Performance assessment of polypropylene fiber-reinforced engineered cementitious composites using experimental investigation and ANN modeling

Abstract

The use of engineered cementitious composites (ECC) in structural repair and rehabilitation is becoming more and more common because of their strain-hardening behavior, controlled cracking, and better durability. Polypropylene fiber-reinforced ECC has even more benefits like being resistant to corrosion and saving money, however its mechanical performance is very much affected by the changes in mix composition and the amount of fiber used in the mix. The traditional approaches in mix design are mostly based on either carrying out large-scale experimental trials or using complex micromechanics-based formulations, which are both time-consuming and not very applicable in practice. On the other hand, although the use of Artificial Neural Networks (ANNs) for modeling the behavior of different cement-based materials has been successful, its use in the case of polypropylene fiber-reinforced ECC is still limited. The present study aims to develop an ANN-based predictive framework to predict compressive and flexural strength of polypropylene fiber-reinforced ECC by using the parameters of the mix design. A detailed experimental program with compressive strength and three-point bending tests was performed with different polypropylene fiber contents to study the ECC mixtures. ECC mixtures containing 0–2.0% PP fibers were experimentally tested at curing ages of 14, 28, and 56 days. The results indicate that an optimum PP fiber content of 1.5% yielded the highest mechanical performance, achieving compressive strengths of 31.95 MPa, 47.95 MPa, and 50.61 MPa, and corresponding flexural strengths of 13.41 MPa, 17.50 MPa, and 24.12 MPa at 14, 28, and 56 days, respectively. For making the model more robust and generally applicable, the experimental results were merged with the data from published literature that was carefully selected for ANN training and validation. An ANN model trained using 105 datasets (experimental and literature-based) demonstrated strong predictive capability, with coefficients of determination (R²) of 0.955 and 0.963 for compressive and flexural strength during training, and 0.953 and 0.975 during testing, respectively. The mean absolute percentage error remained below 8.2% for both strength parameters. The proposed model shows a very strong agreement between the values of strength predicted and those measured in the experiments, as it is able to capture the non-linear relationship between the composition of ECC and its mechanical performance. The framework that has been developed is a practical tool for the optimization of ECC mixes and assessment of their performance, particularly for applications involving the rehabilitation of buildings where accurate predictions and efficient design of materials are of utmost importance.