Mixture Design and Multiobjective Optimization of Industrial Wastes toward Geotechnical Applications

Abstract

This study introduces a novel optimization methodology for industrial waste soil stabilizers combining the mixture design (MD), the back-propagation neural network (BPNN), and the nondominated sorting genetic algorithm II (NSGAII). Industrial wastes including blast furnace slag (BFS), steel slag (SS), and phosphogypsum (PG) mixed with ordinary portland cement (OPC) were applied to formulate a composite cementitious material for geotechnical use in underground improvement. The unconfined compressive strength (UCS) of stabilized soil samples cured for 7 days and 28 days was analyzed to investigate the influence of the precursors on the strength. A BPNN-NSGAII model was developed using experimental data to solve the multiobjective optimization (MOO) problem. Then, the macroscopic and microscopic characteristics of the optimized solidified soil were evaluated. The results demonstrate the substantial role of PG in strength enhancement and identify an optimal range of the initial-strength ratio of BFS to OPC. Statistical analyses and experimental validation confirm that the MD-BPNN-NSGAII model can evaluate the quantitative relation between variables and responses and effectively predict the UCS of treated soils with superior accuracy. The enhanced solidified soil exhibits better macro- and microproperties and environmental impacts than the conventional cement soil does. The proposed framework is universally applicable and versatile, and it is designed to guide the initial mix design and formulation of industrial waste slag-based cementitious materials and their proportional composition.