Passive determination of anisotropic compressive strength of 3D printed concrete using multiple neural networks enhanced with explainable machine learning (XML).

3D Concrete Printing (3DCP) offers significant advantages over traditional construction such as faster construction time, reduced material wastage, and enhanced ability to execute complex architectural designs. The incorporation of various fibres and industrial wastes into 3DCP can improve performance and sustainability but introduces non-linear effects on compressive strength (CS) that are difficult to predict with standard laboratory methods. This study aims to develop reliable prediction models for the CS of 3DCP by employing advanced neural network and deep learning algorithms such as Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), and Radial Basis Functional Neural Network (RBFNN). A comprehensive database of 200 experimental instances of CS of 3DCP was collected from published literature. The database includes mixture constituents of 3DCP as inputs and CS as the output. The trained algorithms were validated by means of k-fold validation, error metrics, and residual assessment. Among the tested algorithms, the CNN model exhibited the highest predictive performance with a testing R² value of 0.95, demonstrating its robustness in modelling the complex behaviour of 3DCP. To enhance interpretability, Shapley (SHAP) and Individual Conditional Expectation (ICE) analyses were performed, identifying the water-to-cement ratio, loading direction, and fibre content as key factors influencing compressive strength. Finally, a graphical user interface (GUI) has been developed for stakeholders to implement the findings of this study practically.