High-ductility concrete batching performance table data increasing and expanding method and system

Abstract

The invention relates to the technical field of electrical digital data processing using machine learning, and provides a high-ductility concrete batching performance table data increasing and expanding method and system.The high-ductility concrete batching performance table data increasing and expanding method comprises the following steps that S1, a threshold is set; s2, generating a first data table; s3, testing in groups; s4, performing model training; s5, generating a second data table; s6, grouping an increasing and expanding table; and S7, carrying out outlier screening. Compared with the prior art, the method has the advantages that the test data can be increased and expanded at a remarkably low cost, and a large amount of performance data can be effectively acquired with a small test quantity, so that the acquisition cost of the performance table data of the high-ductility concrete ingredients is greatly reduced, and an effective technical support is provided for in-depth research on preparation of the high-ductility concrete from the machine-made sand and the stone powder.

Description

Technical Field

[0001] The invention relates to a method and system for increasing and expanding data of a high-ductility concrete batching performance table, and belongs to the technical field of electrical digital data processing using machine learning. Background Art

[0002] At present, there are few studies on the preparation of high ductility concrete using machine-made sand and stone powder, and only a few literature reports. WuHL conducted a comprehensive experimental study on the influence of machine-made sand particle morphology parameters on the mechanical properties of HDCC, analyzed the fiber dispersion in HDCC, and established an empirical equation between the morphology parameters of machine-made sand particles and the mechanical properties. Luo Baifu studied the influence of coarse sand on the toughness and ductility of HDCC. The results showed that the fresh performance, toughness and ductility of coarse sand HDCC were significantly lower than those of fine sand HDCC. Based on the principle of the most compact particle packing, a certain proportion of stone powder was added, and the toughness and ductility of coarse sand and stone powder HDCC were significantly improved, and stone powder helped to improve the flexural and compressive strength of coarse sand HDCC.

[0003] However, at present, there are relatively more studies on the application of machine-made sand and stone powder in ordinary concrete, but fewer studies on their application in high-ductility concrete. The effects of adding machine-made sand and an appropriate amount of stone powder on concrete can be summarized as improving the mechanical properties of concrete, but it will cause the fluidity of concrete to deteriorate. In high-ductility concrete, poor fluidity will make fiber dispersion difficult, further affecting the mechanical properties of high-ductility concrete. Therefore, it is necessary to conduct in-depth research on the preparation of high-ductility concrete using machine-made sand and stone powder.

[0004] However, actual research requires a large amount of actual test data support. The actual test generally uses each component combination to cast into a concrete block with a side length of 0.5m~1m to form a test unit, and the test unit is tested by instruments. On the one hand, the pouring and solidification time is relatively long, and a large number of tests means extremely high labor costs. On the other hand, the atmospheric environment interference needs to be properly excluded during the test, which requires that it be carried out in the open space of the factory. A large number of tests means a large area of ​​land. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a method and system for augmenting data of a high-ductility concrete batch performance table. The method and system for augmenting data of a high-ductility concrete batch performance table can augment test data at a significantly low cost, and effectively obtain a large amount of performance data with a small test amount, thereby greatly reducing the cost of obtaining data of the high-ductility concrete batch performance table, and providing effective technical support for in-depth research on the preparation of high-ductility concrete using machine-made sand and stone powder.

[0006] The present invention is achieved through the following technical solutions.

[0007] The present invention provides a method for expanding data of a high-ductility concrete batching performance table, comprising the following steps: S1. Setting thresholds: prompting the user to set the upper and lower limits of stone powder fineness, stone powder content, PVA fiber aspect ratio, PVA fiber content, PP fiber aspect ratio, and PP fiber content and waiting for input, and forming a data limit table of threshold values ​​with the user input values; S2. Generate a first data table: within the range of the data restriction table, set 100 to 150 groups of different values ​​of stone powder fineness, stone powder addition amount, PVA fiber aspect ratio, PVA fiber addition amount, PP fiber aspect ratio, and PP fiber addition amount to form a first data table, and output the first data table; S3, group test: prompt the user to add stone powder, PVA fiber and PP fiber to each group of machine-made sand of constant weight according to the first data table grouping for group test, measure the truncated cone fluidity, flow cone fluidity, consistency, segregation resistance, bonding strength, flexural tensile strength, compressive strength, static compressive elastic modulus, splitting tensile strength, water absorption, wear resistance for each group, pause the process and wait for user input, obtain the corresponding input value of the user and combine them to obtain the measured data table; S4, model training: using the first data table and the measured data table as training data to train a multi-branch prediction network; S5. Generate a second data table: within the range of the data restriction table, set at least 1000 groups of different values ​​of stone powder fineness, stone powder addition, PVA fiber aspect ratio, PVA fiber addition, PP fiber aspect ratio, and PP fiber addition to form a data table to be tested; S6, grouping augmented table: using a multi-branch prediction network, calculate the prediction augmented table consisting of truncated cone fluidity, flow cone fluidity, consistency, segregation resistance, bonding strength, flexural tensile strength, compressive strength, static compressive elastic modulus, splitting tensile strength, water absorption, and wear resistance according to the data table to be tested, and combine the data table to be tested and the prediction augmented table in groups to obtain a parameter data set; S7, outlier screening: Use the KNN algorithm to perform clustering calculations on the parameter data set based on the Euclidean distance, retain the largest cluster data and delete other cluster data, obtain the augmented data set and output it.

[0008] The multi-branch prediction network consists of a feature processing network, a backbone network and multiple sub-item prediction networks. The input of the feature processing network is a combination of added value values. Each sub-item prediction network outputs one of the measured value combinations. The input of each sub-item prediction network is all the outputs of the backbone network.

[0009] The feature processing network is a feedforward neural network, the backbone network is a convolutional neural network, and the item prediction network is a ResNet network.

[0010] The backbone network adopts the VGG19 network and removes the softmax layer.

[0011] In step S2, the values ​​in the first data table are randomly distributed within the range of the data restriction table.

[0012] The random distribution follows a Gaussian distribution with a variance greater than 2.0.

[0013] In step S5, the values ​​in the data table to be tested are evenly distributed within the range of the data restriction table.

[0014] The present invention also provides a high-ductility concrete batching performance table data augmentation system, comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described above; and further comprising a display, wherein in steps S1 and S3, a user is prompted by displaying on the display, and in steps S2 and S7, output is performed in two ways: generating a list file on the memory and displaying on the display.

[0015] The beneficial effects of the present invention are that the test data can be expanded at a significantly low cost, and a large amount of performance data can be effectively obtained with a smaller test volume, thereby greatly reducing the cost of obtaining the performance table data of high-ductility concrete ingredients, and providing effective technical support for in-depth research on the preparation of high-ductility concrete using machine-made sand and stone powder. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of a process of an embodiment of the present invention; Figure 2 yes Figure 1 Model architecture diagram of the multi-branch prediction network used in the model training step. DETAILED DESCRIPTION

[0017] The technical solution of the present invention is further described below, but the scope of protection claimed is not limited to the description.

[0018] A first embodiment of the present invention relates to Figure 1 A method for augmenting data of a high ductility concrete batching performance table is shown, comprising the following steps: S1. Setting thresholds: prompting the user to set the upper and lower limits of stone powder fineness, stone powder content, PVA fiber aspect ratio, PVA fiber content, PP fiber aspect ratio, and PP fiber content and waiting for input, and forming a data limit table of threshold values ​​with the user input values; S2. Generate a first data table: within the range of the data restriction table, set 100 to 150 groups of different values ​​of stone powder fineness, stone powder addition amount, PVA fiber aspect ratio, PVA fiber addition amount, PP fiber aspect ratio, and PP fiber addition amount to form a first data table, and output the first data table; S3, group test: prompt the user to add stone powder, PVA fiber and PP fiber to each group of constant component machine-made sand according to the adding value table for group test. Each group is measured for truncated cone fluidity, flow cone fluidity, consistency, segregation resistance, bonding strength, flexural tensile strength, compressive strength, static compressive elastic modulus, splitting tensile strength, water absorption, wear resistance, pause the process and wait for user input, obtain the corresponding input value of the user and combine it to obtain the measured data table; Generally, the constant weight of each group of machine-made sand is 1kg, and the machine-made sand used in common concrete can be selected; the truncated cone fluidity and flow cone fluidity are measured based on the "Technical Specifications for Application of Cement-based Grouting Materials" (GB / T 50448-2015) standard; the consistency and segregation resistance are measured based on the "Standard Test Method for Performance of Ordinary Concrete Mixtures" (GB / T50080-2016) standard; the bond strength, flexural tensile strength, compressive strength, static compressive elastic modulus, splitting tensile strength, water absorption, and wear resistance are measured based on the "Standard Test Method for Physical and Mechanical Properties of Concrete" (GB / T 50081-2019) standard; S4, model training: using the first data table and the measured data table as training data to train a multi-branch prediction network; S5. Generate a second data table: within the range of the data restriction table, set at least 1000 groups of different values ​​of stone powder fineness, stone powder addition, PVA fiber aspect ratio, PVA fiber addition, PP fiber aspect ratio, and PP fiber addition to form a data table to be tested; S6, grouping augmented table: using a multi-branch prediction network, calculate the prediction augmented table consisting of truncated cone fluidity, flow cone fluidity, consistency, segregation resistance, bonding strength, flexural tensile strength, compressive strength, static compressive elastic modulus, splitting tensile strength, water absorption, and wear resistance according to the data table to be tested, and combine the data table to be tested and the prediction augmented table in groups to obtain a parameter data set; S7, outlier screening: Use the KNN algorithm to perform clustering calculations on the parameter data set based on the Euclidean distance, retain the largest cluster data and delete other cluster data, obtain the augmented data set and output it.

[0019] Based on this, this implementation is essentially to train a model based on the data actually measured by the user, and then use the trained model to predict performance data, wherein step S3 is to set the content that needs to be input, generally by displaying the name of the corresponding item on the screen, and displaying a numerical input box after the name, and the user needs to spend a certain amount of time to conduct experiments and measurements according to the prompts to obtain the measured data, and after the user fills in the corresponding numerical value in the numerical input box, the measured data can be obtained by obtaining the numerical value input by the user; when prediction is performed in step S6, the numerical item distribution of the predicted expansion table and the measured data table is consistent.

[0020] The second embodiment of the present invention is substantially the same as the first embodiment, mainly in that the multi-branch prediction network is as follows: Figure 2 As shown, it consists of a feature processing network, a backbone network and multiple sub-item prediction networks. The input of the feature processing network is a combination of added value values. Each sub-item prediction network outputs one of the measured value combinations. The input of each sub-item prediction network is all the outputs of the backbone network.

[0021] It is easy to understand that the numerical combination of added materials is a numerical group composed of stone powder fineness, stone powder addition, PVA fiber aspect ratio, PVA fiber addition, PP fiber aspect ratio, and PP fiber addition, and the final output item required is a measurement value combination, namely, truncated cone fluidity, flow cone fluidity, consistency, anti-segregation, bonding strength, flexural tensile strength, compressive strength, static compressive elastic modulus, splitting tensile strength, water absorption, and wear resistance, a total of 11 items, each output item has a different meaning, so 11 sub-item prediction networks are used for calculation respectively, so the feature processing network and the backbone network actually realize the role of feature extraction, among which the feature processing network generally uses a smaller network mainly for data preprocessing, and the backbone network in this case has a larger network mainly used to calculate the characteristic data of the mixture (not expressed in a way that humans can understand), and the sub-item prediction network completes the result prediction calculation. When the model is trained by back propagation, each sub-item prediction network corresponds to a fixed output item. It is worth noting that the data transmitted between the feature processing network, the backbone network, and the sub-item prediction network only involves the internal data of the branch prediction network. The above description of the internal input and output of the feature processing network, the backbone network, and the sub-item prediction network is used to indicate the data flow direction, has no external meaning, and can be flexibly adjusted.

[0022] Furthermore, the feature processing network is a feedforward neural network, the backbone network is a convolutional neural network, and the item prediction network is a ResNet network.

[0023] Furthermore, the backbone network adopts the VGG19 network and removes the softmax layer.

[0024] Further, in step S2, the values ​​in the first data table are randomly distributed within the range of the data restriction table.

[0025] Furthermore, the random distribution follows a Gaussian distribution with a variance greater than 2.0.

[0026] Furthermore, in step S5, the values ​​in the data table to be tested are evenly distributed within the range of the data restriction table.

[0027] The third embodiment of the present invention relates to a data augmentation system for a high-ductility concrete batch performance table, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the first embodiment; and further comprising a display, wherein in steps S1 and S3, a user is prompted by displaying the information on the display, and in steps S2 and S7, the information is output in two ways: generating a list file on the memory and displaying the information on the display.

Claims

1. A method for augmenting data of a high ductility concrete batching performance table, characterized in that: The following steps are involved: S1. Setting thresholds: prompting the user to set the upper and lower limits of stone powder fineness, stone powder content, PVA fiber aspect ratio, PVA fiber content, PP fiber aspect ratio, and PP fiber content and waiting for input, and forming a data limit table of threshold values ​​with the user input values; S2. Generate a first data table: within the range of the data restriction table, set 100 to 150 groups of different values ​​of stone powder fineness, stone powder addition amount, PVA fiber aspect ratio, PVA fiber addition amount, PP fiber aspect ratio, and PP fiber addition amount to form a first data table, and output the first data table; S3, group test: prompt the user to add stone powder, PVA fiber and PP fiber to each group of constant component machine-made sand according to the adding value table for group test. Each group is measured for truncated cone fluidity, flow cone fluidity, consistency, segregation resistance, bonding strength, flexural tensile strength, compressive strength, static compressive elastic modulus, splitting tensile strength, water absorption, wear resistance, pause the process and wait for user input, obtain the corresponding input value of the user and combine it to obtain the measured data table; S4, model training: using the first data table and the measured data table as training data to train a multi-branch prediction network; S5. Generate a second data table: within the range of the data restriction table, set at least 1000 groups of different values ​​of stone powder fineness, stone powder addition, PVA fiber aspect ratio, PVA fiber addition, PP fiber aspect ratio, and PP fiber addition to form a data table to be tested; S6, grouping augmented table: using a multi-branch prediction network, calculate the prediction augmented table consisting of truncated cone fluidity, flow cone fluidity, consistency, segregation resistance, bonding strength, flexural tensile strength, compressive strength, static compressive elastic modulus, splitting tensile strength, water absorption, and wear resistance according to the data table to be tested, and combine the data table to be tested and the prediction augmented table in groups to obtain a parameter data set; S7, outlier screening: Use the KNN algorithm to perform clustering calculations on the parameter data set based on the Euclidean distance, retain the largest cluster data and delete other cluster data, obtain the augmented data set and output it.

2. The method for expanding data of a high ductility concrete batching performance table according to claim 1, characterized in that: The multi-branch prediction network consists of a feature processing network, a backbone network and multiple sub-item prediction networks. The input of the feature processing network is a combination of added value values. Each sub-item prediction network outputs one of the measured value combinations. The input of each sub-item prediction network is all the outputs of the backbone network.

3. The method for expanding data of high ductility concrete batching performance table according to claim 2, characterized in that: The feature processing network is a feedforward neural network, the backbone network is a convolutional neural network, and the item prediction network is a ResNet network.

4. The method for expanding data of a high ductility concrete batching performance table according to claim 3, characterized in that: The backbone network adopts the VGG19 network and removes the softmax layer.

5. The method for expanding data of high ductility concrete batching performance table according to claim 1, characterized in that: In step S2, the values ​​in the first data table are randomly distributed within the range of the data restriction table.

6. The method for expanding data of a high ductility concrete batching performance table according to claim 5, characterized in that: The random distribution follows a Gaussian distribution with a variance greater than 2.

0.

7. The method for expanding data of a high ductility concrete batching performance table according to claim 1, characterized in that: In step S5, the values ​​in the data table to be tested are evenly distributed within the range of the data restriction table.

8. A system for augmenting data of a high ductility concrete batching performance table, comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to claim 1; it also includes a display, in steps S1 and S3, prompting the user by displaying on the screen on the display, and in steps S2 and S7, outputting in two ways: generating a list file in the memory and displaying on the screen on the display.

Images