AUTHOR=Sun Haoyue , Su Xiaoping , Xi Xiao , Zhao Qiyuan , Zhao Guijie , Wei Yucong , Zhang Qilei 

TITLE=Engineering properties, strength mechanisms, and machine learning-based strength prediction of controlled low-strength materials prepared with excavated soil

JOURNAL=Frontiers in Materials

VOLUME=Volume 12 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/materials/articles/10.3389/fmats.2025.1673682

DOI=10.3389/fmats.2025.1673682

ISSN=2296-8016

ABSTRACT=IntroductionThe excellent backfilling performance and significant potential for waste resource utilization make controlled low-strength material (CLSM) an important technical alternative to traditional backfilling methods. The preparation of CLSMs using excavated soil not only enables local material sourcing but also promotes waste resource utilization and reduces backfilling costs.MethodsIn this study, a novel CLSM was developed by incorporating sand, cement, fly ash, high-efficiency plasticizer, and water into excavated soil. The engineering properties—including flowability, setting time, bleeding rate, and density—were evaluated, with a focus on strength characteristics and the establishment of a strength-age relationship model. Multiple characterization methods were used to elucidate the strength development mechanism from the perspectives of hydration product evolution and microstructural changes. A machine learning prediction model based on Newton‒Raphson-Based Optimizer (NRBO)-Light Gradient Boosting Machine (LightGBM) was constructed to achieve high-precision prediction of the relationship between mix proportions and strength.Results and discussionResults show that the prepared CLSM exhibits excellent engineering performance: flowability of 165–257 mm ensures good self-compacting and self-levelling; setting time of 4.6–7.48 h meets rapid construction needs; bleeding rate (≤1.28%) and fresh density (1880–2005 kg/m3) meet engineering standards; and 28-day strength (1.35–2.69 MPa) is suitable for both trenchless and excavatable applications. The strength–age relationship fits a hyperbolic model with accuracy above 0.98. Microstructural analysis reveals that hydration of cement and fly ash produces C-S-H and C-A-H gels, filling pores and densifying the structure. The NRBO-LightGBM model achieved R2 values of 0.995 and 0.966 for training and test sets, respectively, demonstrating high accuracy and stability. Furthermore, by utilizing excavated soil as a replacement for sand in the aggregate, each cubic metre of CLSM can recycle 328–600 kg of dry excavated soil. These findings provide theoretical and technical support for CLSM development using excavated soil.