How is the ground frozen to safely build a subway tunnel?

Cities keep expanding, and so do their subway lines. To link two underground sections, engineers use a method called "artificial ground freezing." Imagine turning the soil into a temporary ice block, making it more stable and resistant to underground water. The process starts with drilling pipes, then cooling them with a refrigerant fluid. Once the ground solidifies, the tunnel is excavated and structural supports are installed.

But it's not all smooth sailing. Cold, mechanical stress, and water interact in complex ways, creating risks like pipe fractures or leaks from the "frozen curtain." To assess these dangers, researchers built a Bayesian network, a statistical model that updates risk probabilities in real time using monitoring data. They also applied the Analytic Hierarchy Process (AHP) to weight the various factors: the excavation and support stage accounts for 53% of the risk, while thawing with grout injection adds 31%. The outcome? An overall risk probability of 46.3%, with the excavation phase being the most critical. These insights let managers act quickly, cutting down accidents and delays. In short, the science of freezing becomes a safety guard for our future subways.

Artificial Ground Freezing (AGF) relies on thermodynamic and heat‑transfer principles. A refrigerant fluid, often calcium chloride or glycol solution, circulates through metal pipes inserted into the soil, absorbing heat and lowering the temperature of the ground below the freezing point of pore water. Ice crystals then form, binding sand grains together, drastically reducing permeability and increasing cohesion.

The process is governed by hydro‑thermal‑mechanical coupling: cooling induces thermal contraction, while groundwater pressure can drive flow toward colder zones, altering the thermal gradient. Modeling these interactions requires Darcy's law for water flow, Fourier's conduction equation for heat, and elasticity equations for mechanical stress.

To assess risks, the authors built a multi‑layer Bayesian network. The root node represents the overall incident probability, with child nodes for observable variables such as fluid temperature, water pressure, ground deformation, and pipe integrity. Conditional probabilities were estimated from historical data and laboratory tests, then updated in real time using field sensors measuring temperature and pressure.

The Analytic Hierarchy Process (AHP) was employed to assign weights to risk criteria. Experts subjectively ranked the importance of each construction phase (drilling, freezing, excavation, support, thaw‑grouting) and performed pairwise comparisons to generate a judgment matrix. Consistency was checked via the Consistency Index (CI), and adjustments were made to reduce inconsistency when needed. Results highlighted excavation and support as the most critical phase (52.94% weight), followed by thaw‑grouting (31.48%).

Integrating AHP with the Bayesian network enables dynamic assessment: any change in monitoring parameters (e.g., a sudden rise in water pressure) recalculates posterior probabilities, instantly flagging vulnerable points. This supports operational decisions such as temporarily halting freezing or increasing grout injection pressure, thereby lowering the risk of pipe rupture or frozen‑curtain leakage. In summary, the combination of thermomechanical modeling, AHP, and Bayesian networks provides a comprehensive framework for managing safety in complex underground environments where geological conditions are variable and error margins are tight.