A design method for steel fiber reinforced concrete materials based on multiple performance requirements
By optimizing the mix proportion of steel fiber reinforced concrete through deep learning and genetic algorithms, the problem that traditional methods are difficult to adapt to the needs of complex engineering projects has been solved. This has enabled high-precision multi-performance prediction and optimization, and promoted the sustainable development of building materials.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEFEI UNIV OF TECH
- Filing Date
- 2024-07-17
- Publication Date
- 2026-05-26
Smart Images
Figure CN119092011B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of steel fiber reinforced concrete design technology, specifically relating to a steel fiber reinforced concrete material design method based on multiple performance requirements. Background Technology
[0002] Concrete, as a fundamental construction material, has a wide range of applications, including buildings, roads, bridges, and other critical infrastructure. With increasingly demanding engineering requirements, the focus of concrete design has shifted from solely pursuing high strength to simultaneously considering workability, mechanical properties, and durability. Against this backdrop, fiber-reinforced concrete, especially steel fiber reinforced concrete, is widely used in the construction of important infrastructure such as highways, bridges, and airports due to its excellent crack resistance, tensile strength, and toughness.
[0003] The workability and mechanical properties of steel fiber reinforced concrete depend not only on its basic components, such as the proportions of cement, water, fine aggregate, and coarse aggregate, but also on various factors including fiber volume content, fiber aspect ratio, fiber type, and fiber spatial distribution. Studies have shown that uniform fiber dispersion contributes to improved mechanical and durability properties of concrete, particularly in inhibiting crack initiation and propagation. Therefore, optimizing the design of steel fiber reinforced concrete materials to meet diverse performance requirements has become a key issue in engineering applications.
[0004] Traditional fiber-reinforced concrete mix design methods often rely on empirical formulas and experimental data, making it difficult to adapt to complex and personalized application needs. Summary of the Invention
[0005] In view of this, the purpose of the present invention is to provide a design method for steel fiber reinforced concrete materials based on multiple performance requirements, in order to solve the above-mentioned technical problems.
[0006] To achieve the above objectives, the present invention provides the following technical solution:
[0007] A design method for steel fiber reinforced concrete materials based on multiple performance requirements, comprising:
[0008] Obtain raw data on the parameters and properties of steel fiber reinforced concrete to determine the training dataset;
[0009] The initial convolutional neural network model is trained based on the training dataset to obtain a convolutional neural network performance prediction model;
[0010] A virtual function for the performance of steel fiber reinforced concrete is constructed using a convolutional neural network performance prediction module as the objective function;
[0011] Obtain the constraints on the basic performance parameters corresponding to the objective function; among which, the constraints satisfy the optimal working and mechanical performance indicators of the designed steel fiber reinforced concrete.
[0012] Based on the genetic algorithm, the concrete mix proportion is optimized for the objective function under each constraint, the optimal solution that meets the preset performance index is selected, and the optimal solution set is determined.
[0013] The optimal solution set was selected using the TOPSIS method to obtain the optimal design scheme for steel fiber reinforced concrete materials.
[0014] Furthermore, raw data on the parameters and properties of steel fiber reinforced concrete were obtained to determine the training dataset, including:
[0015] The raw data of steel fiber reinforced concrete parameters and properties were obtained through literature review or experimental data acquisition; the raw data included the mix design parameters, slump, compressive strength and splitting tensile strength of steel fiber reinforced concrete.
[0016] The original data is cleaned to obtain a cleaned dataset; the cleaned dataset includes 5 types of input parameters and 3 types of output parameters.
[0017] The cleaned dataset is divided into a training set and a test set in an 8:2 ratio; the training dataset includes both the training set and the test set.
[0018] Furthermore, the initial convolutional neural network model is trained using the training dataset to obtain a convolutional neural network performance prediction model, including:
[0019] Construct an initial convolutional neural network model; wherein the construction of the initial convolutional neural network model includes an input layer, two convolutional layers, two ReLU activation layers that non-linearly activate the output of the convolutional layers, a Dropout layer that randomly suppresses neurons with a preset probability to reduce overfitting, a fully connected layer for generating predicted values, and a regression layer as the output layer.
[0020] The initial convolutional neural network model is trained using a training set; the loss function used during model training is mean squared error, and the adaptive time estimation method is used as the optimization algorithm.
[0021] After training, a test set is obtained to predict the results of the initial convolutional neural network model after training. The prediction performance of the initial convolutional neural network model after training is evaluated by the coefficient of determination R2, root mean square error RMSE, mean absolute error MAE, and mean bias error MBE.
[0022] When the prediction results meet the preset conditions, the performance prediction model of the convolutional neural network is determined.
[0023] Furthermore, the constraints on the basic performance parameters corresponding to the objective function include: constraints on the water-cement ratio parameter, constraints on the sand ratio parameter, constraints on the steel fiber volume content parameter, constraints on the steel fiber aspect ratio parameter, and constraints on the water-reducing agent dosage parameter.
[0024] Furthermore, the TOPSIS method is used to screen the optimal solution set, obtaining the optimal design scheme for steel fiber reinforced concrete materials, including:
[0025] Obtain the original decision matrix;
[0026] The original decision matrix is transformed into a normalized decision matrix using the vector normalization method.
[0027] Based on the normalized decision matrix and the corresponding attribute weights, a weighted normalized matrix is constructed;
[0028] Obtain the positive and negative ideal solutions for each attribute in the weighted normalization matrix, and calculate the Euclidean distance from each solution in the optimal solution set to the positive and negative ideal solutions respectively;
[0029] Calculate the comprehensive evaluation index of each scheme based on the corresponding Euclidean distance;
[0030] The scheme with the highest comprehensive evaluation index is obtained as the optimal design scheme for steel fiber reinforced concrete materials.
[0031] The beneficial effects of this invention are as follows:
[0032] 1. This invention integrates deep learning methods with steel fiber reinforced concrete material design, which, compared with traditional methods, not only has higher accuracy and efficiency, but also stronger generalization ability.
[0033] 2. This invention considers the slump, compressive strength and splitting tensile strength of steel fiber reinforced concrete at the same time, which is more in line with engineering requirements than the traditional method that only considers compressive strength.
[0034] Other advantages, objectives, and features of the invention will be set forth in the following description and will be apparent to those skilled in the art in some respects, or may be learned by practice of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.
[0035] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0036] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0037] Figure 1 This is a multi-objective optimization design diagram of a steel fiber reinforced concrete material design method based on multiple performance requirements in an embodiment of the present invention;
[0038] Figure 2 This is a diagram of the convolutional neural network architecture in a steel fiber reinforced concrete material design method based on multiple performance requirements, as described in an embodiment of the present invention.
[0039] Figure 3 This is a comparison chart of slump prediction results in a steel fiber reinforced concrete material design method based on multiple performance requirements in an embodiment of the present invention.
[0040] Figure 4 This is a comparison chart of the predicted compressive strength in a steel fiber reinforced concrete material design method based on multiple performance requirements in an embodiment of the present invention.
[0041] Figure 5 This is a comparison chart of the predicted splitting tensile strength in a steel fiber reinforced concrete material design method based on multiple performance requirements in an embodiment of the present invention.
[0042] Figure 6 This is a comparison chart of evaluation indicators for the slump prediction model of steel fiber reinforced concrete in a steel fiber reinforced concrete material design method based on multiple performance requirements in an embodiment of the present invention.
[0043] Figure 7 This is a comparison chart of the residuals of the slump prediction model for steel fiber reinforced concrete in a steel fiber reinforced concrete material design method based on multiple performance requirements in an embodiment of the present invention.
[0044] Figure 8 This is a comparison chart of evaluation indicators for the compressive strength prediction model of steel fiber reinforced concrete in a steel fiber reinforced concrete material design method based on multiple performance requirements in an embodiment of the present invention.
[0045] Figure 9 This is a comparison chart of errors in the compressive strength prediction model of steel fiber reinforced concrete in a steel fiber reinforced concrete material design method based on multiple performance requirements in an embodiment of the present invention.
[0046] Figure 10 This is a comparison chart of evaluation indicators for the prediction model of splitting tensile strength of steel fiber reinforced concrete in a steel fiber reinforced concrete material design method based on multiple performance requirements in an embodiment of the present invention.
[0047] Figure 11 This is a comparison chart of errors in the prediction model for the splitting tensile strength of steel fiber reinforced concrete in a steel fiber reinforced concrete material design method based on multiple performance requirements in an embodiment of the present invention.
[0048] Figure 12This is a comparison chart of test results and predicted results of steel fiber reinforced concrete in a steel fiber reinforced concrete material design method based on multiple performance requirements in an embodiment of the present invention.
[0049] Figure 13 This is a schematic diagram of the three-dimensional Pareto front and optimal solution in a steel fiber reinforced concrete material design method based on multiple performance requirements in an embodiment of the present invention. Detailed Implementation
[0050] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.
[0051] Please see Figure 1 This invention proposes a design method for steel fiber reinforced concrete materials based on multiple performance requirements, including:
[0052] Obtain raw data on the parameters and properties of steel fiber reinforced concrete to determine the training dataset;
[0053] Obtain raw data on the parameters and properties of steel fiber reinforced concrete, and determine the training dataset, including:
[0054] The raw data of steel fiber reinforced concrete parameters and properties were obtained through literature review or experimental data acquisition; the raw data included the mix design parameters, slump, compressive strength and splitting tensile strength of steel fiber reinforced concrete.
[0055] The original data is cleaned to obtain a cleaned dataset; the cleaned dataset includes 5 types of input parameters and 3 types of output parameters.
[0056] The cleaned dataset is divided into a training set and a test set in an 8:2 ratio; the training dataset includes both the training set and the test set.
[0057] The initial convolutional neural network model is trained based on the training dataset to obtain a convolutional neural network performance prediction model;
[0058] The initial convolutional neural network model is trained using the training dataset to obtain a convolutional neural network performance prediction model, including:
[0059] Construct an initial convolutional neural network model; wherein the construction of the initial convolutional neural network model includes an input layer, two convolutional layers, two ReLU activation layers that non-linearly activate the output of the convolutional layers, a Dropout layer that randomly suppresses neurons with a preset probability to reduce overfitting, a fully connected layer for generating predicted values, and a regression layer as the output layer.
[0060] The initial convolutional neural network model is trained using a training set; the loss function used during model training is mean squared error, and the adaptive time estimation method is used as the optimization algorithm.
[0061] After training, a test set is obtained to predict the results of the initial convolutional neural network model after training. The prediction performance of the initial convolutional neural network model after training is evaluated by the coefficient of determination R2, root mean square error RMSE, mean absolute error MAE, and mean bias error MBE.
[0062] When the prediction results meet the preset conditions, the performance prediction model of the convolutional neural network is determined.
[0063] A virtual function for the performance of steel fiber reinforced concrete is constructed using a convolutional neural network performance prediction module as the objective function;
[0064] Obtain the constraints on the basic performance parameters corresponding to the objective function; among which, the constraints satisfy the optimal working and mechanical performance indicators of the designed steel fiber reinforced concrete.
[0065] The constraints on the basic performance parameters corresponding to the objective function include: constraints on the water-cement ratio parameter, constraints on the sand ratio parameter, constraints on the steel fiber volume content parameter, constraints on the steel fiber aspect ratio parameter, and constraints on the water-reducing agent dosage parameter.
[0066] Based on the genetic algorithm, the concrete mix proportion is optimized for the objective function under each constraint, the optimal solution that meets the preset performance index is selected, and the optimal solution set is determined.
[0067] Under a given objective function, the genetic algorithm optimizes the concrete mix proportion and finds the best solution that meets the predetermined performance index, forming the optimal solution set, i.e., the Pareto front.
[0068] The optimal solution set was selected using the TOPSIS method to obtain the optimal design scheme for steel fiber reinforced concrete materials;
[0069] The optimal solution set was selected using the TOPSIS method to obtain the optimal design scheme for steel fiber reinforced concrete materials, including:
[0070] Obtain the original decision matrix;
[0071] The original decision matrix is transformed into a normalized decision matrix using the vector normalization method.
[0072] Based on the normalized decision matrix and the corresponding attribute weights, a weighted normalized matrix is constructed;
[0073] Obtain the positive and negative ideal solutions for each attribute in the weighted normalization matrix, and calculate the Euclidean distance from each solution in the optimal solution set to the positive and negative ideal solutions respectively;
[0074] Calculate the comprehensive evaluation index of each scheme based on the corresponding Euclidean distance;
[0075] The scheme with the highest comprehensive evaluation index is obtained as the optimal design scheme for steel fiber reinforced concrete materials.
[0076] The working principle of the above technical solution is as follows: To ensure the normal operation of the model and sufficient data storage space, the hardware and software configuration adopted in this invention is as follows: Windows 11 operating system equipped with Python 3.7.2 and Matlab R2021b, providing powerful programming and scientific computing support; internally equipped with an AMD Ryzen 75800H with Radeon Graphics processor, paired with an NVIDIA GeForce RTX 3050-6G graphics processor, ensuring computing power and graphics processing capabilities; in addition, it is equipped with 16GB of memory and a 520GB solid-state drive, ensuring multi-tasking capabilities and efficient data access;
[0077] Step 1: Dataset Acquisition and Preprocessing
[0078] (1) Based on literature review, we obtained the mix design parameters, slump, compressive strength and splitting tensile strength data of 294 groups of steel fiber reinforced concrete.
[0079] (2) Perform data cleaning on the original data, including handling outliers and null values, which involves deleting the data item.
[0080] (3) The dataset has 5 input parameters and 3 output parameters, which are divided into training set and test set in an 8:2 ratio;
[0081] Step 2: Constructing the Convolutional Neural Network Model:
[0082] The architecture diagram of the convolutional neural network constructed in this invention is as follows: Figure 2 As shown;
[0083] Step 3: Result Validation and Analysis of the Convolutional Neural Network Model:
[0084] (1) Results of each performance prediction model
[0085] Training and test set results of a CNN-based collapse prediction model, such as Figure 3 As shown ( Figure 3The first image in the figure represents the training set, and the second image represents the test set. It can be seen that the RMSE of the training set is 7.69 mm, and the RMSE of the test set is 9.87 mm, both with errors less than 10%, indicating high prediction accuracy of the model. The R² values for the training and test sets are 0.98 and 0.97, respectively, indicating a high degree of fit. Furthermore, the MAE values for the training and test sets are 5.02 mm and 7.23 mm, respectively, and the MBE values are -1.67 mm and 0.15 mm, respectively. This demonstrates that the collapse prediction model established in this invention can accurately map the relationship between various factors and collapse. The results for the training and test sets are not significantly different, indicating strong generalization ability.
[0086] Figure 4 ( Figure 4 The first image (training set) and the second image (test set) show the training and test set results of the CNN-based compressive strength prediction model. It can be seen that the RMSE of the training set is 2.14 MPa, and the RMSE of the test set is 2.48 MPa, with an error of less than 5%, which is within the allowable error range (<15%) for compressive strength. The R² values for the training and test sets are 0.98 and 0.96, respectively, indicating a high degree of fit. Furthermore, the MAE values for the training and test sets are 1.21 MPa and 1.69 MPa, respectively, and the MBBE values are 0.13 MPa and 0.80 MPa, respectively.
[0087] Figure 5 ( Figure 5 The first image (training set) and the second image (test set) show the training and test set results of the CNN-based splitting tensile strength prediction model. The comparison reveals that the RMSE of the training and test sets are 0.18 MPa and 0.22 MPa, respectively, with errors less than 5%, both within the allowable error range (<15%) for splitting tensile strength. The R² of the training and test sets are 0.97 and 0.94, respectively, indicating a high degree of fit. Furthermore, the MAE of the training and test sets are 0.11 MPa and 0.19 MPa, respectively, and the MBE is 0.02 MPa and 0.11 MPa, respectively. The results show that the model effectively reflects the relationship between various factors and splitting tensile strength and demonstrates high generalization ability.
[0088] (2) Comparison with traditional models
[0089] Based on the same dataset, this invention uses traditional machine learning algorithms—BP neural network, support vector machine, random forest and radial basis neural network—to predict the slump, compressive strength and tensile strength of steel fiber reinforced concrete, and compares the results with the convolutional neural network model established in this invention.
[0090] Figure 6The evaluation metrics of steel fiber reinforced concrete slump prediction models are compared. The model performance ranking is as follows: CNN > SVR > RBF > BPNN > RF. The CNN model has significantly lower RMSE and MAE values than the other models, and the highest R², indicating the highest accuracy. This is due to the weight sharing mechanism, which greatly reduces the number of parameters in the CNN model compared to fully connected networks, lowering model complexity and overfitting risk. Regarding MBE, the models show both positive and negative results, but the MBE values of CNN, RF, and BPNN are relatively small, indicating smaller prediction bias. A comparison of the training and test sets reveals… Figure 6 ( Figure 6 The first image in the image is the training set, and the second image is the test set. The CNN models have similar shapes and the smallest area formed. The RMSE, MAE, and R2 values of the CNN models are not much different, and the models have the best generalization ability.
[0091] Combination Figure 7 It can be seen that the residuals (true value - predicted value) of the SVR model and the CNN model fluctuate less and are close to 0, indicating that the model performance is better than the other models; however, the SVR model has more outliers (circles in the figure), and the model prediction reliability is poor; therefore, the CNN model is the best choice among these five models.
[0092] Figure 8 ( Figure 8 The first image (training set, second image (test set)) shows the comparison results of evaluation metrics for steel fiber reinforced concrete compressive strength prediction models. It can be seen that the model performance ranking is as follows: CNN > SVR > RBF > RF > BPNN. Since the compressive strength data is relatively abundant, the RF model is no longer the worst performing model. The BPNN model performs the worst due to its inherent limitations. The CNN model has lower RMSE and MAE than other models, and the largest R², indicating the highest model accuracy. From the MBE perspective, the error of the CNN model is overestimated, but within the allowable error range for compressive strength. Comparing the training and test sets reveals... Figure 8 The shapes of the CNN models are more similar, and the RMSE, MAE and R2 values of the CNN models are not much different. The model has the lowest risk of overfitting and the best generalization ability.
[0093] Combination Figure 9 It can be seen that the SVR and CNN models have smaller box heights, indicating smaller residual fluctuations, with the residuals fluctuating around zero, resulting in better model performance compared to other models; however, the SVR model has more outliers, leading to poorer predictive reliability; therefore, the CNN model is the best choice among these five models.
[0094] Figure 10 ( Figure 10The first image (training set, second image (test set)) shows the comparison results of evaluation metrics for the steel fiber reinforced concrete splitting tensile strength prediction model. It can be seen that, similar to the results of the steel fiber reinforced concrete slump prediction model, the model performance ranking is as follows: CNN > SVR > RBF > BPNN > RF. The RF model performs the worst because of the limited splitting tensile strength data. The CNN model has lower RMSE and MAE than other models, and the largest R², indicating the highest model accuracy. From the MBE perspective, the error of the CNN model is overestimated, but within the allowable error range for splitting tensile strength. Comparing the training and test sets reveals... Figure 10 The shapes of the CNN models are more similar, and the RMSE, MAE and R2 values of the CNN models are not much different. The model has the lowest risk of overfitting and the best generalization ability.
[0095] Combination Figure 11 It can be seen that the SVR and CNN models have smaller box heights, indicating smaller residual passives, and the residuals fluctuate around zero, resulting in better model performance compared to other models; however, the SVR model has more outliers, leading to poorer predictive reliability; therefore, the CNN model is the best choice among these five models.
[0096] (3) Experimental verification results
[0097] The experimental results of slump, compressive strength, and splitting tensile strength of steel fiber reinforced concrete were compared with the prediction results of a model based on a convolutional neural network, and the errors were calculated. Figure 12 As shown ( Figure 12 The first graph corresponds to slump, the second to compressive strength, and the third to splitting tensile strength. It can be seen that the slump error of steel fiber reinforced concrete is no greater than 13%, the compressive strength error is no greater than 11%, and the splitting compressive strength error is no greater than 8%, all of which are within the allowable error range. The prediction accuracy and reliability of the model have been verified.
[0098] Step 4: Design method for steel fiber reinforced concrete materials based on multiple performance requirements:
[0099] (1) A performance prediction model for the slump, compressive strength, and splitting tensile strength of steel fiber reinforced concrete was established using a convolutional neural network. Virtual functions were then constructed for each model, denoted as “virtualFunction_SF”, “virtualFunction_FCU”, and “virtualFunction_FST”, respectively. The virtual functions were constructed using Matlab software, and the specific code is shown in Table 1 below. The code composition of each virtual function is similar. These virtual functions can accept input parameters generated by optimization algorithms. The input data is standardized through preprocessing steps and reshaped into a data format suitable for the convolutional neural network model. Based on the convolutional neural network, the slump, compressive strength, and splitting tensile strength of steel fiber reinforced concrete are predicted, and the corresponding performance prediction values are output. The prediction results are inversely normalized to restore the original data range, ensuring the practicality and interpretability of the results.
[0100] The virtual function code is as follows:
[0101] Slump:
[0102]
[0103] Compressive strength:
[0104]
[0105] Splitting tensile strength:
[0106]
[0107]
[0108] (2) Based on the provisions of "Steel Fiber Reinforced Concrete" (JG / T 3064-2015) and "Specification for Mix Proportion Design of Ordinary Concrete" (JGJ55-2019), as well as the research results of basic requirements for concrete mix proportions and related literature; to avoid extremely unreasonable results, it is ensured that the water-cement ratio, sand ratio, steel fiber volume content, steel fiber aspect ratio, and water-reducing agent dosage must be within the reasonable range determined by the specifications or experiments; the range of the determining factors is shown in the table below:
[0109] Table 1 - Constraints for Multi-Objective Optimization Design of Steel Fiber Reinforced Concrete
[0110]
[0111] (3) Taking the minimum and maximum values of slump as 70 mm and 90 mm, the minimum and maximum values of compressive strength as 30 MPa and 40 MPa, and the minimum and maximum values of splitting tensile strength as 3 MPa and 4 MPa, respectively;
[0112] The optimal solution selected by the TOPSIS method and the three-dimensional Pareto front obtained by multi-objective optimization based on genetic algorithm are as follows: Figure 13 As shown in the figure; the blue dots represent the set of all potential optimal solutions, each dot representing a concrete mix design and its performance on three different performance objectives; the green dots highlight the optimal solutions that meet the performance requirements, showing that these solutions not only have superior performance but are also feasible within the preset slump range; the red dots represent the optimal solution with the best overall performance selected from all possible solutions using the TOPSIS method; this optimal solution balances multiple performance objectives and, according to the TOPSIS method's evaluation criteria, is considered the solution that is closest to the ideal solution in all considered factors;
[0113] Table 2 shows the optimal solution set selected by the TOPSIS method, including concrete mix design schemes numbered 1 to 3. All of these schemes successfully met the performance requirements. Scheme number 3 was identified by the TOPSIS method as the solution with the best overall performance, having the highest compressive strength, and meeting the requirements for splitting tensile strength and slump.
[0114] Table 2 - Optimal solution set after TOPSIS method selection
[0115]
[0116]
[0117] The beneficial effects of the above technical solution are as follows: By collecting a large amount of data and constructing a prediction model based on deep learning methods, the workability and mechanical properties of steel fiber reinforced concrete can be accurately predicted. Simultaneously, a multi-objective optimization algorithm is combined to propose a material design method for steel fiber reinforced concrete based on multiple performance requirements. This not only improves the various properties of steel fiber reinforced concrete but also avoids the waste of raw materials caused by extensive trial mixing, providing a theoretical basis for the sustainable development of building materials. Furthermore, this solution constructs a steel fiber reinforced concrete performance database based on literature and experimental data statistics, and establishes prediction models for the slump, compressive strength, and splitting tensile strength of steel fiber reinforced concrete based on convolutional neural networks. The models have high accuracy (R² reaches 0.97, 0.96, and 0.94 respectively) and better generalization ability compared to traditional models (BP neural network, support vector machine, random forest, and radial basis neural network). Finally, a material design method for steel fiber reinforced concrete based on multiple performance requirements is proposed by combining genetic algorithms and the TOPSIS (Technique for Order Preference by Similarity to an Ideal Solution) method. This addresses the problem that traditional fiber reinforced concrete mix design methods often rely on empirical formulas and experimental data, making it difficult to adapt to complex and personalized application needs.
[0118] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.
Claims
1. A design method for steel fiber reinforced concrete materials based on multiple performance requirements, characterized in that, include: Obtain raw data on the parameters and properties of steel fiber reinforced concrete to determine the training dataset; the raw data includes the mix design parameters, slump, compressive strength, and splitting tensile strength of steel fiber reinforced concrete. The initial convolutional neural network model is trained based on the training dataset to obtain a convolutional neural network performance prediction model; A virtual function for the performance of steel fiber reinforced concrete is constructed using a convolutional neural network performance prediction module as the objective function; Obtain the constraints of the basic performance parameters corresponding to the objective function; among them, the constraints satisfy the optimal working performance and mechanical performance index of the designed steel fiber reinforced concrete; the constraints of the basic performance parameters corresponding to the objective function include: constraints of water-cement ratio parameter, constraints of sand ratio parameter, constraints of steel fiber volume content parameter, constraints of steel fiber aspect ratio parameter and constraints of water-reducing agent dosage parameter. Based on the genetic algorithm, the concrete mix proportion is optimized for the objective function under each constraint, the optimal solution that meets the preset performance index is selected, and the optimal solution set is determined. The optimal solution set was selected using the TOPSIS method to obtain the optimal design scheme for steel fiber reinforced concrete materials.
2. The design method for steel fiber reinforced concrete materials based on multiple performance requirements according to claim 1, characterized in that, Obtain raw data on the parameters and properties of steel fiber reinforced concrete, and determine the training dataset, including: The raw data of steel fiber reinforced concrete parameters and properties were obtained through literature review or experimental data acquisition. The original data is cleaned to obtain a cleaned dataset; the cleaned dataset includes 5 types of input parameters and 3 types of output parameters. according to The cleaned dataset is divided into training and test sets according to the ratio; the training dataset includes both training and test sets.
3. The design method for steel fiber reinforced concrete materials based on multiple performance requirements according to claim 1, characterized in that, The initial convolutional neural network model is trained using the training dataset to obtain a convolutional neural network performance prediction model, including: Construct an initial convolutional neural network model; wherein the construction of the initial convolutional neural network model includes an input layer, two convolutional layers, two ReLU activation layers that non-linearly activate the output of the convolutional layers, a Dropout layer that randomly suppresses neurons with a preset probability to reduce overfitting, a fully connected layer for generating predicted values, and a regression layer as the output layer. The initial convolutional neural network model is trained using a training set; the loss function used during model training is mean squared error, and the adaptive time estimation method is used as the optimization algorithm. After training, a test set is obtained to predict the results of the initial convolutional neural network model after training. The prediction performance of the initial convolutional neural network model after training is evaluated by the coefficient of determination R2, root mean square error RMSE, mean absolute error MAE, and mean bias error MBE. When the prediction results meet the preset conditions, the performance prediction model of the convolutional neural network is determined.
4. The design method for steel fiber reinforced concrete materials based on multiple performance requirements according to claim 1, characterized in that, The optimal solution set was selected using the TOPSIS method to obtain the optimal design scheme for steel fiber reinforced concrete materials, including: Obtain the original decision matrix; The original decision matrix is transformed into a normalized decision matrix using the vector normalization method. Based on the normalized decision matrix and the corresponding attribute weights, a weighted normalized matrix is constructed; Obtain the positive and negative ideal solutions for each attribute in the weighted normalization matrix, and calculate the Euclidean distance from each solution in the optimal solution set to the positive and negative ideal solutions respectively; Calculate the comprehensive evaluation index of each scheme based on the corresponding Euclidean distance; The scheme with the highest comprehensive evaluation index is obtained as the optimal design scheme for steel fiber reinforced concrete materials.