Open Source Bioreactor: Agitation design basis
Why the impeller is the size it is, what actually limits agitation in this reactor, and what is still unknown. The sources are in References.
1. What limits agitation
Section titled “1. What limits agitation”Ranked by what the evidence supports:
- Bubble rupture at a sparger or free surface - dominant in any aerated vessel, 10^7-10^9 W/m^3
- Local maximum energy dissipation rate epsilon_max near the impeller - not the vessel mean
- Shear stress in Pa - the practical metric for microalgae, measured directly in Couette devices
- Kolmogorov microscale against cell diameter - real principle, not binding at these scales
- Tip speed - a convenient proxy with no mechanistic standing
Bubble rupture, not impeller shear, is the dominant damage mechanism. Sobczuk et al. studied P. tricornutum and Porphyridium on one rig and concluded that mechanical agitation was not the direct cause of damage - bubbles rupturing at the culture surface were, with the impeller implicated only as the thing that generated them. Independently, for mammalian cells, removing the gas-liquid interface leaves growth and viability unaffected at 100-450 rpm, at 600 rpm in a fully-filled vessel, and at 1500 rpm for hybridomas.
Tip speed is not a design criterion. It does not have the dimensions of either shear rate (s^-1) or shear stress (Pa), a point Nienow makes in 2006 and repeats in 2021. The widely-quoted 1.5 m/s limit traces to a single measurement on human melanoma cells in 5 % serum, generalised into a universal rule in a paper that two sentences later reports 6 m/s causing no measurable harm to a hybridoma line. No tip-speed limit is asserted anywhere in this model, deliberately.
For Chlorella vulgaris the relevant figure is an optimum, not a threshold. Leupold et al. measured a peak at 1.26 m/s tip speed (0.45 Pa) - 4.0 % higher photosynthetic activity and 48 % higher growth against unstirred - with activity falling back to the unstirred control by 2.03 m/s (0.9 Pa) and 7-8 % below it by 5.89 m/s. No lethal limit was found at any speed tested. Chlorella is a rigid-walled green alga, the most shear-tolerant class.
Assumption, stated once and carried everywhere below. Leupold’s apparatus is a Gust microcosm of 1.75 L, mixed by a spinning plate with skirt 5 cm off the bottom - not a stirred tank with an impeller. Their tip speed is that plate’s rim speed, and the shear it makes lives in a plate-to-wall gap rather than an impeller’s discharge jet. Reading 1.26 and 2.03 m/s as impeller tip speeds is this project’s transfer, not their result. It is the only Chlorella-specific agitation measurement available, so the band below is built on it anyway - but every speed, every D/T argument and the motor selection itself inherit that assumption. See References for what the paper actually reports, including that its calibration is direct only to 250 rpm and that the quoted Pa figures are the stress the authors call secondary in their own experiment.
2. This reactor, in the units that matter
Section titled “2. This reactor, in the units that matter”Computed on the 70 mm Rushton turbine in the 10 L jar, with Np = 4.17 +/- 0.14 measured (Kaiser 2016) and x = 12. Mean dissipation is over the 8.17 L this build derives from
culture_fill_fraction, not over the full jar - which holds 9.45 L brim full - and it is one
impeller’s power, the floor for the pair (below). head() reports the two registered-drive rows;
the others are computed from the same relations at the speeds the Chlorella band asks for.
| shaft speed | tip speed | Re | mean epsilon (W/m^3) | epsilon_max (W/kg) |
|---|---|---|---|---|
| 320 rpm - registered drive, rated | 1.17 m/s | 26,000 | 130 | 27.1 |
| 344 rpm - Chlorella optimum | 1.26 m/s | 28,000 | 162 | 33.6 |
| 420 rpm - registered drive, no-load | 1.54 m/s | 34,200 | 294 | 61.2 |
| 554 rpm - break-even | 2.03 m/s | 45,100 | 675 | 141 |
The pair draws more than the table. McCabe, Smith & Harriott (1993 p. 253) put two turbines at
least one impeller diameter apart at about 1.9x one; these are exactly one apart, so the pair’s
mean at the rated speed is nearer 250 W/m^3 than 130. head() reports one impeller’s power as the
floor and both summed as the torque’s ceiling, and does not pick a factor between them.
The power number is Kaiser’s, not the textbook’s. McCabe’s Table 9.3 gives 5.75 for the standard disc turbine, from tanks with four baffles at T/10. Kaiser measured 4.17 at laboratory scale. This vessel matches neither exactly: four at T/12 rather than T/10. The model keeps the measured figure, and the difference is 38 % on every power and torque figure: at 5.75 the rated speed draws 1.46 W, 179 W/m^3, and the pair stays under about 0.087 N-m of the 0.490 rating, so nothing about the drive turns on which is right. A bench measurement of the printed part settles it (#133).
The vessel is fully turbulent throughout the band. Every speed above clears both the textbook Re > 10^4 threshold and Nienow’s stricter 2x10^4.
The margin against damage is large. At the rated 320 rpm the impeller’s peak dissipation is ~2.7x10^4 W/m^3 against bubble rupture at 10^7-10^9 W/m^3 and a CHO lethal range of 10^6-10^8 W/m^3 - two to five orders of magnitude. Agitation is not what threatens this culture.
Aeration dominates the power budget too. The one peer-reviewed worked example for a microalgal stirred tank puts aeration at 29.4-49.1 W/m^3 against stirring at 0.65-1.30 W/m^3.
The operating point
Section titled “The operating point”On the 70 mm impeller the band worth aiming at is wider than the drive:
| target | tip speed | shaft speed |
|---|---|---|
| Chlorella growth optimum | 1.26 m/s | 344 rpm |
| break-even, stirring stops paying | 2.03 m/s | 554 rpm |
The registered drive is the 36PG-555PM-14-EN, 14:1, rated 320 rpm and no-load 420 rpm - 1.17 to 1.54 m/s. It was chosen on the alternative 94.5 mm pitched pair, where that span was 1.58 to 2.08 m/s and it was the only motor examined that reached the break-even end of the band. On the standard turbine it brackets the optimum and stops short of break-even, so the optimum-against-break-even comparison wants either a faster motor or the larger impeller back (Baseline).
It was taken over a 19:1 sibling whose entire rated-to-no-load span, 265-385 rpm, sat inside the pitched pair’s band. That containment is a workaround for having no speed feedback: with nothing measuring the shaft, where it settles between rated and no-load is set by load, so both ends have to be safe. This motor carries a magnetic encoder instead - 2 channels, 12 PPR at the motor shaft, which past the 14:1 is 672 quadrature counts per output revolution and resolves about 0.9 rpm over a 100 ms window, against a band 210 rpm wide. Commanding a speed and measuring it beats choosing a motor that cannot miss.
Torque is not what limits the choice. The impeller pair draws under 0.063 N-m at rated and under
0.109 N-m at no-load against a 0.490 N-m rating - 13 to 22 % of it - so the shaft runs nearer
the no-load end than the rated one. That is why the registry carries rated torque and head()
reports the comparison at each speed: it is the fact that says where an unmeasured shaft settles.
The registries carry output speed, reduction ratio and rated torque, so head() computes Re, tip
speed, power, torque and both dissipation figures at render rather than leaving them to be worked
out by hand. No-load and rated speed are registered separately: across the ten ratios of the
36PG-3429 table, rated runs 0.68-0.71 of no-load, so treating a catalogue speed as an operating
point overstates it by about half. An unqualified vendor figure is a no-load figure.
3. Impeller geometry
Section titled “3. Impeller geometry”The standard turbine
Section titled “The standard turbine”Two six-blade Rushton turbines in the standard turbine’s proportions: D = T/3, blades D/5 high and D/4 long (McCabe, Smith & Harriott 1993 pp. 242-243, Fig. 9.7, after Rushton, Costich & Everett 1950), on a 3D/4 disc (Middleton & Ramshaw, US 5198156; McCabe gives no disc diameter). It is the configuration the stirred-tank literature compares against, which is why the reference build runs it; Baseline says which settings are held at a standard and which later work moves.
A Rushton’s flat blades have no hand, so both positions take the same part, and it prints as two halves (Building it).
Diameter
Section titled “Diameter”impeller_bore_ratio is undef and takes the Rushton row’s 1/3: 70 mm in jar_10L’s 210 mm bore.
It is measured against the vessel’s wetted bore - the internal diameter, not the outside of the
glass. This matters: multiplying the outer diameter lets the real ratio drift with wall thickness,
which across this project’s vessel registry meant 0.468 to 0.489 for one nominal 0.45.
1/3 is near the bottom of the 0.3-0.5 band the model echoes. For solids suspension McCabe gives D/T 1/3 as often recommended, “though some prefer” 0.4 (p. 268), and a larger impeller disperses more gas at a given speed (p. 272). Section 4 is the case for larger; the pitched pair below ran at 0.45.
jar_6p5gal_305x470 is the vessel the mouth binds. Its 137 mm mouth on a 280.8 mm bore caps
the ratio at 0.4879 with the impeller exactly filling the neck. At 1/3 the impeller is 93.6 mm and
passes with 43.4 mm to spare; at 0.45 it would leave 10.64 mm. The model asserts that what the impeller sweeps
can pass the vessel’s opening.
Clearance and spacing
Section titled “Clearance and spacing”The lower impeller is T/3 off the floor, 70 mm on jar_10L and exactly 1 D: McCabe’s E/Dt of
1/3 for one turbine, and his “about one impeller diameter above the bottom” for the lowest of
several (p. 241), which at D = T/3 is the same height.
The upper is a further T/3 up, 1 D apart. That is the least spacing McCabe gives for a pair whose power nearly doubles (p. 253), and it is the equal thirds of Rutherford et al. 1996’s merging flow at H = T, as Li et al. 2007 Fig. 1 relays it. Rutherford’s alternatives are parallel flow (lower T/4, the pair T/2 apart) and diverging flow (lower 0.15 T); the thresholds between them are not claimed here.
On jar_10L that leaves 63 mm under the lower impeller’s blades, 56 mm between the two impellers’
blades, and 92.5 mm - 1.32 D - of culture over the upper.
Two impellers, at H/T 1.09. McCabe keeps one turbine to a liquid depth equal to or somewhat over T and adds more for deeper (p. 241), so this column sits at the edge of needing a second. Two is on the advice of the microalgae researcher advising the project.
The twisted paddle
Section titled “The twisted paddle”impeller_twisted_paddle_4 is a constant-pitch helicoid, so the blade angle beta from the plane of
rotation varies with radius, tan(beta) = P/(2 pi r), running 83 deg at the hub to 53 deg at the tip,
and no correlation reaches it: Medek’s envelope stops at 60 deg, and the twist papers tested at most
20 deg of twist where this is 30 deg (References).
That is steeper than 45 deg everywhere - a twisted paddle biased toward radial pumping, not the
axial impeller the D/T guidance is written about. Its blade height has no source either. Its hub is 10 mm radius (0.212 D) to carry the set screws,
within the 0.2-0.33 D hubs carry in practice; a local boss leaving the blade alone was rejected
because it prints with an overhang and has to be indexed against the blades; the joint was the
binding problem, and a slipping impeller costs the whole run.
It stays registered and drawn; a bench measurement of its power number is future work.
The pitched pair
Section titled “The pitched pair”The alternative is two 45 deg four-blade pitched turbines, impeller_pbt_45_4, at D/T 0.45, the lower
0.9 D off the floor and the pair 1 D apart. That layout is the row’s own tank field, so it is one
head_impeller_lower_type / head_impeller_upper_type setting away, and a Rushton under a pitched
blade is future work. For that layout:
-
Po and flow number are correlated. Medek’s correlation (Fort et al. 2002) gives them from the geometry and names the conditions a vessel breaks - on
jar_10L, T/D and H/T. -
Diameter. 0.45 sits mid-band on every source for an axial impeller: 0.3-0.5 generally, 0.4-0.5 in cell culture, 0.44-0.46 as a manufacturer’s “most preferred”. Above roughly 0.5 an axial impeller loses its strong axial motion.
-
Clearance 0.9 D, after Fort. He tested pitched blades at C/D 0.5 and 1.0 and found hydraulic efficiency higher at 1.0, tying low clearances to solids suspension and high ones to blending. Medek’s correlation reproduces it, since
Poscales as(C/D)^-0.165andN_Qas(C/D)^0.254:C/D Po N_Q N_Q/Po vs 0.6 0.6 1.596 0.800 0.502 - 0.8 1.522 0.861 0.566 +12.8 % 0.9 1.492 0.887 0.594 +18.5 % 1.0 1.467 0.911 0.621 +23.9 % Not 1.0, which is the correlation’s own C/D limit and drops coverage over the upper impeller below half a diameter on
jar_10L. The quarter-to-third of tank diameter usual for an axial impeller maps to C/D 0.556-0.741 in that bore; no source held here states it, so no band function encodes it. -
Oldshue’s 1-2 D spacing is a fluidfoil statement. His passage is about “these fluidfoil impellers”, both the allowance and the short-circuiting caveat. The band
head()echoes has no primary (Fitschen relays it from Davis, whose sources do not carry it), so it is reported and never asserted. At 1 D the pitched pair sits below the ~1.55 D at which pitched-blade pairs act independently, and inside the 0.33-0.5 T Nienow specifies for a stacked pair. -
The 0.5 D coverage floor is reasoned, not cited. Oldshue names the concern and gives no number; it is where a down-pumping impeller starts drawing its own discharge back off the surface.
head()warns rather than asserts. -
Blade count. Measured on otherwise identical folded-blade axial impellers, Po runs 0.79 / 0.99 / 1.34 for 3 / 4 / 6 blades, so six costs about 35 % more power than four at the same speed and diameter.
-
The two are mirror images, not the same part turned over: pumping direction follows blade handedness and no rotation changes it.
4. Why large and slow
Section titled “4. Why large and slow”At fixed tip speed, peak dissipation falls as the impeller grows:
epsilon_max = 1.04 * x * Po^(3/4) * N^3 * D^2 with v_tip = pi N D fixed: epsilon_max scales as 1/DDoubling impeller diameter halves the maximum local energy dissipation, because at fixed tip speed
the power grows only as D^2 while the swept volume it is spread over grows as D^3. At constant power per volume in a fixed vessel,
v_tip scales as D^(-2/3) while pumping Q scales as D^(4/3) - a larger impeller runs slower at the tip and pumps
more. Peak-to-mean dissipation scales as (T/D)^3, so moving D/T from 0.33 to 0.50 cuts it by ~3.5x.
This is why shear-sensitive culture wants a large impeller turning slowly, and it runs opposite to the intuition that gentler means smaller and faster.
One caution: “low shear” impellers are not reliably gentle. A low power number can coexist with a higher peak dissipation than a Rushton turbine. Size the impeller; do not trust the label.
The reference build runs at D/T 1/3 anyway. The case above is for the knob, not the baseline:
the first paper’s results have to compare with the literature, so the build holds the standard
turbine and a larger impeller is later work (Baseline). At equal power and diameter a
Rushton is the gentler of the two blades, about 0.56x a pitched blade’s peak, since epsilon_max goes as
x * Po^(-1/4) (Grenville 2017’s x and Po, Table 1).
5. Agitation mode per vessel
Section titled “5. Agitation mode per vessel”Not every registered jar can be agitated the same way, and the reasons are geometric rather than preferential. Three of the five swept vessels carry no baffles, and they are unbaffled for two different reasons.
| vessel | mouth | bore | mouth/T | baffles | why not |
|---|---|---|---|---|---|
jar_1gal_180x197 |
148 | 170 | 0.871 | 4 | - |
jar_10L_220x305 |
142.2 | 210 | 0.677 | 4 | - |
jar_6p5gal_305x470 |
137 | 280.8 | 0.488 | 0 | the port circle falls inside the impeller |
jar_1gal_155x251 |
95.8 | 149.3 | 0.642 | 0 | mouth below the flange floor |
jar_1p5L_112x215 |
88.4 | 103.5 | 0.854 | 0 | mouth below the flange floor |
The absolute floor (Port layout): below about 98 mm no baffle fits beside
two dia 16 Atlas probes at any port count. That is jar_1p5L and jar_1gal_155. Note that jar_1p5L
has a generous mouth for its bore - 0.854 - so its problem is absolute size, not proportion.
The ratio bound: jar_6p5gal’s mouth is small relative to its BORE, so the
port circle its lid can offer sits inside the impeller’s own sweep. Measured by adding a
three-baffle set and rendering: the flanges fit - 7, 8 and 9-port sets all clear at a 137 mm mouth,
and Oldshue sanctions three baffles at equal projected area - and the vessel then failed on
something else, “a 126.36 mm impeller leaves no room for a baffle on a 107 mm port circle”. No port
count reaches it; D/T would have to fall to about 0.20 against a 0.3 floor. At D/T 0.45 this jar
could not carry a lid-hung baffle at all, and the port table is not what stopped it. The
measurement was taken on the pitched pair; by the same 0.20 bound the Rushton’s 1/3 does not reach
either, though the set has not been rendered again at it.
The three modes
Section titled “The three modes”Baffled stirred - jar_10L, jar_1gal_180. The reference configuration; nothing here changes it.
jar_10L’s builds, the reference and the magnetic, carry the four on the wall that the table counts;
jar_1gal_180 hangs three plates from the lid.
Unbaffled eccentric stirred - jar_6p5gal. It is the family’s worst case today: the largest
vessel, unbaffled and centred, which Montante measured at flow number 0.25, 65 % below the same
impeller baffled. It cannot be baffled, but it builds, it carries a top-entry drive, and it has
room to offset one. head() prices that room: 9.8 mm today, e/T 0.035, worth 15.8 % of the
centred blend time by Karcz; about 25 mm with the mount at the dia 36 floor its own gearbox
faceplate allows, e/T 0.090, worth 31 %. This is a mount change on a jar that already works, not a new mode of
agitation.
What rides with it, and none of it is small: six of the eight Karcz departures fire here and Reynolds
is one of them; Galletti reports macro-mixing in an eccentric unbaffled vessel is UNSTEADY, the
vortex oscillating slowly rather than sitting; Cabaret reports power RISING with eccentricity; and
Hall’s measured e = 0.2 T stays out of reach at 0.090.
Gas-driven - jar_1p5L, jar_1gal_155. For these two the eccentric-versus-centred question does
not arise, because neither can carry a shaft to centre or offset. Four things point the same way:
- Both build failures are assertions about the top-entry drivetrain. The mount overlaps the port
flanges by 12.5 mm on
jar_1p5L; the pH probe runs 6.29 mm through the lower impeller onjar_1gal_155. Neither assert exists in a configuration without that drivetrain, which the model expresses asdrive_name = "none"; both jars are registered as the airlift built on it. jar_1p5L’s impeller is already negated at the design aeration rate. Even a Rushton, which needs only 3x the gas stream’s power to hold its pattern against Oldshue’s 8x for an axial blade, holds 0.5 vvm there only above 399 rpm, and the drive is rated at 320, where the ceiling is 0.258 vvm. It is the only vessel in the family that floods at rated speed; the others hold 0.5 vvm down to 227-297 rpm. At intended conditions it is a bubble column with a spinning obstruction in it.- Measured, on this organism, at this scale. Uyar 2024 ran C. sorokiniana in 2 L side-illuminated columns and moved the stirrer from 100 to 200 rpm for 15 % on kLa and 18 % on mixing time, while the sparger moved kLa five-fold. Productivity ranked bubble column 0.097, airlift 0.072, stirred tank 0.064 gdw/(L day). The impeller is not what makes a vessel of this size work.
- Swirl is specifically wrong for a side-illuminated column. Molina Grima: average irradiance “considers only the total length of the dark and the light periods, not the frequency of switch”. Solid-body rotation preserves a cell’s radius, so it is close to the worst available flow here - the failure is not slow blending, it is that the light/dark cycle never happens.
An AIRLIFT rather than a plain bubble column, and the reason is this project’s own objection to leaning on aeration: aeration that mixes as a side effect is an uncontrolled variable. A draft tube gives a riser/downcomer ratio, a circulation velocity and a circulation time - quantities Uyar measured at 10.9-12.1 cm/s and 6.1-6.8 s, and which a bubble column does not have at all; that paper computes circulation time “only for ALR which has a circulation loop, it is not applicable to BCR nor STR”. Uyar’s bubble column did out-produce its airlift, but the authors attribute that to the sparger - microporous against a single 0.8 mm orifice - so it is not evidence that a column beats an airlift at equal sparging.
Nor is the airlift expected to out-grow the column. Sanchez Miron 2000 ran a bubble column, a split cylinder and a draft-tube airlift side by side on one alga at one power input and found “no significant difference in culture performance”. What the draft tube buys is the loop - a cell goes up the core and down the lit annulus - not biomass.
drive_name = "airlift" is that mode: the no-drive build plus a draft tube standing on the floor
round everything that hangs from the lid, sparged inside by the ring, which also centres it. On
jar_1p5L the tube is 85.6 mm across with A_d/A_r 0.508; on jar_1gal_155, 93 mm and 1.72 - the
narrower mouth caps the tube while the bore sets the annulus. The gaps under and over it give the
turn the downcomer’s own area, inside Chisti & Moo-Young’s 1.65 cap. Chisti’s circulation model is
not run: its data start at 1.36 m tall and these loops are under 0.2 m. The DO probe hangs straight
here, because its lean is for the shaft’s shadow and inside the riser there is none.
Eccentric mounting
Section titled “Eccentric mounting”Eq. (6) below is Karcz’s, from his entry in References, and Theta(0) is its up-pumping branch at e = 0.
What the geometry says, with Karcz’s correlation applied to it. Eccentric mounting is real and
well supported, and this architecture can barely use it. The motor mount is a dia 56 body centred on the
lid and the port flanges reach inward to head_port_circle_radius(mouth) - 14.1, so the room for an
offset is Rpc - flange - lid_holes_offset - mount_radius. On jar_1p5L that is already -12.5 mm
and on jar_1gal_155 -8.30 mm - an unasserted collision until head() gained the check for it,
masked before that because the port-spacing assert fires first.
The deeper problem is that eccentricity is referenced to the tank diameter while the room for it is
set by the mouth. Hall’s e = 0.2 T needs 20.7 mm on jar_1p5L and 42.0 mm on jar_10L; the widest
offset any of these lids can give, with the mount shrunk to the dia 36 floor its own gearbox faceplate
allows and every port on a mini flange, is 0.95 mm and 28.4 mm respectively. e/T = 0.2 is out of
reach on every registered vessel - jar_1gal_180 comes closest at 0.181 and misses it narrowly.
What is reachable, by Karcz’s equation:
| vessel | mouth/T | e/T today | gain | e/T at best | gain |
|---|---|---|---|---|---|
jar_1p5L |
0.854 | 0 | 0% | 0.009 | 5% |
jar_1gal_155 |
0.642 | 0 | 0% | 0.031 | 14% |
jar_6p5gal |
0.488 | 0.035 | 16% | 0.090 | 31% |
jar_10L |
0.677 | 0.059 | 24% | 0.135 | 37% |
jar_1gal_180 |
0.871 | 0.090 | 31% | 0.181 | 41% |
T is the BORE here, vessel_outer_diameter - 2 x vessel_wall_thickness, which is what T means
everywhere else in this project: head.scad’s D/T is cut from it and so is the H/T the model echoes.
The offset itself, Rpc - flange - lid_holes_offset - mount_radius, is lid geometry in mm.
The gain columns are computed from eq. (6), up-pumping branch: gain = 1 - Theta(2*e/T) / Theta(0), with Theta(0) = 91.32, so this table can be re-derived rather than trusted.
The e/T today columns are ECHOED BY THE MODEL, not transcribed. head() computes the offset
the mount leaves - the slack above the minimum its mount-versus-flange assert clears - divides it by
the bore, and reports the gain with stirred_tank_eccentric_gain(), so those two columns can be
read off a render rather than trusted. The figures above are that echo: 9.8 mm, 12.4 and 15.3, at
e/T 0.035, 0.059 and 0.090. The two jars showing 0 render no echo at all, because an assert stops
them first - the mount on jar_1p5L, and the pH probe on jar_1gal_155x251, which fails on the
probe before it reaches the mount it would also fail. Their 0 is reasoning, not a render, and it is
the same fact said another way: a lid with no room for the mount has none for an offset.
The e/T at best columns are still by hand, because they describe a lid that does not exist:
the mount shrunk to the dia 36 floor its own gearbox faceplate allows, with every port on a mini
flange. Nothing renders that, so nothing can echo it. They are eq. (6) evaluated at an offset
someone worked out, and they carry exactly that much authority.
And they carry one digit less than they show. Those e/T were re-derived from the superseded
column, which was quoted to three decimals, so each is uncertain by about +/-0.0005 - the offset in
millimetres was never written down for three of the five rows. Run the interval through eq. (6) and
four rows are unmoved: 31 %, 37 %, 41 % and jar_1gal_155’s 14 % hold across it. jar_1p5L
does not - it is 4.48 to 4.98 %, so 4 or 5, and which of the two cannot be said from what was recorded. It is written
as 5, the midpoint rounded, and it is the figure in this table that a re-derivation should expect
to move.
The result is the wrong way round: the vessels with room for eccentricity are the ones already wide enough to carry baffles, and the two that need it have none. Off-centring is not a rescue for the small jars. It is, separately, a cheap 25-31% on the large ones if they ever run unbaffled. -> Port layout
What outranks all three
Section titled “What outranks all three”The sparger, and it applies family-wide. The five-fold kLa spread Uyar measured came entirely from bubble size: 1.56 mm from a microporous sparger against 4.18-4.25 mm from orifice spargers. The spargers drill 1.2 mm holes, which make 3.76 mm bubbles at formation where 3 mm would make 5.10 mm, still coarser than the 0.8 mm orifice that gave them 4 mm bubbles. Whatever mode a jar runs, a finer sparger is worth more than the choice between modes. Fouling does not argue for larger holes: photos of past cultures show almost no buildup on the sparger, or on anything else that deep. It collects higher in the culture, in the headspace and on the lid.
6. The design basis as it stands
Section titled “6. The design basis as it stands”Every figure below is what head() reports for jar_10L_220x305 at the reference fill; the
transcript is tests/echo/head__jar_10L_220x305.txt. What was believed before and got wrong is in
Decisions.
Sparger
Section titled “Sparger”Three spargers share the riser from the air_in port, and a build names one (sparger_name): the
arm is the default under the shaft and magnetic drives, and the ring under no drive and
the airlift; the loop is chosen, never defaulted. What head() reports for each follows; the
ring’s figures are a jar_10L build with sparger_name set, which no transcript holds, and the
loop’s are the reference build’s, tests/echo/build__jar_10L_220x305.txt.
The arm
Section titled “The arm”- One straight run under the lower impeller, holes down. The ring’s feed without the ring: the same socket and elbow, then a run turned inward from the port circle at r 56.5 to end 5 mm short of the shaft at r 9, 53 mm off the floor - 10 mm under the lower impeller’s blades, so the bubbles rise into it. That is where the stirred-tank literature puts a sparger; Birch & Ahmed’s argument for a ring between the impellers stays with the ring below.
- Eight 1.2 mm holes, the ring’s size, at 3.32 hole diameters along the 31.9 mm of bore
between the elbow and the end plug, 7.53 m/s each at 0.5 vvm - past Barbosa’s 0.4-5.4 m/s, and
head()warns: the run fits no more at the pitch floor. Bubbles are 3.76 mm at formation, as the ring’s; nothing about the arm changes what a 1.2 mm hole makes. - The bore is not a departure: 5.42 m/s in the feed run at an open area ratio of 0.72, and the 94 Pa a hole takes outweighs the bore’s 18 Pa of velocity head.
- The gas supply has to beat 2066 Pa before anything bubbles: 186.5 mm of culture over the arm plus 240 Pa of capillary - 509 Pa more than the ring asks, since the arm sits 52 mm lower.
- The arm gives the baffles back the width the ring took. It runs in between two plates rather than outboard of them, so the ring’s bound goes with the ring: the plates widen from 10.35 to 17.09 mm, where the lock bore stops them (below).
The loop
Section titled “The loop”- The arm’s riser, socket and height, hooked round into a C about the shaft, so the gas rises into the lower impeller all the way round rather than on one side of it. It is the pigtail sparger under a turbine, one tube with no tee: the elbow turns inward from the port circle at r 56.5, a horizontal bend of the elbow’s radius hooks it onto the curl tangent from outside, and the C runs 300 deg round from there, 53 mm off the floor, 10 mm under the lower impeller’s blades, holes down, to a plugged end.
- 76.54 mm across, 1.09 D, and the hook places it. Birch & Ahmed’s 1.4 D would be 99 mm, but the C can be no larger than its hook allows (below). 1.09 D is inside the 1-2 D band. The mouth and the shaft are far off, 29.36 and 30.81 mm. Under lid plates instead of the cage the loop sits among them and they place it: it stops 1.25 mm inside their inner edge after the 2.63 mm they can lean at its height and the 2.63 mm its own riser can, by the same coupling rule, and the hook, swinging off the riser’s bearing toward the next plate, keeps 3.15 mm to the nearest once both have leaned.
- The hook has to start where the elbow ends, a bend inside the riser, and meet the C from
outside, which costs about another bend: the C can be no more than
sparge_loop_largest(), 38.3 mm in radius at r 56.5. That is what places it onjar_10L, and on the others, wherehead()warns it is under the band: 25.2 mm (0.73 D) onjar_1p5L, 31.9 mm (0.64 D) onjar_1gal_155and 71.5 mm (0.76 D) onjar_6p5gal. - Eleven of the ring’s 1.2 mm holes, with the same 3.76 mm bubble floor, spread over the 291 deg
of C before the end plug’s solid, 14.7 hole diameters apart. One bore carries all of it to a
dead end, where the ring has two paths, and with sixteen the far end bubbled weaker on the
bench: an open area ratio of 1.44, and each hole’s 24 Pa only just over the bore’s 18 Pa of
velocity head. Eleven bring the open area to 0.99 and the push through a hole to 50 Pa, at
5.47 m/s a hole, a hair past Barbosa’s 5.4, and
head()flags that: the two multiply to the bore’s own 5.42 m/s, so no size or count clears both. The ring keeps sixteen, since an odd count puts a hole in its split. - It hangs on the feed riser alone, through the arm’s boss and screw: 0.374 N/mm at the loop,
29.6 mm under the riser’s end; under lid plates, 0.467 N sideways closes its 1.25 mm to a baffle,
and
head()warns. Nor is that the only way to close it: the impeller’s swirl puts a torque on a C round the shaft, which turns it about the riser against the port’s seal, and as little as 1.27 deg of that does the same (1.25 mm over the riser’s 56.5 mm from the axis). What the flow pushes and turns with is not known; this is the bench’s question. - Lid plates keep the width the arm gives them, 17.09 mm, and the gas has the arm’s 2066 Pa to beat, from the same height.
The ring
Section titled “The ring”- A tube ring at 1.90 D, placed by the mouth. Birch & Ahmed 1997 tested a ring at 1.4 D - the
radius at which the annulus outboard of the impeller encloses the volume it sweeps - and found
better power draw and delayed flooding; Rewatkar & Joshi 1993 recommend a large ring outright.
So the ring goes as far out as the mouth allows, 132.8 mm on this jar, with 1.25 mm to the
baffles inboard and 1.25 mm to the mouth on the way in. On the 70 mm impeller that is 1.90 D,
inside the 1-2 D band
head()echoes. It is a 6.4 mm octagonal tube on a 4 mm bore, split opposite the feed and plugged with two screws so a brush passes through it. - It sits midway between the impellers, 105 mm off the floor, 35 mm from each. That place was chosen for the pitched pair, which converged on it: Birch & Ahmed put a ring above an up-pumping blade and below a down-pumping one. Two Rushtons both discharge radially, so neither pumps toward it, and nothing held here places a ring between two radial turbines.
- Sixteen 1.2 mm holes, for bubble size; not for even flow. Rewatkar & Joshi: “hole size and number of holes have negligible effect when the sparger is located near the impeller.” Even flow is unreachable anyway: 240 Pa of capillary pressure to launch a bubble against 24 Pa to push gas through a hole, so a fixed tolerance hurts small holes most. At 0.5 vvm the holes run 3.76 m/s, inside the 0.4-5.4 m/s Barbosa 2003 actually ran; he establishes no critical velocity. 1 mm holes came out too small on a print.
- What it cannot do is make small bubbles. Diameter at formation goes as the cube root of hole size - 3.76 mm here - so the 1.56 mm of a microporous sparger would want an 86 um orifice, which is why those are sintered. Bubble size is the strongest lever on kLa this design does not have; it is the open item.
- The bore sets the limit. The feed run carries the whole flow at 5.42 m/s, which is the holes’
velocity times their open area: 1.44 for the feed and 0.72 for each of the ring’s two paths. No
hole size clears both Barbosa’s 5.4 m/s and an open area of 1 while the bore runs that fast, and
head()flags the open area and the bore; the fix is a wider riser, which is one designation and a cascade of fits (future work). - The risers are structure. Nothing else in the vessel touches the ring, so the five 4 x 0.5 mm 316 tubes that drop from the tube ports - one feeding, four blind - are what hold it: 1.02 N/mm a support, 0.98 mm of sway per newton against 1.25 mm to the baffles. The feed’s socket is the ring’s own tube standing up; the supports’ are round, which is the only thing that tells them apart at the bottom of a jar.
- The gas supply has to beat 1557 Pa before anything bubbles: 134.5 mm of culture over the ring plus 240 Pa of capillary. The line’s own losses are an order of magnitude more, and the sterile filter’s slope is extrapolated (#131).
Impeller
Section titled “Impeller”- Two six-blade Rushton turbines at the standard, D = T/3, the lower T/3 off the floor and the upper a further T/3 up (section 3). Chosen to be comparable, not for what it does best.
- Po 4.17 +/- 0.14 measured (Kaiser 2016) and flow number 0.65, Grenville’s typical. McCabe gives 5.75 and 1.3; section 2 says why the model keeps the measured power number.
- Radial, and what it costs the circulation. A photobioreactor’s criterion is Molina Grima’s
frequency of switch between light and dark, which is pumping. At equal power and diameter flow
goes as
Fl * Po^(-1/3), which on Grenville’s Table 1 (Rushton 0.65 and 5.0, 45 deg pitched blade 0.80 and 1.5) puts the pitched blade about 1.8x ahead; his Fig. 3 agrees at about 2x the fluid pumped per joule. With McCabe’s own pair for the Rushton, 1.3 and 5.75, the two come out nearly even, so the gap is as uncertain as the flow number. - The gas does not bind on this jar. A radial impeller holds its flow pattern against a gas
stream with 3x the stream’s power (Oldshue), so the pair holds 0.5 vvm down to 242 rpm, under
the drive’s rated 320; at 320 the ceiling is 1.16 vvm and at 420 it is 2.61.
jar_1p5Lis the vessel where it binds (section 5). - What the drive does to the culture, at the rated 320 rpm: Re 26,000, tip 1.17 m/s, 1.06 W
into 8.17 L = 130 W/m^3 mean and 27.1 W/kg peak for one impeller, the pair under 0.063 N-m of a
0.49 N-m rating; blend to 95 % in 5.9 s; kLa 0.0081 s^-1 coalescing. Both transfer correlations
are extrapolated - van’t Riet is fitted over 500-10,000 W/m^3 and Ruszkowski over 0.01-10 m^3 -
and
head()says so. - No energy dissipation rate limit exists for any microalga. The parameter the physics says governs has no citable number for these organisms.
- Speed feedback is specified but not wired. The drive’s encoder resolves 0.89 rpm over 100 ms, 672 counts per output turn, and nothing reads it yet: tip speed and Re are computed from a commanded speed.
Baffles
Section titled “Baffles”Both jar_10L builds, the reference and the magnetic, carry the standard four on the wall, on a
cage: baffle_name = "auto" takes it wherever it fits. jar_1gal_180 hangs plates from the lid,
and the six-port lids carry no baffles (Decisions). The first group of bullets is the
cage; the rest are the plates as head() reports them on jar_10L with baffle_name = "lid",
which no registered build carries and no transcript holds.
-
Four at T/12, T/72 off the glass, floor to surface. The blades are 17.5 mm wide at 0, 90, 180 and 270 deg, out of the lit windows, 2.92 mm off the wall, and 229.5 mm of each is under the culture: 1.00 of Oldshue’s four-at-T/12 reference area. T/12 is McCabe’s standard turbine (p. 243). The gap is Dynamix’s J/6 (read), which is T/72, inside the Handbook of Industrial Mixing’s T/50 to T/72 (relayed through a trade article, not read); McCabe’s Fig. 9.7 draws them flush.
-
The cage changes shape to pass the mouth. The blades’ outer edges are 204 mm apart and the mouth is 142.2 mm, so no rigid cage goes in. Two hoops of r 64-70 pass it lying flat, 1.1 mm a side. Each blade hangs between them on a radial bar at each end that slides in a dovetail channel in the hoop: slid in 30.6 mm to a stop, closed under the bar’s flange at its outer end, the cage goes down with the hoops bent where the pads catch; inside, each blade is pushed out to 0.4 mm off the glass at its pads. Three detents on each bar hold it: two straddle the hoop to hold it out both ways, and one holds it in on the way down. Building it has the parts and the assembly.
-
The lid moves one probe. The hoops are Cs, and the probes hang in the gap between the 0 and 90 deg blades. The pH probe moves from the 120 deg port to the 30 deg port a plate freed; 120, 150 and 270 deg take blanks. Any port takes any function on this lid, so the move is a re-plug.
-
The top hoop stands 1 mm over the culture, out of the flow, and carries an arm that catches the air inlet’s riser, the tube the sparger hangs from, which stops the cage turning: at the no-load 420 rpm each blade carries 0.29 N and the riser 1.93 N at r 56.5, by the same
T / (n * r)the plates use, reasoned, not cited. Friction will not do it: the three prints’ 109 cm^3 of PETG weigh about 29 g in water, about 5 N-mm at mu 0.25 (assumed) on the bottom hoop’s r 67, against the riser’s 63 N-mm at the rated 320 rpm. -
Bench-decided, not modelled: the slide fit, how hard a detent pushes through, and whether one holds a blade out under flow.
-
Three at 120 deg, one fewer than Oldshue’s reference count, and the vessel is under-baffled. His reference is four plates at T/12 running the liquid depth (McCabe’s standard turbine has four at T/12 too, p. 243), and he allows three only at the same total projected area. These hang from the lid to the floor limit, 17.09 x 274 mm with 223.5 mm submerged, and reach 0.713 of the reference projected area; four would reach 0.951. The fourth plate’s port went to a sampling line (Port layout). Both levers are spent: depth is at the floor, and the next count that spaces equally on twelve ports is four, at the cost of that port. Under-baffling lets the vessel swirl rather than mix.
-
The width is capped by the lock bore. The 70 mm impeller leaves room, so the plate keeps 12.96 mm nominal to the blades and 10.51 mm running, after the 2.44 mm of lean the coupling’s 0.2 mm of play allows at the lower impeller. That lean is a worst case, bore play alone with the flange’s face contact ignored. A ring build pays the ring’s bound, 10.35 mm and 0.433 of reference, since the room outside a baffle does not grow with the mouth; the arm builds do not. On
jar_1gal_180the plates are the same 17.09 mm and reach 0.847 of reference. -
What sized the plate was dynamics, not strength and not collision. Root stress is about 1 % of yield, and the plate bends tangentially while the impeller sweeps a circle, so deflection never closes the radial gap. What binds is the first bending mode, which falls as 1/L^2, against blade passing, which a six-blade turbine sweeps from 0 to 42 Hz on the way to 420 rpm. Stiffness goes as t^3 and the mode as t^1.5, so thickening raises the crossing toward the 320-420 rpm band. The plate is 10 mm. On the 17.09 mm plate the mode is 17.9 Hz, crossed by blade passing at 179 rpm and by the shaft at 1072, clearing the band by 44 %; the tip deflects 0.92 mm under 0.64 N, under the tenth-of-width limit
head()warns on. The ring build’s 10.35 mm plate deflects 1.52 mm and draws that warning. -
The load is reasoned from two directions and neither is cited. Each plate reacts its share of the impeller’s torque,
T / (n * r); the dynamic pressure of a tangential stream at 0.3 of tip speed agrees within 26 %. No source held here loads a baffle. Added mass is not optional in the mode: the entrained water is over twice the PETG’s own, and leaving it out overstates the frequency by about 80 %. -
Partial and inboard is uncited. The plates span about r 48-65 mm in a 210 mm bore, so the annulus outboard of them is about 40 mm wide and unobstructed - a consequence of passing every part through a 142.2 mm mouth. Inboard baffles are characterised in the literature, not recommended; the relevant papers are paywalled and unread (References).
-
The plate prints in two dovetailed pieces, and the joint is in the deflection above: 4.2 mm of the plate’s 10 crosses the joint plane, 0.074 of its second moment, 0.13 mm of the 0.92 mm at the tip. Why it splits where it does, which way the slide runs and what the joint leaves unmodelled is in Building it.
