HomeFeaturesExamplesTutorialsHelpTheoryVerificationPricingContact
BoreTherm / TutorialsTutorial 06 / 19
Workflow and theory · Ground data & uncertainty

Thermal response test evaluation

How a thermal response test (TRT) becomes the ground conductivity λ, the effective borehole resistance Rb* (the thermal resistance between the fluid and the borehole wall) and the undisturbed ground temperature T0; the quality checks, the evaluation methods and how BoreTherm picks one; and how the values go into the design. The numbers are those of two tests. The synthetic test, stored in the example project (an office building with 16 boreholes), heats a 150 m borehole for 72 h at about 6 kW with a 1.5 h stop, followed by 24 h of recovery; it was generated with λ = 2.60 W/(m K) and Rb* = 0.085 m K/W, so the result can be checked against the truth. The Beier sandbox test is the published laboratory test of Beier, Smith & Spitler (2011): one 18.3 m borehole heated for about 52 h at 1056 W.

01 / The one idea

The one idea: λ from the slope, Rb* from the level

A TRT puts a constant heat rate into one borehole for two to three days and logs the inlet and outlet temperatures. After the first hours the mean fluid temperature Tf (the mean of the inlet and outlet temperature) rises on a straight line against the logarithm of time. The slope of that line depends only on the ground conductivity. Its level depends on the undisturbed temperature and on the thermal resistance between the fluid and the borehole wall.

TRT righeater, pump, flow meter, logger:time, Tin, Tout, flow, powerTinToutHgroundλ, C, T0boreholerb, Rb*Beier sandbox: mean fluid temperature1251020502831343740time since the heating start [h], log scaleT [°C]evaluation windowslope k = Q̄ / (4π λ H)level → Rb* (with T0)measured
Left: the test. Right: the Beier sandbox test (1056 W, 18.3 m): after about 10 h the points lie on a straight line in ln t (M1 fit, solid in the window, dashed before it; M1 is the straight-line method).
PowerQ = ρ cp V̇ (Tin − Tout)
Mean fluid temperatureTf = (Tin + Tout) / 2
Conductivity (slope k)λ = Q̄ / (4π H k)
Borehole resistance (level m)Rb* = (m − T0) / q̄ − [ln(4α / rb²) − γ] / (4πλ)
Superposition (any power history)Tf = T0 + Rb* q̃ + Σ Δqk G(t − tk) / (2πλ)

Tf = k ln t + m over the window; Q̄ is the mean heat rate, H the borehole length and q̄ = Q̄/H; C is the volumetric heat capacity of the ground, α = λ/C, rb the borehole radius; γ = 0.5772. G is the response of the borehole to a unit heat rate, modelled as a line source (heat released along the borehole axis, infinite or finite in length) or as a hollow cylinder at the borehole wall, and q̃ the heat rate after a first-order lag of 1 h, the heat capacity of fluid and grout.

02 / The workflow

The workflow

The evaluation is a wizard with six steps, started from Tools › TRT evaluation…. Two questions decide whether the result is fit for the design: does the test pass its acceptance checks, and is there a fit the program can recommend? A fit is one evaluation method applied to the data; the Monte Carlo of step 5 repeats the evaluation many times with the sensor, power and borehole data varied within their errors.

1 Importrig file + column map, a .trt.json, or the stored test2 Test and borehole databorehole, pipes, fluid; T0 method; power source3 Data quality: does anacceptance check fail?yesReview the dataexclude faulty intervals (raw data stay),check sensors; maybe repeat the testnocontinue with the findings4 Evaluationall methods, window, sequential plotIs a fit recommendedand plausible?noAdjust and evaluate againwindow, forward models, fit C or T0;read the warnings and findingsyes5 Comparison and uncertaintyall fits with 95 % intervals,Monte Carlo of the input errors6 Apply to projectλ, T0 → surface temperature,Rb* → calibrated grout conductivitythen Calculate or Size borehole length with the measured values
A failed check does not stop the evaluation: the fits are still calculated and carry the findings. A numerical fit is not test acceptance.
TRT evaluationAdvanced analysis:the test in the projectTRT evaluation wizard (Tools › TRT evaluation…)1 Import2 Test data3 Quality4 Evaluation5 Comparison6 ApplyGroundλ; surface temperaturefrom T0; C if estimatedBorehole cross-sectiongrout conductivity(calibrated) or fixed Rb*GroundwaterDarcy velocity of M3(stored only, on request)Also in the wizard: Store the test in the project, TRT report (PDF / HTML), Save test (.trt.json).Command line: boretherm-cli trt import | evaluate | apply.
The TRT evaluation page under Advanced analysis shows the test stored in the project and opens the wizard. Only step 6 changes the project, in one undoable step.
  • Apply defaults: λ (matched to the effective value over the test length in layered ground) and the calibrated grout conductivity are on. The surface temperature from T0, C (only when estimated) and the Darcy velocity are options. Fixed Rb* is the alternative of the classic monthly method to the grout calibration.
03 / How the calculation works

How the calculation works

1 Power and fluid temperatureQ measured, or ρ cp · flow · ΔT;Tf = (Tin + Tout) / 22 Screeningspikes, operating events, gaps,interruptions: flags, data kept3 Quality checksVDI 4640-5 / ASHRAE limits:pass, warning, fail6 Diagnostics and choice95 % intervals, n_eff,identifiability, BIC rule5 Fit the methodsM1 slope, M2 superposition,M3 groundwater, M5 recovery4 T0 and the windowT0 before the heating; windowfrom max(10 h, 5 rb²/α) to the end
The raw records are never changed: screening and exclusions only set flags. The heat injected during an excluded interval still counts in the load history.
  1. Power and Tf. The power is the measured column, or ρ cp V̇ ΔT with the fluid properties at the record’s mean temperature. The heating end is the last sample above half the typical power.
  2. Screening. A Hampel filter (local median ± 4 × 1.4826 × MAD, the median absolute deviation) flags spikes; level changes in flow and power are operating events. Flagged temperatures are left out of the fits by default.
  3. Quality checks (section 6) give pass, warning or fail, and each fit gets a suitability: suitable, review required or unsuitable.
  4. T0 is the mean fluid temperature of the circulation before the heating (or manual, or the mean of a depth profile). The window starts at max(10 h, 5 rb²/α).
  5. Methods. M1: least squares on (ln t, Tf), plus the sequential evaluation (up to 40 window ends and 40 window starts). M2: λ and Rb* (optionally C, T0) by Levenberg–Marquardt on the superposition model, which follows any power history and the recovery, with three kernels: infinite line (ILS), hollow cylinder (CHS), finite line (FLS). M3: moving line source for groundwater. M5: the recovery after the stop.
  6. Diagnostics and choice. 95 % intervals, the residual autocorrelation and the effective number of independent observations neff, identifiability (whether the data determine the parameters), then the rule below.
1 Your choiceif it succeeded, isidentifiable, not unsuitable2 M2 forward modelILS is the reference; CHSor FLS only if ΔBIC ≥ 103 M3 groundwaterreplaces it if identifiableand ΔBIC ≥ 10 (+1 param.)4 Fallbackselse M1 if identifiable,else M5 if identifiable5 Nothingno fit is recommended;Apply is refusedA fit that is not identifiable (a parameter at its bound, |ρ| too high, the λ interval wider than ±20 %)is shown as a sensitivity but never recommended. M5 is the independent check while M2 succeeds.
The recommendation rule, first match wins. BIC is the Bayesian information criterion: ΔBIC = neff ln(SSRref / SSRalt) − Δm ln neff, with SSR the sum of squared residuals, Δm the number of extra parameters and neff the effective number of independent observations.
04 / The theory behind the window and the methods

The theory behind the window and the methods

Why the window starts late: the sequential evaluation

The line-source slope is only valid when t ≫ rb²/α and the heat capacity of the borehole no longer matters. The sequential plot shows it: M1 is repeated with a moving window end and a moving window start. A good window is where λ is flat.

λ of M1 against the window end10203040502.02.53.03.54.0window end [h] (start 10 h)λ [W/(m K)]λ of M1 against the window start0102030402.02.53.03.54.0window start [h] (end 51.8 h)λ [W/(m K)]M2 ILS (recommended) 2.83published 2.88late starts: few points left,the slope scattersshort windows: start-upeffects still in the slope
Beier sandbox. Left: λ settles once the window is longer than about a day. Right: with a late start too few points remain. If λ keeps rising with the window end (by more than 5 %), groundwater flow (M3) or a drifting power is the likely cause.

Why superposition: a test is never perfectly constant

M1 assumes a constant power from the start. A stop, a drift or a daily swing bends the curve and biases the slope. M2 replays the actual power history step by step (temporal superposition, evaluated by FFT), so it can fit through a stop and into the recovery.

Heating power [kW]06stop 1.5 hMean fluid temperature [°C]0122436486072849610141822time since the heating start [h]measuredM1: one straight line in ln tM2 CHS model follows thestop and the recovery
Synthetic test with a 1.5 h stop after 28 h. The M2 model (cylinder source, recommended) follows the dip and the recovery (root mean square error, RMSE, 0.017 K); the M1 line cannot (RMSE 0.64 K), and its λ moves by 49 % between window starts: M1 is unsuitable here.

Which borehole model, and how sure is λ?

The line source (ILS) is the reference. The hollow cylinder (CHS) puts the heat in at rb but lets the borehole store none, so over usual windows it reads λ some 4–9 % lower than the ILS. A real borehole stores heat in its fluid, pipes and grout, which is closer to the line source; the CHS replaces it only when the data show it clearly (ΔBIC ≥ 10). The synthetic test was generated with the cylinder model, and the rule finds it.

Synthetic test2.42.62.83.03.2known 2.60M1 slopeM2 ILSM2 CHSM2 FLSM5 recoveryBeier sandbox2.42.62.83.03.2published 2.88M1 slopeM2 ILSM2 CHSM2 FLSFilled dot: λ with its 95 % interval (often thinner than the dot). Open dot: unsuitable. Orange bar: Monte Carlo P5–P95 of the input errors.
λ by method. Synthetic: the recommended CHS fit is −0.1 % from the known 2.60. Beier: the recommended ILS fit is −1.7 % from the published 2.88; the CHS reads −6.8 % lower.
The 95 % interval of a fit is not the uncertainty of λ. The residuals of a TRT are strongly correlated (Beier: r1 = 0.988, so 2260 observations count as 14 independent ones), and the interval ignores the errors of the sensors, the power and the borehole data. For the Beier ILS fit the interval is ±0.3 %; the Monte Carlo of the input errors (200 samples, default errors) gives P5–P95 2.68–3.00 W/(m K), about ±6 %. Use that range, not the fit interval, as the uncertainty of λ in a Monte Carlo design.
05 / How each result is derived

How each result is derived

The synthetic test (150 m, rb 76 mm, recommended fit: M2 superposition (CHS)).

ResultHow BoreTherm derives itSynthetic test
Undisturbed temperature T0mean Tf of the circulation before the heating (or manual, or a depth profile)11.30 °C (initial circulation (30 samples before the heating start))
Mean power Q̄ and q̄ = Q̄ / Hfrom flow and ΔT: ρ cp V̇ (Tin − Tout) at the mean fluid temperature5877 W, 39.2 W/m
Evaluation windowfrom max(10 h, 5 rb² / α) to the heating end; recovery from the window start after the stop10–72 h, recovery from 82 h
λ and Rb* (M2)Levenberg–Marquardt on the superposition model of section 1 (T-13.2) over the window and the recoveryλ 2.599 W/(m K), Rb* 0.0850 m K/W
95 % interval of λcovariance s² (JᵀJ)−1, Student t; assumes independent residuals2.597–2.601 (±0.07 %)
Effective observations neffn (1 − r1) / (1 + r1), r1 = lag-1 autocorrelation of the residualsCHS 1498, ILS 49 of 2278
Model choiceΔBIC = neff ln(SSRILS / SSRCHS) with the larger r1 of the two; ≥ 10 replaces the ILS49.5 · ln((0.0967 / 0.0169)²) = 172.4 → CHS
M1 and M5 as checksM1: slope over the window (unsuitable across the stop); M5: recovery with the kernel of M2M1 2.715, M5 2.593 (T0 11.29 °C)
Grout conductivitysolves Rb*(kg; test borehole, test flow, mean Tf) = Rb*TRT with the multipole method1.60 → 2.48 W/(m K)
Surface temperatureshifted so that the undisturbed temperature over the test length equals T09.00 → 9.57 °C

Into the design. Applied to the example project (16 boreholes), λ goes from 2.20 to 2.599 W/(m K), the grout and the surface temperature change as above, and the Rb* of the design is recalculated for its own flow and borehole (0.1198 → 0.0942 m K/W). The length from Calculation › Size borehole length goes from 154.5 m to 130.1 m (−16 %). The grout calibration transfers a material property: the project borehole differs from the test borehole, and its Rb* still follows the design flow and length.

06 / The example in numbers

The example in numbers

CheckLimitSyntheticBeier sandbox
Heating duration≥ 48 h, failpass: 72.0 hpass: 51.8 h
Power: sd / mean (10-min means)≤ 1.5 %, failfail: 14.6 %pass: 0.5 %
Power: largest deviation / mean≤ 10 %, failfail: 100.8 %pass: 1.7 %
Specific heat rate Q̄ / H30–80 W/m, warningpass: 39.2 W/mpass: 57.7 W/m
Flow: sd / mean≤ 5 %, warningpass: 0.0 %pass: 0.1 %
Reynolds number≥ 2300, warningpass: 19121pass: 13221
Energy balance: measured vs flow · ΔT≤ 10 %, failpass: 0.0 %pass: 0.1 %
Heating interruptionsnone, warningwarning: 1pass: 0
Operating events (level changes)none, warningwarning: 2 eventswarning: 3 events
T0 against Ts + q_geo depth / λ≤ 2 K, warningpass: −0.1 Kn/a: –

The synthetic test fails the power checks on purpose: the 1.5 h stop. Its suitability is review-required. The Beier test passes the acceptance checks but has 3 operating events, so it is review-required too.

Methodλ synth.Rb* synth.fit, statusλ BeierRb* Beierfit, status
M1 slope2.7150.09170.636 K, neff 2 · unsuitable2.9230.15790.036 K, neff 29 · unsuitable
M2 ILS2.8290.09560.097 K, neff 49 · review2.831 (rec.)0.15480.053 K, neff 14 · review
M2 CHS2.599 (rec.)0.08500.017 K, neff 1498 · review2.6390.14570.060 K, neff 18 · review
M2 FLS2.8200.09550.097 K, neff 49 · review2.7460.15360.051 K, neff 15 · review
M5 recovery2.593–0.014 K, neff 421 · review––not run

λ in W/(m K), Rb* in m K/W. Known values of the synthetic test: λ 2.60, Rb* 0.085. Beier: published λ 2.88; no recovery recorded, so M5 is not run. Recommended: synthetic M2 CHS (ΔBIC 172.4 over the ILS); Beier M2 ILS (the FLS fits slightly better, but not significantly).

07 / Limits to keep in mind

Limits to keep in mind

  • A TRT measures λ as the effective value over the test borehole’s active length. In layered ground BoreTherm can match one layer so that the effective value over that length is right; it cannot resolve the layers from one test.
  • The choice of the borehole model (ILS or CHS) changes λ by several percent, more than the noise of a good test. Keep the recommended model unless you have a reason.
  • The fit interval is optimistic. Use the Monte Carlo of the input errors (wizard step 5) for the uncertainty of λ and Rb*, and carry it into a Monte Carlo of the design.
  • Groundwater flow, a bounded test domain or a drifting power all look like a λ that keeps rising. M3 is recommended only when it is identifiable and fits significantly better; the flow direction cannot be found from one borehole.
  • Screening flags are heuristics. Keep the raw file, read the operating log, and exclude intervals only for a reason.
  • Apply changes the design inputs, not the test: the report of the evaluation (wizard step 5, TRT report) is the evidence to keep with the design.

Numbers calculated with BoreTherm. Theory: BoreTherm theory manual (VDI 4640 Part 5, ISO 17628, ASHRAE). Beier data: Beier, Smith & Spitler (2011), Geothermics 40(1).