Terra Multiphysics

Coupled ground-physics simulation, at the speed of a slider.

Validated surrogate  R² 0.9997
Mean error  0.009 °C
Speed-up vs full FEM  ~1,000,000×
Trained on  200 real coupled-FEM runs
Design & site inputs
60 W/m/K
0.050 kg/s
8.0 °C
2.50 W/m/K
1100 J/kg/K
Predicted pipe outlet temperature
18.40 °C
inlet 8.0 °C, fluid warms 10.4 °C along the loop
AI surrogate
0.1 ms
Full FEM solver
~81 s
⚡~1,000,000× faster, designed for real-time iteration
Site (fixed), then map every design at once
Each point is one design (colour = wall conductance hZ). The red frontier is the efficient trade-off between heat delivered and source temperature (higher → better heat-pump COP). Mapped instantly by the surrogate.
Global sensitivity, which design inputs drive each performance output (Sobol total-effect indices, analytic, cross-checked vs Monte-Carlo). The surrogate computes over the whole design space instantly.
Read it: outlet temperature is driven mostly by the wall coupling hZ; heat delivered and pumping cost are driven by flow rate, pressure drop almost entirely so. A sensitivity study that would take thousands of full FEM runs, returned as one chart.
Outlet temp, 95% confidence
15.5 – 20.1 °C
Heat rate, 95% confidence
0.9 – 7.6 kW
Pressure drop, 95% confidence
20 – 757 Pa
Method
Sobol + Monte-Carlo
📊commercial-grade uncertainty quantification, in milliseconds, fully offline
Describe the job in plain language
sand · 80 kW · standard pipe clay · 200 kW · premium pipe limestone · 120 kW