A neural network-based model for assessing 3D printable concrete performance in robotic fabrication

Abstract

• A backpropagation neural network predicted 3D-printed concrete components. • Flowability and strength were key factors in prediction accuracy. • Specific mix parameters ensured good printability of concrete. 3D concrete printing has advanced rapidly in the construction industry, offering benefits such as faster construction, reduced labor demands, and minimized material waste. Nevertheless, the mixture design process for 3D printable concrete remains largely dependent on trial-and-error methods, which are both time-consuming and resource-intensive. This study proposes a neural network-based approach to predict the performance of 3D printable concrete. Key mixture parameters, including the water-to-binder ratio, sand-to-binder ratio, and fly ash content, are identified, and experiments are conducted to assess printability and mechanical properties across various formulations. Based on experimental results, a back-propagation neural network (BPNN) model is developed and trained using multiple inputs, including flowability, setting time, extrusion width, printing height, compressive strength, and flexural strength, to predict optimal mixture proportions. Additionally, grey correlation analysis is employed to evaluate the influence of each input parameter on the model's predictions. The sensitivity analysis highlights that flowability and mechanical strength are the most critical factors affecting prediction accuracy, while extrusion width and printing height exert a comparatively smaller influence.