Every design calculation of a borehole field rests on the step response of the ground: the temperature at the borehole wall after a constant heat rate of 1 W/m has been switched on at . Made dimensionless it is the g-function (Eskilson 1987). The design model obtains it from finite line sources superposed in space. The 3D analysis obtains the same step response with the finite element method (FEM) on a mesh of the ground, so that layer contrasts, borehole sections, groundwater flow and inclined boreholes enter as they are instead of through effective properties. The loads are applied afterwards by temporal superposition, in exactly the same way as for the analytic g-function. One FEM run therefore serves every load case, and the difference between the two results measures what the analytic model leaves out for the project at hand.
The FEM result is a second opinion next to the design. It never replaces the design results: the commands Calculate, Size borehole length, Optimize, the parametric study and the Monte Carlo analysis use the analytic g-function. The FEM model of this chapter is linear (constant properties, zero initial and boundary values of the temperature change) and treats the boreholes as quasi-steady elements; the direct, time-stepped simulation with the actual loads and transient boreholes is described in the Full FEM chapter.
22.1 Problem, domain and the numerical g-function
The temperature change of the ground (relative to the undisturbed state) satisfies the heat equation with line sources along the active length of every borehole :
with the volumetric heat capacity, the thermal conductivity (both per layer), and the line Dirac function of borehole . The top face is the ground surface held at the undisturbed temperature, which is the physical meaning of the mirror source of the finite line source. The four sides and the bottom of the block lie far enough from the field that the temperature change there is negligible. Their distance is the larger of two rules:
- the heat-travel margin ( the thermal diffusivity, the last time of the g-function grid), at which the temperature rise is below % of its value at the boreholes;
- the field-size rule: in the automatic setting the block is three times the plan extent of the field wide in both horizontal directions and twice the depth of the deepest borehole deep. It leaves room for the temperature field around the field (maps, plume, property lines). A custom domain takes the user's factors; a block smaller than the recommended heat-travel margin is reported.
With groundwater flow the block is extended downstream by the drift of the heat over the period. The g-function follows from the wall temperature of the unit load:
with the length-weighted mean wall temperature under a uniform heat rate (UHTR) or the common wall temperature under a uniform borehole wall temperature (UBWT), and the reference conductivity of the analytic model. The result has the same time scale, reference properties and time grid as the analytic g-function, so the monthly and hourly simulation, the short-time model and the sizing use it without change.
22.2 Mesh
22.2.1 Hexahedral grid for vertical boreholes
The ground block is a tensor-product grid of hexahedral cells. Each axis is a graded one-dimensional distribution built from attractors with a local cell size : in the horizontal directions the borehole axes, in the vertical direction the surface, the borehole tops and bottoms, the layer interfaces and the boundaries of borehole sections. Between fixed nodes the cell size follows
with the attractor position, the number of uniform cells it keeps on each side and the growth ratio; nodes are placed through the cumulative integral of , so neighbouring cells differ by about . The cap is the axial size over the depth of the boreholes and elsewhere, and in the vertical direction at most a third of the thickness of every ground layer and borehole section, so that thin layers are resolved. Each borehole axis has two uniform cells of on each side, and every borehole top and bottom one uniform cell above and below: the cells where a line source ends are then cubes. (The trilinear stiffness couples the two nodes of a vertical edge positively when a cell is taller than wide; under the end of a line source this produced a small undershoot of the temperature below zero, which the cubic end cells remove at a cost of 5–20 % more unknowns.)
| Preset | Growth | Time steps per level | |||
|---|---|---|---|---|---|
| Coarse | 1.0 m | 10 m | 40 m | 1.3 | 6 |
| Medium | 0.5 m | 5 m | 25 m | 1.3 | 6 |
| Fine | 0.25 m | 2.5 m | 15 m | 1.3 | 6 |
| Very fine | 0.25 m | 2.5 m | 15 m | 1.15 | 12 |
| Auto (default) | from the project | from the project | from the project | 6 |
The cell at the axis must be larger than the borehole. The well index of Section 22.4 needs the equivalent radius of the axis cell to exceed the borehole radius , so is raised to at least (0.52 m for = 55 mm, 0.66 m for 70 mm). For typical boreholes the Medium and Fine presets therefore share the axis cell and differ in the rest of the block. A finer mesh does not mean a finer borehole; it refines the far field and, with Very fine, the time steps.
22.2.2 Tetrahedral mesh for inclined boreholes
A tensor grid refines whole planes: an inclined borehole of 200 m at 30° drifts 100 m sideways and would need the fine horizontal spacing over its whole projected band at every depth. As soon as one borehole is inclined the program therefore uses an adaptive conforming mesh of linear tetrahedra refined towards the true 3D axes; everything downstream of the element matrices (borehole coupling, solvers, time levels, energy balance, short-time handling) is shared with the hexahedral path. The mesh is built without third-party meshing code:
- A background lattice of cubes covers the block; every cube is split into the six Kuhn tetrahedra, which share the main diagonal and form a conforming mesh.
- Tetrahedra are refined by newest-vertex bisection in the tagged form of Maubach (1995). Three bisections reproduce the Kuhn tetrahedra of the half-size cube, so the element shapes stay in a few similarity classes and the smallest dihedral angle of every mesh is that of the Kuhn tetrahedron, 45°.
- A tetrahedron is bisected while it is larger than the size field asks for at the nearest point it could contain: the line size near the axes growing with the distance from them, a size near layer interfaces, and a third of every layer thickness near the field.
- After each sweep every element with a hanging node is bisected until none is left (the minimal conforming refinement). A tetrahedron cut by a layer interface takes the volume-weighted properties of the parts on each side.
22.3 Discretisation and time stepping
The ground is discretised by the Galerkin method with trilinear shape functions on the hexahedra and linear shape functions on the tetrahedra. The stiffness of a brick of sizes is available in closed form,
( for the same node, for the neighbour along that axis), and for a tetrahedron of volume it is . An anisotropic ground with horizontal and vertical conductivities and uses in the horizontal and in the vertical terms. The heat capacity is lumped by row sums ( per brick node, per tetrahedron vertex). Time is integrated with the backward Euler method,
which is symmetric positive definite with the fixed faces. The step is constant inside a time level, so one matrix factorisation serves the level, and every level is twice as long as the one before, with six equal steps per level (twelve with Very fine). A g-function over 25 years needs about 17 levels and 100 steps. Backward Euler lags by up to half a step for the modes active at the end of the step, so each step's value is assigned to , the shift with the smallest error against 64 steps per level; the g-function at the requested times is interpolated linearly in .
Systems up to about half a million unknowns are solved with a sparse direct solver (one symbolic analysis, one numerical factorisation per level, one solve per step); larger ones with conjugate gradients and an incomplete factorisation, warm-started from the previous step. The heat balance (injected heat = change of the stored heat + heat through the faces) is checked at every step; it closes to below (direct) and (iterative) relative.
22.4 Borehole coupling: line points and the well index
Each borehole is represented by line points: on the hexahedral grid the nodes of its axis column, on tetrahedra two Gauss points on every piece of the axis inside a tetrahedron. Point carries the length (the lengths sum to exactly), the shape-function weights of the element containing it, and the local properties; the load is the consistent finite element load of a uniform line source.
A line source in a mesh has a logarithmically singular exact solution; the finite element value at the line node is a field average over the cells around it. Following the argument of Peaceman (1978), in the steady two-dimensional field of a line source the node temperature equals the exact solution at an equivalent radius . The wall temperature is therefore
with the cell size across the axis. On the hexahedral grid is a property of the cell pattern, calibrated once by a steady two-dimensional run: = 0.1161, 0.1154, 0.1151 and 0.1150 for 1, 2, 3 and 4 uniform axis cells, independent of within 0.1 %. An inclined axis passes through its tetrahedra at every possible position, so on tetrahedra is calibrated per point: a steady problem on a patch around the axes, with the exact free-space solution of all finite lines on its boundary, gives the discrete line temperature against the exact wall temperature , and
This takes in the tilt, the local sizes, the borehole ends and the neighbouring boreholes.
Boundary conditions. Under UHTR the wall temperature is the length-weighted mean of (22.6). Under UBWT and under a uniform mean fluid temperature (UFT) one further unknown , the common wall or fluid temperature, is coupled to every point through ( the local fluid-to-wall resistance under UFT, zero under UBWT) and carries the total heat :
( the capacity terms of the step). The bordered system stays symmetric positive definite.
Near-field correction. The well index is exact for the steady field. While the temperature front is still within a few cells of the axis, the line node lags behind the true response. On the hexahedral grid this lag is a property of the plane discretisation around the axis and is measured by a plane run with the same cell pattern, lumping and time levels, compared with the exact self response (the infinite line or cylinder source):
tends to zero at long times. On tetrahedra a plane cannot represent the 3D pattern around an inclined axis; the correction is the transient counterpart of the per-point calibration, a run on the same patch with the exact transient finite line source on its boundary, and with groundwater the exact moving finite line source built from the continuous moving point source of Carslaw & Jaeger (1959). The corrections are weighted by a small local model of the boreholes and their network, each wall connected to the undisturbed ground through its own resistance; under UHTR this is the length-weighted mean, under UBWT on layered ground the better-conducting layers take more heat and more weight.
Short times. Below
(about 17 h at = 0.52 m, 2.5 days at = 1 m) the mesh cannot resolve the borehole. There the g-function follows the shape of the analytic self response (from the local model), scaled to the FEM value at . Peaks of a few hours therefore see the analytic shape; the FEM decides from on.
22.5 Heat exchangers, networks, layers and groundwater
Heat exchangers and the network. Under the network boundary condition (MIFT, mixed inlet fluid temperature: the fluid enters the field at one inlet temperature and the connection network routes and mixes it) every line point carries the nodes of the borehole cross-section: the fluid channels (one per leg of a U-tube; the inner pipe and the annulus of a coaxial pipe), the grout nodes and the wall, connected by the conductances of the thermal resistance and capacity model (TRCM) of Bauer et al. (2011) and Diersch et al. (2011). The wall couples to the ground through the well index. Along each leg the fluid is advected by upwind finite volumes with an exponential fit of the exchange,
with , exact for the outflow towards fixed node temperatures. Series stages pass the outlet of one borehole to the next, parallel groups mix their outlets by flow, and one row closes the field with its heat rate. The g-function element is quasi-steady (no fluid or grout capacity). Its result is an effective-wall g-function that includes the effective resistance of the network, as in the analytic network model, so the simulation and the sizing take it unchanged.
Layers and borehole sections. Every cell takes the conductivity, heat capacity and groundwater velocity of its layer; the cells end at the layer interfaces. Borehole sections put mesh nodes at their boundaries, give each point its own radius in the well index and, under the network condition, their own grout, sleeves, casing and diameter in the cross-section, so a borehole resistance that varies with depth enters the field solution directly.
Groundwater advection. A horizontal heat-transport velocity per layer (the Darcy velocity times ) adds the term . It is discretised by Galerkin advection stabilised by the streamline-upwind diffusion of Brooks & Hughes (1982),
with the element length along the flow. The stabilisation acts only along the streamlines and has zero row sums, so it adds no heat; the heat carried out through the faces is booked in the energy balance, which closes to . The system becomes nonsymmetric and is solved by a sparse LU factorisation or by flexible GMRES, still one factorisation per time level.
22.6 Mesh choice, cost and convergence
Before a run the program builds only the mesh axes (milliseconds) and predicts the unknowns, the memory and the run time from a cost model calibrated on measured runs; the prediction is within about 10 % on an idle machine. The Auto preset (the default) and the Suggest command choose the sizes from the project at a mesh scale (: accuracy first):
- , so that the analytic short-time shape ends at the earliest time that matters (, default 10 h) as far as the radius allows;
- and ;
- growth , because the grading of the far field is part of the resolution.
The expected relative error of is modelled from convergence runs against the exact analytic UHTR g-function of a 3 × 2 field,
( the steps per level). A single borehole converges two to five times faster and large fields somewhat slower, so is an order of magnitude, not a bound. The scale is reduced in steps of 0.8 while exceeds the target error (0.5 %) and the estimate fits the time budget, or increased until the estimate fits. Check mesh runs the chosen mesh and one 1.5 times coarser over at most five years and reports the largest relative change of and its Richardson estimate of the error of the chosen mesh.
The time part of the error is small: doubling the steps per level (6 to 12) improves a 3 × 2 field by about 0.1 percentage points, which is why Very fine combines a gentler grading with twice the steps. A second-order time integrator was evaluated and not adopted, because it could gain at most a few hundredths of a percentage point over doubling the steps.
22.7 The 3D analysis: comparison and temperature field
The 3D analysis computes, for the current design, the FEM g-function , the fluid temperatures with it, and the fluid temperatures of the same project with the analytic g-function evaluated on the same time grid, so that only the g-function differs. Both use the monthly superposition of the design,
( the load steps per metre, the undisturbed ground temperature over the borehole, the effective borehole resistance), and the result page lists the minimum and maximum per year and over the period with .
Temperature field. The run keeps the ground part of its solution at every time step: the unit response , the temperature change at every node for a step of 1 W/m in every borehole, distributed as the boundary condition distributes it. The ground temperature over the design period follows by the same superposition of the monthly base loads,
with the undisturbed profile from the surface temperature and the geothermal heat flux . is interpolated linearly in between the stored frames. Peak loads are not in the field: they last hours and reach only centimetres to decimetres into the ground. From the field the program draws plan maps (by default averaged over the active depth), a vertical section along the groundwater flow, isotherms of the temperature change at user thresholds, the area and reach of each isotherm, how far the plume reaches downstream of the field, and the largest warming and cooling along property lines and at neighbouring wells, with the first month at which each threshold is reached. The field can be exported in the VTK unstructured-grid format.
What needs a new run. The FEM g-function and the field are step responses; the loads enter only through the superposition. A change of the loads, the limits, the thresholds or the evaluation lines therefore redoes only the superposition, in milliseconds. A cryptographic fingerprint of every input that can change the FEM run (geometry, ground and layers, groundwater, borehole construction and sections, fluid and flow, network, boundary condition, mesh and domain, period and the time grid) decides whether saved or cached results can be reused; any change of those inputs makes the results out of date.
Sizing with the FEM engine (available in batch runs, not in the interactive design) keeps the analytic interpolation nodes of the sizing and corrects them: at three lengths the ratio is computed with one FEM run each and interpolated quadratically in . The ratio is smooth and within a few per mille of one, so the interpolation adds less than 0.2 %.
22.8 Example: one borehole in homogeneous ground
A single 100 m borehole with a single U-pipe in homogeneous ground ( = 2.0 W/(m K)), heating-only monthly loads of 6.7 MWh a year with 6-hour heating peaks of up to 4.2 kW, a design period of 2 years, UBWT, Coarse mesh. The mesh has 23 × 23 × 40 = 21 160 nodes and 16 759 unknowns; the heat-travel margin m = 37.2 m and the field-size rule give a block of 74.5 × 74.5 × 208 m. The axis cell is = 1.0 m with = 0.1154, so = 83 h; the run takes 12 time levels with 69 steps to 3.3 years and closes its energy balance to . At the last time the FEM g-function is 5.075 against 5.097 analytic (−0.43 %).
| Mesh | 3D | against 2D | |
|---|---|---|---|
| Coarse | −0.43 % | −2.909 °C | 0.025 K |
| Medium | −0.44 % | −2.908 °C | 0.026 K |
| Fine | −0.44 % | −2.908 °C | 0.025 K |
| Very fine | −0.48 % | −2.907 °C | 0.027 K |
| Auto | −0.43 % | −2.908 °C | 0.025 K |
| Year | 2D | 3D | min | 2D | 3D | max |
|---|---|---|---|---|---|---|
| 1 | −2.11 °C | −2.08 °C | 0.023 K | 8.60 °C | 8.61 °C | 0.004 K |
| 2 | −2.93 °C | −2.91 °C | 0.025 K | 8.34 °C | 8.35 °C | 0.006 K |
For one borehole in homogeneous ground both models describe the same physics, and every preset gives the same answer within a few hundredths of a per cent. The remaining 0.4 % comes from the discretisation and from the boundary condition (UBWT solved on the mesh and with the segments of the analytic model); it moves by 0.025 K against a margin of 0.07 K to the design limit of −3.0 °C, so the 3D analysis confirms the design. With strong layer contrasts, groundwater flow, borehole sections or inclined boreholes the analytic model relies on effective properties or approximations; there is the check of the design, and Check mesh should be used before is trusted for a field.
22.9 Verification
| Check | Result |
|---|---|
| 3 × 2 field, 25 years, UHTR, against the exact analytic g-function | Medium 0.43 %; Fine with growth 1.15 and 12 steps per level (Very fine) 0.19 % |
| the same field under UBWT | Medium 0.79 %; Very fine 0.58 % |
| single borehole (100 m, r_b 55 mm) | UHTR 0.07–0.10 %; UBWT 0.30–0.43 % for every preset |
| inclined boreholes 0–60°, single borehole, UHTR, 2 years, tetrahedra | within 0.22 % |
| vertical field on tetrahedra against the hexahedral grid | single 0.31 %, 3 × 2 field 0.49 % |
| ring of six boreholes at 15°; 200 m at 30° against the inclined line source | 0.64 %; 0.44 % |
| field-size rule of the block against the heat-travel margin alone | g changes by less than 0.012 % |
| groundwater on tetrahedra with the advective near-field patch (Coarse / Medium) | late-time deviation 3.1 % → 0.8 % / 1.3 % → 0.06 % |
| energy balance per step | below (direct), (tetrahedra) |
The single-borehole UBWT difference of 0.3–0.45 % does not shrink with the mesh or the steps: it is an end effect of the borehole ends that scales with , given here for = 100 m and = 55 mm. The remaining error of a field is the grading of the far field, which is why a gentler growth ratio is more effective than a finer axis cell.
22.10 Limitations
- The 3D analysis checks a design; design, sizing, optimisation and the studies stay analytic in the interactive program. If is large after a mesh check, the design should be changed so that the 3D fluid temperatures stay inside the limits.
- Peaks shorter than (83 h at an axis cell of 1 m) follow the scaled analytic shape. The temperature field contains the monthly base loads only.
- The model is linear: constant properties, no freezing, superposed loads, quasi-steady boreholes. The direct simulation with the actual loads is Full FEM.
- Coarse is a preview for fields; hexahedra take vertical boreholes only, inclined fields use tetrahedra.
- On the hexahedral grid the well index and the plane correction assume conduction around the axis; at an axis-cell Péclet number of 0.4 (31 m/a at = 1 m) the late-time g-function is about 1 % low.
- Above the memory limit of the field (default 1 GB) only every k-th time step is kept and the values between them are interpolated.
- Fields with fewer than three axis cells between two boreholes get a warning: the mesh cannot resolve the gap, and a model with a per-borehole ground coupling (the BTES model) is the better choice.