Piezoresistive Prediction of CNTs-Embedded Cement Composites via Machine Learning Approaches

Conductive cementitious composites are innovated materials that have improved electrical conductivity compared to general types of cement, and are expected to be used in a variety of future infrastructures with unique functionalities such as self-heating, electromagnetic shielding, and piezoelectricity. In the present study, machine learning methods that have been recently applied in various fields were proposed for the prediction of piezoelectric characteristics of carbon nanotubes (CNTs)-incorporated cement composites. Data on the resistivity change of CNTs/cement composites according to various water/binder ratios, loading types, and CNT content were considered as training values. These data were applied to numerous machine learning techniques including linear regression, decision tree, support vector machine, deep belief network, Gaussian process regression, genetic algorithm, bagging ensemble, random forest ensemble, boosting ensemble, long short-term memory, and gated recurrent units to estimate the time-independent and -dependent electrical properties of conductive cementitious composites. By comparing and analyzing the computed results of the proposed methods, an optimal algorithm suitable for application to CNTs-embedded cementitious composites was derived.