Integration of Artificial Intelligence and Electrical Resistivity for the Prediction of Compressive Strength in Steel Fiber-Reinforced Concrete

Artificial intelligence (AI) has become a powerful tool for machine-learning-based forecasting from available data. This study evaluates several artificial neural network (ANN) architectures and the traditional multiple linear regression (MLR) method to predict the compressive strength of steel-fiber-reinforced concrete (SFRC). The input parameters considered in the models included electrical resistivity, concrete age, water-to-cement ratio (w/c), and cement content. Fifty-four concrete mixes were designed by varying the w/c ratio (0.45, 0.50 and 0.60), the nominal maximum size of the coarse aggregate (1″, 3/4″ and 1/2″) and the type of metallic fiber (Sika® Fiber CHO 65/35 [F1] and Sika® Fiber CHO 80/60 [F2]). Cylindrical specimens were cured in accordance with ASTM C31 and tested at 7, 14, and 28 days. Compressive strength was determined in accordance with ASTM C39. Electrical resistivity was measured at 7, 14 and 28 days using the Wenner method. Using this dataset, six ANN architectures were trained and the multiple linear regression (MLR) equation was calculated using Matlab R2018a software. The ANN models outperformed the MLR approach in predictive accuracy. Optimal performance was achieved with a three-layer ANN comprising 50 neurons in the first hidden layer, 20 in the second, and a single output neuron. The activation functions used were f(s) = tanh(s) for the first two layers and g(s) = s for the third layer. This ANN architecture achieved a correlation coefficient (R) of 0.98157 and the lowest error metrics, reported as percentages: mean absolute error (MAE), mean absolute percentage error (MAPE), mean squared error (MSE), and root mean squared error (RMSE) of 2.37%, 2.52%, 0.124%, and 3.52%, respectively. These findings demonstrate that ANN models can accurately predict the compressive strength of metal fiber reinforced concrete from electrical resistivity measurements and the variables mentioned above.