Evaluation of the compressive strength and Cl⁻ content of the blast furnace slag-soda sludge-based cementitious material using machine-learning approaches

Solid waste-based cementitious material is friendly to the environment. The machine-learning technique brings the advantage of high efficiency to the study of physical properties of cementitious materials. In this study, the activation mechanism of soda sludge (SS) on blast furnace slag (BFS) was investigated using X-ray diffraction and thermal gravimetric; the compressive strength and Cl⁻ of the BFS-SS-based cementitious material was evaluated using back propagation (BP) neural network model and nine other machine-learning models (Tree, Bagged Trees, Boosted Trees, Linear Support Vector Machine, Quadric Support Vector Machine, Cubic Support Vector Machine, Gaussian Support Vector Machine, Linear Regression, and Gaussian Process Regression). The potential correlation between the compressive strength and Cl⁻ was investigated using Python 3.6. Results show that the hydraulicity phase of the BFS-SS-based cementitious material was CaAl₂Si₂O₈·4H₂O, 3CaO·Al₂O₃·CaCl₂·10H₂O, and 3CaO·Al₂O₃·3CaSO₄·32H₂O; when the BFS to SS was 40:60, the compressive strength and the solidification ratio of the Cl⁻ were the highest with 4.04 MPa and 5.36% at 2 days and 10.47 MPa and 22.58% at 30 days. The BP neural network model with LM training algorithm is the lowest on mean squared error for the compressive strength and Cl⁻, with 0.0013 and 0.0061 at 2 days and 0.0794 and 0.4794 at 30 days, which has a best predictive ability comparing to the other machine-learning approaches motioned in this study. Pearson correlation coefficient was 0.9749, indicating that the compressive strength and the solidification ratio of the Cl⁻ is a positive and extremely strong correlation.