SCADA & MES: the control loop and its memory

An ordinary factory process can be tuned once and left alone; a bioreactor cannot, because its contents are alive and keep editing their own environment. Cells consume oxygen, excrete CO₂, acidify or alkalise the medium, generate heat, and change viscosity as they grow — the plant is non-stationary by construction, while the quality of the product depends on holding temperature, pH and dissolved oxygen inside narrow bands for days. Holding a setpoint against a drifting plant, around the clock, with consistent response time, is precisely what a control system does and a human cannot. That is the whole justification for the SCADA layer: not monitoring as a courtesy, but feedback as a necessity.

Cascades: spending actuators in order

Dissolved oxygen illustrates the standard strategy. The controller cannot push a variable directly; it must choose actuators, and they are not equivalent. Raising agitator speed improves gas transfer without changing the gas; enriching with oxygen costs money and complicates safety; sparging harder strips CO₂ and foams the medium. So the loop is a cascade: agitation is spent first, and only when the cheap actuator saturates does the expensive one engage. pH control shows the same asymmetry with a worse complication: the actuators have long dead time — base pumped in takes minutes to reach the pH probe, mix, and register — and control loops with dead time oscillate if tuned aggressively. Tuning is therefore not a commissioning detail but a quality attribute: an hour-long pH excursion is invisible to the vessel and legible in the product’s glycosylation pattern.

The record is part of the loop

The electronic batch record is usually described as compliance; mechanically it is the process’s memory. A controlled run must be reconstructible — every setpoint change, alarm and operator intervention timestamped and attributed — for two reasons that are engineering, not bureaucracy. First, diagnosis: when a batch drifts, the reconstruction is how the drift is traced to a cause. Second, detection: review-by-exception is a comparison of the current run’s trajectories against the family of previous runs, and only a complete, structured record makes the comparison automatic. The recipe standards in this space (ISA-88 for batch control) encode the same idea structurally: a process defined as named phases and parameters is portable between vessels and comparable between batches. The record turns each run into training data for controlling the next one.

Measurement is the ceiling

Both the loop and the record are limited by what can be measured, when. Probes drift and require calibration; critical variables like cell density and product quality are historically off-line assays with hours of lag, so the controller runs partially blind on those axes and interpolates. The real content of process-analytical technology is moving measurements inside the loop — in-line spectroscopy estimating composition continuously so that feeding reacts to the culture rather than to the calendar. What the software layer can never compensate for is a variable nobody measures; the honest boundary of bioprocess automation is the boundary of its sensor suite.