Soil conditioning optimization of sandy pebble soil with different fine contents and prediction by machine learning.

Proper soil conditioning is critical for earth pressure balance (EPB) shield tunneling to mitigate problems such as mud cake formation and cutter headwear by enhancing muck flow plasticity, particularly in sandy pebble soils. This research investigates the effect of conditioning agents-bentonite slurry and foam-on sandy pebble soils with varying fines contents, considering both concentration and injection ratio. Slump tests were conducted to evaluate the workability of conditioned soil. As fines content increases from 10 to 80%, the slump value of unconditioned soil decreases from 238 to 47 mm because of poor grading. Bentonite-conditioned soils with 10%-20% fines content decrease in slump value as slurry injection ratio (SIR) rises to 8% and 6%. This increase is attributed to improved grading and interlocking from the clay particles. As SIR and fines content increase, clay particles and free water form a film around the coarse particles and act as a lubricant, increasing slump value. In soils with 60% and 80% fines contents, foam significantly enhances workability, with slump value rising to 200 mm as FIR increases. To predict slump values and evaluate workability for improved engineering efficiency, this study employs a deep learning algorithm, considering the nonlinear relationship between slump values under different conditions. This model, with an accuracy exceeding 95%, contributes to developing rapid and intelligent solutions for soil conditioning in urban tunnel projects.