Scrubber Test Bench

Sources & methods

How the scrubber test bench computes what you see, which numbers come from where, and how well it matches published tests. This is an educational model, calibrated to a handful of public data points. It is not a certified test and must not be used for dive planning.

1 · The virtual test

The bench follows the usual unmanned canister-endurance procedure (EN 14143 style). A breathing machine pushes a sinusoidal breath around the loop, and CO₂ is metered into the exhale side at a constant mass rate. The canister sits in a temperature-controlled water bath at the test pressure. The analyser watches the inspired PCO₂ at the canister outlet, and the canister's duration is the time until it reaches the endpoint (0.5 kPa = 5 mbar by default).

The loop re-breathes whatever slips through, so the gas entering the canister carries the injected CO₂ plus the slip. With slip fraction s and a first-order mixing lag (6 L loop volume):

ΔC = V̇CO₂ / (RMV · P) C_in → ΔC / (1 − s) P_ICO₂ = s · C_in · P P_ECO₂ = C_in · P

The flow through the canister is integrated over four points of the half-breath (peak flow = π × RMV), so fast mid-breath gas, which slips most, is weighted correctly.

2 · Bed geometry and flow split

The bed is divided into stream tubes: 6 bulk rings plus a half-grain wall layer in an axial bed, or 6 bulk height layers plus the two layers against the end plates in a radial bed. Every bulk slice is made of 4 interleaved sub-channels (see packing non-uniformity below), and every tube is split into 36 cells along the gas path. Tubes share the same pressure drop, so the flow splits by each tube's Ergun resistance:

Δp/L = 150 μ (1−ε)² u / (ε³ (φ d_p)²) + 1.75 ρ (1−ε) u² / (ε³ φ d_p) φ = 0.8

Wall layer: it has a higher voidage (by packing: +0.05 / +0.08 / +0.14), so it carries extra flow. This is wall channelling, and it matters when the canister is less than about 25 grains wide.
Packing non-uniformity: within every slice, the gas splits between 4 interleaved sub-channels running at 1 ± σ·(0.45, 1.35) of the mean velocity (σ = 0.22 × 0.6 / 1 / 1.6 for tight / good / loose). The fast paths break through first, which is what makes real breakthrough curves gradual. The sub-channels sit at the same place, share their heat, and are drawn as their average.
Radial settling gap: a loose radial bed settles by 0.4 % of its height. The gap at the top acts as a parallel-plate channel straight to the centre tube and bypasses the absorbent.
Axial inlet jetting: short, fat axial beds get extra flow through the centre.
Two canisters in series (sidemount units such as the KISS Sidewinder): the gas path is split into two identical axial beds. The gas leaving the first canister mixes completely (in a Sidewinder it passes through the counterlung) before it enters the second, so a fast channel in the first bed does not line up with one in the second. The mixed gas also loses about half of its excess heat to the water on the way. Each canister has its own end caps, inlet jetting and port losses, and no heat conducts between them.

3 · CO₂ uptake: shrinking-core granules

Each granule is treated as a shrinking core of fresh lime inside a growing shell of carbonate. Per m² of outer granule surface, three resistances act in series:

1/K = 1/k_f + (R/D_e)·(1/ξ − 1) + 1/(k_r ξ²) ξ = (1 − X/0.97)^1/3, R = d_p/2

Gas film: k_f from the Wakao–Funazkri correlation, Sh = 2 + 1.1 Sc^1/3 Re^0.6. It is slower at depth (the diffusivity D ∝ 1/P) and faster in helium.
Product shell: D_e = D_e0 · (D_gas/D_{air,1 bar})^α · exp(−E_d/R (1/T − 1/298)). The fraction α is gas-phase pore diffusion. This is where depth, coarse grain and helium matter most.
Surface reaction: k_r = k_r0 · exp(−E_r/R (1/T − 1/298)).

Each cell removes the fraction 1 − e^−NTU of the CO₂ reaching it, where NTU = K · a_v · V_cell / Q_tube and a_v = 6(1−ε)/d_p. Utilisation is not fixed: it emerges from how long the gas stays against each granule (dwell time) against how fast the shell chokes it.

4 · Heat

Each cell balances four heat terms:

reaction heat of 113 kJ/mol CO₂ (liquid-water basis; any water that evaporates takes its latent heat away with the gas)
downstream transport by the gas, both sensible and latent (saturated vapour, about 2.4 mbar/K at 30 °C, so it dominates at the surface and matters little at depth)
conduction through the bed, k_eff = 0.28 W/m·K
loss through the wall (U = 70 / 30 / 4 W/m²K for aluminium / acetal / insulated) and the end caps (5 W/m²K)

The balance is solved with an unconditionally stable exponential update. Exhaled gas enters the canister at T_water + 0.35 × (32 °C − T_water), because the hoses cool it on the way.

5 · Work of breathing

WOB = ∫ Δp(Q) · Q dt / ∫ Q dt over a sinusoidal half-breath, Δp = bed (Ergun, parallel tubes) + ports (K = 1.5, 28 mm ID)

This is the resistive work of the canister alone, in J/L. For comparison, the whole-rig limit is about 2 J/L (EN 14143 class). The canister should use only a small share of it.

6 · Parameters

Parameter	Value	Basis
Chemical capacity	200 L CO₂/kg	Full conversion of the Ca(OH)₂ in soda lime. Real use: Sofnolime "typical usage capacity" 150 L/kg [1]; Harvey measured 153 L/kg warm [2]; Clarke says under about 50 % of granules react completely [6].
Heat of reaction	113 kJ/mol	CO₂ + Ca(OH)₂ → CaCO₃ + H₂O(l) (standard enthalpies). The CFD model [8] uses 16.4 kcal/mol (vapour basis).
Grain sizes	1.5–3.4 mm	Mesh grades. 797 is 8–12 mesh (1.5–2.5 mm), Spherasorb/408 is 4–8 mesh (2.5–5 mm) [2].
Bulk density · voidage	0.82–0.88 kg/L · 0.37–0.40	Typical soda lime. Harvey: 2.64 kg of 797 vs 2.38 kg of 4–8 in the same canister [2].
k_r0 · E_r/R	0.034 m/s · 6240 K	Fitted (§7)
D_e0 · E_d/R · α	3.3·10⁻⁷ m²/s · 3360 K · 0.17	Fitted (§7)
Packing spread σ	0.22	Fitted to the gradual 0.5 → 1 kPa rise [3]. Shearwater cites 5–10 % run-to-run variance between packings [4].
Wall-heat multiplier	1.04	Fitted (≈ 1, so the physical U values stand)
Test standards	4 °C · 40 L/min · 1.6 L/min · 5 mbar	EN 14143 unmanned endurance test [4]. VCO₂ is about 4 % of RMV [1].

7 · Calibration and validation

The six kinetic and heat parameters were fitted together (Nelder–Mead, log-error) to the cases below. Canister shapes for [2] and [3] are approximations of the real canisters.

Test	Published	Model
Inspiration, 2.64 kg Sofnolime 797, surface, dry bench 19 °C, 45 L/min, 2 L/min CO₂, to 1 kPa [2]	202 min	191 min	✓
Same, 4–8 mesh grain (Spherasorb, 2.38 kg; also a different chemistry) [2]	138 min	153 min	~
O2ptima, 2.08 kg Sofnolime, 18 °C water, 45/2.0, to 0.5 kPa [3]	120 min	143 min	~
… extra time from 0.5 to 1 kPa (shape of the breakthrough curve) [3]	+16 %	+6 %	~
JJ-size axial, 2.44 kg, EN 14143 cold test	≈180 min	170 min	✓
L/kg at 20 vs 40 L/min RMV (InDepth rule of thumb) [1]	+19 %	+18 %	✓
Depth 40 → 100 m, 4 °C, air (manufacturer) [7]	−11 %	−14 %	✓
Depth 58 → 92 msw, 10 °C, coarse 408 grain (US Navy) [7]	−17 %	−12 %	✓
Same, fine 812 grain (US Navy) [7]	≈0 %	−7 %	~
Depth 15 → 40 m, 4 °C (two manufacturers) [7]	−27 / −39 %	−10 %	✗
5× the workload [4]	≈10× shorter	7.8× shorter	~
Water 4 → 15 °C (rule of thumb) [4]	+50 %	+12 %	✗

The previous version of this model (a fixed temperature-limited conversion depth) matched the reference points just as well. But it was almost blind to depth (−1 %), breathing rate (+1 %) and workload (5.3×), and its breakthrough was a sudden step. The shrinking-core model fixed those.

8 · Known limitations

Cold water is under-penalised. The bed's own reaction heat (35–45 °C at the front) dominates the kinetics, so the bath temperature has little effect. Real canisters are thought to lose much more in cold water, but published data is scarce: Clarke notes that the US Navy's cold-water duration data is restricted [5]. Treat warm-vs-cold comparisons as optimistic.
Shallow-depth trend is weaker than the manufacturer data. The depth effect from 15 to 40 m is weaker than the two manufacturer figures. Those figures also conflict with the Navy's "no loss" result for fine grain [7].
Coarse grain is treated only as grain size. The model does not capture differences in absorbent chemistry (e.g. Spherasorb vs Sofnolime), moisture content (15–20 % is best [4]), or drying of the bed when it runs hot.
No gas mixing across stream tubes. CO₂ that creeps along a cold wall stays near the wall. Clarke's carbonate deposits show exactly that pattern [6], but it may overstate channelling in narrow canisters.
Work of breathing is untested against measurements. The Ergun bed resistance is textbook physics, but it has not been checked against measured values. Harvey saw only 10 % less loop pressure swing with coarse grain, for the whole circuit including long hoses [2].

References

InDepth — Estimating your scrubber duration (Sofnolime TUC 150 L/kg; 95 / 85 / 80 % of it at 20 / 30 / 40 L/min; VCO₂ ≈ 4 % of RMV).
Harvey D, Pollock NW, Gant N, Hart J, Mesley P, Mitchell SJ. The duration of two carbon dioxide absorbents in a closed-circuit rebreather diving system. Diving Hyperb Med 2016;46(2):92–97 (ResearchGate).
Performance of cartridge and granular carbon dioxide absorbents in a closed-circuit diving rebreather. Diving Hyperb Med 2019 (O2ptima, ExtendAir vs Sofnolime 797).
Shearwater — The CO₂ scrubber (EN 14143 conditions, temperature and workload rules of thumb, moisture, packing variance).
J. Clarke — Diving a rebreather in frigid water: canister concerns.
J. Clarke — A look inside rebreather scrubber canisters, part 1 and part 2 (moving thermal front, carbonate along cold walls).
J. Clarke — How does your rebreather scrubber handle the deep? (Navy and manufacturer depth trends).
Cunningham S, Burke A, Kelly G. CFD modeling of breakthrough in closed circuit rebreather scrubbers. Aust J Basic Appl Sci 2013;7(5):116–127 (grain size, wall channelling, heat of reaction).
Ergun S. Fluid flow through packed columns. Chem Eng Prog 1952;48:89–94 · Wakao N, Funazkri T. Chem Eng Sci 1978;33:1375–1384 · Levenspiel O. Chemical Reaction Engineering, shrinking-core model.