Deep learning-based method for predicting long-term loss of prestress in staged tensioning

Abstract

The application discloses a kind of based on deep learning's staged tension prestressed long-term loss prediction method, belong to deep learning field, this method solves the problem that traditional prestress determination method prediction result robustness is poor, including: based on standard drawing set and bridge design specification obtains bridge original data;Based on construction environment condition and concrete age, calculate concrete material attribute;Based on bridge original data and concrete material attribute, utilize finite element software to simulate the staged tension process of prestressed reinforcement under different working conditions, generate bridge prestressed reinforcement staged tension loss data set;Based on multilayer perception machine neural network, construct prestressed loss prediction model, based on prestressed loss prediction model, bridge prestressed reinforcement staged tension loss data set is handled, and the long-term loss prediction result that prestressed reinforcement is caused by shrinkage and creep is output.The application can effectively predict the stress loss of prestressed reinforcement, improve the robustness and accuracy of prediction.

Description

Technical Field

[0001] This invention belongs to the field of deep learning technology, and in particular relates to a method for predicting long-term losses of staged tensioned prestressing based on deep learning. Background Technology

[0002] Prestressed concrete beams undergo shrinkage deformation during concrete hardening, and further creep occurs under the continuous prestress of the prestressing steel bars. This shrinkage and creep shorten the main beam along its length, reducing the elongation of the prestressing steel bars and causing prestress loss. This results in a decrease in the effective prestress provided by the prestressing steel bars, reducing the crack resistance and stiffness of the member, and potentially causing the flexural capacity to fail to meet design requirements. For long-span bridges, long-term creep can lead to deflection far exceeding design values ​​during later operation. To address these issues, a phased tensioning method can be adopted. In the early stages of strength development, a portion of the stress is tensioned to accelerate shrinkage and creep, and then the beam is tensioned to the control stress according to specifications. This method can offset some of the shrinkage and creep effects and has been widely used. For example, the paper "Experimental Study on Staged Tensioning Technology of Prestressed Concrete Beams" published in the journal *China Journal of Highway and Transport* conducted an experimental study on staged tensioning technology for prestressed concrete rectangular beams. Prestressed steel bars were tensioned twice at different ages, and prestress loss was measured. The results showed that early-age staged prestressing technology can effectively reduce prestress loss in concrete, increase the cracking load and yield load of the beam, and does not affect the flexural failure mode of the beam. Chinese patent CN108396661B, entitled "Construction Method of Staged Prestressing Tensioning Based on Eliminating Concrete Creep," proposes a method to partially offset the creep effect by tensioning prestress three times: after the hollow slab is hoisted, during the second-stage paving, and after the completion of the entire bridge deck construction, effectively reducing creep deformation. However, staged tensioning technology cannot completely eliminate all shrinkage and creep, and the measurement of prestress loss is still necessary. For example, Chinese patent CN119756648B, entitled "A Bridge Prestress Loss Measurement System Based on Acoustic Emission Technology," proposes a method to calculate prestress loss by collecting acoustic emission signals during the prestress loss process of bridges and correcting errors by combining environmental data. Another example is the paper "Monitoring of Prestress Loss in Concrete Beams Using Quasi-Distributed FBG Steel Strands," published in the journal *Journal of Railway Science and Engineering*, which proposes a technique to couple multi-point quasi-distributed FBGs to the center wires of prestressed steel strands in concrete beams to monitor prestress loss and its distribution. In addition, traditional prestress measurement methods include the reaction force method and the strain gauge method. These methods can reflect the prestress loss situation of bridges to a certain extent.

[0003] In existing technologies, staged tensioning technology involves long construction cycles. Traditional prestress measurement methods rely on specialized equipment, consume significant manpower, cannot provide continuous monitoring, depend on the experience of on-site personnel, lack accuracy, and suffer from data lag. Prestress loss can only be corrected through supplementary tensioning. Obtaining original bridge data through on-site experiments is difficult, costly, and yields limited data, resulting in poor robustness of prediction results. Therefore, this invention proposes a deep learning-based method for predicting long-term prestress loss during staged tensioning. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a deep learning-based method for predicting long-term losses in staged tensioned prestressing, thereby resolving the issues present in the existing technologies.

[0005] To achieve the above objectives, this invention provides a deep learning-based method for predicting long-term losses in staged tensioned prestressing, comprising:

[0006] Original bridge data was obtained based on standard drawing sets and bridge design specifications.

[0007] Calculate the material properties of concrete based on construction environment conditions and concrete age;

[0008] Based on the original bridge data and the concrete material properties, the prestressed steel reinforcement stage tensioning process under different working conditions was simulated using finite element software to generate a dataset of prestressed steel reinforcement stage tensioning loss of the bridge.

[0009] A prestress loss prediction model is constructed based on a multilayer perceptron neural network. The prestress loss prediction model is then used to process the prestressed steel reinforcement stage tensioning loss dataset of the bridge, and the long-term loss prediction results of the prestressed steel reinforcement caused by shrinkage and creep are output.

[0010] Optionally, the concrete material properties include: concrete elastic modulus, concrete shrinkage coefficient, and concrete creep coefficient;

[0011] The process of calculating the material properties of concrete includes:

[0012] Based on the concrete age and cement type, the elastic modulus of concrete is calculated using the compressive strength development function.

[0013] Based on the concrete age and the theoretical thickness of the component, the shrinkage coefficient of the concrete is calculated using a shrinkage development function;

[0014] Based on the loading age and the average annual humidity of the environment, the creep coefficient of concrete was calculated using viscoelastic theory and statistical regression analysis.

[0015] Optionally, the formula for calculating the concrete shrinkage coefficient is:

[0016] ;

[0017] In the formula, express arrive The coefficient of concrete shrinkage development; This indicates the concrete age at the time of calculation, in days. This indicates the age of the concrete at the onset of shrinkage, expressed in days. This indicates the theoretical thickness of the component, in mm. This indicates the theoretical thickness, expressed in mm.

[0018] Optionally, the expression for calculating the concrete creep coefficient is:

[0019] ;

[0020] In the formula, express arrive The concrete creep coefficient at any given time; This indicates the concrete age at the time of calculation, in days. Indicates the loading age; This indicates the annual average relative humidity of the environment; This represents the nominal creep coefficient of concrete obtained from the nominal creep coefficient table; Indicates the ambient baseline humidity; This indicates the theoretical thickness of the component, in mm. This indicates the theoretical thickness, expressed in mm.

[0021] Optionally, the process of generating a dataset of prestressed steel reinforcement staged tensioning losses for the bridge based on the original bridge data and the concrete material properties includes:

[0022] Based on the original bridge data, a finite element model of the bridge containing coupled concrete elements and steel reinforcement elements was established using the finite element modeling method.

[0023] Based on the concrete shrinkage coefficient in the aforementioned concrete material properties, the concrete shrinkage strain value is obtained by using the shrinkage strain calculation method, and the concrete shrinkage strain value is converted into an equivalent temperature load.

[0024] Based on the concrete creep coefficient in the aforementioned concrete material properties, the material parameters are updated using a creep parameter calculation method.

[0025] Based on the equivalent temperature load and the updated material parameters, the bridge finite element model is solved to obtain prestress loss data under various working conditions.

[0026] Based on the prestress loss data, a standardized dataset with unified dimensions is obtained by using a data standardization processing method.

[0027] The standardized dataset is split into training, validation, and test sets to complete the construction of the bridge prestressed steel reinforcement stage tensioning loss dataset.

[0028] Optionally, the expression for updating the material parameters based on the concrete creep coefficient in the concrete material properties is as follows:

[0029] ;

[0030] In the formula, express Momentary elastic modulus of concrete Indicates the concrete creep coefficient at the time of calculation; This indicates the concrete creep coefficient considering the time preceding the calculation time. This indicates the age at the time of calculation, in days. Consider the age of the time preceding the calculation time, in days; This indicates the updated material parameters.

[0031] Optionally, a dual-segmentation strategy can be used to segment the standardized dataset to obtain a dataset of bridge prestressed steel reinforcement stage tensioning loss.

[0032] The dataset of bridge prestressed steel reinforcement stage tensioning loss includes a training set, a validation set, and a test set.

[0033] Optionally, the process of processing the bridge prestressed steel reinforcement staged tensioning loss dataset based on the prestressed loss prediction model and outputting the long-term loss prediction results of prestressed steel reinforcement caused by shrinkage and creep includes:

[0034] Based on the training set, a multilayer perceptron neural network is used for model training, the predicted value is calculated through forward propagation, and the mean squared error loss function is used to evaluate the prediction bias.

[0035] Based on the validation set and prediction error, the model hyperparameters are optimized using early stopping and learning rate scheduling strategies.

[0036] Based on the hyperparameters of the optimized model and the test set, the generalization performance of the prestress loss prediction model is evaluated using mean square error, mean absolute error and coefficient of determination to obtain a well-trained prestress loss prediction model.

[0037] The trained prestress loss prediction model is applied to new input data to output the predicted prestress loss value.

[0038] The present invention also provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described thereon.

[0039] The present invention also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.

[0040] Compared with the prior art, the present invention has the following advantages and technical effects:

[0041] This invention provides a deep learning-based method for predicting long-term prestressing losses during staged tensioning. Through data acquisition, material property calculation, finite element simulation, data preprocessing, and integrated prediction modeling, a reliable long-term prediction model for staged prestressing reinforcement in bridges is constructed, providing an important decision-making method for prestressed beam tensioning compensation. This invention employs machine learning, finite element analysis, and other technologies, combined with various data processing and model optimization methods, to effectively predict prestressing reinforcement stress losses, improving the robustness and accuracy of the predictions, and providing important scientific basis for tensioning management in precast beam yards. Attached Figure Description

[0042] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0043] Figure 1 This is a flowchart illustrating the long-term prediction process for staged tensioning of prestressed steel bars according to an embodiment of the present invention.

[0044] Figure 2 This is a flowchart illustrating the process of obtaining the original prediction data using the finite element method in an embodiment of the present invention.

[0045] Figure 3 This is a loss convergence graph of the training process according to an embodiment of the present invention;

[0046] Figure 4 This is a learning rate change curve according to an embodiment of the present invention;

[0047] Figure 5 This is a comparison chart of the actual and predicted loss values ​​of the N1 rebar test set in this embodiment of the invention;

[0048] Figure 6 This is a comparison chart of the actual and predicted loss values ​​of the N1 rebar training set in this embodiment of the invention.

[0049] Figure 7 This is a comparison chart of the actual and predicted loss values ​​of the N2 rebar test set in this embodiment of the invention;

[0050] Figure 8 This is a comparison chart of the actual and predicted loss values ​​of the N2 rebar training set in this embodiment of the invention.

[0051] Figure 9 This is a comparison chart of the actual and predicted loss values ​​of the N3 rebar test set in this embodiment of the invention;

[0052] Figure 10 This is a comparison chart of the actual and predicted loss values ​​of the N3 rebar training set in this embodiment of the invention.

[0053] Figure 11 This is a distribution diagram of the loss prediction error of the N1 rebar test set in this embodiment of the invention;

[0054] Figure 12 This is a distribution diagram of the loss prediction error of the N1 rebar training set in this embodiment of the invention;

[0055] Figure 13 This is a distribution diagram of the loss prediction error of the N2 rebar test set in an embodiment of the present invention;

[0056] Figure 14 This is a distribution diagram of the loss prediction error of the N2 rebar training set in this embodiment of the invention;

[0057] Figure 15 This is a distribution diagram of the loss prediction error of the N3 rebar test set in this embodiment of the invention;

[0058] Figure 16 This is a distribution diagram of the loss prediction error of the N3 rebar training set in this embodiment of the invention. Detailed Implementation

[0059] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0060] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0061] Deep learning technology can extract complex mapping features between input data and output target data to predict prestress loss. Finite element simulation technology breaks through formula limitations and significantly improves the accuracy of long-term predictions. The deep learning-based method for predicting long-term prestress loss during staged tensioning features high accuracy, strong practicality, and low cost. It can quickly predict the over-tensioning correction value for long-term prestress loss caused by shrinkage and creep during staged tensioning, greatly improving construction speed and efficiency.

[0062] like Figure 1 As shown, this embodiment provides a method for predicting long-term losses of staged tensioned prestressing based on deep learning, including the following steps S101 to S105.

[0063] Step S101: Data Acquisition. Obtain original bridge data for various T-beams, slab beams, and box girders from standard drawing sets and highway reinforced concrete and prestressed concrete bridge and culvert design specifications, including main beam section type, area, moment of inertia, centroid location, rebar diameter, rebar resultant force point location, and beam length.

[0064] The original bridge data in this embodiment can be obtained from the standard bridge drawing set. The data is supplemented on the basis of the standard drawing set to meet the needs of most precast beam long components. It is necessary to extract the necessary data features for finite element analysis modeling. In this embodiment, box girders and T-beams, two common precast beam types, are extracted.

[0065] Step S102: Considering shrinkage and creep, calculate the material properties of concrete at different initial tensioning ages. Concrete material properties include the concrete modulus of elasticity, the concrete shrinkage coefficient, and the concrete creep coefficient.

[0066] Specifically, the annual average humidity, initial and final tensioning ages, and strength combinations under construction conditions are obtained. Based on the concrete material properties and construction environment conditions, the development strength and elastic modulus of the concrete over time are calculated. Using existing viscoelastic theory and calculation formulas obtained from statistical regression analysis, the shrinkage and creep coefficients of the concrete are determined.

[0067] This embodiment refers to the CEB-FIP model, and the formula for predicting the elastic modulus of concrete is as follows:

[0068]

[0069]

[0070]

[0071] in, It is a function of compressive strength over time; This indicates the concrete age at the time of calculation. This is a coefficient related to the type of cement: 0.2 for rapid-hardening high-strength cement, 0.25 for ordinary cement and rapid-hardening cement, and 0.38 for slow-hardening cement. The concrete age is The elastic modulus of the weather; It is the elastic modulus of concrete at 28 days of age; It is the elastic modulus.

[0072] Concrete shrinkage and creep are strongly correlated with air humidity. Lower humidity significantly increases the drying shrinkage value of concrete and reduces its internal moisture, affecting its viscoelastic behavior. Before calculation, the average annual humidity of the area where this method is applied must be investigated. According to specifications, the average annual humidity (RH) and the nominal shrinkage coefficient of concrete are calculated. When RH is between 40% and 70%, RH = 55%, and the nominal shrinkage coefficient is 0.529; when RH is between 70% and 90%, RH = 80%, and the nominal shrinkage coefficient is 0.310. In this example, the average annual humidity in Guangxi Zhuang Autonomous Region is 90%, the calculated humidity is 80%, and the nominal shrinkage coefficient is taken as 0.310.

[0073] Referring to the CEB-FIP model, the formula for calculating the concrete shrinkage coefficient is as follows:

[0074]

[0075] in, express arrive The coefficient of concrete shrinkage development; This indicates the concrete age at the time of calculation. Indicates the age of the concrete at which shrinkage begins; Indicates the theoretical thickness of the component; Indicates the theoretical thickness.

[0076] The creep coefficient of concrete is calculated using existing viscoelastic theory and statistical regression analysis. The calculation formula is as follows:

[0077]

[0078] in, express arrive The concrete creep coefficient at any given time; Indicates the loading age; This indicates the concrete age at the time of calculation. This indicates the annual average relative humidity of the environment; The nominal creep coefficient of concrete is obtained from the nominal creep coefficient table in Table 1, and interpolation may be used if necessary. This indicates the ambient baseline humidity.

[0079] Table 1

[0080]

[0081] Step S103: Based on the bridge foundation information and concrete material properties obtained in steps S101 and S102, and since it is impossible to traverse all beam combinations in a real-world scenario, it is necessary to use the finite element software ANSYS to calculate the shrinkage and creep loss data of prestressed steel bars under different working conditions. The data is then preprocessed to obtain a dataset of bridge prestressed steel bar staged tensioning losses. The preprocessing process includes data segmentation and elimination of dimensional differences. The specific dataset creation process is as follows: Figure 2 As shown, it includes the following steps.

[0082] Parameter calculation: Based on the linear creep theory, the shrinkage strain value of concrete is calculated using the concrete shrinkage coefficient and the nominal shrinkage coefficient of concrete. The calculation formula is as follows:

[0083]

[0084] in, express arrive Concrete shrinkage strain value at any given time; express arrive The coefficient of concrete shrinkage development; This represents the nominal shrinkage coefficient of concrete.

[0085] The equivalent temperature method is used to apply initial strain to the material, converting concrete shrinkage strain into temperature load. The calculation formula is as follows:

[0086]

[0087] in, This indicates the calculation of temperature change; This represents the coefficient of thermal expansion of concrete.

[0088] There is no direct method for calculating concrete creep in ANSYS. To simulate concrete creep in ANSYS, it is necessary to consider concrete creep according to the principle of metal creep. The aging theory of metal creep states that at a constant temperature, there is a certain relationship between creep deformation, stress, and time. The stress-strain equation for concrete creep is similar to that for metal creep, so metal creep can be used as an approximation to calculate concrete creep.

[0089] The stress-strain equation for concrete considering creep is shown in equation (8):

[0090]

[0091] in, Indicates the concrete age as The elastic modulus of the weather; express arrive Constant concrete creep strain; express arrive Constant concrete creep stress; express arrive The creep coefficient of concrete at any given time.

[0092] In ANSYS, the explicit integration method is used, and equation number 6 for calculating the material is selected. The stress-strain equation is shown in formula (8):

[0093]

[0094] in, This represents the strain of the concrete at time t; The stress in the concrete at time t represents the stress at time t; T represents the ambient temperature at time t. Indicates the material coefficient. , , Let represent the first constant coefficient, the second constant coefficient, and the third constant coefficient, respectively, and e be the natural constant.

[0095] In this embodiment, linear theory is adopted, where material strain is directly proportional to stress in a linear fashion. Therefore, we take... =1; Assuming the strain rate during creep depends only on the strain in the material, the strain hardening criterion is adopted, i.e. =0; and since the effect of temperature on creep is not considered in the calculation, therefore =0.

[0096] The material coefficients for the corresponding calculation times can be obtained from formulas (8) and (9). The creep coefficient at the corresponding time point can be obtained using the following formula:

[0097]

[0098] in, express Momentary elastic modulus of concrete Indicates the concrete creep coefficient at the time of calculation; This indicates the concrete creep coefficient considering the time preceding the calculation time. Indicates the age at the time of calculation; Consider the age of the time preceding the calculation time.

[0099] Finite element simulation: Using the command flow function in ANSYS APDL, a phased tensioning calculation model for prestressed bridges is established, changing the elastic modulus and material coefficients of concrete at each calculation time. Initial shrinkage strain was applied. By iteratively traversing the combinations of prestressing stage tensioning ages and different spans and main beam cross-sections, the shrinkage and creep loss data of prestressed steel bars were calculated, and the calculated data were extracted to construct a dataset for use in neural networks to extract digital features.

[0100] This embodiment employs the solid reinforcement method, dividing the concrete and prestressed steel strands into different elements, and linking the degrees of freedom of the solid elements and the reinforcement elements through coupling. In this embodiment, the concrete uses a bilinear hardening model, i.e., the MISO model, to simulate more realistic concrete deformation under prestressing. SOLID65 elements are used to simulate the main beam concrete. LINK180 elements are used to simulate the prestressed steel strands, and the equivalent temperature method is used to cool the tensioned prestressed equivalent paired reinforcement components. The temperature load calculation formula is as follows:

[0101]

[0102] in, This represents the equivalent temperature value of prestressing; This indicates the prestressed force applied to the prestressed steel strands; This indicates the elastic modulus of the steel strand; is the coefficient of linear expansion of the steel strand; The cross-sectional area of ​​the steel strand; The conversion factor is used to account for prestress loss.

[0103] Data Segmentation: The obtained dataset of bridge prestressed steel reinforcement staged tensioning losses was segmented using a double-segmentation strategy to ensure the reliability and robustness of the model evaluation. First, the data was split into an 80% training set and a 20% test set, with the data order randomly shuffled. Second, the training set was further split into a sub-training set and a validation set, with 80% of the original training set used as the sub-training set and 20% as the validation set.

[0104] Ultimately, the data was divided into a 64% training set for learning model parameters; a 16% validation set for model selection and hyperparameter tuning; and a 20% test set for evaluating the model's generalization ability, which was not involved in the model development process.

[0105] Data standardization: Standardization methods are used to eliminate differences in the dimensions of different features, ensuring that all features contribute information at the same scale. The bridge prestressed steel reinforcement staged tensioning loss dataset includes data such as beam length, beam width, beam height, cross-sectional area, cross-sectional moment of inertia, cross-sectional centroid, height of the resultant force point of the reinforcement, eccentricity of the resultant force point, initial tension concrete strength, and prestressing loss of the reinforcement. These data often have significantly different scales, leading to large differences in gradient update amplitudes and affecting model performance. Therefore, to eliminate the influence of different dimensions, it is necessary to standardize the dataset features.

[0106] This embodiment uses Z-score normalization to scale the original data features to the same scale. The normalization formula is as follows:

[0107]

[0108] in, Represents the original feature data; Represents the characteristic mean. The standard deviation represents the characteristic.

[0109] This method adjusts the mean to 0 and standardizes the variance to 1 without altering the basic shape of the data distribution, thus making the loss function smoother and the optimization path more direct. This effectively avoids computational instability caused by extreme values ​​and ensures that the input values ​​fall within the sensitive region of the activation function.

[0110] Step S104: Construction of Prestress Loss Prediction Model. This step uses a multilayer perceptron (MLP) neural network to construct a prestress loss prediction regression model. This regression model accurately predicts the long-term prestress loss caused by shrinkage and creep in concrete beams using staged tensioning technology.

[0111] This embodiment constructs a multilayer perceptron (MLP) neural network model. Dropout + L2 multiple regularization is used to prevent overfitting and improve generalization ability. Batch Norm stabilizes the data distribution between layers, and Adam adaptive learning rate accelerates training convergence and improves training stability. The model with the best performance on the validation set is selected as the optimal prediction model. This model aims to accurately predict the long-term prestress loss caused by shrinkage and creep in concrete beams using staged tensioning technology.

[0112] This embodiment trains a neural network model based on the training set obtained in S103. First, it configures the shared layer and task-specific head, initializes the optimizer and learning rate scheduler, and calculates the predicted value and model loss through forward propagation. The loss function is the mean squared error loss (MSE), as shown in the following formula:

[0113]

[0114] in, This represents the mean squared error loss; This represents the true value of the i-th sample; This represents the predicted value of the i-th sample; n represents the number of samples in the current batch.

[0115] Dropout regularization is used to randomly discard neurons during training, preventing neuron co-adaptation and improving generalization ability. The formula is as follows:

[0116]

[0117] in, This represents the neuron's raw output value; This represents the output value after Dropout; Indicates the probability of Dropout; This represents the scaling factor.

[0118] At the same time, L2 regularization is used to penalize large weight values ​​to prevent overfitting. The formula is as follows:

[0119]

[0120] in, Indicates the total loss; This represents the regularization strength coefficient; This represents all trainable weight parameters of the model.

[0121] The cost function is optimized using gradient descent, and the weights and biases are updated via backpropagation. At each training step, the Adam optimizer is used to adaptively adjust the parameter update magnitude to reduce model loss. The gradient descent optimization formula is as follows:

[0122]

[0123] in, This indicates an update to the weights; Indicates the original weights; Indicates the learning rate; This represents the gradient of the loss function.

[0124] Among them, learning rate Dynamic learning rate scheduling is employed; the learning rate is reduced when the validation loss stagnates to prevent oscillations and reduce the risk of overfitting. This embodiment stores the learning rate and loss for each training step and labels the optimal model loss, such as... Figures 3-4 As shown.

[0125] Step S105: Evaluate and validate the model using evaluation metrics and an independent test set. The evaluation metrics include MSE (mean squared error), MAE (mean absolute error), and R² (coefficient of determination). Calculate the individual metric for each output dimension. The formulas for calculating the evaluation metrics are as follows:

[0126] Mean Square Error (MSE):

[0127]

[0128] Mean Absolute Error (MAE):

[0129]

[0130] Absolute coefficient R²:

[0131]

[0132] in, This represents the true value of the i-th sample; This represents the predicted value of the i-th sample; This represents the average of the true values.

[0133] In this embodiment, the evaluation indicators for each steel bar loss prediction result are shown in Table 2.

[0134] Table 2

[0135]

[0136] Additionally, the scatter plots and error distribution plots for the test set and prediction set are shown below. Figure 5 , Figure 6 , Figure 7 , Figure 8 , Figure 9 , Figure 10 , Figure 11 , Figure 12 , Figure 13 , Figure 14 , Figure 15 , Figure 16 As shown.

[0137] The present invention has the following beneficial effects:

[0138] (1) Improve the speed and accuracy of prestress loss prediction: The construction cycle of staged tensioning technology is long. Traditional prestress measurement methods rely on special equipment, consume a lot of manpower, cannot continuously detect prestress, and rely on the experience of on-site personnel, which lacks accuracy and the detection data is lagging. Prestress loss can only be corrected by supplementary tensioning. However, the prestress loss prediction method based on deep learning adopted in this invention can predict the long-term prestress loss caused by shrinkage and creep in advance, and make corrections for the control stress of construction tensioning in advance, with high accuracy and stability.

[0139] (2) Overcoming the difficulty of obtaining original data: The method of obtaining original bridge data through field experiments is difficult, costly, and results in a small amount of data, leading to poor robustness of the prediction results. In contrast, this invention extracts a large number of bridge types and uses finite element analysis to calculate the original data of the prestress loss prediction model, thereby improving the robustness of the calculation results.

[0140] (3) Employing multiple regularization strategies to prevent overfitting and improve generalization ability: This invention effectively suppresses overfitting through Dropout + L2 regularization, ensuring stable model performance under unknown conditions. The Batch Norm technology stack stabilizes the distribution of inputs at each layer, accelerates training, and improves model performance. The Adam optimizer is used for training, utilizing its adaptive learning rate characteristics to accelerate convergence and improve the stability of the training process.

[0141] (4) Possesses engineering practicality and wide applicability: The dataset of this invention covers mainstream bridge types such as T-beams and box girders, and the model can be adapted to predict prestress loss for different main beam types. Finite element calculations only need to be performed once during the data preparation stage; the trained model can output prediction results in real time, avoiding repeated time-consuming simulations and greatly improving the efficiency of design iteration. In actual engineering, it can quickly predict prestress loss and provide guidance for construction design.

[0142] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting long-term losses of staged tensioned prestressing based on deep learning, characterized in that, Includes the following steps: Original bridge data was obtained based on standard drawing sets and bridge design specifications. Calculate the material properties of concrete based on construction environment conditions and concrete age; Based on the original bridge data and the concrete material properties, finite element method (FEM) software is used to simulate the prestressed steel reinforcement staged tensioning process under different working conditions, generating a prestressed steel reinforcement staged tensioning loss dataset for the bridge. The process includes: establishing a bridge finite element model containing coupled concrete and steel reinforcement elements based on the original bridge data; calculating the concrete shrinkage strain value using a shrinkage strain calculation method based on the concrete shrinkage coefficient in the concrete material properties, and converting the concrete shrinkage strain value into an equivalent temperature load; updating the material parameters using a creep parameter calculation method based on the concrete creep coefficient in the concrete material properties; solving the bridge finite element model based on the equivalent temperature load and the updated material parameters to obtain prestressed loss data under each working condition; obtaining a standardized dataset with unified dimensions using a data standardization processing method based on the prestressed loss data; and dividing the standardized dataset into training, validation, and test sets to complete the construction of the bridge prestressed steel reinforcement staged tensioning loss dataset. A prestress loss prediction model is constructed based on a multilayer perceptron neural network. The prestress loss prediction model is then used to process the prestressed steel reinforcement stage tensioning loss dataset of the bridge, and the long-term loss prediction results of the prestressed steel reinforcement caused by shrinkage and creep are output.

2. The method for predicting long-term losses of staged tensioned prestressing based on deep learning according to claim 1, characterized in that, The concrete material properties include: concrete elastic modulus, concrete shrinkage coefficient, and concrete creep coefficient; The process of calculating the material properties of concrete includes: Based on the concrete age and cement type, the elastic modulus of concrete is calculated using the compressive strength development function. Based on the concrete age and the theoretical thickness of the component, the shrinkage coefficient of the concrete is calculated using a shrinkage development function; Based on the loading age and the average annual humidity of the environment, the creep coefficient of concrete was calculated using viscoelastic theory and statistical regression analysis.

3. The method for predicting long-term losses of staged tensioned prestressing based on deep learning according to claim 2, characterized in that, The formula for calculating the concrete shrinkage coefficient is as follows: ； In the formula, express arrive The coefficient of concrete shrinkage development; This indicates the concrete age at the time of calculation, in days. This indicates the age of the concrete at the onset of shrinkage, expressed in days. This indicates the theoretical thickness of the component, in mm. This indicates the theoretical thickness, expressed in mm.

4. The method for predicting long-term losses of staged tensioned prestressing based on deep learning according to claim 3, characterized in that, The formula for calculating the concrete creep coefficient is as follows: ； In the formula, express arrive The concrete creep coefficient at any given time; This indicates the concrete age at the time of calculation, in days. Indicates the loading age; This indicates the annual average relative humidity of the environment; This represents the nominal creep coefficient of concrete obtained from the nominal creep coefficient table; Indicates the ambient baseline humidity; This indicates the theoretical thickness of the component, in mm. This indicates the theoretical thickness, expressed in mm.

5. The method for predicting long-term losses of staged tensioned prestressing based on deep learning according to claim 1, characterized in that, The expression for updating the material parameters based on the concrete creep coefficient in the aforementioned concrete material properties is as follows: ； In the formula, express Momentary elastic modulus of concrete Indicates the concrete creep coefficient at the time of calculation; This indicates the concrete creep coefficient considering the time preceding the calculation time. This indicates the age at the time of calculation, in days. Consider the age of the time preceding the calculation time, in days; This indicates the updated material parameters.

6. The method for predicting long-term losses of staged tensioned prestressing based on deep learning according to claim 5, characterized in that, A dual-segmentation strategy was used to segment the standardized dataset to obtain a dataset of bridge prestressed steel reinforcement stage tensioning loss. The bridge prestressed steel reinforcement stage tensioning loss dataset includes a training set, a validation set, and a test set.

7. The method for predicting long-term losses of staged tensioned prestressing based on deep learning according to claim 6, characterized in that, The process of processing the staged tensioning loss dataset of the bridge prestressed steel bars based on the prestress loss prediction model and outputting the long-term loss prediction results of the prestressed steel bars due to shrinkage and creep includes: Based on the training set, a multilayer perceptron neural network is used for model training, the predicted value is calculated through forward propagation, and the mean squared error loss function is used to evaluate the prediction bias. Based on the validation set and prediction error, the model hyperparameters are optimized using early stopping and learning rate scheduling strategies. Based on the hyperparameters of the optimized model and the test set, the generalization performance of the prestress loss prediction model is evaluated using mean square error, mean absolute error and coefficient of determination to obtain a well-trained prestress loss prediction model. The trained prestress loss prediction model is applied to new input data to output the predicted prestress loss value.

8. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1-7.

9. A storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-7.

Images