total number of acquisitions. The acquired multi-dimensional state data is preprocessed, including outlier removal based on the Laida criterion and missing value imputation based on linear interpolation, to obtain a standardized data set D={σ,w,f,T,H,P|i=1,2,...,n}. S2. Feature Extraction and Weight Allocation: Based on the standardized dataset D obtained from the preprocessing in S1, the statistical and trend features of each dimension of the data are extracted. The statistical features include mean, variance, peak factor and kurtosis. The trend features include linear fitting slope and trend term deviation. S3. Construction and Inference of Safety Status Assessment Model: A fusion assessment model based on an improved Long Short-Term Memory (LSTM) network and an Extreme Learning Machine (ELM) is constructed. The feature vectors extracted in S2 and their corresponding weights are input into the fusion assessment model to obtain the bridge safety status assessment results. The improved Long Short-Term Memory (LSTM) network optimizes the weight allocation of the input gate, forget gate, and output gate by introducing an attention mechanism. The attention weight calculation function is as follows:

[0008] In the formula, s represents the similarity between the input feature and the hidden layer state at time t; the output of the improved LSTM is used as the input of the ELM, and the output layer function of the ELM is:

[0009] In the formula, L represents the number of hidden layer nodes, β represents the weight from the hidden layer to the output layer, ω represents the weight from the input layer to the hidden layer, b represents the hidden layer node bias, and g(·) represents the activation function, which is the Sigmoid function. S4. Assessment Result Correction and Output: Historical bridge structural defects and environmental impact coefficients are introduced to correct the initial assessment results obtained in S3. The environmental impact coefficient k comprehensively considers temperature, humidity, rainfall, and corrosion levels, and is determined using the fuzzy comprehensive evaluation method. The formula for calculating the corrected safety status assessment value S is as follows:

[0010] In the formula, y represents the initial output value of the fusion assessment model, and y represents the assessment value based on historical disease data; the safety level is divided according to the assessment value S, including four levels: safe (S≥0.8), relatively safe (0.6≤S<0.8), critical state (0.4≤S<0.6), and dangerous (S<0.4), and an assessment report is output.

[0011] Preferably, the sensing devices described in S1 include fiber optic stress sensors, laser deflection sensors, acceleration sensors, temperature and humidity sensors, and dynamic weighing sensors. The sampling frequency of each sensing device is synchronously set to 100Hz, and data transmission adopts 5G+edge computing mode to realize real-time data acquisition and preprocessing.

[0012] As a preferred embodiment, the improved analytic hierarchy process described in S2 introduces a fuzzy complementary judgment matrix to replace the traditional judgment matrix. The fuzzy complementary judgment matrix A satisfies a+a=1 and a=0.5. It is converted into a fuzzy consistent matrix using the fuzzy consistent matrix conversion formula, thereby improving the accuracy of weight calculation.

[0013] The feature extraction and weight allocation described in S2 employs a combined weighting model that combines the improved Analytic Hierarchy Process (AHP) with the entropy weighting method to calculate the weights of features in each dimension. The specific calculation process is as follows: S2.1 Constructing a hierarchical model: The target layer is the weight allocation for bridge safety assessment, the criterion layer is the features of each dimension, and the scheme layer is the samples of each collected data. S2.2 Hierarchical Single Sorting and Consistency Check: Construct a judgment matrix A through expert scoring, calculate the maximum eigenvalue λ and the corresponding eigenvector ω of the judgment matrix A, and use the consistency index CI=(λ-m) / (m-1) for consistency check, where m is the number of features in the criterion layer. When CI<0.1, the judgment matrix meets the consistency requirements; otherwise, the judgment matrix is ​​reconstructed. S2.3 Entropy Weight Method for Calculating Objective Weights: Calculate the information entropy E of the j-th feature using the following formula:

[0014] In the formula, p represents the normalized value of the j-th feature of the i-th sample, p = x / Σx, where x is the original value of the j-th feature of the i-th sample; the objective weight ω is calculated based on information entropy, and the formula is:

[0015] S2.4 Combined Weight Calculation: The subjective weights ω obtained from the improved Analytic Hierarchy Process (AHP) and the objective weights ω obtained from the entropy weight method are linearly combined to obtain the final weights ω of each dimension feature. The formula is as follows:

[0016] Where α is the weighting coefficient, α∈[0.4,0.6], and the optimal value is determined by cross-validation. Preferably, the training process of the fusion evaluation model described in S3 includes: collecting historical status data and safety assessment results of different types of highway bridges, constructing training datasets and test datasets; inputting the training datasets into the fusion evaluation model after normalization, using the adaptive momentum optimization algorithm (Adam) to optimize the model parameters, using mean squared error (MSE) as the loss function, and stopping training when the loss function value converges to a preset threshold or the number of training iterations reaches its maximum value.

[0017] A highway bridge safety status assessment system, comprising: Data acquisition module: used to collect multi-dimensional status data of the bridge during operation from sensing devices deployed at key structural parts of highway bridges. The sensing devices include fiber optic stress sensors, laser deflection sensors, acceleration sensors, temperature and humidity sensors, and dynamic weighing sensors. Data preprocessing module: Connected to the data acquisition module, it is used to remove outliers and fill in missing values ​​in the acquired multi-dimensional state data to obtain a standardized dataset; Feature extraction and weight allocation module: Connected to the data preprocessing module, it is used to extract the statistical and trend features of the standardized dataset, and calculate the weights of each dimension of features using a combined weighting model of improved analytic hierarchy process (AHP) and entropy weighting method. Fusion evaluation module: Connects the feature extraction and weight allocation modules, and has a built-in fusion evaluation model based on improved LSTM and ELM. It is used to input the extracted feature vectors and corresponding weights into the fusion evaluation model to obtain the initial security status evaluation results. Results Correction and Output Module: Connects to the fusion assessment module, used to import historical bridge structural defects data and environmental impact coefficients to correct the initial assessment results, classify safety levels, and output an assessment report; Database module: Connects to the data acquisition module, data preprocessing module, feature extraction and weight allocation module, fusion evaluation module, and result correction and output module respectively. It is used to store the acquired raw data, preprocessed data, feature data, weight data, model parameters, historical disease data, and evaluation results.

[0018] Preferably, it also includes a remote monitoring and early warning module and a connection result correction and output module, which are used to monitor the bridge safety status assessment results in real time. When the assessment results are in a critical state or dangerous level, an audible and visual early warning is automatically triggered, and the early warning information is pushed to the relevant management terminal.

[0019] Preferably, the data acquisition module adopts a distributed deployment method, with sensing devices deployed in key structural parts of the bridge such as the main beam, piers, supports and bridge deck. Each sensing device achieves synchronous data transmission through a wireless sensor network, with a transmission delay of ≤50ms.

[0020] Preferably, the database module adopts a cloud-edge collaborative storage architecture, where edge nodes store real-time collected raw and pre-processed data, and cloud nodes store historical data, model parameters, and evaluation reports, supporting data backup and traceability.

[0021] Compared with the prior art, the safety status assessment method and system for highway bridges provided by the present invention have the following beneficial effects: (1) Data collection and preprocessing are more comprehensive and accurate; This solution overcomes the limitations of existing technologies that rely on a single data acquisition dimension. By distributing multiple high-precision sensors across key structural components such as the bridge's main girder, piers, and bearings, it simultaneously collects core mechanical parameters like structural stress, deflection, and vibration frequency, along with multi-dimensional data such as environmental temperature and humidity, and vehicle loads, achieving comprehensive awareness of the bridge's operational status. Furthermore, a combined preprocessing strategy employing the Laida criterion for outlier removal and linear interpolation to fill in missing values, coupled with real-time transmission via 5G+edge computing, ensures the integrity and accuracy of the standardized dataset. This provides high-quality data support for subsequent evaluations and effectively addresses the issue of low data quality in traditional solutions.

[0022] (2) Feature extraction and weight allocation are more scientific; Compared to existing technologies that only focus on statistical characteristics and suffer from strong subjectivity in weight allocation, this solution simultaneously extracts statistical and trend characteristics from data across all dimensions, comprehensively capturing the state change patterns and long-term evolution trends of bridge structures. In weight calculation, it innovatively employs a combined weighting model that integrates an improved analytic hierarchy process (AHP) with entropy weighting. By optimizing the subjective judgment bias of traditional AHP through a fuzzy complementary judgment matrix and combining the objective data-driven advantages of entropy weighting, the final weights are obtained through linear combination. This approach balances expert experience with respect for the essence of the data, resulting in a consistency index (CI) far below 0.1, smaller variance in feature contribution, and significantly improved rationality and accuracy of weight allocation.

[0023] (3) The evaluation model and result correction are more adapted to the actual scenario; This solution overcomes the performance limitations of single models by constructing an improved LSTM and ELM fusion evaluation model. It optimizes the gating weight allocation of LSTM through an attention mechanism, strengthening the focus on key features. Simultaneously, it leverages the rapid learning capability of ELM to improve model inference efficiency. Through Adam algorithm optimization and MSE loss function constraints, the model's convergence speed and evaluation accuracy significantly outperform traditional single LSTM models. Furthermore, it innovatively introduces historical bridge damage data and environmental impact coefficients, quantifying the influence of environmental factors such as temperature and corrosion through fuzzy comprehensive evaluation. A weighted formula is used to double-correct the initial evaluation results, making the evaluation results more reflective of the actual safety status of the bridge. Evaluation errors are significantly reduced under complex conditions such as vehicle overloading and extreme high temperatures, demonstrating greater adaptability.

[0024] (4) The system architecture and functional design are more in line with engineering requirements; The assessment system designed in this scheme features a robust modular architecture and practical functional expansion. The database adopts a cloud-edge collaborative storage architecture, with edge nodes storing real-time data to ensure processing efficiency, and cloud nodes backing up historical data to support traceability analysis, resolving the contradiction between data storage and access in traditional systems. Simultaneously, a new remote monitoring and early warning module is added. When the assessment results reach a critical or dangerous level, it can automatically trigger audible and visual warnings and push information to the management terminal, achieving closed-loop management from assessment to early warning. The system's modules collaborate efficiently, with data transmission latency ≤50ms. Data preprocessing and model inference efficiency far exceed existing technologies, meeting the engineering needs for real-time monitoring and refined management of bridge safety, and providing a scientific and timely basis for maintenance decisions. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the process for assessing the safety status of highway bridges according to the present invention; Figure 2 This is a schematic diagram of the architecture of a highway bridge safety status assessment system according to the present invention; Figure 3 This is a schematic diagram comparing the rationality of weight allocation in embodiments of the present invention; Figure 4 This is a schematic diagram comparing the data processing efficiency of embodiments of the present invention; Figure 5 This is a schematic diagram comparing the adaptability of the present invention to extreme working conditions in an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0027] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0028] Example 1, please refer to Figures 1 to 5 As shown: Existing technologies employ a single LSTM model combined with the traditional AHP weighting method for safety assessment, which suffers from low data processing accuracy, strong subjectivity in weight allocation, and failure to consider the coupling effects of historical defects and the environment in the assessment results. This embodiment compares the method of this invention with existing technologies to verify the superiority of the proposed solution.

[0029] Sensor deployment: Sensors are distributed across the main girder at mid-span, quarter-span, pier tops, bearings, and key stress areas of the bridge deck, with the specific configuration as follows:

[0030] All sensing devices are synchronously set to a sampling frequency of 100Hz, and data is transmitted using 5G+edge computing mode. The actual transmission latency is 32~45ms, which meets the design requirement of ≤50ms.

[0031] Data storage and processing equipment: Industrial-grade servers are used for edge nodes, and cloud server clusters are used for cloud nodes to achieve cloud-edge collaborative storage. Bridge operation data was continuously collected for 60 days, with a total of n = 60 × 24 × 3600 × 100 = 51,840,000 collections. Multi-dimensional status data were obtained, including: structural stress data σ (range: -850-920 με), deflection data w (range: 0.3-12.8 mm), vibration frequency data f (range: 825 Hz), environmental temperature and humidity data T / H (temperature: 538℃; humidity: 40%-95%RH), and vehicle load data P. Simultaneously, 12 years of historical damage data for the bridge were collected, including: two instances of minor crack damage and one instance of bearing aging damage. Historical damage treatment records and subsequent monitoring data are complete. The safety level was determined according to step four of this invention, with the modified safety status assessment value S as the evaluation index. The specific standards are as follows: Table 1 is a statistical table of evaluation indicators;

[0032] Example 2, the specific implementation process described above is as follows: Step 1: Data preprocessing; First, outlier removal was performed. Based on the Raida criterion (3σ criterion), outlier detection was performed on the collected multi-dimensional data. A total of 326 outliers in stress data, 189 outliers in deflection data, 98 outliers in vibration frequency data, 45 outliers in temperature and humidity data, and 123 outliers in vehicle load data were removed. The outlier rate of all data was less than 0.001%.

[0033] Subsequently, missing value imputation was performed. Linear interpolation was used to fill in the missing data generated after outlier removal. After imputation, the data integrity reached 100%, resulting in a standardized dataset D={σ,w,f,T,H,P|i=1,2,...,51840000}.

[0034] Step 2: Feature extraction and weight allocation; In the feature extraction stage, statistical features (mean, variance, peak factor, kurtosis) and trend features of each dimension of data are extracted from the standardized dataset D, resulting in a total of 36 feature indicators in 6 dimensions × 6 features.

[0035] The weight calculation employs a combined weighting model using an improved AHP (Advanced Hierarchical Method) and entropy weighting. In the improved AHP subjective weight calculation, a hierarchical model is constructed: the target layer represents the weight allocation for bridge safety assessment, the criterion layer comprises six dimensional features, and the scheme layer consists of collected data samples. A fuzzy complementary judgment matrix A (satisfying a_ij+a_ji=1, a_ii=0.5) is constructed through scoring by five bridge engineering experts. After conversion to a fuzzy consistency matrix, the maximum eigenvalue λ=6.23 is calculated, and the consistency index CI=(6.23-6) / (6-1)=0.046<0.1, satisfying the consistency requirement, thus obtaining the subjective weight ω_Aj. In the entropy weighting objective weight calculation, the information entropy E_j (range: 0.82~0.95) of each feature is calculated according to the information entropy formula, and the objective weight ω_Ej is calculated based on E_j. Finally, the weight coefficient α = 0.52 was determined by cross-validation. The final weight ω_j was calculated according to the combined weight formula, where the combined weight of structural stress σ is 0.266, deflection w is 0.260, vibration frequency f is 0.185, ambient temperature and humidity T / H is 0.125, and vehicle load P is 0.164.

[0036] Step 3: Integrate the evaluation model training and inference; When constructing the dataset, 80% of the standardized data was selected as the training dataset and 20% as the test dataset. The training dataset also included data samples from historical disease occurrence periods.

[0037] For model training, an improved LSTM and ELM fusion evaluation model was constructed. The improved LSTM introduced an attention mechanism to optimize gating weight allocation, and the ELM had 128 hidden layer nodes (L=128) and used the Sigmoid function as the activation function. The Adam optimization algorithm was used to optimize the model parameters, with MSE as the loss function. The number of training iterations was set to 200. When the iteration reached 126, the loss function value converged to the preset threshold of 0.0012, and training was stopped.

[0038] In the initial evaluation phase, the feature vectors of the test dataset and the corresponding weights are input into the trained fusion evaluation model to obtain the initial evaluation result y_0 (range: 0.65~0.92).

[0039] Step 4: Evaluation result correction and output; First, using the fuzzy comprehensive evaluation method, considering temperature, humidity, rainfall, and corrosion degree, the environmental impact coefficient k = 0.68 was determined. Then, based on historical disease data and treatment effects, the weighted average method was used to calculate the historical disease assessment value y_h = 0.72. According to the corrected formula S = k·y_0 + (1 k)·y_h, the corrected safety status assessment value S (range: 0.67~0.89) is calculated.

[0040] Based on the S-value, the safety level of the bridge was classified. Over the 60 days, the safety level distribution was as follows: 42 days were at a safe level (S≥0.8), accounting for 70%; 18 days were at a relatively safe level (0.6≤S<0.8), accounting for 30%. There were no critical or dangerous periods. The final output is an assessment report including data trends, feature analysis, evaluation results, and maintenance recommendations.

[0041] Comparative Example 1 uses the same test bridge, data acquisition cycle, and evaluation criteria as this invention, only changing the evaluation method. It employs a single LSTM model and the traditional AHP weighting method (without introducing a fuzzy complementary judgment matrix), without considering historical disease data and environmental impact coefficient corrections. Comparative Example 2 uses a fusion model of LSTM and ELM, with the traditional AHP weighting method, only considering the correction of the environmental impact coefficient (without incorporating historical damage data). The comparative indicators include assessment accuracy, the degree of consistency with the actual bridge condition verified by manual inspection, the rationality of weight allocation (measured by the consistency index CI and feature contribution variance), data processing efficiency, data preprocessing and model inference time per unit time, and adaptability to extreme working conditions. Table 2 is a statistical table comparing the rationality of weight allocation;

[0042] The consistency index (CI) of this invention is 0.046, and the variance of feature contribution is 0.0032; the CI of Comparative Example 1 is 0.189, and the variance of feature contribution is 0.0087; the CI of Comparative Example 2 is 0.175, and the variance of feature contribution is 0.0073. The CI value of this invention is much lower than 0.1, and the variance of feature contribution is even smaller, indicating that the weight allocation is more consistent and reasonable, avoiding the subjective biases of traditional AHP.

[0043] Table 3 is a statistical table comparing data processing efficiency;

[0044] The data preprocessing time of this invention is 0.8 hours / day, and the model inference time is 2.3 ms / sample. Comparative Example 1's data preprocessing time is 1.2 hours / day, and the model inference time is 3.8 ms / sample; Comparative Example 2's data preprocessing time is 1.0 hour / day, and the model inference time is 2.9 ms / sample. This invention employs a 5G+edge computing mode and cloud-edge collaborative storage, resulting in higher efficiency in data preprocessing and model inference, meeting real-time evaluation requirements.

[0045] Table 3 shows the comparison of adaptability under extreme operating conditions; Under vehicle overload (120t) conditions, the evaluation error of this invention is 0.032, compared to 0.087 in Comparative Example 1 and 0.065 in Comparative Example 2; under extreme high temperature (45℃) conditions, the evaluation error of this invention is 0.028, compared to 0.079 in Comparative Example 1 and 0.058 in Comparative Example 2; under heavy rain (200mm / day) conditions, the evaluation error of this invention is 0.035, compared to 0.092 in Comparative Example 1 and 0.069 in Comparative Example 2. The evaluation error of this invention under extreme conditions is significantly lower than that of the comparative examples, indicating that the fusion model and correction mechanism enhance adaptability in complex environments.

[0046] In summary, this embodiment, through testing on actual highway bridges, demonstrates that the highway bridge safety status assessment method and system of the present invention possess multiple advantages. Firstly, the assessment accuracy is high (96.67%), significantly superior to existing single models and traditional weighting methods, accurately reflecting the bridge's safety status. Secondly, the weight allocation is reasonable, and data processing efficiency is high, meeting the needs of real-time monitoring and assessment. Thirdly, it exhibits strong adaptability to extreme working conditions, considering the coupled effects of historical defects and the environment, making the assessment results more reliable. Fourthly, the system deployment is flexible, with cloud-edge collaborative storage supporting data backup and traceability, and remote early warning functionality enabling timely response to critical states, providing a scientific basis for bridge maintenance. This embodiment fully verifies the feasibility and superiority of the present invention in practical applications and can be widely extended to various highway bridge safety status assessment scenarios.

[0047] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0048] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for assessing the safety status of highway bridges, characterized in that, Includes the following steps: S1. Multi-source data acquisition: Sensing devices are deployed at key structural parts of highway bridges to collect multi-dimensional state data during bridge operation. The multi-dimensional state data includes structural stress data (σ), deflection data (w), vibration frequency data (f), environmental temperature and humidity data (T / H), and vehicle load data (P). The collected multi-dimensional state data is preprocessed, including outlier removal based on the Laida criterion and missing value imputation based on linear interpolation, to obtain a standardized data set D={σ,w,f,T,H,P|i=1,2,...,n}. S2. Feature Extraction and Weight Allocation: Based on the standardized dataset D obtained from the preprocessing in S1, the statistical and trend features of each dimension of the data are extracted. The statistical features include mean, variance, peak factor and kurtosis. The trend features include linear fitting slope and trend term deviation. S3. Construction and Inference of Safety Status Assessment Model: A fusion assessment model based on an improved Long Short-Term Memory (LSTM) network and an Extreme Learning Machine (ELM) is constructed. The feature vectors extracted in S2 and their corresponding weights are input into the fusion assessment model to obtain the bridge safety status assessment results. The improved Long Short-Term Memory (LSTM) network optimizes the weight allocation of the input gate, forget gate, and output gate by introducing an attention mechanism. The attention weight calculation function is as follows: In the formula, s represents the similarity between the input feature and the hidden layer state at time t; the output of the improved Long Short-Term Memory (LSTM) network is used as the input of the Extreme Learning Machine (ELM), and the output layer function of the Extreme Learning Machine (ELM) is: In the formula, L represents the number of hidden layer nodes, β represents the weight from the hidden layer to the output layer, ω represents the weight from the input layer to the hidden layer, b represents the hidden layer node bias, and g(·) represents the activation function, which is the Sigmoid function. S4. Assessment Result Correction and Output: Historical bridge structural defects and environmental impact coefficients are introduced to correct the initial assessment results obtained in S3. The environmental impact coefficient k comprehensively considers temperature, humidity, rainfall, and corrosion levels, and is determined using the fuzzy comprehensive evaluation method. The formula for calculating the corrected safety status assessment value S is as follows: In the formula, y represents the initial output value of the fusion assessment model, and y represents the assessment value based on historical disease data; the safety level is divided according to the assessment value S, including four levels: safe, relatively safe, critical state and dangerous, and an assessment report is output.

2. The method for assessing the safety status of highway bridges according to claim 1, characterized in that, The sensing devices described in S1 include fiber optic stress sensors, laser deflection sensors, acceleration sensors, temperature and humidity sensors, and dynamic weighing sensors. The sampling frequency of each sensing device is synchronously set to 100Hz, and data transmission adopts 5G+edge computing mode to realize real-time data acquisition and preprocessing.

3. The method for assessing the safety status of highway bridges according to claim 1, characterized in that, The improved analytic hierarchy process described in S2 introduces a fuzzy complementary judgment matrix to replace the traditional judgment matrix. The fuzzy complementary judgment matrix A satisfies a+a=1 and a=0.

5. It is converted into a fuzzy consistent matrix through the fuzzy consistent matrix transformation formula, thereby improving the accuracy of weight calculation. The feature extraction and weight allocation described in S2 employs a combined weighting model that combines the improved Analytic Hierarchy Process (AHP) with the entropy weighting method to calculate the weights of features in each dimension. The specific calculation process is as follows: S2.1 Constructing a hierarchical model: The target layer is the weight allocation for bridge safety assessment, the criterion layer is the features of each dimension, and the scheme layer is the samples of each collected data. S2.2 Hierarchical Single Sorting and Consistency Check: Construct a judgment matrix A through expert scoring, calculate the maximum eigenvalue λ and the corresponding eigenvector ω of the judgment matrix A, and use the consistency index CI=(λ-m) / (m-1) for consistency check, where m is the number of features in the criterion layer. When CI<0.1, the judgment matrix meets the consistency requirements; otherwise, the judgment matrix is ​​reconstructed. S2.3 Entropy Weight Method for Calculating Objective Weights: Calculate the information entropy E of the j-th feature using the following formula: In the formula, p represents the normalized value of the j-th feature of the i-th sample, p=x / Σx, where x is the original value of the j-th feature of the i-th sample; The objective weight ω is calculated based on information entropy, using the following formula: S2.4 Combined Weight Calculation: The subjective weights ω obtained from the improved Analytic Hierarchy Process (AHP) and the objective weights ω obtained from the entropy weight method are linearly combined to obtain the final weights ω of each dimension feature. The formula is as follows: Where α is the weighting coefficient, α∈[0.4,0.6], and the optimal value is determined by cross-validation.

4. The method for assessing the safety status of highway bridges according to claim 1, characterized in that, The training process of the fusion evaluation model described in S3 includes: collecting historical status data and safety assessment results of different types of highway bridges, constructing training datasets and test datasets; inputting the normalized training datasets into the fusion evaluation model, using the adaptive momentum optimization algorithm (Adam) to optimize the model parameters, using mean squared error (MSE) as the loss function, and stopping training when the loss function value converges to a preset threshold or the number of training iterations reaches its maximum value.

5. A highway bridge safety status assessment system, applicable to the highway bridge safety status assessment method according to any one of claims 1-4, characterized in that, include: Data acquisition module: used to collect multi-dimensional status data of the bridge during operation from sensing devices deployed at key structural parts of highway bridges. The sensing devices include fiber optic stress sensors, laser deflection sensors, acceleration sensors, temperature and humidity sensors, and dynamic weighing sensors. Data preprocessing module: Connected to the data acquisition module, it is used to remove outliers and fill in missing values ​​in the acquired multi-dimensional state data to obtain a standardized dataset; Feature extraction and weight allocation module: Connected to the data preprocessing module, it is used to extract the statistical and trend features of the standardized dataset, and calculate the weights of each dimension of features using a combined weighting model of improved analytic hierarchy process (AHP) and entropy weighting method. Fusion evaluation module: Connects the feature extraction and weight allocation modules, and has a built-in fusion evaluation model based on improved long short-term memory network (LSTM) and extreme learning machine (ELM). It is used to input the extracted feature vectors and corresponding weights into the fusion evaluation model to obtain the initial security status evaluation results. Results Correction and Output Module: Connects to the fusion assessment module, used to import historical bridge structural defects data and environmental impact coefficients to correct the initial assessment results, classify safety levels, and output an assessment report; Database module: Connects to the data acquisition module, data preprocessing module, feature extraction and weight allocation module, fusion evaluation module, and result correction and output module respectively. It is used to store the acquired raw data, preprocessed data, feature data, weight data, model parameters, historical disease data, and evaluation results.

6. The highway bridge safety status assessment system according to claim 5, characterized in that, It also includes a remote monitoring and early warning module and a connection result correction and output module, which are used to monitor the bridge safety status assessment results in real time. When the assessment results are in a critical state or dangerous level, an audible and visual early warning is automatically triggered and the warning information is pushed to the relevant management terminal.

7. The highway bridge safety status assessment system according to claim 5, characterized in that, The data acquisition module adopts a distributed deployment method, with sensing devices deployed on key structural parts of the bridge such as the main beam, piers, supports, and bridge deck. Each sensing device achieves synchronous data transmission through a wireless sensor network, with a transmission delay of ≤50ms.

8. The highway bridge safety status assessment system according to claim 5, characterized in that, The database module adopts a cloud-edge collaborative storage architecture, with edge nodes storing real-time collected raw and pre-processed data, and cloud nodes storing historical data, model parameters, and evaluation reports, supporting data backup and traceability.

Images