Advanced soft computation methods to predict the mechanical properties of engineered cementitious composites (ECC).

Abstract

Engineered Cementitious Composites (ECC) are famous for their enhanced mechanical and durability properties throughout the world. However, its mix design is based on extensive experimentation due to unavailability of mix design guides which is a costly and timeconsuming process. This study. aims to develop machine learning based models that could predict the mix design specific to ECC constituents. A dataset of 176 datapoints composed of mix design and their associated stress strain curves was collected from the published literature. This paper uses Gradient Boost Regressor and Extra Trees Regressor models incorporating 11 input parameters containing all the constituents of the mix design and the properties of the fibers used to forecast the complete tensile stress strain curve of ECC. The results of both models are evaluated using RMSE, R2 , and MAE which yields promising accuracy of the models in prediction of the parameters. Finally, the performance of the model was revalidated by employing mixes which were not part of training and testing data. The results show that these models possess a high accuracy and can help to design ECC without extensive experimentation, which can help in advancement in the commercialization potential of this robust composite.