A design method, system and medium for HPC-RC combined eccentrically compressed columns

Abstract

This disclosure relates to the field of bridge engineering, specifically to a design method, system, and medium for HPC-RC combined eccentrically compressed columns. The method includes: defining a design space; selecting optimal design data combinations based on engineering specification constraints, bearing capacity constraints, and cost-effectiveness values ​​to construct a design data sample set; constructing a two-branch heterogeneous neural network model, the model including an input layer, a shared feature extraction layer, a diameter prediction classification branch, and a reinforcement area prediction regression branch; constructing a loss function composed of focal loss and mean square error loss; training the two-branch heterogeneous neural network model using the design data sample set; and using the trained two-branch heterogeneous neural network model to predict the diameter and total reinforcement area of ​​the eccentrically compressed column. This disclosure improves the stability and accuracy of the design, reduces the computational burden, increases computational speed, and reduces computational complexity.

Description

Technical Field

[0001] This disclosure relates to the field of bridge engineering, and in particular to a design method, system and medium for HPC-RC combined eccentrically compressed columns. Background Technology

[0002] Bridge piers, as crucial components of bridge engineering, play a vital role in transferring superstructure loads to the foundation. They are susceptible to damage from water erosion, freeze-thaw cycles, and ice floes. As bridge engineering progresses towards higher elevations, longer spans, and deeper water areas, traditional bridge piers are showing limitations in durability, load-bearing capacity, and seismic resistance. In recent years, HPC-RC composite eccentrically loaded columns, constructed from high-performance concrete (HPC) and reinforced concrete (RC), have attracted widespread attention due to their superior mechanical properties and durability. High-performance concrete can serve as permanent high-performance concrete during construction and also improves the load-bearing efficiency of structural members over long-term use.

[0003] However, the design of HPC-RC combined eccentric compression columns has a much higher computational complexity than traditional bridge piers, and presents the following technical challenges in design: The high-performance concrete (HPC)-RC composite eccentrically compressed columns, in addition to traditional design variables such as column diameter and reinforcement ratio, also include the key variable of high-performance concrete shell thickness. The calculation of high-performance concrete must consider its compressive and tensile properties at the same time. There are complex nonlinear coupling relationships between different parameters, which makes the calculation very difficult.

[0004] Traditional design relies heavily on human experience, and the design parameter space is huge. It is difficult for manual design to explore all feasible solutions, which often leads to overly conservative designs, resulting in material waste and difficulty in finding a balance between safety and economy.

[0005] Traditional mechanical calculation methods are time-consuming, resulting in low computational efficiency and high computational requirements when real-world engineering projects require rapid response to multi-condition design needs.

[0006] In recent years, with the rapid development of deep learning technology, it has been widely applied in the field of bridge engineering technology. However, there are the following technical bottlenecks in applying this technology to the design of HPC-RC composite eccentric compression columns: (1) Poor stability: It only works well for the design of conventional HPC-RC combined eccentric compression columns, i.e., it is accurate and reliable. However, it is poor for the design of uncommon HPC-RC combined eccentric compression columns, and the probability of obtaining the truly optimal design is low.

[0007] (2) The computational burden of implementing the design is large: the amount of computation is large and / or the computational efficiency is low.

[0008] Therefore, there is an urgent need to design a method, system, and medium for HPC-RC combined eccentrically compressed columns. Summary of the Invention

[0009] Therefore, it is necessary to provide a design method, system, and medium for HPC-RC combined eccentric compression columns to address the problems of poor stability and high computational burden in existing HPC-RC combined eccentric compression column designs.

[0010] To solve the above problems, the present disclosure adopts the following technical solution: In a first aspect, the present invention provides a design method for an HPC-RC combined eccentrically compressed column, comprising: Step 1: Define the design space, which includes the design requirement space and the design parameter space. For each design data item in the design requirement space, perform stratified sampling, and combine the stratified sampling results to generate the design requirement data combination. The design data in the design requirements includes the axial pressure design value of the eccentrically compressed column, the eccentricity of the eccentrically compressed column, the calculated length of the eccentrically compressed column, and the thickness of the high-performance concrete shell. The design data in the design parameters includes the diameter of the eccentrically compressed column, the diameter of the reinforcing bar, and the number of reinforcing bars. Step 2: For each combination of design requirements data, the optimal design data combination is selected based on engineering specification constraints, load-bearing capacity constraints, and cost-effectiveness values. A design data sample set is then constructed based on the optimal design data combination. Step 3: Construct a dual-branch heterogeneous neural network model. The model includes an input layer for receiving and normalizing a set of design data samples, a shared feature extraction layer for extracting high-dimensional nonlinear features from the normalization results, a diameter prediction classification branch for predicting the diameter of eccentrically compressed columns based on the high-dimensional nonlinear features, and a reinforcement area prediction regression branch for predicting the total reinforcement area of ​​eccentrically compressed columns based on the high-dimensional nonlinear features. Step 4: Construct a loss function consisting of focus loss and mean squared error loss, and train the dual-branch heterogeneous neural network model using the designed data sample set; Step 5: Obtain the design requirements of the eccentrically compressed column to be designed as input to the trained bi-branch heterogeneous neural network model, and use the trained bi-branch heterogeneous neural network model to predict the diameter of the eccentrically compressed column and the total reinforcement area of ​​the eccentrically compressed column.

[0011] In a preferred embodiment, step 2 includes: Step 201: For each design requirement data combination, the design data in the design parameters are traversed and the engineering specifications are checked. The design data combination that conforms to the engineering specifications is retained according to the engineering specifications constraints. The engineering specifications checks include slenderness ratio check, reinforcement ratio check, and rebar spacing check. Step 202: Calculate the bearing capacity of the design data combination that meets the engineering specification constraints, and retain the design data combination that meets the bearing capacity constraints. Step 203: Calculate the cost-effectiveness value of the design data combination that meets the bearing capacity constraints, and select the design data combination with the highest cost-effectiveness value as the preferred design data combination.

[0012] In a preferred embodiment, the calculation process in step 202 is as follows: calculate the eccentricity amplification factor; establish a nonlinear equilibrium equation for the half-pressure angle and solve it using the bisection method to obtain the half-pressure angle of the eccentrically compressed member; calculate the ultimate bearing capacity of the section using the eccentricity amplification factor and the half-pressure angle; determine whether the calculated ultimate bearing capacity of the section meets the bearing capacity constraint conditions, and if it does, retain the design data combination corresponding to the ultimate bearing capacity of the section.

[0013] In a preferred embodiment, the nonlinear equilibrium equation is: In the formula: in, It is the half-pressure angle; This is the eccentricity amplification factor; The eccentricity of the eccentrically compressed column; This represents the balance error between the eccentricity of internal forces and the eccentricity of external forces within a cross section. This represents the normalized bending moment integral term; This represents the normalized axial force integral term; The core concrete radius; The radius of the circumference where the reinforcing bars are distributed; This refers to the reinforcement ratio; For high-performance concrete reinforcement ratio; The radius of the center of the high-performance concrete shell; This refers to the design value of the core concrete compressive strength. This is the design value for the strength of the reinforcing steel. This refers to the design value of the compressive strength of high-performance concrete. Design value of tensile strength of high-performance concrete; This is the height coefficient of the tension zone reinforcement relative to the limiting compression zone. The formula for calculating the ultimate bearing capacity of a cross section is: in, The ultimate bearing capacity of the section. The area of ​​the core concrete cross section.

[0014] In a preferred embodiment, the design data sample set includes: the axial pressure design value of the eccentrically compressed column in the preferred design data combination, the eccentricity of the eccentrically compressed column in the preferred design data combination, the calculated length of the eccentrically compressed column in the preferred design data combination, the thickness of the high-performance concrete shell in the preferred design data combination, the diameter of the eccentrically compressed column in the preferred design data combination, and the total reinforcement area of ​​the eccentrically compressed column determined based on the data in the preferred design data combination.

[0015] In a preferred embodiment, the loss function is: in, This represents the total loss of the model; These are the regression task weight coefficients; This represents the true value of the regression task. This represents the model prediction for the regression task. This represents the mean squared error loss of the regression task. , For the sample number, For the real category, For the model to predict the category, For the sample The true category, For the sample The model predicts the category. For batch size, For the classification task weight coefficients, For the true labels of the classification task, To predict probabilities for models used in classification tasks. This represents the focus loss for the classification task. , For the model to predict the first Each sample belongs to the true category. The probability, This is the category balance coefficient. For focusing parameters.

[0016] In a preferred embodiment, the method further includes a step of verifying whether the prediction results of the trained dual-branch heterogeneous neural network model meet preset requirements.

[0017] In a preferred embodiment, the shared feature extraction layer includes four sequentially arranged first fully connected layers, each of which includes a batch normalization layer and a ReLU activation function layer, and the first three first fully connected layers include a random deactivation layer.

[0018] Secondly, the present invention provides an HPC-RC combined eccentrically compressed column design system, comprising: The space definition module is used to define the design space, which includes the design requirements space and the design parameter space. The design data in the design requirements includes the axial pressure design value of the eccentrically compressed column, the eccentricity of the eccentrically compressed column, the calculated length of the eccentrically compressed column, and the thickness of the high-performance concrete shell. The design data in the design parameters includes the diameter of the eccentrically compressed column, the diameter of the reinforcing bar, and the number of reinforcing bars. The stratified sampling module is used to perform stratified sampling on each piece of design data in the design requirement space, and to combine the stratified sampling results to generate a combination of design requirement data. The data sample construction module is used to select the optimal design data combination for each design requirement data combination based on engineering specification constraints, load-bearing capacity constraints, and cost-effectiveness values, and to construct a design data sample set based on the optimal design data combination. The model building module is used to build a dual-branch heterogeneous neural network model. The model includes an input layer for receiving and normalizing a set of design data samples, a shared feature extraction layer for extracting high-dimensional nonlinear features from the normalization results, a diameter prediction classification branch for predicting the diameter of an eccentrically compressed column based on the high-dimensional nonlinear features, and a reinforcement area prediction regression branch for predicting the total reinforcement area of ​​an eccentrically compressed column based on the high-dimensional nonlinear features. The model training module is used to construct a loss function consisting of focus loss and mean squared error loss, and to train the dual-branch heterogeneous neural network model using the designed data sample set. The acquisition and prediction module is used to acquire the design requirements of the eccentrically compressed column to be designed as input to the trained bi-branch heterogeneous neural network model, and to use the trained bi-branch heterogeneous neural network model to predict the diameter of the eccentrically compressed column and the total reinforcement area of ​​the eccentrically compressed column.

[0019] Thirdly, the present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium includes instructions that, when executed on a computer, cause the computer to perform the HPC-RC combined eccentrically compressed column design method described in the first aspect.

[0020] The aforementioned design method, system, and medium for HPC-RC combined eccentrically compressed columns utilize stratified sampling of each design data item in the design requirement space. The stratified sampling results are then combined to generate a design requirement data set. Optimal design data sets are selected based on engineering code constraints, bearing capacity constraints, and cost-effectiveness values ​​to construct a design data sample set. This is coupled with a neural network model employing a loss function consisting of focus loss and mean square error loss. This design improves the stability and accuracy of the design, providing accurate designs for all HPC-RC combined eccentrically compressed columns. Furthermore, by designing and training a dual-branch heterogeneous neural network model, multi-branch tasks can be collaboratively optimized and predicted. The shared feature extraction layer accelerates the model's processing speed, enabling collaborative training of different task categories. The model can simultaneously output the reinforcement area and the diameter of the HPC-RC combined eccentrically compressed column, resulting in a low computational burden, increased computational speed, and reduced computational complexity. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating a method in one embodiment of the present disclosure; Figure 2 A simplified diagram for calculating the eccentric compression bearing capacity of HPC-RC combined eccentrically compressed columns; Figure 3 Another simplified diagram for calculating the eccentric compression bearing capacity of HPC-RC combined eccentrically compressed columns; Figure 4 This is a network architecture diagram of a two-branch heterogeneous neural network model. Figure 5 This is a convergence curve showing the total loss of the model during training. Figure 6 This is a convergence curve of the regression loss during the training process; Figure 7 Discrete plot of regression prediction for total steel reinforcement area; Figure 8 A graph for evaluating the accuracy of diameter classification for eccentrically compressed columns; Figure 9 This is a schematic diagram of the system structure in one embodiment of the present disclosure. Detailed Implementation

[0022] The technical solutions of this disclosure will now be described in detail with reference to the accompanying drawings and preferred embodiments.

[0023] See Figure 1 This embodiment provides a design method for an HPC-RC combined eccentrically compressed column, including the following steps: Step 1: Define the design space, which includes the design requirement space and the design parameter space. The design data in the design requirements include the axial pressure design value of the eccentrically compressed column, the eccentricity of the eccentrically compressed column, the calculated length of the eccentrically compressed column, and the thickness of the high-performance concrete shell. The design data in the design parameters include the diameter of the eccentrically compressed column, the diameter of the reinforcing bars, and the number of reinforcing bars. For each design data item in the design requirement space, stratified sampling is performed, and the stratified sampling results are combined to generate the design requirement data combination. Step 2: For each combination of design requirements data, the optimal design data combination is selected based on engineering specification constraints, load-bearing capacity constraints, and cost-effectiveness values. A design data sample set is then constructed based on the optimal design data combination. Step 3: Construct a dual-branch heterogeneous neural network model. The model includes an input layer for receiving and normalizing a set of design data samples, a shared feature extraction layer for extracting high-dimensional nonlinear features from the normalization results, a diameter prediction classification branch for predicting the diameter of eccentrically compressed columns based on the high-dimensional nonlinear features, and a reinforcement area prediction regression branch for predicting the total reinforcement area of ​​eccentrically compressed columns based on the high-dimensional nonlinear features. Step 4: Construct a loss function consisting of focus loss and mean squared error loss, and train the dual-branch heterogeneous neural network model using the designed data sample set; Step 5: Obtain the design requirements of the eccentrically compressed column to be designed as input to the trained bi-branch heterogeneous neural network model, and use the trained bi-branch heterogeneous neural network model to predict the diameter of the eccentrically compressed column and the total reinforcement area of ​​the eccentrically compressed column.

[0024] The method will be described in detail below.

[0025] Step 1: Define the design space, which includes the design requirement space and the design parameter space. That is, define the design requirement space and the design parameter space. Perform stratified sampling on each requirement in the design requirement space, and combine the stratified sampling results to generate the design requirement combination.

[0026] The design space includes design data, some of which belongs to the design requirements space and some of which belongs to the design parameters space.

[0027] Design space refers to the multidimensional data set consisting of all freely adjustable design variables and their allowed value ranges during the design process.

[0028] The design requirement space refers to the set of design requirements proposed before the design of HPC-RC combined eccentrically compressed columns based on the working conditions of the upper load, geological conditions, and the preset thickness of the HPC shell. It represents the specific combination of working conditions for the column. The design parameter space refers to the set of adjustable variables that meet design requirements, which constitute a specific design scheme for a column.

[0029] The design requirements (i.e., the design requirement space) include at least four requirements, which means at least four design data points, including: the design value of the axial pressure of the eccentrically compressed column. Eccentricity of an eccentrically compressed column Calculated length of eccentrically compressed column High-performance concrete shell thickness ; The design data in the design parameters (i.e., the design parameter space) includes the diameter of the eccentrically compressed column. , diameter of reinforcing bars Number of steel bars The specific parameter meanings and calculation diagrams are as follows: Figure 2 and Figure 3 As shown; To ensure the comprehensiveness of the training samples and the model's ability under extreme conditions, this embodiment employs a stratified sampling strategy, as an example rather than a limitation, to sample the axial pressure design value. Divided into 4 intervals, eccentricity Divide into 4 intervals and calculate the length. Divided into 3 sections, high-performance concrete shell thickness The system is divided into four intervals. For each requirement in the design requirements, samples are taken from each interval. These samples are then combined to form a design requirement combination. In this embodiment, stratified sampling is performed within each interval layer, generating 12,000 design requirement combinations. This ensures that the samples uniformly cover various working conditions, including large and small eccentricities and high and low axial pressures.

[0030] The data in each design requirement combination includes the axial pressure design value of the eccentrically compressed column sampled in an axial pressure design value range, the eccentricity of the eccentrically compressed column sampled in an eccentricity range, the calculated length of the eccentrically compressed column sampled in a calculated length range, and the thickness of the high-performance concrete shell sampled in a high-performance concrete shell thickness range.

[0031] Step 2: For each design requirement data combination, the design data is screened to obtain the optimal design data combination based on engineering specification constraints, load-bearing capacity constraints, and cost-effectiveness values. The design data sample set is then constructed based on the optimal design data combination and serves as the training dataset for subsequent steps.

[0032] The design data in the filter design data refers to the design data in the design requirements space and the design data in the design parameter space.

[0033] The design data combination is the combination of design requirement data and the corresponding design parameter combination.

[0034] In this embodiment, as a preferred embodiment rather than a limitation, the design data sample set includes the diameter of the eccentrically compressed column, the total reinforcement area of ​​the eccentrically compressed column, and the design data in the design requirements, that is, the axial pressure design value of the eccentrically compressed column, the eccentricity of the eccentrically compressed column, the calculated length of the eccentrically compressed column, the thickness of the high-performance concrete shell, the diameter of the eccentrically compressed column, and the total reinforcement area of ​​the eccentrically compressed column. Understandably, in one embodiment, the data in the preferred design data set includes the axial compressive design value of the eccentrically compressed column, the eccentricity of the eccentrically compressed column, the calculated length of the eccentrically compressed column, the thickness of the high-performance concrete shell, the diameter of the eccentrically compressed column, and the total reinforcement area of ​​the eccentrically compressed column. The data in the design data sample set includes: the axial compressive design value of the eccentrically compressed column in the preferred design data set, the eccentricity of the eccentrically compressed column in the preferred design data set, the calculated length of the eccentrically compressed column in the preferred design data set, the thickness of the high-performance concrete shell in the preferred design data set, the diameter of the eccentrically compressed column in the preferred design data set, and the total reinforcement area of ​​the eccentrically compressed column determined based on the data in the preferred design data set.

[0035] In other embodiments, the design data sample set includes design data from design requirements and design data from design parameters.

[0036] It is understood that the eccentrically loaded column is an abbreviation for HPC-RC combined eccentrically loaded column.

[0037] In this embodiment, selecting the preferred design data combination includes the following steps: Step 201: For each design requirement data combination, iterate through the design data in the design parameters, perform engineering specification checks, and retain the data that meets the structural requirements according to the engineering specification constraints. That is, retain the design data combinations that comply with the engineering specification constraints. The engineering specification checks include: slenderness ratio check, reinforcement ratio check, and rebar spacing check.

[0038] Step 202: Calculate the bearing capacity of the design data combination that meets the engineering specification constraints, and retain the design data combination that meets the bearing capacity constraints.

[0039] A simplified structural diagram for calculating the eccentric compression bearing capacity of the HPC-RC composite eccentrically compressed column is shown below. Figure 2 and Figure 3 As shown, Figure 2 The axial pressure (corresponding to the design value of axial pressure) is shown in the figure. And the equivalent eccentricity after considering second-order effects. It still shows the eccentricity of the axial force with respect to the centroidal axis of the cross section. Core concrete radius Circumferential radius of reinforcing bar distribution High-performance concrete shell thickness , diameter of reinforcing bars . Figure 3 The diagram shows the compression and tension zones of the HPC-RC combined eccentrically compressed column, as well as the rounded corner corresponding to the half-compression angle. .

[0040] The calculation process in step 202 strictly follows these steps: Based on mechanical theory, the eccentricity amplification factor is calculated. The formula is as follows: in, This is the eccentricity amplification factor; The calculated length of the component; The effective height of the cross-section; For the cross-sectional height, The diameter of the eccentrically compressed column in the HPC-RC combination is... For high-performance concrete shell thickness; Eccentricity, also known as the eccentricity of an eccentrically compressed column, or the eccentricity of the axial force about the centroidal axis of the cross section. The influence coefficient of load eccentricity on cross-sectional curvature: The coefficient representing the influence of the slenderness ratio of the component on the curvature of the cross section: in, Describes the minimum value function; A nonlinear equilibrium equation for the half-pressure angle was established and solved using the bisection method to obtain the half-pressure angle of the eccentrically compressed member. The accuracy of the bisection method was controlled within 10°. -5 The equilibrium equation comprehensively considers the contributions of the internal ordinary concrete, high-performance concrete, and longitudinal reinforcement to the bearing capacity, and the equation is as follows: In the formula: in, It is the half-pressure angle; This is the eccentricity amplification factor; This represents the balance error between the eccentricity of internal forces and the eccentricity of external forces within a cross section. This represents the normalized bending moment integral term; This represents the normalized axial force integral term; The eccentricity of the axial force about the centroidal axis of the cross section; The core concrete radius; The radius of the circumference where the reinforcing bars are distributed; For reinforcement ratio, This represents the total cross-sectional area of ​​the reinforcing bars. The core concrete cross-sectional area; For high-performance concrete reinforcement ratio, The cross-sectional area of ​​the high-performance concrete shell; The center radius of the high-performance concrete shell, For high-performance concrete shell thickness; This refers to the design value of the core concrete compressive strength. This is the design value for the strength of the reinforcing steel. This refers to the design value of the compressive strength of high-performance concrete. Design value of tensile strength of high-performance concrete; It is the height coefficient of the tension zone reinforcement relative to the limiting compression zone.

[0041] Calculate the ultimate bearing capacity of the section using the eccentricity amplification factor and the half-pressure angle. The formula is as follows: in, The core concrete cross-sectional area; This refers to the design value of the core concrete compressive strength. The half-pressure angle is obtained by the bisection method; This is the height coefficient of the tension zone reinforcement relative to the limiting compression zone. The core reinforcement ratio; This is the design value for the strength of the reinforcing steel. For high-performance concrete reinforcement ratio; This refers to the design strength value for high-performance concrete. To achieve high-performance concrete compressive strength, This refers to the tensile strength of high-performance concrete.

[0042] Determine whether the calculated ultimate bearing capacity of the section meets the bearing capacity constraint conditions. If it does, retain the design data combination corresponding to the ultimate bearing capacity of the section; otherwise, do not retain the design data combination.

[0043] In other words, only design data combinations that satisfy the safety mechanics range are retained. Understandably, a certain design requirement may not have a design data combination that satisfies the load-bearing capacity constraint; that is, there is no solution for that design requirement combination.

[0044] Step 203: Calculate the cost-effectiveness index (CPI) of the design data combinations that meet the bearing capacity constraints. Select the design data combination with the highest CPI as the optimal design sample for this design requirement, i.e., the preferred design data combination. Record the diameter of the eccentrically compressed column in the HPC-RC combination. Index, total reinforcement area of ​​eccentrically compressed columns .

[0045] The axial pressure design value corresponding to the optimal design sample eccentricity Calculated length of HPC-RC combined eccentrically compressed column High-performance concrete shell thickness HPC-RC combined eccentric compression column diameter The total reinforcement area of ​​an eccentrically compressed column Forming a combination involves constructing a design data sample set for HPC-RC combined eccentrically compressed columns, which will be used for training in subsequent steps.

[0046] Step 3: Construct a two-branch heterogeneous neural network model. The network architecture is as follows: Figure 4 As shown.

[0047] The two-branch heterogeneous neural network model consists of an input layer, a shared feature extraction layer, and an output layer. The two branches refer to a classification branch and a regression branch, specifically a diameter prediction classification branch and a reinforcement area prediction regression branch. The diameter prediction classification branch is used to predict the diameter of the eccentrically compressed column based on high-dimensional nonlinear characteristics, while the reinforcement area prediction regression branch is used to predict the total reinforcement area of ​​the eccentrically compressed column based on high-dimensional nonlinear characteristics.

[0048] The input layer is used to receive a set of design data samples and to normalize the received data to obtain a normalized result in order to eliminate the influence of dimensions and accelerate convergence. The shared feature extraction layer is used to extract high-dimensional nonlinear features from the normalized processing result output by the input layer. The shared feature extraction layer includes: multiple first fully connected layers as a whole, which are used to increase the dimensionality of the normalized processing result through multiple transformations.

[0049] The first fully connected layer includes a batch normalization layer and a ReLU activation function layer. The batch normalization layer in each first fully connected layer forces the input distribution of each layer to be stable, preventing internal covariate shifts and providing a regularization effect. The ReLU activation function layer in each first fully connected layer introduces nonlinear computation using the ReLU activation function to fit complex nonlinear computational relationships, as shown in the following formula: in, This is the linear output value of the previous fully connected layer. Indicates taking and The maximum value in.

[0050] In this embodiment, a total of four sequentially arranged first fully connected layers are used, with the dimensions changing sequentially from 4 to 64, 64 to 128, 128 to 256, and 265 to 128. This means the shared feature extraction layer has a 4-dimensional input and a 128-dimensional output. When the model is used for prediction, the first fully connected layer has a 4-dimensional input, corresponding to the design requirements data in the design data sample set, namely the axial pressure design value of the eccentrically loaded column, the eccentricity of the eccentrically loaded column, the calculated length of the eccentrically loaded column, and the thickness of the high-performance concrete shell. Preferably, the first fully connected layer also includes a Dropout layer (also known as a random deactivation layer). Specifically, Dropout layers are only present in the first three first fully connected layers. In this embodiment, 20% of the neurons are randomly deactivated during training to force the network to not depend on specific paths, thereby improving the model's generalization ability and preventing overfitting.

[0051] In one embodiment, the nonlinear high-dimensional features can represent the nonlinear complex high-order coupling relationship between design parameters and the bending moment and axial force that the column can bear, which is mainly expressed by the bearing capacity calculation formula.

[0052] The output layer includes a diameter prediction classification branch and a reinforcement area prediction regression branch. The diameter prediction classification branch predicts the standard diameter based on high-dimensional nonlinear characteristics and outputs the prediction probability for each standard diameter category. The reinforcement area regression branch predicts the total reinforcement area based on high-dimensional nonlinear characteristics and outputs the total reinforcement area of ​​the eccentrically compressed column. Continuous predicted values.

[0053] The diameter prediction classification branch is specifically used to output the probability of each standard diameter using the Softmax formula. Let the logits vector (logits representing log odds) output by the classification head be... ,in, The number of categories for the standard diameter of HPC-RC combined eccentrically compressed columns. The category indicating the standard diameter of the HPC-RC combined eccentrically compressed column is: Category The logarithmic probability value, The category indicating the standard diameter of the HPC-RC combined eccentrically compressed column is: Category The logarithmic probability value, The category indicating the standard diameter of the HPC-RC combined eccentrically compressed column is: Category The logarithmic probability value, The category indicating the standard diameter of the HPC-RC combined eccentrically compressed column is category. The log-odds value of the category Predicted probability The calculation formula is as follows: in, The total number of categories with standard diameters for HPC-RC combined eccentrically compressed columns; Category coding for standard diameters, It also indicates the category code for standard diameter.

[0054] The diameter prediction classification branch includes a second fully connected layer. The reinforcement area regression branch includes two third fully connected layers, both used for dimensionality reduction, with dimensionality reductions from 128 to 64 and from 64 to 1, respectively.

[0055] Step 4: Construct a loss function and train a two-branch heterogeneous neural network model using the designed data sample set; the loss function consists of focus loss and mean squared error loss.

[0056] In this embodiment, the design data sample set is divided into a training set, a validation set, and a test set in a 6:2:2 ratio. The training set is used to train a dual-branch heterogeneous neural network model, the validation set is used to validate the model, and the test set is used to test the model.

[0057] The loss function can be called a hybrid loss function. A hybrid loss function is constructed to collaboratively optimize discrete diameter selection tasks and continuous steel reinforcement tasks.

[0058] The regression task uses the mean squared error (MSE) loss function to calculate the loss between the predicted value and the true mechanical value; the classification task uses the focal loss function to address the problem of the highly uneven distribution of effective and excellent designs in the design parameter samples, as shown in the following formula: in, This represents the total loss of the model; These are the regression task weight coefficients; This represents the true value of the regression task. This represents the model prediction for the regression task. This represents the mean squared error loss of the regression task, used in the regression branch. Specifically, , The sample number, For the real category, For the model to predict the category, For the sample The true category, For the sample The model predicts the category. Batch size; For the classification task weight coefficients, For the true labels of the classification task, To predict probabilities for models used in classification tasks. This represents the focus loss for a classification task, used in the classification branch. Specifically, , For the model to predict the first Each sample belongs to the true category. The probability, This is the category balance coefficient. For focusing parameters.

[0059] Preferred setting Prioritize ensuring the accuracy of the cross-sectional diameter.

[0060] In this embodiment, a dynamic learning rate adjustment strategy is adopted for training the dual-branch heterogeneous neural network model. Specifically, the Adam optimizer (adaptive moment estimation) is used for parameter updates, with an initial learning rate of 0.001. The ReduceLROnPlateau (Reduce Learning Rate on Plateau, a type of learning rate scheduler) mechanism reduces the learning rate to 0.5 times its original value when the validation loss does not decrease within 10 epochs, until the minimum learning rate is reached. An early stopping mechanism is introduced, which completes training when the validation set loss does not improve within 100 epochs, and the model with the lowest validation set loss is taken as the best model to ensure the model's generalization ability.

[0061] In this embodiment, batch size Set to 64; Focus parameter Set to 2.0; Classification task weight coefficient Set to 2.0, regression task weight coefficient Set it to 1.0 to prioritize the accuracy of cross-section selection.

[0062] The convergence curve during training is shown in the figure. Figure 5 and Figure 6 As shown, Figure 5 This is a convergence curve for the total loss of the model. Figure 6 This is a convergence curve for the regression task loss (regression loss).

[0063] In this embodiment, the regression prediction discrete graph of the total steel reinforcement area (both horizontal and vertical axes are in square meters) and the column diameter classification accuracy evaluation graph (both horizontal and vertical axes are diameters, both in meters) are shown as follows: Figure 7 and Figure 8 As shown.

[0064] Figure 7 The distribution of the actual and predicted values ​​of the steel reinforcement area in the test set is shown. The dashed line represents the predicted value being exactly equal to the actual value. The vast majority of the scattered points closely surround the ideal fitting line, indicating that the model has successfully learned the complex sublinear mapping relationship between the design requirement parameters and the optimal steel reinforcement area.

[0065] Figure 8 The confusion matrix for the diameter classification task is shown below. The horizontal axis represents the diameter labels predicted by the model, and the vertical axis represents the true optimal diameter labels. The diagonal elements are significantly higher than the off-diagonal elements, indicating that the model has a high recognition accuracy and almost no serious cross-class errors, which verifies that the model has learned profound mechanical laws.

[0066] Step 5, Model Application: Obtain the axial pressure design value of the HPC-RC combined eccentrically compressed column to be designed. eccentricity Calculate length High-performance concrete shell thickness Data preprocessing is performed; the preprocessed data is then input into the dual-branch heterogeneous neural network model trained in step 4, and this model is used to predict the diameter of the HPC-RC combined eccentrically compressed column. Reinforcement area of ​​HPC-RC combined eccentrically compressed columns The optimal design is the design output by the model.

[0067] Preferably, the design method further includes step 6: verifying whether the prediction results of the trained dual-branch heterogeneous neural network model meet preset requirements. If the preset requirements are met, the final design scheme is obtained; otherwise, the design fails, requiring manual review and adjustment of the design parameters by the designer.

[0068] Specifically, the predicted optimal design is parameter-mapped, and the results are verified to ensure it meets engineering specifications and design safety requirements. The parameter mapping involves calculating parameters for verification based on the model's input and output from step 5. The specific content of the design safety requirements is not limited here.

[0069] More specifically, the preset requirements are engineering specification constraints and bearing capacity constraints, that is, to verify whether the prediction results meet the engineering specification constraints and bearing capacity constraints. In other words, the design safety requirements include bearing capacity constraints.

[0070] See Figure 9 This embodiment provides an HPC-RC combined eccentric compression column design system, including: The space definition module is used to define the design space, which includes the design requirements space and the design parameter space. The design data in the design requirements includes the axial pressure design value of the eccentrically compressed column, the eccentricity of the eccentrically compressed column, the calculated length of the eccentrically compressed column, and the thickness of the high-performance concrete shell. The design data in the design parameters includes the diameter of the eccentrically compressed column, the diameter of the reinforcing bar, and the number of reinforcing bars. The stratified sampling module is used to perform stratified sampling on each piece of design data in the design requirement space, and to combine the stratified sampling results to generate a combination of design requirement data. The data sample construction module is used to select the optimal design data combination for each design requirement data combination based on engineering specification constraints, load-bearing capacity constraints, and cost-effectiveness values, and to construct a design data sample set based on the optimal design data combination. The model building module is used to build a dual-branch heterogeneous neural network model. The model includes an input layer for receiving and normalizing a set of design data samples, a shared feature extraction layer for extracting high-dimensional nonlinear features from the normalization results, a diameter prediction classification branch for predicting the diameter of an eccentrically compressed column based on the high-dimensional nonlinear features, and a reinforcement area prediction regression branch for predicting the total reinforcement area of ​​an eccentrically compressed column based on the high-dimensional nonlinear features. The model training module is used to construct a loss function consisting of focus loss and mean squared error loss, and to train the dual-branch heterogeneous neural network model using the designed data sample set. The acquisition and prediction module is used to acquire the design requirements of the eccentrically compressed column to be designed as input to the trained bi-branch heterogeneous neural network model, and to use the trained bi-branch heterogeneous neural network model to predict the diameter of the eccentrically compressed column and the total reinforcement area of ​​the eccentrically compressed column.

[0071] In this embodiment, the data sample construction module is specifically used to: traverse the design data in the design parameters for each design requirement data combination, perform engineering specification checks, and retain design data combinations that comply with engineering specification constraints; calculate the bearing capacity of design data combinations that comply with engineering specification constraints, and retain those that comply with bearing capacity constraints; calculate the cost-effectiveness value of design data combinations that comply with bearing capacity constraints, and select the design data combination with the highest cost-effectiveness value as the preferred design data combination; the engineering specification checks include slenderness ratio checks, reinforcement ratio checks, and rebar spacing checks.

[0072] In this embodiment, the step of calculating the bearing capacity of the design data combination that conforms to the engineering specification constraints and retaining the design data combination that conforms to the bearing capacity constraints specifically involves: calculating the eccentricity amplification factor; establishing a nonlinear equilibrium equation about the half-pressure angle and solving it using the bisection method to obtain the half-pressure angle of the eccentrically compressed member; calculating the ultimate bearing capacity of the section using the eccentricity amplification factor and the half-pressure angle; and determining whether the calculated ultimate bearing capacity of the section meets the bearing capacity constraints. If it does, the design data combination corresponding to the ultimate bearing capacity of the section is retained.

[0073] The system also includes a verification module, which is used to verify whether the prediction results of the trained dual-branch heterogeneous neural network model meet the preset requirements.

[0074] In specific implementation, the system can be designed as an HPC-RC combined eccentric compression column by referring to the method in any of the above embodiments. The specific implementation steps will not be repeated here.

[0075] The method according to this embodiment can implement an electronic device, the electronic device including: a memory; one or more processors; one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for executing an HPC-RC combined eccentric compression column design method according to any of the above embodiments.

[0076] This embodiment also provides a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the steps of the HPC-RC combined eccentrically compressed column design method described in any of the above embodiments.

[0077] The effects of the HPC-RC combined eccentric compression column design method, system, and medium of the present invention are as follows: By performing stratified sampling on each design data item in the design requirement space, and combining the stratified sampling results to generate a design requirement data combination, and then selecting the optimal design data combination through engineering specification constraints, bearing capacity constraints, and cost-effectiveness values ​​to construct a design data sample set, coupled with a neural network model with a loss function consisting of focus loss and mean square error loss, the stability and accuracy of the design are improved. Accurate designs can be provided for all HPC-RC combined eccentrically compressed columns. By designing and training a two-branch heterogeneous neural network model, multi-branch tasks can be collaboratively optimized and predicted. The shared feature extraction layer accelerates the model's processing speed, enabling collaborative training of different types of tasks. The model can simultaneously output the reinforcement area and the diameter of the HPC-RC combined eccentrically compressed column, reducing the computational burden, improving the computational speed, and reducing the computational load.

[0078] Specifically, it has the following effects: (1) Improved design efficiency: Without relying on human experience and a large number of iterative calculations, the trained neural network can quickly (in milliseconds) provide the optimal design, which greatly improves the design efficiency of HPC-RC combined eccentric compression column, and takes into account both low computational load and computational accuracy.

[0079] (2) The design's excellence was ensured: Through full parameter traversal, screening of mechanical and cost-effectiveness indicators, and engineering specification constraints, the safety and economy of the model design were guaranteed, and it met the specification requirements. Furthermore, a rigorous and accurate mechanical calculation method was constructed to ensure the authenticity and reliability of the load-bearing capacity label.

[0080] (3) Multi-branch task collaborative optimization: A dual-branch heterogeneous neural network model is established, which uses a shared feature extraction layer to accelerate the processing speed of the model and realize collaborative training of different types of tasks. The model can simultaneously output the reinforcement area and the diameter of the HPC-RC combined eccentric compression column, and the output results are highly practical.

[0081] (4) High design accuracy: The stratified sampling method is used when generating training samples to ensure the robustness of the model design; the focus loss function is introduced to improve the design capability for rare optimal solutions and improve the design accuracy.

[0082] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0083] The embodiments described above are merely illustrative of several implementations of this disclosure, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this disclosure, and these all fall within the protection scope of this disclosure. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A design method for an HPC-RC combined eccentrically compressed column, characterized in that, include: Step 1: Define the design space, which includes the design requirement space and the design parameter space. For each design data item in the design requirement space, perform stratified sampling, and combine the stratified sampling results to generate the design requirement data combination. The design data in the design requirements includes the axial pressure design value of the eccentrically compressed column, the eccentricity of the eccentrically compressed column, the calculated length of the eccentrically compressed column, and the thickness of the high-performance concrete shell. The design data in the design parameters includes the diameter of the eccentrically compressed column, the diameter of the reinforcing bar, and the number of reinforcing bars. Step 2: For each combination of design requirements data, the optimal design data combination is selected based on engineering specification constraints, load-bearing capacity constraints, and cost-effectiveness values. A design data sample set is then constructed based on the optimal design data combination. Step 3: Construct a dual-branch heterogeneous neural network model. The model includes an input layer for receiving and normalizing a set of design data samples, a shared feature extraction layer for extracting high-dimensional nonlinear features from the normalization results, a diameter prediction classification branch for predicting the diameter of eccentrically compressed columns based on the high-dimensional nonlinear features, and a reinforcement area prediction regression branch for predicting the total reinforcement area of ​​eccentrically compressed columns based on the high-dimensional nonlinear features. Step 4: Construct a loss function consisting of focus loss and mean squared error loss, and train the dual-branch heterogeneous neural network model using the designed data sample set; Step 5: Obtain the design requirements of the eccentrically compressed column to be designed as input to the trained bi-branch heterogeneous neural network model, and use the trained bi-branch heterogeneous neural network model to predict the diameter of the eccentrically compressed column and the total reinforcement area of ​​the eccentrically compressed column.

2. The HPC-RC combined eccentrically compressed column design method according to claim 1, characterized in that, Step 2 includes: Step 201: For each design requirement data combination, the design data in the design parameters are traversed and the engineering specifications are checked. The design data combination that conforms to the engineering specifications is retained according to the engineering specifications constraints. The engineering specifications checks include slenderness ratio check, reinforcement ratio check, and rebar spacing check. Step 202: Calculate the bearing capacity of the design data combination that meets the engineering specification constraints, and retain the design data combination that meets the bearing capacity constraints. Step 203: Calculate the cost-effectiveness value of the design data combination that meets the bearing capacity constraints, and select the design data combination with the highest cost-effectiveness value as the preferred design data combination.

3. The design method for an HPC-RC combined eccentrically compressed column according to claim 2, characterized in that, The calculation process in step 202 is as follows: calculate the eccentricity amplification factor; establish a nonlinear equilibrium equation for the half-pressure angle and solve it using the bisection method to obtain the half-pressure angle of the eccentrically compressed member; calculate the ultimate bearing capacity of the section using the eccentricity amplification factor and the half-pressure angle; determine whether the calculated ultimate bearing capacity of the section meets the bearing capacity constraint conditions, and if it does, retain the design data combination corresponding to the ultimate bearing capacity of the section.

4. The HPC-RC combined eccentrically compressed column design method according to claim 3, characterized in that, The nonlinear equilibrium equation is: In the formula: in, It is the half-pressure angle; This is the eccentricity amplification factor; The eccentricity of the eccentrically compressed column; This represents the balance error between the eccentricity of internal forces and the eccentricity of external forces within a cross section. This represents the normalized bending moment integral term; This represents the normalized axial force integral term; The core concrete radius; The radius of the circumference where the reinforcing bars are distributed; This refers to the reinforcement ratio; For high-performance concrete reinforcement ratio; The radius of the center of the high-performance concrete shell; This refers to the design value of the core concrete compressive strength. This is the design value for the strength of the reinforcing steel. This refers to the design value of the compressive strength of high-performance concrete. Design value of tensile strength of high-performance concrete; This is the height coefficient of the tension zone reinforcement relative to the limiting compression zone. The formula for calculating the ultimate bearing capacity of a cross section is: in, The ultimate bearing capacity of the section. The area of ​​the core concrete cross section.

5. The design method for an HPC-RC combined eccentrically compressed column according to claim 1, characterized in that, The design data sample set includes: the axial pressure design value of the eccentrically compressed column in the preferred design data combination, the eccentricity of the eccentrically compressed column in the preferred design data combination, the calculated length of the eccentrically compressed column in the preferred design data combination, the thickness of the high-performance concrete shell in the preferred design data combination, the diameter of the eccentrically compressed column in the preferred design data combination, and the total reinforcement area of ​​the eccentrically compressed column determined based on the data in the preferred design data combination.

6. The design method for an HPC-RC combined eccentrically compressed column according to claim 1, characterized in that, The loss function is: in, This represents the total loss of the model; These are the regression task weight coefficients; This represents the true value of the regression task. This represents the model prediction for the regression task. This represents the mean squared error loss of the regression task. , For the sample number, For the real category, For the model to predict the category, For the sample The true category, For the sample The model predicts the category. For batch size, For the classification task weight coefficients, For the true labels of the classification task, To predict probabilities for models used in classification tasks. This represents the focus loss for the classification task. , For the model to predict the first Each sample belongs to the true category. The probability, This is the category balance coefficient. For focusing parameters.

7. The design method for an HPC-RC combined eccentrically compressed column according to claim 1, characterized in that, The method further includes a step of verifying whether the prediction results of the trained dual-branch heterogeneous neural network model meet preset requirements.

8. The design method for an HPC-RC combined eccentrically compressed column according to claim 1, characterized in that, The shared feature extraction layer includes four sequentially arranged first fully connected layers. Each first fully connected layer includes a batch normalization layer and a ReLU activation function layer. The first three first fully connected layers include a random deactivation layer.

9. A design system for an HPC-RC combined eccentrically compressed column, characterized in that, include: The space definition module is used to define the design space, which includes the design requirements space and the design parameter space. The design data in the design requirements includes the axial pressure design value of the eccentrically compressed column, the eccentricity of the eccentrically compressed column, the calculated length of the eccentrically compressed column, and the thickness of the high-performance concrete shell. The design data in the design parameters includes the diameter of the eccentrically compressed column, the diameter of the reinforcing bar, and the number of reinforcing bars. The stratified sampling module is used to perform stratified sampling on each piece of design data in the design requirement space, and to combine the stratified sampling results to generate a combination of design requirement data. The data sample construction module is used to select the optimal design data combination for each design requirement data combination based on engineering specification constraints, load-bearing capacity constraints, and cost-effectiveness values, and to construct a design data sample set based on the optimal design data combination. The model building module is used to build a dual-branch heterogeneous neural network model. The model includes an input layer for receiving and normalizing a set of design data samples, a shared feature extraction layer for extracting high-dimensional nonlinear features from the normalization results, a diameter prediction classification branch for predicting the diameter of an eccentrically compressed column based on the high-dimensional nonlinear features, and a reinforcement area prediction regression branch for predicting the total reinforcement area of ​​an eccentrically compressed column based on the high-dimensional nonlinear features. The model training module is used to construct a loss function consisting of focus loss and mean squared error loss, and to train the dual-branch heterogeneous neural network model using the designed data sample set. The acquisition and prediction module is used to acquire the design requirements of the eccentrically compressed column to be designed as input to the trained bi-branch heterogeneous neural network model, and to use the trained bi-branch heterogeneous neural network model to predict the diameter of the eccentrically compressed column and the total reinforcement area of ​​the eccentrically compressed column.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes instructions that, when executed on a computer, cause the computer to perform a design method for an HPC-RC combined eccentrically compressed column as described in any one of claims 1 to 8.

Images