A 1 m³ box with a 90 m³/h circulator is nominally “15× turnover per hour.” That number is a lie about local reality. Turnover is a tank-mixing fiction. The gas at the sensor port (usually on the return plenum, 1 cm from the blower suction) sees 15–20×/h. The gas behind the spin coater, under the bottom shelf, and in the 12 cm dead corner at the rear-left sees 0.3–1.2×/h — measured by tracer-SF₆ decay mapping in real lab boxes.
The hydrodynamics math
Circulator 90 m³/h through a 1 m³ chamber: mean face velocity at the outlet diffuser (200×200 mm = 0.04 m²) = 90/0.04/3600 = 0.625 m/s. That jet penetrates ~0.5–0.7 m before dissipating into recirculation eddies (Re ≈ 40,000 at 25 °C N₂, turbulent). Beyond that penetration cone, flow collapses to buoyancy-driven roll cells at 0.02–0.05 m/s.
A 5 µm particle in the rear dead zone: settling velocity 0.021 mm/s (Stokes, computed earlier). Local horizontal advection 0.03 m/s would sweep it out in ~30 s — but if local velocity is 0.005 m/s (measured behind a 2-tier tray stack), residence time jumps to 3–5 min per 1 m travel, and vertical settling dominates: particle falls 1.26 mm/min, lands on the rear shelf, re-entrains when gloves move. Net: rear-corner particle count 4–7× the sensor-zone count in the same box at the same LCD reading. Perovskite PL maps from substrates placed rear-left vs center-front differ in shunt density by 8–15% — not from gas purity, from local residence time of contaminated parcels.
Purifier contact efficiency is not uniform
Copper/molecular-sieve column sees the mixed return stream, but the column itself has channeling. A 5 kg bed in a 150 mm canister: superficial velocity 90 m³/h / (π×0.075²) = 5.7 m/s through the bed. At that velocity, breakthrough of H₂O on 4Å sieve is textbook ~900 g — but only if flow distributes evenly. Real canisters with top-entry single diffuser show 20–30% flow maldistribution (CFD, vendor teardowns): one side of the bed does 70% of the work, poisons first, then O₂ slips while LCD (on return line, post-bed) still reads low because the averaged exit is fine. The slip is local and time-dependent; static decay test (doors closed, no work) never excites it.
Glove motion as a transient CFD event
A slow push (hand 5 cm/s) displaces 577 mL (computed prior round) over ~3 s = 0.19 L/s jet into the chamber. That jet, at the glove port mouth (220 mm ring), is 0.19 / (π×0.11²) = 5 m/s peak — an order of magnitude above circulator face velocity. It punches a vortex pair into the chamber that takes 40–90 s to dissipate (PIV studies on glove-box analogs). During that window, the rear dead zone gets scoured (good) but the sensor-zone gets a slug of glove-surface-desorbed H₂O/O₂ (bad, transient +0.2–0.5 ppm O₂ spike that averages out). A lab logging 1-min O₂ traces sees the spike; a lab reading the LCD at lunch doesn’t.
Field contrast (two “identical” boxes, same argon)
Lab U: blower outlet diffuser aimed at center-front, spin coater parked rear-left, trays stacked to back wall. SF₆ tracer: rear-left residence time 52 min effective (0.19×/h), sensor-zone 18×/h. Perovskite shunt density center-front 4/cm², rear-left 11/cm². O₂ LCD identical to sister box.
Lab V: same model, diffuser redirected with a 30° baffle to throw flow along side walls, spin coater on a 50 mm standoff from rear wall, trays on cantilever brackets leaving 80 mm rear clearance, small 20 m³/h auxiliary fan on rear wall (anti-dead-zone). Tracer residence rear-left 6 min (10×/h). Shunt density uniform 4–5/cm² across the whole work surface.
Gas spend identical. Purifier identical. Difference: hydrodynamic layout, zero SKU change.
The ledger you should actually keep (numbered, measurable)
- Do one SF₆ or CO₂ tracer map per box/year: inject 1 mL at glove port, sample 5 fixed points (sensor zone, rear-left floor, under bottom shelf, behind spin coater, top-front corner) at 10 s cadence for 10 min. Compute local e-fold time. Any point >5× the sensor-zone time = dead zone.
- Baffle the diffuser so the jet throws along a side wall, not into the center pile of equipment. $20 sheet-metal job.
- Standoff all equipment 50–80 mm from walls/floor; rear clearance matters most (buoyancy sink).
- Add a 15–20 m³/h auxiliary wall fan (filtered, on the loop) pointed at the known dead corner. Cuts residence 5–10×.
- Log O₂ at 1-min cadence for 24 h (cheap data logger on the analyzer output). Spikes >0.3 ppm correlated with glove motion = hydrodynamic slugging, not leaks.
mate 6. Quarterly “corner strip” check: resazurin/H₂O strip pinned rear-left vs center-front. If rear strip is pink and front is blue at same LCD −78 °C, you have a dead zone, not a gas problem.