The problem space
A bioreactor is a noisy place. Temperature wobbles, pH drifts within control bands, a sensor reads a touch higher than it did last week. Each signal in isolation looks fine; the pattern across signals is where a developing problem lives. Operators are good at recognising the patterns they've seen before — but the rare ones, the early ones, and the ones that only matter in combination tend to slip through until they cost a campaign.
When a fault does land, the response matters more than the alert. Knowing who needs to do what, in what order, with which data in hand — that's the slow part of a fast response.
What we're exploring
Modica's adaptive digital twin is being built to compare each new run against the model's expectation of normal behaviour for this process, at this scale, in this configuration. The intent isn't to replace the operator's intuition — it's to give the model a memory that doesn't forget the rare patterns, and to surface them in time to act.
Like the rest of the platform, this is early-stage work; some of it is funded under our FENG project.