Prediction of concrete strength using response surface function modified depth neural network.

Area of Science:

• Civil Engineering
• Materials Science
• Computational Science

Background:

• Deep Neural Networks (DNNs) often struggle with discrete input and training data, limiting their predictive accuracy.
• Existing models may lack robustness and generalization capabilities when dealing with complex material properties.

Purpose of the Study:

• To develop a novel deep neural network model that overcomes data discreteness issues.
• To enhance the predictive accuracy and generalization ability of deep learning models for material property prediction.

Main Methods:

• Utilizing a multivariable response surface function (MRSF) to revise discrete input and training data.
• Deriving a new loss function based on response surface data.
• Establishing a Multivariable Response Surface Function Deep Neural Network (MRSF-DNN) model.

Main Results:

• The MRSF-DNN model demonstrated high prediction accuracy for recycled brick aggregate concrete compressive strength.
• A high correlation coefficient (0.9882) was achieved between real and forecasted values.
• The model exhibited a low relative error, ranging from -0.5% to 1%.

Conclusions:

• The MRSF-DNN model offers superior prediction accuracy and stability compared to standard DNNs.
• The developed model shows enhanced generalization capabilities, making it suitable for complex material science applications.
• This approach effectively addresses data discreteness challenges in deep learning for engineering applications.