Reinforced Concrete Life Prediction System and Method Based on Material Degradation Simulation

Abstract

The present invention relates to the field of durability assessment and life prediction of reinforced concrete structures, and particularly to a life prediction system and method for reinforced concrete based on material degradation simulation, including an erosion front tracking module that extracts chloride ion distribution image features using a three-dimensional convolutional neural network integrating topological invariance to generate three-dimensional coordinates and depth information of the steel bar corrosion front; a transfer learning parameter inversion module that inputs the three-dimensional coordinates of the steel bar corrosion front into a finite element simulation platform to perform parameter inversion calculations and generate a corrosion rate evolution curve; a corrosion rate prediction module that uses a long short-term memory network to generate a corrosion rate prediction curve; and an online correction module that collects internal temperature and humidity information of the concrete structure, calculates the influence weight of environmental factors on the corrosion rate, corrects the corrosion rate prediction curve, and updates the parameters of the long short-term memory network to accurately identify and track the steel bar corrosion front and accurately capture complex three-dimensional corrosion patterns.

Description

Technical Field

[0001] The present invention relates to the field of durability assessment and life prediction of reinforced concrete structures, and particularly to a reinforced concrete life prediction system and method based on material degradation simulation, which is especially applicable to the life prediction and health monitoring of concrete structures in areas with severe chloride ion erosion such as marine environments and saline-alkali regions. Background Art

[0002] During the long-term service process of reinforced concrete structures, due to the erosion of environmental factors, especially chloride ions, material degradation gradually occurs, resulting in a decline in structural performance and ultimately possibly causing structural failure. Therefore, accurately assessing and predicting the service life of reinforced concrete structures is of great significance for ensuring structural safety.

[0003] Currently, the following methods are mainly used for the life prediction of reinforced concrete: traditional empirical formula method, numerical simulation method based on finite element analysis, and statistical analysis method based on monitoring data. The traditional empirical formula method relies on simplified models and cannot accurately reflect the degradation process of concrete in complex environments; the numerical simulation method based on finite element analysis is computationally complex and time-consuming, making it difficult to conduct long-term predictions; the statistical analysis method based on monitoring data overly relies on historical data and has insufficient prediction accuracy for newly built structures.

[0004] The main problems faced by the existing technologies include: first, the accuracy of corrosion front identification is insufficient, making it difficult to accurately capture complex three-dimensional corrosion patterns; second, it is difficult to accurately invert material parameters, resulting in unreasonable parameter settings for prediction models; third, the long-term prediction ability is limited, and it is unable to effectively process non-linear time series data; fourth, there is a lack of an effective online correction mechanism and it cannot adjust prediction results according to real-time environmental changes.

[0005] Therefore, there is an urgent need to develop a reinforced concrete life prediction system and method that can achieve high-precision corrosion front identification, accurate parameter inversion, long-term prediction, and real-time correction. Summary of the Invention

[0006] The purpose of the present invention is to provide a reinforced concrete life prediction system and method based on material degradation simulation, aiming to solve the problems existing in the prior art such as low accuracy of corrosion front identification, inaccurate parameter inversion, weak long-term prediction ability, and lack of an online correction mechanism.

[0007] The present invention proposes a reinforced concrete life prediction system based on material degradation simulation, including:

[0008] An erosion front tracking module, configured to collect chloride ion distribution images of a reinforced concrete specimen, extract feature vectors of the chloride ion distribution images through a three-dimensional convolutional neural network that fuses topological invariance, and generate three-dimensional coordinates and depth information of the steel corrosion front;

[0009] The transfer learning parameter inversion module, which is communicatively connected to the erosion front tracking module, is configured to receive the three-dimensional coordinates and depth information of the steel bar corrosion front, input the three-dimensional coordinates of the steel bar corrosion front into the finite element simulation platform, perform parameter inversion calculations, and generate a corrosion rate evolution curve;

[0010] The corrosion rate prediction module, which is communicatively connected to the transfer learning parameter inversion module, is configured to receive the corrosion rate evolution curve, use the corrosion rate evolution curve as a training set to input into the long short-term memory network, and generate a corrosion rate prediction curve;

[0011] The system online correction module, which is communicatively connected to the corrosion rate prediction module, is configured to collect the temperature and humidity information inside the concrete structure, calculate the influence weight of environmental factors on the corrosion rate, correct the corrosion rate prediction curve, and update the long short-term memory network parameters.

[0012] Preferably, the erosion front tracking module includes:

[0013] The image preprocessing unit, which is configured to receive the chloride ion concentration distribution picture and unify the chloride ion concentration distribution picture into a fixed size and format;

[0014] The topological feature extraction unit, which is configured to apply topological analysis methods to the preprocessed chloride ion concentration distribution picture to extract the connectivity and geometric invariant features of the corrosion front;

[0015] The three-dimensional convolutional network unit, which is configured to perform feature extraction on the chloride ion concentration distribution picture through a designed three-dimensional convolution kernel structure to generate a feature vector;

[0016] The front positioning unit, which is configured to determine the coordinates and depth information of the corrosion front on the three-dimensional cross-section according to the feature vector, and generate a complete three-dimensional corrosion front surface model.

[0017] Preferably, the three-dimensional convolutional network unit includes:

[0018] The convolutional feature extraction layer, which is configured to perform feature extraction on the input image through multiple layers of three-dimensional convolutional kernels;

[0019] The attention mechanism layer, which is configured to weight the extracted features to highlight the feature information in the corrosion front area;

[0020] The differential geometry feature layer, which is configured to calculate the Gaussian curvature and mean curvature in the image to capture the geometric characteristics of the corrosion front;

[0021] The topological feature fusion layer, which is configured to fuse the topological features and convolutional features to generate a comprehensive feature vector.

[0022] Preferably, the transfer learning parameter inversion module includes:

[0023] A finite element modeling unit for establishing a reinforced concrete corrosion model based on the three-dimensional coordinates of the steel bar corrosion front;

[0024] A parameter optimization unit for adjusting the material model parameters of steel bars and concrete and optimizing the objective function;

[0025] A transfer learning unit for using the differences between experimental data and simulation data for model parameter optimization to achieve knowledge transfer;

[0026] A corrosion curve generation unit for generating a corrosion rate evolution curve based on the optimized material parameters.

[0027] Preferably, the parameter optimization unit adopts the Bayesian optimization algorithm, where the optimized material parameters include: the concrete density ρ, the concrete elastic modulus E, and the chloride ion diffusion coefficient D, and the optimized objective function is:

[0028] ,

[0029] where, , is the corrosion rate simulated under given parameters, is the time variable.

[0030] Preferably, the corrosion rate prediction module includes:

[0031] A data preprocessing unit for normalizing the corrosion rate evolution curve and dividing it into a training set and a validation set;

[0032] An LSTM network construction unit for setting the structural parameters of the long short-term memory network, including the number of hidden layer nodes, the network depth, and the learning rate;

[0033] A model training unit for training the long short-term memory network based on the Adam algorithm to generate a prediction model; [[ID=4】

[0034] A prediction result output unit for generating a long-term corrosion rate prediction curve using the trained model.

[0035] Preferably, the system online correction module includes:

[0036] An environment monitoring unit for real-time collecting the temperature and humidity data inside the concrete structure through an embedded sensor;

[0037] A weight calculation unit for calculating the influence weight of environmental factors on the corrosion rate according to the temperature and humidity data;

[0038] A model correction unit for correcting the predicted corrosion rate value according to the influence weight;

[0039] A parameter update unit for updating the parameters of the long short-term memory network according to the corrected result.

[0040] Preferably, the weight calculation unit calculates the influence weight of environmental factors based on the following formula:

[0041] ,

[0042] where is the influence weight, is the temperature, is the humidity, is the proportionality coefficient, and are the weight exponents, calibrated by measured data.

[0043] Preferably, the system further includes a multi-scale fusion module, communicatively connected to the erosion front tracking module and the transfer learning parameter inversion module, for correlating the microscopic corrosion mechanism with the macroscopic deterioration performance to achieve unified expression and processing of multi-scale information.

[0044] A method for predicting the service life of reinforced concrete based on material degradation simulation, comprising the following steps:

[0045] Collect the chloride ion distribution image of the reinforced concrete specimen, preprocess the chloride ion distribution image, input the preprocessed chloride ion distribution image into a three-dimensional convolutional neural network that fuses topological invariance, extract the feature vector, determine the coordinates and depth of the corrosion front on the three-dimensional cross-section, and generate the three-dimensional surface of the steel bar corrosion front;

[0046] Input the three-dimensional surface of the steel bar corrosion front into the finite element simulation platform, adjust the material parameters of the steel bar and the concrete, perform parameter inversion calculation, and obtain the corrosion rate evolution curve;

[0047] Use the corrosion rate evolution curve as the training set to input into the long short-term memory network, set the network parameters, train the network model based on the stochastic gradient descent algorithm, and generate the corrosion rate prediction curve;

[0048] Collect the temperature and humidity information inside the concrete structure, calculate the influence weight of environmental factors on the corrosion rate, correct the corrosion rate prediction curve based on the influence weight, and update the parameters of the long short-term memory network.

[0049] The beneficial effects of the present invention include:

[0050] 1. The high-precision identification and tracking of the steel bar corrosion front are achieved through a 3D convolutional neural network that integrates topological invariance, which can accurately capture complex 3D corrosion morphologies and improve the accuracy of corrosion state assessment.

[0051] 2. By adopting the transfer learning parameter inversion technology, the effective conversion from experimental data to model parameters is realized, the inversion accuracy of material parameters is improved, and reliable basic data for subsequent prediction are provided.

[0052] 3. By introducing a long short-term memory network to process long-period non-linear time series data, the long-term prediction ability of the corrosion rate is significantly improved, and the prediction accuracy is increased by about 30% compared with the existing technology.

[0053] 4. An online correction mechanism based on real-time temperature and humidity monitoring is established, which can dynamically adjust the prediction model parameters to make the prediction results more conform to the actual environmental changes.

[0054] 5. Through the collaborative work of multiple modules, a complete closed-loop system from corrosion front identification, parameter inversion, rate prediction to online correction is formed, realizing the full-process intelligence of the service life prediction of reinforced concrete structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 is the overall architecture diagram of the reinforced concrete service life prediction system based on material degradation simulation of the present invention;

[0056] Figure 2 is the structural schematic diagram of the erosion front tracking module of the present invention;

[0057] Figure 3 is the structural schematic diagram of the 3D convolutional neural network of the present invention;

[0058] Figure 4 is the workflow diagram of the transfer learning parameter inversion module of the present invention;

[0059] Figure 5 is the structural schematic diagram of the corrosion rate prediction module of the present invention;

[0060] Figure 6 is the working principle diagram of the system online correction module of the present invention;

[0061] Figure 7 is the flowchart of the reinforced concrete service life prediction method based on material degradation simulation of the present invention;

[0062] Figure 8 is the schematic diagram of the temperature and humidity weight calculation method of the present invention;

[0063] Figure 9 is the structural schematic diagram of the multi-scale fusion module of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0064] Please refer to the attached Figures 1-9 , and hereinafter, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Those of ordinary skill in the art should understand that the embodiments described herein are only a part of the embodiments of the present invention and do not limit the scope of the rights of the present invention.

[0065] As Figure 1 shown, the reinforced concrete life prediction system based on material degradation simulation provided by the present invention includes: an erosion front tracking module 1, a transfer learning parameter inversion module 2, a corrosion rate prediction module 3, and a system online correction module 4. Preferably, the system may further include a multi-scale fusion module 5.

[0066] As Figure 2 shown, the erosion front tracking module 1 is used to collect the chloride ion distribution image of the reinforced concrete specimen, extract the feature vector of the chloride ion distribution image through a three-dimensional convolutional neural network integrating topological invariance, and generate the three-dimensional coordinates and depth information of the steel bar corrosion front.

[0067] In an embodiment of the present invention, the erosion front tracking module 1 includes: an image preprocessing unit 11, a topological feature extraction unit 12, a three-dimensional convolutional network unit 13, and a front positioning unit 14.

[0068] The image preprocessing unit 11 is used to receive the chloride ion concentration distribution picture and unify the chloride ion concentration distribution picture into a fixed size and format. Preferably, the present invention uses a high-resolution scanning technology to obtain the cross-sectional image of the reinforced concrete specimen, and the image resolution is preferably 1024×1024 pixels to ensure capturing sufficient detailed information. The image preprocessing includes operations such as size normalization, denoising, and contrast enhancement. The size of the standardized image is preferably 224×224 pixels, which meets the input requirements of the deep learning model.

[0069] The topological feature extraction unit 12 is used to apply a topological analysis method to the preprocessed chloride ion concentration distribution picture to extract the connectivity and geometric invariant features of the corrosion front. The present invention innovatively introduces the topological invariance theory and characterizes the topological features of the corrosion front by calculating the persistent homology group. The mathematical model for topological feature extraction is as follows:

[0070] ,

[0071] where represents the k-dimensional homology group, represents the k-dimensional cycle group, Denotes the k - dimensional boundary group. By calculating the homology groups of different dimensions, topological invariants such as the connectivity of the corrosion front and the number of holes can be captured. These features remain stable under image deformation and noise interference, greatly improving the robustness of corrosion front recognition.

[0072] The three - dimensional convolutional network unit 13 is used to extract features from the chloride ion concentration distribution picture through the designed three - dimensional convolutional kernel structure and generate feature vectors. As Figure 3 shown, the three - dimensional convolutional network unit 13 includes a convolutional feature extraction layer 131, an attention mechanism layer 132, a differential geometry feature layer 133, and a topological feature fusion layer 134.

[0073] The convolutional feature extraction layer 131 adopts a multi - layer three - dimensional convolutional structure. The number of convolutional kernels in each layer increases from 64 to 256. The size of the convolutional kernel is 3×3×3, the stride is 1, and the padding is 1. After each convolutional layer, a batch normalization layer and a ReLU activation function are connected to ensure the effective transmission of signals. Preferably, the convolutional layer adopts a residual connection structure to effectively alleviate the gradient vanishing problem in the deep network.

[0074] The attention mechanism layer 132 adopts a spatial attention mechanism to weight the feature map by learning weight coefficients and highlight the feature information in the corrosion front region. The attention weight calculation formula is as follows:

[0075] ,

[0076] where, denotes the attention weight at the i - th position, denotes the feature mapping function, and n denotes the size of the feature map. Through the attention mechanism, the model can adaptively focus on the regions related to the corrosion front in the image and filter out irrelevant information.

[0077] The differential geometry feature layer 133 is used to calculate the Gaussian curvature and mean curvature in the image and capture the geometric characteristics of the corrosion front. The calculation formulas for the Gaussian curvature K and the mean curvature H are as follows:

[0078] ,

[0079] ,

[0080] where, E, F, G are the coefficients of the first fundamental form, and L, M, N are the coefficients of the second fundamental form. These differential geometry features can effectively characterize the local surface characteristics of the corrosion front and provide important information for accurate positioning.

[0081] The topological feature fusion layer 134 is used to fuse the topological features with the convolutional features to generate a comprehensive feature vector. The fusion adopts an attention - weighted mechanism, and the fusion formula is as follows:

[0082] ,

[0083] Among them, represents the fused feature vector, represents the convolutional feature, represents the topological feature, represents the geometric feature, 、 and are weight coefficients, obtained through network learning. Preferably, the initial weights can be set to = 0.5, = 0.3, = 0.2, and are automatically adjusted through subsequent network training.

[0084] The front positioning unit 14 is used to determine the coordinates and depth information of the corrosion front on the three-dimensional section according to the feature vector, and generate a complete three-dimensional corrosion front surface model. The positioning adopts the evolution equation based on geodesic flow to realize the representation from discrete points to continuous surfaces:

[0085] ,

[0086] Among them, C represents the corrosion front curve, t represents the evolution time, κ represents the curvature, represents the unit normal vector, and φ represents the potential field of the feature vector. By solving the above partial differential equation, a stable corrosion front curve can be obtained, and then a complete three-dimensional corrosion front surface can be constructed.

[0087] As Figure 4 shown, the transfer learning parameter inversion module 2 is communicatively connected to the erosion front tracking module 1, and is used to receive the three-dimensional coordinates and depth information of the steel bar corrosion front, input the three-dimensional coordinates of the steel bar corrosion front into the finite element simulation platform, perform parameter inversion calculation, and generate a corrosion rate evolution curve.

[0088] In an embodiment of the present invention, the transfer learning parameter inversion module 2 includes: a finite element modeling unit 21, a parameter optimization unit 22, a transfer learning unit 23, and a corrosion curve generation unit 24.

[0089] The finite element modeling unit 21 is used to establish a reinforced concrete corrosion model according to the three-dimensional coordinates of the steel bar corrosion front. Preferably, in the present invention, the ABAQUS finite element software is used to establish the model. The steel bar is modeled by solid elements, and the concrete is modeled by solid elements. The interface characteristics between the two are simulated by contact elements. The mesh size is preferably 2mm to balance the calculation accuracy and efficiency.

[0090] The parameter optimization unit 22 is used to adjust the material model parameters of steel bars and concrete and optimize the objective function. The parameter optimization unit 22 adopts the Bayesian optimization algorithm, and the optimized material parameters include: concrete density ρ, concrete elastic modulus E, and chloride ion diffusion coefficient D. The optimized objective function is:

[0091] ,

[0092] where, represents the measured corrosion rate at time point , represents the corrosion rate obtained by simulation under given parameters, represents the time variable, represents the number of sampling points. The Bayesian optimization process includes constructing a Gaussian process surrogate model, calculating the acquisition function (such as the expected improvement EI), selecting the next set of parameters to be evaluated, and iterating until convergence. Preferably, the initial range of the concrete density ρ is 2200 - 2500 kg / m³, the initial range of the elastic modulus E is 25 - 35 GPa, and the initial range of the chloride ion diffusion coefficient D is 1×10⁻¹² - 1×10⁻¹ 0 m² / s.

[0093] The transfer learning unit 23 is used to utilize the differences between experimental data and simulation data for model parameter optimization to achieve knowledge transfer. Transfer learning uses the domain adaptation method to address the distribution difference problem between the source domain (laboratory data) and the target domain (actual engineering). Preferably, the maximum mean discrepancy (MMD) is used as the inter-domain distance metric:

[0094] MMD( ) = ,

[0095] where, and represent the data distributions of the source domain and the target domain respectively, and represent the means of the two distributions in the reproducing kernel Hilbert space respectively. By minimizing the MMD distance, the feature alignment between the two domains is achieved, and the generalization ability of the model in the target domain is improved.

[0096] The corrosion curve generation unit 24 is used to generate the corrosion rate evolution curve according to the optimized material parameters. The curve generation is based on the finite element simulation results and combined with the calibration of experimental data to obtain the corrosion rate curve varying with time. Preferably, the corrosion rate calculation is based on Fick's second law and the theory of electrochemical corrosion, considering the combined effects of chloride ion concentration gradient and potential difference:

[0097] ,

[0098] ,

[0099] Among them, C represents the chloride ion concentration, D represents the diffusion coefficient, represents the corrosion rate, represents the chloride ion concentration on the steel bar surface, represents the potential difference, and are proportionality coefficients.

[0100] As Figure 5 shown, the corrosion rate prediction module 3 is communicatively connected to the transfer learning parameter inversion module 2, and is configured to receive the corrosion rate evolution curve, and use the corrosion rate evolution curve as a training set to input into the long short-term memory network to generate a corrosion rate prediction curve.

[0101] In an embodiment of the present invention, the corrosion rate prediction module 3 includes: a data preprocessing unit 31, an LSTM network construction unit 32, a model training unit 33, and a prediction result output unit 34.

[0102] The data preprocessing unit 31 is configured to perform normalization processing on the corrosion rate evolution curve and divide it into a training set and a validation set. The normalization adopts the Z-score method:

[0103] ,

[0104] Among them, represents the normalized value, represents the original value, represents the mean value, represents the standard deviation. The training set and the validation set are divided at a ratio of 8:2 to ensure the effectiveness and generalization ability of model training.

[0105] The LSTM network construction unit 32 is configured to set the structural parameters of the long short-term memory network, including the number of hidden layer nodes, the network depth, and the learning rate. Preferably, the present invention adopts a bidirectional LSTM network structure, which can consider both past and future information at the same time. The number of hidden layer nodes is set to 128, the network depth is 3 layers, the initial value of the learning rate is set to 0.001, and a learning rate decay strategy is adopted, and it decays to 0.9 times the original every 50 epochs.

[0106] The model training unit 33 is configured to train the long short-term memory network based on the Adam algorithm to generate a prediction model. Preferably, the present invention adopts an Adam optimizer and combines an early stopping strategy to prevent overfitting. The loss function adopts the mean square error (MSE):

[0107] ,

[0108] Among them, represents the true value, represents the predicted value, represents the number of samples. During the training process, the model performance is evaluated on the validation set every 5 epochs. If the performance of the validation set does not improve for 10 consecutive times, the training is stopped.

[0109] The prediction result output unit 34 is used to generate a long-term corrosion rate prediction curve by using the trained model. The prediction adopts a sliding window strategy, the window size is 24 time steps, the corrosion rate of the next time step is predicted, and then the window is slid to perform iterative prediction. Preferably, in order to improve the prediction accuracy, an ensemble learning method is adopted to perform weighted averaging on the prediction results of 5 independently trained LSTM models, and the weights are determined by the performance of the validation set.

[0110] As Figure 6 shown, the system online correction module 4 is communicatively connected to the corrosion rate prediction module 3, and is used to collect the temperature and humidity information inside the concrete structure, calculate the influence weight of environmental factors on the corrosion rate, correct the corrosion rate prediction curve, and update the long short-term memory network parameters.

[0111] In an embodiment of the present invention, the system online correction module 4 includes: an environmental monitoring unit 41, a weight calculation unit 42, a model correction unit 43, and a parameter update unit 44.

[0112] The environmental monitoring unit 41 is used to collect the temperature and humidity data inside the concrete structure in real time through an embedded sensor. Preferably, a distributed sensing network is adopted, and a node is arranged every 1 meter. Each node includes a temperature sensor and a humidity sensor. The temperature sensor uses a thermocouple, the measurement range is -40°C to 80°C, and the accuracy is ±0.5°C; the humidity sensor uses a capacitive sensor, the measurement range is 0% to 100%RH, and the accuracy is ±3%RH. The data collection frequency is once per hour, and the data is transmitted to the data processing center through a wireless network.

[0113] The weight calculation unit 42 is used to calculate the influence weight of environmental factors on the corrosion rate according to the temperature and humidity data. The influence weight calculation formula is as follows:

[0114] ,

[0115] Among them, represents the influence weight, represents the temperature (°C), represents the humidity (%RH), represents the proportionality coefficient, and Represents the weight index. Through experimental calibration, generally takes a value of 0.01, takes a value of 0.8, takes a value of 1.2. This exponential relationship can accurately reflect the non-linear influence of temperature and humidity on the corrosion rate. For example, in a high-temperature and high-humidity environment, the corrosion rate increases significantly.

[0116] The model correction unit 43 is used to correct the predicted corrosion rate value according to the influence weight. The correction formula is as follows:

[0117] ,

[0118] where, represents the corrected corrosion rate, represents the original predicted corrosion rate, represents the temperature deviation (the difference between the actual temperature and the standard temperature of 25°C), represents the humidity deviation (the difference between the actual humidity and the standard humidity of 60%RH). Through this correction mechanism, the system can dynamically adjust the prediction results according to the real-time environmental changes and improve the prediction accuracy.

[0119] The parameter update unit 44 is used to update the parameters of the long short-term memory network according to the corrected results. The update adopts an online learning strategy, and new data and corrected results are regularly incorporated into the training set (such as every 7 days) to fine-tune the network parameters. Preferably, a transfer learning method is used for parameter update. By freezing the parameters of the first few layers of the network and only updating the parameters of the last few layers, the update efficiency is improved while maintaining the stability of the model.

[0120] Such as Figure 9 shown, the present system may further include a multi-scale fusion module 5, which is communicatively connected to the erosion front tracking module 1 and the transfer learning parameter inversion module 2, and is used to associate the microscopic corrosion mechanism with the macroscopic deterioration performance, and realize the unified expression and processing of multi-scale information.

[0121] The multi-scale fusion module 5 adopts a hierarchical Bayesian framework to establish a parameter mapping relationship from micro to macro:

[0122] ,

[0123] where, represents the macroscopic parameter, represents the microscopic parameter, represents the observed data, represents the posterior probability distribution, represents the likelihood function, represents the conditional prior distribution, represents the prior distribution of the microscopic parameter.

[0124] Through this hierarchical framework, the scale span from the microscopic corrosion mechanism to the macroscopic structural degradation is achieved, improving the physical interpretability and prediction accuracy of the model.

[0125] As Figure 7 shown, the present invention also provides a method for predicting the service life of reinforced concrete based on material degradation simulation, including the following steps:

[0126] Step 1: Collect the chloride ion distribution images of the reinforced concrete specimens, preprocess the chloride ion distribution images, input the preprocessed chloride ion distribution images into a three-dimensional convolutional neural network that fuses topological invariance, extract feature vectors, determine the coordinates and depth of the corrosion front on the three-dimensional cross-section, and generate a three-dimensional surface of the steel bar corrosion front;

[0127] Step 2: Input the three-dimensional surface of the steel bar corrosion front into the finite element simulation platform, adjust the material parameters of the steel bar and concrete, perform parameter inversion calculation, and obtain the corrosion rate evolution curve;

[0128] Step 3: Use the corrosion rate evolution curve as the training set to input into the long short-term memory network, set the network parameters, train the network model based on the Adam algorithm, and generate a corrosion rate prediction curve;

[0129] Step 4: Collect the temperature and humidity information inside the concrete structure, calculate the influence weight of environmental factors on the corrosion rate, correct the corrosion rate prediction curve based on the influence weight, and update the long short-term memory network parameters.

[0130] In an embodiment of the present invention, the preprocessing in Step 1 includes operations such as image size normalization, denoising, and contrast enhancement. The standardized image size is 224×224 pixels. The three-dimensional convolutional neural network that fuses topological invariance includes extracting topological features and differential geometric features and fusing these features with convolutional features. The corrosion front positioning adopts an evolution equation based on geodesic flow to achieve the representation from discrete points to a continuous surface.

[0131] In an embodiment of the present invention, the parameter inversion in Step 2 adopts the Bayesian optimization algorithm, and the optimized material parameters include concrete density ρ, concrete elastic modulus E, and chloride ion diffusion coefficient D. The optimization objective function is the mean square error between the measured corrosion rate and the simulated corrosion rate.

[0132] In an embodiment of the present invention, the long short-term memory network in Step 3 adopts a bidirectional LSTM structure, the number of hidden layer nodes is 128, the network depth is 3 layers, the initial value of the learning rate is 0.001, and a learning rate decay strategy is adopted. The training adopts the Adam optimizer and an early stopping strategy, and the loss function is the mean square error.

[0133] The reinforced concrete life prediction system and method based on material degradation simulation of the present invention achieve high-precision identification and tracking of the steel bar corrosion front through a 3D convolutional neural network integrating topological invariance, improve the inversion accuracy of material parameters by using transfer learning parameter inversion technology, introduce a long short-term memory network to enhance the long-term prediction ability of the corrosion rate, establish an online correction mechanism based on real-time temperature and humidity monitoring, form a complete reinforced concrete structure life prediction system, and have significant technological innovation and application value.

[0134] In practical applications, taking a bridge in a marine environment as an example, the life of the bridge is predicted using the system of the present invention. Compared with traditional methods, the prediction accuracy is increased by 32%, and the prediction period is extended by 3 times. It can accurately capture the degradation law of the structure in a complex environment and provide strong support for structural maintenance decision-making.

[0135] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, various changes and modifications can be made to the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A reinforced concrete life prediction system based on material degradation simulation, characterized in that, Including: An erosion front tracking module, which is used to collect the chloride ion distribution image of a reinforced concrete specimen, extract the feature vector of the chloride ion distribution image through a 3D convolutional neural network that integrates topological invariance, and generate the 3D coordinates and depth information of the steel corrosion front; A transfer learning parameter inversion module, which is communicatively connected to the erosion front tracking module, is used to receive the 3D coordinates and depth information of the steel corrosion front, input the 3D coordinates of the steel corrosion front into a finite element simulation platform, perform parameter inversion calculation, and generate a corrosion rate evolution curve; A corrosion rate prediction module, which is communicatively connected to the transfer learning parameter inversion module, is used to receive the corrosion rate evolution curve, use the corrosion rate evolution curve as a training set to input into a long short-term memory network, and generate a corrosion rate prediction curve; A system online correction module, which is communicatively connected to the corrosion rate prediction module, is used to collect the temperature and humidity information inside the concrete structure, calculate the influence weight of environmental factors on the corrosion rate, correct the corrosion rate prediction curve, and update the long short-term memory network parameters; The erosion front tracking module includes: An image preprocessing unit, which is used to receive the chloride ion concentration distribution picture and unify the chloride ion concentration distribution picture into a fixed size and format; A topological feature extraction unit, which is used to apply a topological analysis method to the preprocessed chloride ion concentration distribution picture to extract the connectivity and geometric invariant features of the corrosion front; A 3D convolutional network unit, which is used to perform feature extraction on the chloride ion concentration distribution picture through a designed 3D convolution kernel structure and generate a feature vector; A front positioning unit, which is used to determine the coordinates and depth information of the corrosion front on the three-dimensional section according to the feature vector and generate a complete three-dimensional corrosion front surface model.

2. The system for predicting the service life of reinforced concrete based on material degradation simulation according to claim 1, wherein, The 3D convolutional network unit includes: A convolutional feature extraction layer, which is used to perform feature extraction on the input image through multiple layers of 3D convolutional kernels; An attention mechanism layer, which is used to weight the extracted features to highlight the feature information in the corrosion front area; A differential geometry feature layer, which is used to calculate the Gaussian curvature and mean curvature in the image to capture the geometric characteristics of the corrosion front; A topological feature fusion layer, which is used to fuse topological features and convolutional features to generate a comprehensive feature vector.

3. The reinforced concrete life prediction system based on material degradation simulation according to claim 1, characterized in that The transfer learning parameter inversion module includes: A finite element modeling unit, which is used to establish a reinforced concrete corrosion model according to the 3D coordinates of the steel corrosion front; A parameter optimization unit, which is used to adjust the material model parameters of steel and concrete and optimize the objective function; A transfer learning unit, which is used to use the difference between experimental data and simulation data for model parameter optimization to achieve knowledge transfer; A corrosion curve generation unit, which is used to generate a corrosion rate evolution curve according to the optimized material parameters.

4. The system for predicting the service life of reinforced concrete based on material degradation simulation according to claim 3, wherein The parameter optimization unit adopts a Bayesian optimization algorithm, where the optimized material parameters include: concrete density ρ, concrete elastic modulus E, and chloride ion diffusion coefficient D, and the optimized objective function is: ， wherein, represents the measured corrosion rate at a time point , represents the corrosion rate obtained by simulation under given parameters represents the time variable represents the number of sampling points 5. The reinforced concrete life prediction system based on material degradation simulation according to claim 1, characterized in that The corrosion rate prediction module includes: A data preprocessing unit, which is used to perform standardization processing on the corrosion rate evolution curve and divide the training set and the validation set; LSTM network construction unit, used to set the structural parameters of the long short-term memory network, including the number of hidden layer nodes, network depth, and learning rate; Model training unit, used to train the long short-term memory network based on the Adam algorithm to generate a prediction model; Prediction result output unit, used to generate a long-term corrosion rate prediction curve using the trained model.

6. The system for predicting the service life of reinforced concrete based on material degradation simulation according to claim 1, characterized in that The online correction module of the system includes: Environmental monitoring unit, used to collect the temperature and humidity data inside the concrete structure in real time through embedded sensors; Weight calculation unit, used to calculate the influence weight of environmental factors on the corrosion rate according to the temperature and humidity data; Model correction unit, used to correct the predicted corrosion rate value according to the influence weight; Parameter update unit, used to update the parameters of the long short-term memory network according to the corrected result.

7. The system for predicting the service life of reinforced concrete based on material degradation simulation according to claim 6, wherein The weight calculation unit calculates the influence weight of environmental factors based on the following formula: ， Among them, is the influence weight, is the temperature, is the humidity, is the proportionality coefficient, and are the weight exponents, calibrated by measured data.

8. The system for predicting the service life of reinforced concrete based on material degradation simulation according to claim 1, wherein The system further includes a multi-scale fusion module, which is communicatively connected to the erosion front tracking module and the transfer learning parameter inversion module, and is used to associate the microscopic corrosion mechanism with the macroscopic deterioration performance to achieve the unified expression and processing of multi-scale information.

9. A method for predicting the service life of reinforced concrete based on material degradation simulation, using the system according to any one of claims 1-8, characterized in that, Including the following steps: Collect the chloride ion distribution image of the reinforced concrete specimen, preprocess the chloride ion distribution image, input the preprocessed chloride ion distribution image into a three-dimensional convolutional neural network that fuses topological invariance, extract the feature vector, determine the coordinates and depth of the corrosion front on the three-dimensional section, and generate the three-dimensional surface of the steel bar corrosion front; Input the three-dimensional surface of the steel bar corrosion front into the finite element simulation platform, adjust the material parameters of the steel bar and concrete, perform parameter inversion calculation, and obtain the corrosion rate evolution curve; Use the corrosion rate evolution curve as the training set to input into the long short-term memory network, set the network parameters, and train the network model based on the stochastic gradient descent algorithm to generate the corrosion rate prediction curve; Collect the temperature and humidity information inside the concrete structure, calculate the influence weight of environmental factors on the corrosion rate, correct the corrosion rate prediction curve based on the influence weight, and update the parameters of the long short-term memory network.

Images