Automation

Automation covers the sensors that measure the system, the controllers that act on those measurements, the alarms that summon a person, and the records that make the whole history reviewable. In aquaculture it spans a range from simple standalone alarms to fully supervised control of every process in the loop.

• Continuous monitoring detects fast-acting problems — oxygen, flow, level — in the interval between routine checks, which is when they mostly occur.
• Alarms convert a detected problem into a human response, which is the only part that actually saves fish.
• Logging turns operation into evidence, so trends are visible and post-event analysis is possible.
• Control stabilises parameters more tightly than manual intervention can, improving both fish performance and energy use.
• Automation reduces the number of routine tasks competing for operator attention, so attention goes to the things that need judgement.

• Fail-safe by design. Every control loop must have a defined behaviour on sensor failure, signal loss and power loss, and that behaviour must be the safe one for the fish.
• Independence of critical alarms. The oxygen alarm should not depend on the same controller, power supply or network as the oxygen control.
• Redundant measurement for critical parameters, so a single drifting sensor cannot silently misrepresent the system.
• Alarm escalation with acknowledgement. An unacknowledged alarm must escalate to another person rather than repeating into an empty building.
• Plausibility checking. A reading that is physically impossible should raise a sensor fault, not drive a control action.
• Manual override on every automated function, accessible without the control system.

• Identify the critical parameters first — oxygen, flow, level, temperature, power — and design their monitoring, alarming and redundancy before anything else.
• Provide alarm annunciation that reaches a person off site, with escalation and acknowledgement, and test the full path.
• Provide uninterruptible power to the monitoring and alarm system, independent of the plant it monitors.
• Place sensors where the measurement means something, and where they can be reached for cleaning and calibration.
• Specify enclosures, cabling and materials for a wet, corrosive environment.
• Keep the log locally as well as remotely, so a network failure does not create a gap in the record.
• Document the control logic in a form the operators can read.

• Calibrate sensors on a written schedule and record every calibration.
• Cross-check automated readings against a manual measurement regularly.
• Test alarms functionally, including the escalation path and the out-of-hours route, rather than checking that they are enabled.
• Review the alarm log for repeated nuisance alarms; alarm fatigue is the mechanism by which a real alarm gets ignored.
• Keep manual operating procedures current, and practise running the plant without the control system.
• Review trend data as a scheduled activity, because the value of logging is realised in review rather than in collection.

• Clean and calibrate sensors on schedule; fouling is continuous in culture water.
• Test the alarm and escalation path functionally at a fixed interval, including out-of-hours delivery.
• Verify uninterruptible supplies by loading them, not by checking their indicator lights.
• Check backup and retention of the logged data.
• Review and update control documentation whenever logic is changed, and keep a change record.
• Inspect enclosures, glands and cabling for water ingress and corrosion.

• Automating control before establishing reliable, calibrated measurement.
• Powering the alarm system from the same supply as the plant it monitors.
• Setting alarm limits so tight that nuisance alarms train everyone to ignore them.
• Trusting a single sensor for a critical parameter.
• Assuming a control command was executed, without independent verification.
• Testing that alarms are enabled rather than testing that they arrive.
• Collecting logs that nobody ever reviews.
• Leaving no documented manual procedure for running the plant without automation.

Predictive and model-based control uses the system's own history to anticipate load rather than react to it — for instance adjusting oxygenation ahead of a scheduled feeding rather than after the resulting demand appears. The benefit is tighter parameter stability and lower energy use; the requirement is a trustworthy measurement record long enough to characterise the system.

Connectivity introduces a security consideration that a purely local system does not have. Remote access is genuinely valuable for out-of-hours response, and it also creates a path into a control system that is holding a live biomass. Network segregation, authentication and controlled remote access belong in the design.

Automation in a recirculating system is life support. Design it around failure: defined fail-safe behaviour, independent critical alarms, redundant measurement, escalation until acknowledged, and manual override throughout. Calibrate relentlessly, test the alarm path rather than the alarm, and review the log — because the record is only worth what someone reads from it.