S3Pulse — a biomimetic computational algorithm with its hands on a real freeze-dryer.

Sensor-first authority

The product temperature authority is not chamber air and not a shelf setpoint: it is the warmest of three dipped Lyoprester probes (top / mid / bottom), read over an ADAM-4019 thermocouple module. Nucleation triggers, rebound confirmation and governor decisions all key off real product.

Hardware in the loop

A Waveshare Modbus-TCP relay board carries all sixteen channels; the release valve is operator-only with explicit confirmation — the software never vents the chamber on its own. Pressure paths fail visibly: an unanswered relay aborts the sequence instead of pretending.

Honesty as architecture

Sensor dropouts freeze displays at their last truth and mark it; pressure peaks above gauge range are labelled UNMEASURED, never estimated into a chart; the computational-intelligence stack is advisory by construction and cannot touch a recipe without the operator's explicit apply.

Sensing

• 3 dipped product thermocouples (top / mid / bot) via ADAM-4019 over serial
• Inficon capacitance manometer (voltage-tapped)
• Barocel chamber gauge, 0–1000 hPa absolute
• Condenser and HTF temperature channels
• FUYU chamber camera — event-triggered captures

Around the chamber

• WLED strip, 300 addressable LEDs — cold-trap and chamber status lighting
• View-only TightVNC mirror to a bench PC — remote input blocked server-side, lab-LAN only
• Delta PLC vial crimper on the lab network (separate project, monitored read-only)

Interlocks that bite

• Argon and FCA valves can never both be open — both are closed before either opens
• CN-burst mutex against septum-settle; pumpdown exit needs elapsed time AND a live pressure reading
• Cleanup discipline closes both gas valves and releases the mutex on every path, including abort

What a recipe is

A lyophilization recipe (cycle) is the program the dryer executes: shelf-temperature ramps and holds, chamber-pressure setpoints, and the rules for moving from freezing to primary drying (sublimation of ice) to secondary drying (desorption of bound water). Developing one means finding the conditions a specific formulation tolerates — the published design-space tradition frames this around the product's critical temperature (Tang, Nail & Pikal, 2005).

Open loop vs closed loop

An open-loop (fixed) recipe follows the clock: setpoints change at pre-set times whatever the product is doing. A closed-loop (adaptive) cycle responds to measurements — product temperature, pressure-rise tests — as they happen. Regulatory authors document both tracks in current human-drug applications and inspections (Rhieu et al., 2025).

Where automated control began

The foundational automated cycle-development work used manometric temperature measurement (MTM): brief pressure-rise interruptions that estimate product temperature mid-run and feed a control algorithm (Gieseler, Kramer & Pikal, 2007). A single-machine trial with equipment caveats the paper states itself — the category's starting point, caveats included.

1 · Data loggers & compliance

Record, archive and document cycles for audit trails. They observe the run; they do not steer it. Example: Ellab ValSuite (vendor description).

2 · PLC / SCADA execution

Production-line controllers that run a fixed recipe faithfully: setpoint in, setpoint held, alarms on deviation. The timer-driven end of the spectrum — most installed freeze-dryer control lives here.

3 · Offline modelling tools

Simulate a cycle before running it; the output is a proposed recipe, not control. LyoPRONTO is open source with published accuracy figures (Shivkumar et al., 2019); SP Scientific's LyoModelling is the vendor counterpart (vendor description).

4 · In-loop adaptive control

Adjusts the running cycle from live measurements: the SP Scientific SMART/MTM lineage (Gieseler et al., 2007), Millrock's LyoPAT/AutoDry tooling (vendor descriptions) — and S3Pulse, the sensor-gated implementation this page documents.

5 · Digital twins

Model-based soft sensors and scenario optimization run alongside the process. Published studies report large scenario gains; those are model results for specific equipment, not category-wide outcomes (Klepzig et al., 2020).

6 · Consumer & DIY controllers

Custom-cycle settings on home units (Harvest Right app documentation) and aftermarket vacuum controllers (DigiVac FYRA) — a different league, included because the queries overlap.