Postfire residual capacity of steel fiber reinforced volcanic scoria concrete using PSO-BPNN machine learning

Volcanic scoria (VS) is a potential green aggregate, as it is abundant around the world. Replacing normal aggregate with vS aggregate improves economic efficiency, ecological benefits in the construction industry. In this study, an artificial neural network (ANN) is employed to predict the behavior of steel fiber reinforced volcanic scoria concrete (SFVSC). Compression strength (CS) and tensile splitting strength (STS) tests on SFVSC was first conducted to obtain 240 groups of training data. Based on the training data, appropriate neural network structures and training processes was established using several neural network models. The behavior of steel fiber volcanic scoria concrete is predicted with various parameters. The prediction accuracy of is appraised using the mean absolute percentage error (MAPE), root mean square error (RMSE), and correlation coefficient (R2) for different machine learning algorithm back propagation neural network (BPNN), genetic algorithm (GA-BPNN), and particle swarm optimization (PSO-BPNN). The main conclusion is that the PSO-BPNN model outperforms the other models in prediction accuracy. The other results show that both GA-BPNN and PSO-BPNN exhibit good accuracy and suitable in predicting the mechanical properties of SFVSC. Using machine learning, the effect of different vS aggregate replacement levels (VR) (30, 50, 70 %), steel fiber dosage (Vst) (0, 0.5, 1, 1.5 %), water-to -cement ratios (w/b) (0.4, 0.5), and temperatures (t) (20, 200, 400, 600, 800 degrees C) on the mechanical properties of SFVSC is also studied.