Bridge steel bar corrosion dynamic monitoring system based on fiber grating sensor

Abstract

The application provides a bridge steel bar corrosion dynamic monitoring system based on a fiber grating sensor, relates to the field of bridge structure health monitoring and civil engineering safety technology, and comprises a fiber grating sensor array, the fiber grating sensor array is arranged on the surface of steel bars at key positions, is used for collecting radial expansion pressure, point temperature and macroscopic axial stress, the fiber grating sensor array is connected with a fiber demodulator, the fiber demodulator is connected with a data acquisition and communication unit, the fiber demodulator is used for calculating and obtaining the data collected by the fiber grating sensor array, and the data is sent to the data acquisition and communication unit, the data acquisition and communication unit is connected with a data processing unit, and the data processing unit is used for performing bridge steel bar corrosion dynamic monitoring according to the obtained data.The application realizes millimeter-level quantitative diagnosis of steel bar corrosion, significantly improves monitoring precision and timeliness, and provides reliable technical support for long-term operation and maintenance of bridge structures.

Description

Technical Field

[0001] This invention relates to the field of bridge structural health monitoring and civil engineering safety technology, and in particular to a dynamic monitoring system for bridge steel reinforcement corrosion based on fiber optic grating sensing. Background Technology

[0002] Steel corrosion is one of the main factors leading to decreased durability, loss of load-bearing capacity, and even catastrophic damage in reinforced concrete bridges. Traditional steel corrosion monitoring methods, such as the half-cell potential method and the linear polarization method, have inherent drawbacks, including susceptibility to electromagnetic interference, significant environmental influence on measurement results, difficulty in long-term stable operation, and the inability to provide only point-based information and quantitative measurement of corrosion rate and products. Fiber Bragg grating sensors, on the other hand, offer significant advantages such as resistance to electromagnetic interference, corrosion resistance, small size, ease of distributed deployment, absolute measurement capabilities, and good long-term stability, making them ideal for long-term monitoring in harsh environments like bridges. However, current technologies largely rely on simple threshold judgments of physical quantity changes, exhibiting limitations such as a single sensing dimension, strong model dependence, low intelligence level, and inability to provide quantitative assessments. For example, they typically depend on only a single parameter like strain or temperature, making it difficult to accurately distinguish corrosion. The strain is caused by load and temperature changes; traditional decoupling methods rely on precise physical models, but actual bridge structures are complex, subject to numerous environmental interferences, making it difficult to establish precise physical models; there is a lack of ability to mine the deep features of monitoring data, making it impossible to accurately identify the corrosion state, provide early warnings, and predict trends; it is difficult to directly and accurately convert sensor signals into key quantitative indicators such as steel corrosion depth and corrosion rate, especially as existing monitoring methods have significant shortcomings in data processing, failing to effectively combine advanced deep learning algorithms for multi-source signal decoupling, and lacking effective automated hyperparameter optimization mechanisms in model optimization, resulting in limited prediction accuracy and delayed early warning response, failing to meet the urgent needs of modern bridge intelligent operation and maintenance. Therefore, there is an urgent need for a new method for steel corrosion monitoring that can overcome the above shortcomings and achieve intelligent, quantitative, and dynamic real-time monitoring. Thus, designing a dynamic monitoring system for bridge steel corrosion based on fiber optic grating sensing is essential. Summary of the Invention

[0003] To overcome the shortcomings of the existing technology, the purpose of this invention is to provide a dynamic monitoring system for bridge steel reinforcement corrosion based on fiber optic grating sensing.

[0004] To achieve the above objectives, the present invention provides the following solution:

[0005] This invention provides a dynamic monitoring system for bridge steel reinforcement corrosion based on fiber Bragg grating (FBG) sensing, comprising: a FBG sensor array, a fiber demodulator, a data acquisition and communication unit, and a data processing unit. The FBG sensor array is disposed on the surface of the steel reinforcement at key locations to collect radial expansion pressure, measuring point temperature, and macroscopic axial stress. The FBG sensor array is connected to the fiber demodulator, which is connected to the data acquisition and communication unit. The fiber demodulator is used to process the data collected by the FBG sensor array and send it to the data acquisition and communication unit. The data acquisition and communication unit is connected to the data processing unit, which sends the acquired data to the data processing unit. The data processing unit is used to perform dynamic monitoring of bridge steel reinforcement corrosion based on the acquired data.

[0006] The key areas include the water level fluctuation zone and splash zone of the bridge pier, the area below the drainage outlet on the bridge deck, and the area around the expansion joint.

[0007] Preferably, the fiber optic grating sensor array includes an optical fiber and a radial expansion pressure sensor, a measuring point temperature sensor, and a macroscopic axial stress sensor connected in series on the optical fiber, and the optical fiber is connected to the optical fiber demodulator.

[0008] Preferably, the data processing unit is used to perform dynamic monitoring of bridge steel reinforcement corrosion based on the acquired data, specifically:

[0009] Step 1: Obtain radial expansion pressure, measuring point temperature, and macroscopic axial stress data, and preprocess them;

[0010] Step 2: Construct a dynamic monitoring model for steel reinforcement corrosion in bridges;

[0011] Step 3: Build the dataset;

[0012] Step 4: Train the dynamic monitoring model for bridge steel reinforcement corrosion based on the dataset to obtain the trained dynamic monitoring model for bridge steel reinforcement corrosion.

[0013] Step 5: Input the radial expansion pressure, measuring point temperature and macroscopic axial stress data to be monitored into the trained bridge steel reinforcement corrosion dynamic monitoring model to obtain corrosion indicator factors.

[0014] Step 6: Calculate the corrosion depth of the reinforcing steel based on the corrosion indicator factor;

[0015] Step 7: Dynamic early warning based on the depth of steel corrosion.

[0016] Preferably, in step 2, the dynamic monitoring model for steel reinforcement corrosion of the bridge is constructed as follows:

[0017] A dynamic monitoring model for steel bar corrosion of bridges is constructed based on an improved CNN-LSTM network structure. The specific structure of the improved CNN-LSTM network structure includes an input layer, two CNN modules, two LSTM modules, and a fully connected layer. Each of the two CNN modules includes a convolutional layer and a pooling layer. The convolutional layer of the first CNN module includes 32 filters and a kernel size of 3, while the convolutional layer of the second CNN module includes 64 filters and a kernel size of 3.

[0018] Each of the LSTM modules has 64 hidden units.

[0019] Preferably, in step 3, constructing the dataset specifically involves:

[0020] Make multiple steel bars identical to those in step 1 and cast them into concrete test blocks;

[0021] Experiments were conducted on the test blocks based on a controllable accelerated corrosion environment;

[0022] During the experiment, radial expansion pressure, measuring point temperature, macroscopic axial stress data, and actual corrosion depth were collected.

[0023] The radial expansion pressure, measuring point temperature, and macroscopic axial stress data for each time period are aligned with the actual corrosion depth measured at the corresponding time. Each segment of radial expansion pressure, measuring point temperature, and macroscopic axial stress data is assigned a corrosion indicator factor as a label.

[0024] The dataset is constructed based on the aligned data.

[0025] Preferably, step 4: training the dynamic monitoring model for bridge steel reinforcement corrosion based on the dataset, specifically as follows:

[0026] The dataset is divided into three parts: 70% as the training set, 15% as the validation set, and 15% as the test set.

[0027] Based on the training and validation sets, the IGWO algorithm was used to train the dynamic monitoring model for steel corrosion of bridges to obtain the optimal hyperparameters;

[0028] The dynamic monitoring model for bridge steel corrosion was re-initialized based on the optimal hyperparameters. The training set and validation set were merged, and the initialized dynamic monitoring model for bridge steel corrosion was finally trained based on the merged dataset.

[0029] The final trained dynamic monitoring model for bridge steel corrosion was evaluated based on the test set.

[0030] Preferably, in step 6, the corrosion depth of the reinforcing steel is calculated based on the corrosion indicator factor, specifically as follows:

[0031] Construct a physical-guided Gaussian process regression model;

[0032] It is trained based on a pre-defined rust indicator factor - steel bar rust depth dataset;

[0033] The corrosion indicator factor to be detected is input into the trained Gaussian process regression model to obtain the corrosion depth of the steel bars.

[0034] Preferably, in step 7, dynamic early warning is performed based on the depth of steel bar corrosion, specifically as follows:

[0035] A four-level early warning system is established. If the depth of steel bar corrosion is less than 0.1 mm, it is judged as a level one early warning. The data processing unit records the depth of steel bar corrosion without performing any other processing.

[0036] If the corrosion depth of the steel bar is less than 0.3 mm and 0.1 mm, it is judged as a level two warning. The data processing unit sends it to the monitoring platform for staff to view and shortens the monitoring interval of this steel bar to increase the monitoring frequency.

[0037] If the steel bar corrosion depth is less than 0.5mm and less than 0.3mm, it is judged as a level three warning. The data processing unit sends it to the monitoring platform, the monitoring platform generates an operation and maintenance report, and sends it to the maintenance management department for maintenance.

[0038] If the depth of steel bar corrosion is ≥0.5mm, it is judged as a level four warning. The data processing unit sends it to the monitoring platform, and the monitoring platform notifies the relevant departments to take traffic control measures and immediately carry out reinforcement and repair.

[0039] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0040] This invention provides a dynamic monitoring system for bridge steel reinforcement corrosion based on fiber Bragg grating (FBG) sensing, comprising a FBG sensor array, a fiber demodulator, a data acquisition and communication unit, and a data processing unit. The FBG sensor array is installed on the surface of the steel reinforcement at key locations to collect radial expansion pressure, measuring point temperature, and macroscopic axial stress. The FBG sensor array is connected to the fiber demodulator, which is connected to the data acquisition and communication unit. The fiber demodulator is used to process the data collected by the FBG sensor array and send it to the data acquisition and communication unit. The data acquisition and communication unit is connected to the data processing unit, which sends the acquired data to the data processing unit. The data processing unit is used to perform dynamic monitoring of bridge steel reinforcement corrosion based on the acquired data. This invention achieves real-time, online, and quantitative diagnosis and prediction of bridge steel reinforcement corrosion status by deploying a fiber optic grating sensor array integrating radial expansion pressure, temperature measurement, and macroscopic axial stress sensing functions on the surface of steel reinforcement in key areas, and by incorporating an improved CNN-LSTM deep learning model and IGWO optimization algorithm. It boasts significant advantages such as high precision and anti-interference capabilities, quantitative monitoring, dynamic real-time and forward-looking prediction, intelligence and self-learning, and strong engineering applicability. Specifically, by decoupling the multi-physics FBG sensor array and deep learning signal, the interference of temperature and load strain on corrosion signals is effectively eliminated, significantly improving the accuracy and reliability of monitoring. It achieves a direct and quantitative mapping from physical signals to key engineering parameters such as steel reinforcement corrosion depth and rate, surpassing traditional qualitative or semi-quantitative judgments. The improved Grey Wolf optimization algorithm and automated model hyperparameters significantly enhance model convergence speed and prediction accuracy. A four-level early warning mechanism enables closed-loop management of the entire process, from daily monitoring to emergency response. The entire system possesses self-learning capabilities; with the accumulation of monitoring data, it can continuously optimize the model through online learning, improving adaptability and prediction accuracy across different bridges and environments. The sensing system boasts high durability, a clear methodology, and is easily integrated into existing bridge health monitoring systems, achieving automated and intelligent corrosion management. Compared to existing technologies, this invention can detect steel corrosion earlier and more accurately, reducing early warning response time by approximately 40% and improving prediction accuracy by over 30%. It provides a scientific basis for the safe operation and maintenance and lifespan prediction of bridges, effectively preventing sudden structural damage caused by corrosion, and possesses significant social and economic value. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 A schematic diagram illustrating the usage process of the dynamic monitoring system for bridge steel reinforcement corrosion based on fiber optic grating sensing provided in an embodiment of the present invention;

[0043] Figure 2 A schematic diagram of the improved CNN-LSTM network structure;

[0044] Figure 3 This is a schematic diagram of the LSTM model structure. Detailed Implementation

[0045] 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] The purpose of this invention is to provide a dynamic monitoring system for bridge steel reinforcement corrosion based on fiber optic grating sensing, which realizes millimeter-level quantitative diagnosis of steel reinforcement corrosion, significantly improves monitoring accuracy and timeliness, and provides reliable technical support for the long-term operation and maintenance of bridge structures.

[0047] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0048] This invention provides a dynamic monitoring system for bridge steel reinforcement corrosion based on fiber Bragg grating (FBG) sensing, comprising: a FBG sensor array, a fiber demodulator, a data acquisition and communication unit, and a data processing unit. The FBG sensor array is disposed on the surface of the steel reinforcement at key locations to collect radial expansion pressure, measuring point temperature, and macroscopic axial stress. The FBG sensor array is connected to the fiber demodulator, which is connected to the data acquisition and communication unit. The fiber demodulator is used to process the data collected by the FBG sensor array and send it to the data acquisition and communication unit. The data acquisition and communication unit is connected to the data processing unit, which sends the acquired data to the data processing unit. The data processing unit is used to perform dynamic monitoring of bridge steel reinforcement corrosion based on the acquired data.

[0049] The key areas include the water level fluctuation zone and splash zone of the piers, the area below the drainage outlets on the bridge deck, and the periphery of the expansion joints. Based on cost-effectiveness, it is not necessary to install them on every single pier; only the key areas of the bridge most prone to corrosion need to be selected.

[0050] The fiber optic grating sensor array includes an optical fiber and a radial expansion pressure sensor, a measuring point temperature sensor, and a macroscopic axial stress sensor connected in series on the optical fiber. The optical fiber is connected to the optical fiber demodulator.

[0051] The optical fiber is provided with a flexible protective sleeve to provide physical protection.

[0052] FBG P (Radial expansion pressure sensor) is encapsulated in a flexible material (such as polyurethane), making it highly sensitive to radial expansion pressure caused by steel corrosion;

[0053] FBG T (Temperature sensor at measuring point): Encapsulated in a rigid material (such as a thin steel sheet) and directly attached to the reinforcing bar, it is used to accurately measure the temperature at the measuring point;

[0054] FBG ε (Strain): Encapsulated in a rigid material and directly attached to the reinforcing bar, used to monitor the macroscopic axial strain of the reinforcing bar;

[0055] The fiber optic grating sensor array is firmly bonded to the surface of the reinforcing steel using high-strength epoxy resin adhesive.

[0056] The main technical parameters of the fiber optic demodulator used in this invention include:

[0057] 1. Number of channels: Supports simultaneous connection of multiple trunk optical cables (such as 4 channels, 8 channels or 16 channels) to meet the needs of distributed monitoring;

[0058] 2. Scanning frequency: It has high-frequency scanning capability (e.g., ≥1Hz) and can capture dynamic signals of rust development and load.

[0059] 3. Wavelength accuracy: High precision (e.g., ±1 pm) ensures reliable measurement of minute strain and pressure changes;

[0060] 4. Connection method: Physically connect to the main single-mode optical cable led out from the field junction box via optical fiber flange / interface.

[0061] The data acquisition and communication unit used in this invention can be composed of an industrial-grade computer or an embedded industrial control unit, and can be connected to an optical fiber demodulator via a network port / serial port. It has a built-in 4G / 5G wireless communication module or Ethernet interface for remote data transmission.

[0062] Since the fiber optic demodulator and the data acquisition and communication unit are both relatively mature technologies in the prior art, this invention does not limit them, as long as the functions can be achieved.

[0063] like Figure 1As shown, the core of this invention lies in the data processing unit's dynamic monitoring of bridge steel reinforcement corrosion based on the acquired data. A detailed description follows:

[0064] Step 1: Obtain radial expansion pressure, measuring point temperature, and macroscopic axial stress data, and preprocess them;

[0065] Step 2: Construct a dynamic monitoring model for steel reinforcement corrosion in bridges;

[0066] Step 3: Build the dataset;

[0067] Step 4: Train the dynamic monitoring model for bridge steel reinforcement corrosion based on the dataset to obtain the trained dynamic monitoring model for bridge steel reinforcement corrosion.

[0068] Step 5: Input the radial expansion pressure, measuring point temperature and macroscopic axial stress data to be monitored into the trained bridge steel reinforcement corrosion dynamic monitoring model to obtain corrosion indicator factors.

[0069] Step 6: Calculate the corrosion depth of the reinforcing steel based on the corrosion indicator factor;

[0070] Step 7: Dynamic early warning based on the depth of steel corrosion.

[0071] In step 1, radial expansion pressure, measuring point temperature, and macroscopic axial stress data are acquired and preprocessed, specifically as follows:

[0072] 1. Data cleaning and outlier handling

[0073] Algorithms based on existing technologies can identify and process invalid and abnormal data points caused by sensor momentary failures, signal transmission interference, and other reasons.

[0074] Invalid and abnormal data points are handled as follows:

[0075] (1) Handling invalid values: Directly remove or mark missing signals (such as zero or null values ​​caused by demodulator lockout);

[0076] (2) Physical range filtering: Set reasonable thresholds based on the physical characteristics of the sensor. For example, the wavelength drift of a certain FBG should theoretically not exceed ±5nm. Data that exceeds this range is considered abnormal and is rejected.

[0077] (3) Statistical anomaly detection: For more subtle anomalies, the isolation forest or Z-score method is used for detection. For example, the Z-score of each sensor data stream is calculated, and data points with an absolute value greater than 3 are regarded as extreme outliers and processed.

[0078] (4) Filling: For the few data points that have been removed, linear interpolation or the mean of the data before and after the time step is used to fill them in in order to maintain the continuity of the time series.

[0079] 2. Data alignment and synchronization

[0080] This is used to ensure that data from different FBG sensors on the same rebar are strictly synchronized in terms of timestamps;

[0081] A fiber optic demodulator is used to provide a unified timestamp for all channels. During preprocessing, it is necessary to verify and ensure that the three data streams are aligned at every sampling moment. If a slight misalignment occurs due to data transmission, interpolation and resampling are performed based on the timestamp to align all sensor data on the same time grid.

[0082] 3. Signal smoothing and noise reduction

[0083] It is used to filter out signal jitter caused by high-frequency electronic noise, random environmental vibrations, etc., and retain low-frequency trend signals caused by real physical processes such as corrosion, temperature, and load.

[0084] This invention uses the existing stable wavelet transform for denoising. The stable wavelet transform is an excellent non-stationary signal processing method. It can effectively separate noise components and useful components in a signal and preserve the edges and features of the signal to the maximum extent during reconstruction, which is superior to traditional low-pass filters.

[0085] In addition, if the data noise is not severe, a moving average filter or a Savitzky-Golay filter can be used, the latter of which can better preserve the local extremum characteristics of the signal while smoothing it.

[0086] 4. Data standardization

[0087] This invention uses RobustZ-Score normalization to process data, eliminating dimensional differences between different physical quantities (pressure, temperature, strain) and scaling the data to a range suitable for deep learning model training, thereby accelerating model convergence.

[0088] In step 2, a dynamic monitoring model for steel reinforcement corrosion of the bridge is constructed, specifically as follows:

[0089] A dynamic monitoring model for steel bar corrosion of bridges is constructed based on an improved CNN-LSTM network structure. The input of the model is three-dimensional time-series data, namely radial expansion pressure, measuring point temperature and macroscopic axial stress data; the output is corrosion indicator factor.

[0090] Among them, such as Figure 2As shown, the improved CNN-LSTM network structure consists of an input layer, two CNN modules, two LSTM modules, and a fully connected layer. The input layer receives three-dimensional temporal data, and the fully connected regression layer maps the LSTM output to the final corrosion indicator. The two CNN modules and two LSTM modules are described in detail below:

[0091] Each CNN module contains one convolutional layer, one activation function layer, and one pooling layer. The first convolutional layer contains 32 filters with a kernel size of 3, and the second convolutional layer contains 64 filters with a kernel size of 3. The design of these two convolutional layers can effectively capture relevant features, improve the feature learning ability, and provide high-quality feature input for the subsequent temporal modeling of the LSTM network.

[0092] The LSTM used is a special type of recurrent neural network (RNN) designed to capture long-term dependencies in sequential data. By introducing memory units and gating mechanisms, it effectively solves the gradient vanishing problem of traditional RNNs when processing long sequences. The relevant formulas are as follows:

[0093] (1)

[0094] (2)

[0095] (3)

[0096] (4)

[0097] (5)

[0098] In the formula, x t h t－1 c represents the input and output of the LSTM model at times t and t-1, respectively; t c t－1 These represent the cell states at times t and t-1, respectively; i t f t o t , respectively, are the outputs of the input gate, forget gate, and output gate at time t; σ() is the sigmoid function; ⊙ indicates element-wise multiplication; W and b are the parameters of the model (different subscripts represent different parameters);

[0099] The structure of the LSTM model is as follows Figure 3 As shown, there are two LSTM modules, each with 64 hidden units. The multi-layer LSTM structure can improve the temporal modeling effect and enhance the generalization ability of the model.

[0100] Step 3 involves constructing the dataset, specifically as follows:

[0101] Make multiple steel bars identical to those in step 1 and cast them into concrete test blocks;

[0102] Experiments were conducted on the test blocks based on a controllable accelerated corrosion environment;

[0103] During the experiment, radial expansion pressure, measuring point temperature, and macroscopic axial stress data were collected. At predetermined, unequal intervals of critical time points (e.g., 0 hours, 100 hours, 300 hours, 600 hours of corrosion, etc.), one or more specimens were taken for destructive testing to obtain the actual corrosion depth. Specifically:

[0104] Break open the concrete and directly measure the corrosion depth (unit: mm) on the surface of the reinforcing steel using vernier calipers or a laser scanning microscope.

[0105] After removing the rust products from the surface of the reinforcing bars, the mass loss of the reinforcing bars is measured with a precision balance, and then converted into the average section loss rate or the depth of rust on the reinforcing bars using a formula.

[0106] Next, the continuous FBG data is labeled according to the corrosion depth, specifically including:

[0107] 1. Define the reference point:

[0108] The corrosion depth D0 at the start time of the test (t0, 0 hours) is defined as 0;

[0109] The end time of the experiment (t) n The corrosion depth D (e.g., after 1000 hours) n Defined as 1;

[0110] A sequence was established from [D0, D n A linear mapping from [0, 1] to [0, 1].

[0111] 2. Linear interpolation assigns true values:

[0112] For the two destructive detection time points t i and t j For any FBG sampling time t between these points, assume that the corrosion depth increases linearly;

[0113] Calculate the true value Y of the corrosion indicator factor at time t using a linear interpolation formula. true (t) is:

[0114] Y true (t)=(D(t)-D0) / (D n -D0), where D(t) = D i +(tt) i ) (D)j -D i ) / (t j -t i );

[0115] The entire corrosion process is normalized to a scale of 0 to 1, where 0 represents brand new and 1 represents severe corrosion at the end of the test. Each moment in between is assigned a value between 0 and 1 based on its progress of corrosion from the start and end points.

[0116] The final dataset is obtained.

[0117] In step 4, the dynamic monitoring model for steel reinforcement corrosion of bridges is trained based on the dataset, specifically as follows:

[0118] The dataset is divided into three parts: 70% as the training set, 15% as the validation set, and 15% as the test set.

[0119] Based on the training and validation sets, the IGWO algorithm was used to train the dynamic monitoring model for steel corrosion of bridges to obtain the optimal hyperparameters;

[0120] The dynamic monitoring model for bridge steel corrosion was re-initialized based on the optimal hyperparameters. The training set and validation set were merged, and the initialized dynamic monitoring model for bridge steel corrosion was finally trained based on the merged dataset.

[0121] The final trained dynamic monitoring model for bridge steel corrosion was evaluated based on the test set.

[0122] Based on the training and validation sets, the IGWO algorithm was used to train the dynamic monitoring model for steel corrosion of bridges to obtain the optimal hyperparameters, specifically:

[0123] 1. Define the search space: Determine the hyperparameters to be optimized and their range, such as: learning rate, number of LSTM units, number of CNN filters, dropout rate, and batch size.

[0124] 2. IGWO Optimization:

[0125] Encode a set of hyperparameters to the position of a "gray wolf";

[0126] Perform fitness calculations for each wolf (each combination of hyperparameters):

[0127] (1) Instantiate a model: Create a new, untrained improved CNN-LSTM model with the current hyperparameter combination;

[0128] (2) Short training: The model is trained on the training set in a small number of rounds;

[0129] (3) Evaluation: The RMSE of the model is calculated on the validation set as a performance metric, where fitness = -RMSE;

[0130] The wolf pack updates its position based on fitness (α, β, δ wolves) and searches for the next generation;

[0131] 3. Output: After the optimization is completed, output the globally optimal combination of hyperparameters.

[0132] The IGWO algorithm needs to be introduced here, specifically:

[0133] The Grey Wolf Optimization (GWO) algorithm arises from the cooperative and competitive relationships among individuals within a grey wolf pack. It utilizes the actions of grey wolves to search for the optimal solution to a problem. The GWO algorithm process encompasses multiple stages, including social hierarchy classification, hunting prey, attacking prey, and searching for prey. The IGWO algorithm is an improvement upon the GWO algorithm; therefore, the GWO algorithm will not be described in detail here. To address the problems of the GWO algorithm, such as lack of population diversity, imbalance between development and exploration, and premature convergence, a dimension learning-based hunting (DLH) search strategy is introduced on the basis of the GWO algorithm. This enhances the balance between local and global search, allowing individual wolf hunting to be learned by their neighbors, thereby maintaining diversity.

[0134] In step 6, the corrosion depth of the reinforcing steel is calculated based on the corrosion indicator factor, specifically as follows:

[0135] Calculating the corrosion depth of steel bars using corrosion indicator factors does not require complex physical formulas with precise parameters. Instead, it employs a hybrid approach that is "data-driven and based on physical mechanisms." Gaussian process regression is a probabilistic nonparametric model that can provide not only predicted values ​​but also the range of uncertainty (confidence interval) of the predictions, which is crucial for risk assessment.

[0136] First, data preparation is required. Similar to the dataset mentioned above, the data here comes from the laboratory. The difference is that the data here is a mapping pair of corrosion indicator factor and steel corrosion depth.

[0137] Next, Gaussian process regression (GPR) is used to learn the complex nonlinear relationship between corrosion indicator factors and steel corrosion depth. To ensure that the model predictions conform to physical laws (e.g., corrosion depth is not negative and is usually monotonically increasing), prior knowledge is introduced into the selection of the GPR kernel function, such as using a kernel function that guarantees non-negative output or adding monotonicity constraints to the loss function. This step is crucial, as it ensures that the data-driven model learns within the framework of physical laws and improves the reliability of extrapolation.

[0138] The GPR model was trained using a laboratory dataset;

[0139] The corrosion indicator factor to be monitored is input into the trained GPR model to obtain the corrosion depth of the steel bars.

[0140] In step 7, dynamic early warning is provided based on the depth of steel reinforcement corrosion, specifically as follows:

[0141] A four-level early warning system is established. If the depth of steel bar corrosion is less than 0.1 mm, it is judged as a level one early warning. The data processing unit records the depth of steel bar corrosion without performing any other processing.

[0142] If the corrosion depth of the steel bar is less than 0.3 mm and 0.1 mm, it is judged as a level two warning. The data processing unit sends it to the monitoring platform for staff to view and shortens the monitoring interval of this steel bar to increase the monitoring frequency.

[0143] If the steel bar corrosion depth is less than 0.5mm and less than 0.3mm, it is judged as a level three warning. The data processing unit sends it to the monitoring platform, the monitoring platform generates an operation and maintenance report, and sends it to the maintenance management department for maintenance.

[0144] If the depth of steel bar corrosion is ≥0.5mm, it is judged as a level four warning. The data processing unit sends it to the monitoring platform, and the monitoring platform notifies the relevant departments to take traffic control measures and immediately carry out reinforcement and repair.

[0145] To verify the effectiveness and advancement of this invention, we took the monitoring of the piers of a large coastal cross-sea bridge as an implementation case. The bridge has been in operation for 8 years and is located in a marine corrosive environment. The piers are severely corroded by chloride ions and are high-risk areas for corrosion. We selected a total of 20 steel bars in the water level fluctuation area and splash zone of the four key piers (numbered P3-P6) of the bridge, with 5 bars for each pier, located at different heights, to comprehensively monitor the development of corrosion.

[0146] 1. System Deployment and Data Acquisition

[0147] The fiber optic grating sensor array was mounted on the steel reinforcement and led to the monitoring station on the top of the bridge pier via armored optical cable. The monitoring system operated continuously for 12 months with a sampling frequency of 2Hz, and obtained approximately 63 million valid data records. The fiber optic demodulator was a 16-channel model with a wavelength accuracy of ±1pm, which fully met the monitoring requirements.

[0148] 2. Data Processing and Model Training

[0149] First, the collected raw data was preprocessed, including data cleaning, outlier handling, signal alignment, stable wavelet transform denoising, and Robust Z-Score normalization. During the 12-month monitoring period, accelerated corrosion tests were carried out simultaneously in the laboratory. Thirty identical specimens were prepared and destructive tests were performed at five key time points (0, 400, 800, 1200, and 1600 hours) to obtain actual corrosion depth data and establish a true value dataset of corrosion indicator factors.

[0150] The training process of the improved CNN-LSTM model uses the IGWO algorithm for hyperparameter optimization. The optimized parameters include learning rate, number of LSTM hidden units, number of CNN filters, dropout rate and batch size. After 200 iterations, the IGWO algorithm converges and finds the optimal combination of hyperparameters. The final model performs well on the test set, and all indicators are significantly better than the comparison model.

[0151] 3. Performance Comparison Analysis

[0152] To objectively evaluate the technical effect of the present invention, the method of the present invention was compared with traditional LSTM, GRU, unoptimized CNN-LSTM and the half-cell potential method commonly used in engineering. The comparison results are shown in Tables 1, 2 and 3.

[0153] The comparison results show that the method of this invention is significantly superior to traditional methods in terms of prediction accuracy, early warning timeliness, and reliability, specifically in the following aspects:

[0154] 1. Improved prediction accuracy: Compared to traditional LSTM, RMSE decreased from 0.048mm to 0.019mm, representing an accuracy improvement of approximately 60%; compared to unoptimized CNN-LSTM, the accuracy improvement is approximately 38%.

[0155] 2. Early warning response: The method of this invention can effectively issue an early warning when the rust depth reaches 0.25mm, which is about 30% earlier than the traditional method, allowing sufficient time for maintenance decisions;

[0156] 3. Significantly reduced false alarm rate: Through IGWO optimization and multi-source signal decoupling, the false alarm rate is reduced from 23.5% in the traditional method to 4.3%, greatly reducing unnecessary maintenance costs.

[0157] In practical application, the system of this invention successfully detected abnormal corrosion development in the No. 3 rebar of pier P4 in the 9th month. The system triggered a level-three warning on the 12th day of the 9th month, indicating a corrosion depth of 0.32mm. On-site verification confirmed the presence of microcracks in the concrete cover at that location, highly consistent with the predicted results. The maintenance department promptly implemented surface sealing and anti-corrosion treatments, effectively controlling corrosion development and preventing potential structural damage. In contrast, the same area monitored using the half-cell potential method did not show an anomaly until the 11th month, at which point the corrosion depth had already reached 0.45mm, significantly increasing the cost and difficulty of treatment.

[0158] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0159] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A dynamic monitoring system for steel reinforcement corrosion in bridges based on fiber optic grating sensing, characterized in that, include: The system comprises a fiber Bragg grating sensor array, a fiber optic demodulator, a data acquisition and communication unit, and a data processing unit. The fiber Bragg grating sensor array is installed on the surface of the reinforcing steel at key locations to collect radial expansion pressure, measuring point temperature, and macroscopic axial stress. The fiber Bragg grating sensor array is connected to the fiber optic demodulator, which is connected to the data acquisition and communication unit. The fiber optic demodulator is used to process the data collected by the fiber Bragg grating sensor array and send it to the data acquisition and communication unit. The data acquisition and communication unit is connected to the data processing unit, which is used to send the acquired data to the data processing unit. The data processing unit is used to perform dynamic monitoring of bridge steel reinforcement corrosion based on the acquired data. The key areas include the water level fluctuation zone and splash zone of the bridge pier, the area below the drainage outlet of the bridge deck, and the periphery of the expansion joint. The data processing unit is used to perform dynamic monitoring of bridge steel reinforcement corrosion based on the acquired data, specifically: Step 1: Obtain radial expansion pressure, measuring point temperature, and macroscopic axial stress data, and preprocess them; Step 2: Construct a dynamic monitoring model for steel reinforcement corrosion in bridges; specifically: A dynamic monitoring model for steel bar corrosion of bridges is constructed based on an improved CNN-LSTM network structure. The specific structure of the improved CNN-LSTM network structure includes an input layer, two CNN modules, two LSTM modules, and a fully connected layer. Each of the two CNN modules includes a convolutional layer and a pooling layer. The convolutional layer of the first CNN module includes 32 filters and a kernel size of 3, while the convolutional layer of the second CNN module includes 64 filters and a kernel size of 3. Each of the LSTM modules has 64 hidden units; Step 3: Build the dataset; Step 4: Train the dynamic monitoring model for bridge steel reinforcement corrosion based on the dataset to obtain the trained dynamic monitoring model for bridge steel reinforcement corrosion. Step 5: Input the radial expansion pressure, measuring point temperature and macroscopic axial stress data to be monitored into the trained bridge steel reinforcement corrosion dynamic monitoring model to obtain corrosion indicator factors. Step 6: Calculate the corrosion depth of the reinforcing steel based on the corrosion indicator factor; Step 7: Dynamic early warning based on the depth of steel corrosion.

2. The bridge steel reinforcement corrosion dynamic monitoring system based on fiber optic grating sensing according to claim 1, characterized in that, The fiber optic grating sensor array includes an optical fiber and a radial expansion pressure sensor, a measuring point temperature sensor, and a macroscopic axial stress sensor connected in series on the optical fiber. The optical fiber is connected to the optical fiber demodulator.

3. The bridge steel reinforcement corrosion dynamic monitoring system based on fiber optic grating sensing according to claim 2, characterized in that, Step 3 involves constructing the dataset, specifically as follows: Make multiple steel bars identical to those in step 1 and cast them into concrete test blocks; Experiments were conducted on the test blocks based on a controllable accelerated corrosion environment; During the experiment, radial expansion pressure, measuring point temperature, macroscopic axial stress data, and actual corrosion depth were collected. The radial expansion pressure, measuring point temperature, and macroscopic axial stress data for each time period are aligned with the actual corrosion depth measured at the corresponding time. Each segment of radial expansion pressure, measuring point temperature, and macroscopic axial stress data is assigned a corrosion indicator factor as a label. The dataset is constructed based on the aligned data.

4. The bridge steel reinforcement corrosion dynamic monitoring system based on fiber optic grating sensing according to claim 3, characterized in that, Step 4: Train the dynamic monitoring model for bridge steel reinforcement corrosion based on the dataset, specifically as follows: The dataset is divided into three parts: 70% as the training set, 15% as the validation set, and 15% as the test set. Based on the training and validation sets, the IGWO algorithm was used to train the dynamic monitoring model for steel corrosion of bridges to obtain the optimal hyperparameters; The dynamic monitoring model for bridge steel corrosion was re-initialized based on the optimal hyperparameters. The training set and validation set were merged, and the initialized dynamic monitoring model for bridge steel corrosion was finally trained based on the merged dataset. The final trained dynamic monitoring model for bridge steel corrosion was evaluated based on the test set.

5. The bridge steel reinforcement corrosion dynamic monitoring system based on fiber optic grating sensing according to claim 4, characterized in that, In step 6, the corrosion depth of the reinforcing steel is calculated based on the corrosion indicator factor, specifically as follows: Construct a physical-guided Gaussian process regression model; It is trained based on a pre-defined rust indicator factor - steel bar rust depth dataset; The corrosion indicator factor to be detected is input into the trained Gaussian process regression model to obtain the corrosion depth of the steel bars.

6. The bridge steel reinforcement corrosion dynamic monitoring system based on fiber optic grating sensing according to claim 5, characterized in that, In step 7, dynamic early warning is provided based on the depth of steel reinforcement corrosion, specifically as follows: A four-level early warning system is established. If the depth of steel bar corrosion is less than 0.1 mm, it is judged as a level one early warning. The data processing unit records the depth of steel bar corrosion without performing any other processing. If the corrosion depth of the steel bar is less than 0.3 mm and 0.1 mm, it is judged as a level two warning. The data processing unit sends it to the monitoring platform for staff to view and shortens the monitoring interval of this steel bar to increase the monitoring frequency. If the steel bar corrosion depth is less than 0.5mm and less than 0.3mm, it is judged as a level three warning. The data processing unit sends it to the monitoring platform, the monitoring platform generates an operation and maintenance report, and sends it to the maintenance management department for maintenance. If the depth of steel bar corrosion is ≥0.5mm, it is judged as a level four warning. The data processing unit sends it to the monitoring platform, and the monitoring platform notifies the relevant departments to take traffic control measures and immediately carry out reinforcement and repair.

Images