A Molecular Dynamics and Machine Learning Study on the Calcium Leaching Mechanism of CSH in Shotcrete for Tunnels

Crystallization-induced clogging in tunnel drainage systems has emerged as a critical challenge, threatening both structural durability and operational safety. This issue becomes particularly acute with the widespread use of shotcrete, where calcium leaching plays a pivotal role in promoting carbonate crystallization. Based on field investigations and sample analyses from representative tunnel projects, this study systematically unveils the causes and evolution of such clogging phenomena. Leaching experiments and microstructural characterizations on CSH, the primary hydration product in shotcrete, reveal that calcium loss disrupts the internal structure and triggers a notable increase in silicate chain polymerization, with the MCL rising from 6.99 to 8.70. Atomic-scale simulations further clarify the thermodynamic dissociation pathways of Cw and Ca. Due to its loosely coordinated environment, high hydration potential, and lower energy barrier, Cw leaches preferentially. In contrast, Ca forms stable complexes with the silicate framework, exhibiting stronger resistance to leaching. Free energy landscape analysis highlights their intrinsic differences in dissociation paths and energy thresholds. Meanwhile, machine learning models were introduced to predict the free energy, among which XGBoost demonstrated excellent nonlinear fitting capability and generalization performance, with prediction accuracy significantly superior to that of the other models. This research provides atomistic insights and practical guidance for understanding tunnel crystallization pathologies and optimizing the design and performance of shotcrete materials. It holds significant implications for engineering applications. • Reveals the formation pathway and material source of CaCO 3 crystallization blockage driven by calcium leaching from shotcrete. • Elucidates the atomic‐scale mechanism of “preferential Cw leaching-silicate backbone reconstruction.” • Demonstrates that XGBoost achieves superior accuracy and generalization, significantly reducing the computational cost of atomistic simulations.