Predicting mechanical properties of engineering cementitious composite reinforced with PVA using artificial neural network

This paper developed an artificial neural network (ANN) models to predict engineered cementitious composites (ECC) mechanical properties such as compressive strength, flexural strength, and direct tensile stress-strain curve. ANN models were created, trained, validated, and tested based on a large data set with variable mix designs. The used data set was 151,76, and 44 test results for compressive strength, flexural strength, and direct tensile stress-strain curve collected from recently published research. Models data analysis showed outstanding predictive performance with accepted accuracy near to 100%. Additional evaluation using an extra experimental data set confirmed the accuracy of the proposed ANN models with minimum relative errors around (0.15:9.40) % for compressive strength, (0.05:4.71) % for flexural strength, and (1.40:5.00) % for the tensile strength. Based on the model's data analysis, additional data sets evaluation, and the statistical tools for the external data set including the absolute fraction of variance (R-2), and the degree of agreement (d) the models were capable of predicting the mechanical strengths of ECC mixtures. Finally, stress-strain relations can be predicted precisely using ANN models with a maximum variance of 7.10%.