Method for predicting compressive strength of ordinary Portland cement based on mineral composition

Abstract

The invention discloses a method for predicting compressive strength of ordinary Portland cement based on mineral composition, which comprises the following steps of: collecting a plurality of groups of cement samples, taking the mineral composition and the specific surface area as independent variables, and taking the 28d compressive strength data as dependent variables, wherein each group of cement samples comprises the mineral composition, the specific surface area and the 28d compressive strength data; a compressive strength prediction model based on a convolutional neural network is constructed, the compressive strength prediction model comprises an input layer, a convolutional layer, a batch normalization layer, a Dropout layer, a full connection layer and a regression layer, the compressive strength prediction model is trained by using a cement sample, mineral components and specific surface area data of to-be-detected cement are input into the compressive strength prediction model, and a 28d compressive strength prediction value is obtained. According to the method, the mechanical property of the cement can be quickly and accurately predicted on the premise of not depending on the traditional 28-day test period, so that the cement quality is pre-judged in advance, and the risk of insufficient concrete strength caused by the use of cement with substandard strength is reduced to the greatest extent.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of engineering materials, and particularly relates to a method for predicting compressive strength of ordinary Portland cement based on mineral composition. BACKGROUND

[0002] Concrete is an important building material widely used in modern construction engineering, and its quality is directly related to the safety and durability of engineering structures. Among the constituent materials of concrete, cement as the basic binding material plays a decisive role in the final strength of concrete. The compressive strength of cement is an important indicator for evaluating its performance and quality, especially for concrete used in load-bearing structures, the strength grade and quality stability of cement are of great importance.

[0003] At present, the commonly used type of cement in construction engineering is ordinary Portland cement. According to the current national standards or industry specifications, the compressive strength of ordinary Portland cement needs to be determined after curing for 28 days (referred to as 28d) through standard test methods, which is used as the basis for final evaluation of its strength grade and whether it is qualified. However, the 28d compressive strength test period is relatively long, and it is often difficult to wait for the completion of the period before deciding whether to use a batch of cement in actual engineering.

[0004] Limited by the storage capacity of cement production enterprises and concrete mixing stations, as well as the requirements of construction site engineering progress and continuity, cement is usually used for pouring of concrete structures before the 28d strength test results are obtained. This "use first and evaluate later" approach has certain advantages in terms of construction efficiency, but it also brings great quality risks: if the 28d strength of cement does not meet the standard, it may lead to insufficient overall strength of the concrete structure, thereby affecting the safety and service life of the project, and even causing rework, economic losses and safety hazards.

[0005] Therefore, how to evaluate the quality of cement as early as possible and accurately, especially to predict its 28d compressive strength, without affecting the construction progress, has become a technical problem to be solved in the field of building material detection and quality control. SUMMARY

[0006] In view of the above technical problems of the prior art, the present application aims to provide a method for predicting the compressive strength of ordinary Portland cement with short time consumption and high accuracy.

[0007] To solve the above technical problems, the technical scheme adopted by the present application is as follows:

[0008] A mineral composition-based ordinary portland cement compressive strength prediction method, a plurality of groups of cement samples are collected, each group of cement samples includes mineral composition, specific surface area and 28d compressive strength data, the mineral composition and the specific surface area are taken as independent variables, and the 28d compressive strength data are taken as dependent variables; a compressive strength prediction model based on a convolutional neural network is constructed, including an input layer, a convolutional layer, a batch normalization layer, a Dropout layer, a full connection layer and a regression layer, the compressive strength prediction model is trained by using the cement samples, the mineral composition and the specific surface area data of the cement to be measured are input into the compressive strength prediction model, and a 28d compressive strength prediction value is obtained.

[0009] As optimization, the mineral composition includes tricalcium silicate, dicalcium silicate, tricalcium aluminate, tetracalcium aluminoferrite and gypsum.

[0010] As optimization, the mineral composition further includes F-glass body, S-glass body and an inactive phase, the F-glass body is an amorphous glass body phase in fly ash, and the S-glass body is an amorphous glass body phase in mineral powder.

[0011] As optimization, the compressive strength prediction model includes at least two convolutional layers and two batch normalization layers, the convolution kernel size of each convolutional layer is same and is less than or equal to 3x1, the number of convolution kernels is 2 n n is a positive integer greater than or equal to 4, the step length of each convolutional layer is 1 and the activation function is ReLU, and the random probability of the Dropout layer is set to (0.1-0.3).

[0012] As optimization, in the training process of the compressive strength prediction model, a plurality of groups of the cement samples are pretreated, a random sequence is generated by using a randperm function to disturb the order of the cement samples, the independent variables in the cement samples are normalized to the (0, 1) interval by using a mapminmax function, and finally the plurality of groups of the cement samples are divided into a training set and a test set.

[0013] As optimization, the data processing mode of the input layer is that the independent variables in the cement samples are converted into a four-dimensional tensor with a dimension of [D, 1, 1] by using a reshape function, wherein D is the parameter number of the independent variables in the cement samples, and the data type of the four-dimensional tensor is converted into a double-precision type by using a double function.

[0014] As optimization, the SGDM algorithm is used to train the compressive strength prediction model, and the initial learning rate is 0.01-0.05, the learning rate is reduced by 50%-90% every N period, 300≤N≤500, and the single training sample amount is less than or equal to 30.

[0015] Compared with the prior art, the present application has the following advantages:

[0016] (1) The prediction model has good fitting degree, high prediction accuracy and fast analysis speed;

[0017] (2) The prediction model has strong generalization ability, fully considers various influencing factors of the 28d compressive strength of cement, and has wide application range;

[0018] (3) The method can quickly and accurately predict the 28d compressive strength of cement, greatly shortens the reaction time of cement whose strength may not meet the standard, and maximally reduces the risk of insufficient strength of concrete caused by using cement whose strength does not meet the standard. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a flowchart of the present application;

[0020] Figure 2 is a prediction result graph of the training set of the present application;

[0021] Figure 3 is a prediction result graph of the test set of the present application. DETAILED DESCRIPTION

[0022] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in connection with the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the scope of protection of the present application.

[0023] The ordinary portland cement compressive strength prediction method based on mineral composition in the specific embodiment collects a plurality of groups of cement samples, each group of cement samples includes mineral composition, specific surface area and 28d compressive strength data, takes the mineral composition and specific surface area as independent variables, and takes the 28d compressive strength data as dependent variable; a compressive strength prediction model based on convolutional neural network is constructed, including an input layer, a convolutional layer, a batch normalization layer, a Dropout layer, a full connection layer and a regression layer, the cement samples are used to train the compressive strength prediction model, the mineral composition and specific surface area data of the cement to be tested are input into the compressive strength prediction model, and a 28d compressive strength prediction value is obtained.

[0024] The mineral components include tricalcium silicate, dicalcium silicate, tricalcium aluminate, tetracalcium aluminoferrite, gypsum, F-glass body, S-glass body and an inactive phase, the F-glass body is an amorphous glass body phase in fly ash, and the S-glass body is an amorphous glass body phase in mineral powder.

[0025] The compressive strength prediction model comprises at least two convolution layers and two batch normalization layers, the convolution kernel size of each convolution layer is equal and less than or equal to 3x1, the number of convolution kernels is 2 n , n is a positive integer greater than or equal to 4, the step length of each convolution layer is 1 and the activation function is ReLU, and the random probability of the Dropout layer is set to (0.1-0.3).

[0026] In the training process of the compressive strength prediction model, a plurality of groups of cement samples are pretreated, a random sequence is generated using a randperm function to disturb the order of the cement samples, the independent variables in the cement samples are normalized to the interval (0, 1) by a mapminmax function, and finally a plurality of the cement samples are divided into a training set and a test set.

[0027] The data processing mode of the input layer is that the independent variables in the cement samples are converted into four-dimensional tensors with dimensions [D, 1, 1] as characteristic variables by a reshape function, wherein D is the parameter number of the independent variables in the cement samples, and the data type of the four-dimensional tensor is converted to double precision by a double function.

[0028] The SGDM algorithm is used to train the compressive strength prediction model, and the initial learning rate is 0.01-0.05, the learning rate is reduced by 50%-90% every N period, 300≤N≤500, and the single training sample amount is ≤30.

[0029] In the specific implementation, the XRD full spectrum fitting quantitative analysis technology is used to collect the cement mineral composition data, the specific surface area of the cement is measured by the cement specific surface area determination method, and the 28d compressive strength of the cement is measured by referring to the cement mortar strength test method. Finally, the cement mineral composition and the specific surface area are used as independent variables, and the 28d compressive strength is used as a dependent variable to form a sample set. In this example, the effective sample amount is 115 groups.

[0030] To avoid the prediction model from depending on the order of data in the training process, the original data needs to be randomly arranged, and a randperm function is used to generate a random arrangement with a length of 115 to disturb the data. At the same time, in order to eliminate the dimensional differences between the characteristics, the original data is normalized by using a mapminmax function to scale the characteristic data to the interval (0, 1). According to the characteristics of the neural network and the size of the sample data set, the randomly arranged data set is divided into 100 training sets and 15 test sets.

[0031] The multiple independent variables and dependent variables in the sample data will be used as feature variables and target variables to construct the input layer of the model. The reshape function is used to convert the feature data in the training set and test set from a two-dimensional matrix to a four-dimensional tensor. The data type of the four-dimensional tensor is converted to double precision by the double function. The target variable does not need to be flattened because it has only one dimension.

[0032] A network structure containing two convolutional layers, two batch normalization layers, a Dropout layer, a fully connected layer, and a regression layer is constructed for regression prediction of input data with a size of [9, 1, 1]. The network parameters of the convolutional layer are shown in Table 1:

[0033]

[0034] Table 1

[0035] The batch normalization layer is used to normalize the output data to improve the training speed and performance. The random probability in the Dropout layer is set to 0.2 to reduce the risk of overfitting. The fully connected layer flattens the output of the convolutional layer into a one-dimensional vector and connects it to a single neuron to generate the output of the network. The regression layer is used to calculate the error between the output value and the target value and propagate it back to the network as a loss function to update the weights of the network.

[0036] The SGDM optimization algorithm is used as the training algorithm to train the above network structure. The hyperparameter settings are shown in Table 2:

[0037]

[0038] Table 2

[0039] In the trainingOptions function, SGDM is specified as the optimization algorithm, and the training set is iterated in batches. The number of samples for a single training is 25, and the training stops after 800 iterations.

[0040] The predict function is used to simulate the prediction of the training set and test set. The predict function uses the trained deep learning model to predict the given input data and performs denormalization. The performance of the prediction model is evaluated by the mean absolute error (MAE), mean relative error (MRE), and root mean square error (RMSE).

[0041] As Figure 2 and Figure 3As shown, from the training set effect diagram, the 28d compressive strength prediction model has good fitting effect and strong explanatory power, wherein R 2 In the test set, the 28d compressive strength prediction value has high degree of agreement with the true value, and the prediction ability of the prediction model is strong.

[0042] Combined with the error indicators in Table 3, the average absolute error and the average relative error of the 28d compressive strength test set are 1.250MPa and 2.607%, respectively, and the prediction results are within the acceptable range, indicating that the prediction model has strong adaptability to different cement 28d compressive strength and high accuracy, and can be used for rapid prediction of the 28d compressive strength of ordinary Portland cement.

[0043]

[0044] Table 3

[0045] Finally, it should be pointed out that the above examples are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described by referring to the preferred embodiments thereof, it should be understood by those skilled in the art that various changes in form and details can be made without departing from the spirit and scope of the present application as defined by the appended claims.

Claims

1. A method for predicting the compressive strength of ordinary Portland cement based on mineral composition, characterized in that: Several sets of cement samples were collected. Each set of cement samples included mineral composition, specific surface area, and 28-day compressive strength data. Mineral composition and specific surface area were used as independent variables, and 28-day compressive strength data was used as the dependent variable. A compressive strength prediction model based on a convolutional neural network was constructed, including an input layer, a convolutional layer, a batch normalization layer, a dropout layer, a fully connected layer, and a regression layer. The compressive strength prediction model was trained using cement samples. The mineral composition and specific surface area data of the cement to be tested were input into the compressive strength prediction model to obtain the predicted value of 28-day compressive strength.

2. The method for predicting the compressive strength of ordinary silicate cement based on mineral composition according to claim 1, characterized in that: The mineral components include tricalcium silicate, dicalcium silicate, tricalcium aluminate, tetracalcium aluminoferrite, and gypsum.

3. The method for predicting the compressive strength of ordinary Portland cement based on mineral composition according to claim 2, characterized in that: The mineral components also include F-glass, S-glass and inactive phases. F-glass is the amorphous glass phase in fly ash and S-glass is the amorphous glass phase in mineral powder.

4. The method for predicting the compressive strength of ordinary silicate cement based on mineral composition according to claim 1, characterized in that: The compressive strength prediction model includes at least two convolutional layers and two batch normalization layers. The kernel size of each convolutional layer is the same, ≤3×1, and the number of kernels is 2 in each layer. n There are 1 convolutional layer with n being a positive integer greater than or equal to 4. Each convolutional layer has a stride of 1 and an activation function of ReLU. The random probability of the Dropout layer is set to (0.1~0.3).

5. The method for predicting the compressive strength of ordinary silicate cement based on mineral composition according to claim 1, characterized in that: During the training process of the compressive strength prediction model, several groups of cement samples are preprocessed. The random sequence generated by the randperm function is used to shuffle the order of the cement samples. The independent variables in the cement samples are normalized to the (0,1) interval by the mapminmax function. Finally, the several cement samples are divided into training set and test set.

6. The method for predicting the compressive strength of ordinary Portland cement based on mineral composition according to claim 1, characterized in that: The data processing method of the input layer is as follows: the independent variables in the cement sample are used as feature variables and converted into a four-dimensional tensor with dimension [D,1,1] through the reshape function, where D is the number of parameters of the independent variables in the cement sample, and the four-dimensional tensor data type is converted into double precision type using the double function.

7. The method for predicting the compressive strength of ordinary Portland cement based on mineral composition according to claim 1, characterized in that: The SGDM algorithm is used to train the compressive strength prediction model, and the following conditions are met: the initial learning rate is 0.01 to 0.05, the learning rate decreases by 50% to 90% every N periods, where 300 ≤ N ≤ 500, and the number of training samples per session is ≤ 30.

Images