Machine learning approach to predict the early-age flexural strength of sensor-embedded 3D-printed structures

Abstract The absence of formwork in 3D-printed concrete, unlike conventional mold-cast concrete, introduces greater variability in curing conditions, posing significant challenges in accurately estimating the early-age mechanical strength. Therefore, common non-destructive techniques such as the maturity method fail to deliver a generalized predictive model for the mechanical strength of 3D-printed structures. In this study, multiple machine learning (ML) algorithms, including linear regression (LR), support vector regression (SVR), and artificial neural network (ANN), were developed to estimate the early-age flexural strength of 3D-printed beams under varying curing conditions, utilizing data collected from embedded sensors. Six input variables were employed for the ML models, including relative permittivity, internal temperature, and curing method. For model development, 144 data points were collected from an extensive experimental study, and multiple statistical metrics were employed to evaluate the proposed models. The ANN model outperformed the other models in predicting early-age strength, achieving a coefficient of determination of 95.1%. Furthermore, the input variable analysis highlighted the curing method as the most influential factor affecting the strength of 3D-printed beams.