Bayesian machine learning-aided approach bridges between dynamic elasticity and compressive strength in the cement-based mortars

Abstract

This study tries to establish a powerful machine learning (ML) model for predicting the compressive strength of cement-based mortars by using dynamic elasticity data. The ML model was developed on the basis of Bayesian theorem, leading to a decrease in the overfitting problem compared with other conventional neural networks. Moreover, for the first time, the empirical equations were embedded in the ML model, enhancing the correlation between dynamic elasticity and compressive strength in the cement-based mortars. The results showed that the ML model efficiently predicted the compressive strength with determination coefficient (R2) of 95.2% and root mean square error (RMSE) of 0.0488 for testing data. The detailed analysis also showed that the excessive increase in the number of empirical-equation inputs in the ML model does not significantly improve the prediction efficiency. The results also unveiled that the correct linkage between dynamic elasticity and compressive strength strongly depends on the weighted portion of each empirical equation in the ML model. Finally, it is concluded that it is feasible to non-destructively capture the compressive strength in a wide range of physical and mechanical features without knowing or giving importance to the compositional variations. Furthermore, the outcomes of this work shed light on the intricate correlation between the dynamic response of the material and the compressive strength.