Biopolymer manufacturing
Seeing the product grade before the batch is over
A biopolymer manufacturer whose demand exceeds capacity wants to know, while a batch still runs, which molecular-weight grade it is heading for.
The challenge
Molecular weight decides the grade, and the grade decides the price. It is set by what happens during fermentation, but confirmed by offline analysis once the batch is finished. By then the result is whatever it is.
When demand exceeds capacity, every off-grade batch is capacity that cannot be recovered: it runs to completion, consumes downstream time and is then sold cheaper, reworked or blended. The drivers — dissolved oxygen, agitation and feeding — are already on the trend charts. The quality attribute they determine is not.
The client
A European manufacturer of hyaluronic acid by bacterial fermentation, exporting most of its output and running at the limit of its capacity.
Our approach
We calibrate a model on historical batches: live signals on one side, the analytical result of each batch on the other. During a run the twin:
- estimates the molecular weight the batch is heading for, with an uncertainty band
- warns when the trajectory leaves the pattern of batches that hit the grade
- names the variable responsible, so the correction is specific
Estimated effects
0
new bioreactors or sensors
The estimate is built from signals the plant already records; more output lands in the grade ordered without adding capacity.
During the run
instead of after analysis
The grade becomes visible while the batch can still be steered, not when it can no longer be changed.
4 weeks
to a measured answer
How much of the molecular-weight spread the recorded signals explain is exactly what Phase 0 measures. A ready-made industrial soft sensor for this attribute does not exist, so this is the research-grade part of the work.
We do not promise a percentage. Success criteria are agreed and frozen in Phase 0, on the manufacturer’s own historical batches, before any model is built.
How the PoC runs
-
01
Phase 0 · 4 weeks
Data assessment; success criteria agreed and frozen.
-
02
Data buffer · 2–4 weeks
Cleaning, digitising and time-aligning the batch data.
-
03
Phase 1 · 8–12 weeks
The model, built and tested on held-out batches.
-
04
Next step
A written reference for a good batch, ready to transfer to a new line and to support new-grade development.
What we asked for
-
Line description
A P&ID or drawing of one fermenter, with measurement points and ranges.
-
Batch trends
Dissolved oxygen, agitation, temperature, pH, feeding, and stirrer power or torque if recorded.
-
Analytical results
Molecular weight and distribution per batch, plus yield — what the model learns to predict.
-
Batch history
Batch number, start and end, target grade, medium lot and any recorded deviations.
What you get
-
More saleable output
More of the plant’s capacity in the grade ordered.
-
Time to react
While the batch can still be steered.
-
A transferable reference
What a good batch looks like, written down for the next line.
-
Fewer development trials
The model ranks what actually moves molecular weight.
Data can come in any format — CSV exports, spreadsheets, PDF, even scanned batch records. No recipe disclosure is needed: medium composition and strain details can stay coded as lot identifiers. No new sensors, IT integration, remote access or production downtime are needed to start.
Have a similar process?
Tell us about it. We’ll tell you where a model pays off first and what data you’d need.
Book a technical call