Concrete compressive strength prediction by artificial neural network approach

Abstract

The structural integrity of concrete structure is explicitly influenced by concrete compressive strength (CS). Timely prediction of concrete compressive strength exhibits a good performance in the field of construction. However, it is very challenging due to the unpredictable physical and mechanical properties of concrete and its constituent ingredients. To mitigate the limitation of the laboratory testing-based experimental method, this manuscript presents optimal artificial neural network (ANN) model to forecast CS. For this purpose, total number of 776 datasets were collected from previous research papers. The preprocess dataset was randomly split into training and tesing set. After that,  optimal artificial neural network (ANN) model was developed by establishing appropriate hyperparameters. The overfitting and validation loss were stabilized by loss function assessment with Adaptive Optimization Algorithms (Adam) optimizer. The ANN output results exhibit good prediction performance with R-squared value of 0.87, and errors such as MAE, MSE, and RMSE with values of 3.419 MPa, 21.909 MPA, and 4.68 MPa, respectively. In addition, SHAP value of the output model shows volume of cement and water has highest positive impact, whereas water and fly have highest negative impact on concrete compressive strength. This manuscript shows the power of machine learning techniques to timely and efficient prediction of concrete compressive strength. Thus, this optimal ANN model is applicable in concrete made infrastructure design and construction industry.