Prediction model for compressive strength of concrete based on neural network

Abstract

The invention belongs to the field of neural network-based concrete compressive strength prediction in civil engineering, and relates to a neural network-based concrete compressive strength prediction model. The prediction model based on the artificial neural network can effectively capture the complex nonlinear relationship between the content of the nano graphene sheets, the concrete mix proportion, the ratio of slurry to aggregate and the compressive strength of the 28-day-old concrete, and has important significance on the accurate compressive strength of the concrete.

Description

Technical Field

[0001] The invention belongs to a strength prediction method of neural network concrete, and in particular relates to a prediction model of the compressive strength of neural network concrete. Background Art

[0002] Nanographene Platelets (NGPs) have attracted much attention in the research field of reinforced concrete materials in recent years due to their excellent mechanical properties and unique nanoscale size effects. As a high-performance nano-reinforcement material, the addition of NGPs can significantly improve the mechanical properties of concrete, such as compressive strength, tensile strength and durability, making concrete perform better in terms of heavy load, earthquake resistance and durability. However, in practical applications, factors such as different nanographene platelet contents, concrete mix ratios, and the ratio of paste to aggregate (i.e., paste-to-aggregate ratio, VP / VA) will have a significant impact on the final performance of concrete. The interaction between these influencing factors forms a complex nonlinear relationship, making it an extremely challenging task to accurately predict the compressive strength of 28-day-old concrete.

[0003] In order to overcome this problem, artificial neural network (ANN) as a powerful nonlinear modeling tool has been widely used in the field of concrete performance prediction. By simulating the structure and function of human brain neurons, ANN can automatically learn and extract complex nonlinear relationships in data, thereby accurately predicting and classifying unknown data. In the ANN-based concrete compressive strength prediction model, the content of nanographene sheets, concrete mix ratio (including the proportion of cement, water, aggregate and other components), mortar-aggregate ratio, etc. can be used as input variables, and the compressive strength of 28-day-old concrete can be used as the output variable. Through a large amount of experimental data training, the ANN model can gradually optimize its internal structure and weight parameters, thereby realizing the accurate prediction of the compressive strength of concrete under different conditions. Summary of the invention

[0004] In view of the limitations of current technology, the core problem that the present invention aims to solve is to develop a new method for predicting concrete compressive strength using an artificial neural network, which can simplify the operation process, shorten the time consumption, and ensure that the error between the predicted results and the actual measured values ​​is small and the accuracy is high.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0006] Step 1: Create a layer from the input layer to the first hidden layer:

[0007] For the jth neuron in the hidden layer:

[0008]

[0009] a j (1) =φ(z j (1) )

[0010] in:

[0011] x i : the i-th input feature;

[0012] w ij (1) : Weights from input layer to hidden layer;

[0013] b j (1) : The bias of the jth neuron in the hidden layer;

[0014] φ: activation function (such as ReLU, Sigmoid);

[0015] Step 2, hidden layer to output layer (assuming there is only one hidden layer):

[0016]

[0017] f c28 =z (2)

[0018] in:

[0019] w j (2) : The weight from the hidden layer to the output layer;

[0020] b (2) : Bias of the output layer;

[0021] Step 3: Get the comprehensive formula

[0022]

[0023] As a further solution of the present invention, the prediction model of the compressive strength of concrete based on a neural network has the following input features: the input feature vector X includes: (1) the amount of cementitious material (kg / m 3 ): cement dosage C, fly ash dosage F, silica fume dosage L, nano-silicon dioxide dosage 0, nano-titanium dioxide dosage T; (2) water-binder ratio Bone ratio (4) Nanographene sheet content α G (expressed as a percentage of the total amount of cementitious materials); (5) recycled aggregate volume ratio (if recycled aggregate is used); (6) Aggregate properties: apparent density, water absorption, and moisture content of coarse and fine aggregates; (7) Amount and type of admixtures; (8) Curing conditions: temperature and humidity.

[0024] As a further solution of the present invention, the prediction model of the compressive strength of concrete based on a neural network sets the content of nanographene sheets α G :0%, 0.05%, 0.1%, 0.2%, 0.5%, 1.0%; Set the mortar-aggregate ratio VP / VA: 0.35, 0.40, 0.45; Set the water-binder ratio W / B: 0.35, 0.40, 0.45; Cementitious material composition: adjust the proportion of cement and mineral admixtures according to design requirements; Select aggregates with different particle sizes and properties; Dispersion of nano-graphene sheets: In order to ensure the uniform dispersion of nano-graphene sheets in concrete, ultrasonic dispersion or the addition of dispersants can be used.

[0025] As a further solution of the present invention, a prediction model for the compressive strength of concrete based on a neural network and a flow chart for predicting the compressive strength of concrete at 28 days of age based on a neural network are provided.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] (1) The present invention utilizes an artificial neural network to effectively capture the complex nonlinear relationship between the content of nanographene sheets, concrete mix ratio, slurry-aggregate ratio and the compressive strength of concrete at 28 days of age.

[0028] (2) The present invention reduces the time for repeated tests, accelerates the test progress, and significantly improves the accuracy and reliability of the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 The present invention is a flow chart for predicting the compressive strength of concrete at 28 days of age based on a neural network. DETAILED DESCRIPTION

[0030] The following will be combined with the specific embodiments of the present invention to elaborate the technical solutions therein in detail and comprehensively. It should be noted that the embodiments described here are only some examples of the present invention, rather than exhaustive of all possible implementation methods. Based on these embodiments, all other embodiments that can be derived by those skilled in the art without the need for creative work should be deemed to fall within the scope of protection of the present invention.

[0031] In this specific implementation, the raw materials for producing the artificial neural network concrete include cement, fly ash, silica fume, nano silicon dioxide, nano titanium dioxide, natural coarse aggregate and natural fine aggregate. In addition, the artificial neural network concrete of the present invention is not limited to the above raw materials during implementation. Among the above raw materials, cement is ordinary silicate cement, and the fine aggregate is medium sand, and the fine aggregate is mixed with the natural coarse aggregate, and the grade is matched.

[0032] 1. Build and train artificial neural network models

[0033] 1. Dataset division:

[0034] Training set: 70%

[0035] Validation set: 15%

[0036] Test set: 15%

[0037] 2. Select the network structure:

[0038] Number of input layer nodes: equal to the number of input features (assuming it is n);

[0039] Hidden layer: You can select one or more hidden layers, and the number of neurons in each layer can be 2n+1 or n;

[0040] Number of output layer nodes: 1 (predicted compressive strength);

[0041] 3. Set training parameters:

[0042] Activation function: The hidden layer uses ReLU, Sigmoid or Tanh, and the output layer uses a linear activation function;

[0043] Loss function: mean square error (MSE);

[0044] Optimization algorithms: Adam, SGD, etc.

[0045] Learning rate: such as 0.001, which can be adjusted according to the training effect;

[0046] Number of training rounds (Epochs): For example, 1000 times, adjusted according to the convergence of the loss function;

[0047] 4. Model training:

[0048] Forward propagation: calculate the predicted output;

[0049] Loss calculation: Calculate the error between the predicted value and the true value;

[0050] Back propagation: calculate gradients, update weights and biases;

[0051] Iterative training: Repeat the above steps until the loss function converges or the maximum number of training rounds is reached.

[0052] 2. Model Validation and Testing

[0053] Evaluate model performance on the validation set: adjust network structure and hyperparameters to prevent overfitting or underfitting.

[0054] Evaluate the generalization ability of the model on the test set: calculate evaluation indicators such as MSE, RMSE, MAE, R 2 value.

[0055] 3. Model application

[0056] Enter new data:

[0057] Input the new concrete mix proportion, nanographene sheet content, mortar-brick ratio, etc. into the model;

[0058] Prediction results:

[0059] The model output is the predicted compressive strength of concrete at 28 days of age.

[0060] 4. Experimental verification

[0061] 1. Experimental Data Collection

[0062] Experimental samples: Collect no less than 100 sets of experimental data with different mix ratios and nanographene sheet contents to ensure the generalization ability of the model.

[0063] 2. Model Performance Evaluation

[0064] Evaluation indicators:

[0065] Mean Square Error (MSE):

[0066]

[0067] Root Mean Square Error (RMSE):

[0068]

[0069] Mean Absolute Error (MAE):

[0070]

[0071] Coefficient of determination (R 2 ):

[0072]

[0073] 3. Result analysis: If R 2 It is close to 1, and the MSE, RMSE, and MAE are small, indicating that the model has high prediction accuracy.

[0074] 5. Practical application verification:

[0075] (1) Predict the compressive strength of the new mix:

[0076] Input new mix ratio and nanographene sheet content and use the trained model to make predictions;

[0077] (2) Experimental verification:

[0078] The corresponding concrete specimens are actually made, the compressive strength is tested, and the measured values ​​are compared with the predicted values.

[0079] Finally, it should be emphasized that the above implementation cases are only intended to illustrate the technical solutions of the present invention, and do not constitute a limitation on its application. Although we have described the present invention in detail by referring to the preferred implementation cases, professionals in this field should recognize that it is entirely possible to make various adjustments and changes in form and details without departing from the core spirit and protection scope of the present invention defined by the attached claims.

Claims

1. A prediction model for the compressive strength of concrete based on a neural network, characterized in that The following steps are involved: Step 1, input layer to the first hidden layer: For the jth neuron in the hidden layer: in: x i : the i-th input feature; w ij (1) : weights from input layer to hidden layer; b j (1) : The bias of the jth neuron in the hidden layer; φ: activation function (such as ReLU, Sigmoid); Step 2, hidden layer to output layer (assuming there is only one hidden layer): f c28 =z (2) in: w j (2) : The weight from the hidden layer to the output layer; b (2) : Bias of the output layer; Step 3: Get the comprehensive formula 2. A prediction model for compressive strength of concrete based on a neural network according to claim 1, characterized in that: Input features: Input The characteristic vector X includes: (1) The amount of cementitious material (kg / m 3 ): cement dosage C, fly ash dosage F, silica fume dosage L, nano-silicon dioxide dosage 0, nano-titanium dioxide dosage T; (2) water-binder ratio (3) Bone to Bone Ratio (4) Nanographene sheet content α G (expressed as a percentage of the total amount of cementitious materials); (5) recycled aggregate volume ratio (if recycled aggregate is used); (6) Aggregate properties: apparent density, water absorption, and moisture content of coarse and fine aggregates; (7) Amount and type of admixtures; (8) Curing conditions: temperature and humidity.

3. A prediction model for compressive strength of concrete based on a neural network according to claim 1, characterized in that: Set the nanographene sheet content α G :0%, 0.05%, 0.1%, 0.2%, 0.5%, 1.0%; set the vesicle-bone ratio VP / VA: 0.35, 0.40, 0.45; Set the water-binder ratio W / B: 0.35, 0.40, 0.45; Composition of cementitious materials: adjust the ratio of cement and mineral admixtures according to design requirements; select aggregates with different particle sizes and properties; Dispersion of nanographene sheets: in order to ensure the uniform dispersion of nanographene sheets in concrete, ultrasonic dispersion or the addition of dispersants can be used.

4. A prediction model for compressive strength of concrete based on a neural network according to claim 1, characterized in that: Flowchart for predicting the compressive strength of concrete at 28 days of age based on neural network.

Images