Inocula: bioprocess scale-up checks

Scale-up breaks on physics.

The strain is the same in a 10 L bench reactor and in a 1,000 L tank. The vessel around it changes, and five ratios move with it.

10 L
It works at the bench.

At 1,200 rpm and 7.2 kW/m³ every ratio stays under 0.8. This is the operating point the process is known to run at.
150 L
The margins go thin.

Holding power per volume would need 658 rpm; the stirrer tops out at 510. Oxygen reaches 0.90, mixing 0.93.
1,000 L
Mixing crosses first.

The tank needs 27.1 s to mix while the cells clear the feed in 15.0 s. Substrate gradients reach 1.81, oxygen 1.36.

10L

E. coli preset, constant P/V from the 10 L base
Every ratio under its limit

Prototype engine, illustrative literature values, not validated. Same engine as the simulator below.

Oxygen transfer, mixing and heat removal do not scale linearly. Most failures are visible on paper before the first 1,000 L batch, if someone does the paper.

Volume10 L → 1,000 L×100

Tank diameter0.19 → 0.86 m×4.6

Wall area per litre (heat removal)∝ V^-1/3×0.22

Stirrer speed at equal P/V∝ V^-2/9×0.36

Mixing time at equal P/V∝ V^2/9×2.8

Geometrically similar stirred tanks, turbulent regime. Exponents follow from P = N_pρN³D⁵ and N·t₉₅ = constant.

Now move it yourself.

The same engine with every control open. Change the organism, the scale-up rule or the oxygen demand, drag the volume, and see which limit is crossed first.

True relative scale, H/T = 2

Prototype engine using published correlations and illustrative literature values. Not validated.

Organism (illustrative preset)

Scale-up rule from the 10 L base

10L

Oxygen uptake rate (OUR) at target biomass

200 mmol/L/h

Runs entirely in your browser; nothing you set here leaves the page. Base case: the 10 L example reactor at its maximum stirrer speed and gas flow, where the process is assumed to work. Vessels between the three example reactors (10, 150, 1,000 L) are interpolated with geometric similarity. All inputs are confidence C (literature or estimate); sources and assumptions are in the method note.

Five ways a fermentation breaks.

Each check is a ratio of demand to capacity. Under 0.8 the margin holds, 0.8 to 1.0 is a warning, above 1.0 the limit is crossed.

Oxygen transfer

Cells consume oxygen faster than the gas can dissolve it. Dissolved oxygen falls below the critical level and metabolism shifts.

OUR / OTR_maxthreshold 1.0, warn from 0.8

{OTR}_{\max} = k_{L} a C^{*} (1 - {DO}_{crit})

Substrate gradients

The feed is consumed before it is mixed in. Cells cycle between excess and starvation; yield drops and by-products appear.

t₉₅ / t_Smixing time over substrate consumption time

t_{S} = \frac{S_{res}}{q_{S} X}

Heat removal

Metabolic heat grows with volume; jacket area grows only with volume to the two-thirds. The temperature runs away.

(Q_met + P) / UAΔTabout 460 kJ released per mol O₂

Q_{met} = OUR \cdot V \cdot Δ H_{O_{2}}

Shear

Holding oxygen transfer at scale pushes impeller tip speed and power input toward what the cells tolerate.

tip / tip_max, P/V / (P/V)_maxthe larger of the two

u_{tip} = π N D

H₂ and off-gas flammability

Gas fermentations feed hydrogen and oxygen together. The feed or the headspace can sit inside the flammable range.

min(H₂/LFL, O₂/LOC)LFL 4% H₂, LOC 5% O₂; screening rule, not a certificate

y_{H_{2}} \geq 0.04 \land y_{O_{2}} \geq 0.05

Your data, a characterised reactor, a verdict.

Three steps, in this order. The reactor side is measured once and reused for every process.

You share kinetics and lab runs
Biomass, uptake rates, residual substrate, the conditions where your process works at 1 to 20 L. Each input is tagged by how it was obtained: measured on this process, measured elsewhere, or estimated.
We map them onto a measured reactor
kLa and mixing time are measured on the vessel itself, and replace the correlations wherever they exist.
You get limits ranked by margin
Which limit is crossed first, at what volume, under which scale-up rule. Assumptions are printed next to each number, together with the inputs worth measuring at bench scale before you buy pilot time.

One report per process, every number traced.

The software comes first. This is the assessment it returns, computed live in this page by the prototype engine for the E. coli preset at constant power per volume.

AssessmentsE. coli fed-batch, glucose-limitedReport

Prototype interface. Illustrative values from literature, not validated.

Inputs

Ameasured on this process Bmeasured elsewhere Cliterature or estimate

Input	Value	Conf.	Source

Reactor family

Check	10 L base	150 L	1,000 L

Sensitivity

First limit with each C input moved by 30%

What to measure at bench

Computed in your browser by the prototype engineRule: constant P/V from the 10 L base at its maximum stirrer speed and gas flow

Published correlationsvan 't Riet for kLa, Grenville and Cooke for mixing time, Henry's law for oxygen solubility, Luong and Volesky for metabolic heat.
Sources labelled by statusEach value is marked verified, secondary or assumption. Assumptions are said to be assumptions.
Sensitivity on every estimateEach literature or estimated input is moved by 30% up and down. If the verdict flips, it goes on the list of things to measure.

Every run tests a prediction.

We plan to build an open pilot hub in Lombardy: 1 to 20 L labs, a 150 L pilot and 1,000 L, for companies and researchers. Before each run the prediction is written down; after it, the prediction is scored.

See the hub plan

Where we are.

Status on 1 November 2026.

SoftwarePrototype engine, internal only. The simulator on this page runs it.

ValidationNot validated. The next step is a test on a real reactor with a partner laboratory.

CustomersNone yet.

HubPlanned, not built. Site candidate: Insubrias BioPark, Gerenzano.

MaturityTRL 1-2.

ProgrammesSemi-finalist, Nucleate Activator 2027.

Two founders and one open seat.

Bioreactor design and plant, funding and partnerships. The biology is the seat we are filling now.

Stefano Orifici

Bioreactors and plant

Mechanical engineer; bioreactor design and plant engineering. Physis, Polimi Energy Boat.

Giacomo Massolini

Grants, finance and partnerships

Biomedical and Electronic Engineering, Politecnico di Milano. President of BEA 2024‑26.

Open seat

Fermentation scientist

You have taken a microbial process past the bench and know where it breaks. You would own the biology inside the engine and the first runs at 150 L and 1,000 L.

We are building the team through Nucleate Activator 2027. If you have run fed-batches and want to see them predicted before the steel is bought, write to us.

Write to us

Have a process that must scale? Test us on it.

We are looking for 3 design partners with existing 1-20 L data. You get the checks on your process; we learn where the engine is wrong.