An Experimental Study on the Mechanical Properties and ANN-Based Prediction of a Tensile Constitutive Model of ECCs

Traditional concrete materials have limitations in terms of load-bearing capacity and ductile failure. In contrast, Engineered Cementitious Composites (ECCs), with their superior strain-hardening behavior and multiple cracking characteristics, have attracted widespread attention in the field of high-performance materials. In this study, ECC specimens incorporating different types of fibers (polyethylene (PE) fibers, polyvinyl alcohol (PVA) fibers) at varying contents were tested to systematically analyze their influence on mechanical properties. Compressive, flexural, and uniaxial tensile strength tests were conducted to evaluate the mechanical performance of ECCs. In addition, scanning electron microscopy (SEM) was employed to examine the fracture surfaces of the fibers, providing deeper insights into the interfacial behavior and fracture morphology of the different fiber-reinforced systems. Fracture surface analysis reveals that the interfacial bonding characteristics between different fibers and the matrix significantly influence fracture behavior. Moreover, as the tensile performance of ECCs is influenced by the interaction of multiple factors, traditional constitutive models exhibit limitations in accurately predicting its complex nonlinear behavior. To address this limitation, an Artificial Neural Network (ANN) approach was adopted to develop a predictive model based on bilinear stress-strain relationships. The model was constructed using ten key input parameters, including matrix composition and fiber properties, and was able to accurately predict the first cracking strain, first cracking stress, ultimate strain, and ultimate stress of ECCs. Sensitivity analysis revealed that fiber tensile strength and fiber content were the most significant factors influencing the tensile behavior. The predicted tensile curves showed strong consistency with the experimental results, thereby confirming the reliability and applicability of the proposed ANN-based model.