Artificial neural network based mechanical and electrical property prediction of engineered cementitious composites

Engineered cementitious composite (ECC) is a type of cement-based material fabricated with a variety of add-in functional fillers, featuring superior properties of strain-hardening, ductility and energy absorption. Proper composition is essential for designing ECC material, which may lead to different mechanical and electrical properties. However the design for ECC is still a complex process on the basis of micro-mechanism followed by numerical and experimental analyses, and there is no simple model yet for practical engineering application. This study presents the prediction of mechanical and electrical properties of ECC based on the artificial neural network (ANN) technique with the aim of providing a gateway for a more efficient and effective approach in ECC design. Specifically, neural network models were developed for ECCs reinforced with polyvinyl alcohol (PVA) fibre or steel fibre (SF) with experimental data collected from other researchers for training. The development, training and validation of the proposed models were discussed. To assess the capability of well-trained ANN models for property prediction, experimental studies were conducted, including compression test, four-point bending test, tensile test and electrical resistance measurement for ECCs of various composition. Excellent consistency between the predicted and tested results is obtained, demonstrating the feasibility of ANN models for property prediction of ECCs.