Model-Based Neural Network for Predicting Strain-Rate Dependence of Tensile Ductility of High-Performance Fibre-Reinforced Cementitious Composite

Abstract

High-performance fibre-reinforced cementitious composite (HPFRCC) has been demonstrated to provide superior tensile ductility and fracture energy compared to normal concrete at both quasi-static and dynamic strain rates. For this reason, this material becomes potential material for application to structures subjected to dynamic loading. However, there is still a lack of accuracy model for estimating strain-rate dependence of tensile ductility of HPFRCCs since most current empirical regression models have been proposed based on individual limited test data. In this study, a model-based neural network has been trained to estimate the strain-rate dependence of tensile ductility of HPFRCCs using 150 tensile test results. There are six input variables: matrix strength, fibre type, fibre length, fibre diameter, and fibre volume content, while strain-rate dependence of tensile ductility is output parameter. The results of prediction showed that the machine learning-based model was an efficient method to estimate strain-rate sensitivity in tensile ductility of HPFRCCs with high accuracy. By performing sensitivity analysis, the relative importance of all influencing factors was determined.