How it works
An inline monitor that never touches the control loop
CANARY is a measurement device. It observes and forecasts; it does not actuate. That distinction governs the whole design, and it is why the unit can be fitted to a working valve without re-qualifying the machine's control behaviour.
It is fitted between the plant and the valve
The existing 12-pin cable from the plant lands on the monitor's input. A short loom runs from the monitor's output to the valve. Supply, enable, command, actual value, enable acknowledgment and the fault-free output all pass straight through. No cutting into the control signal, no re-terminating the valve, no change to the machine's program. Plant-side and valve-side connectors are built to match your valve — 5-pin, 7-pin and 12-pin configurations are available, specified from the valve data sheet at order.
It measures continuously while the machine runs
The monitor reads the valve's electrical behaviour and its mechanical response at the same time: supply voltage and current through a precision shunt, the command the plant is sending, the actual spool position the valve reports back, and the vibration and acoustic emission the valve body radiates. Measurement is continuous, not sampled on a maintenance round.
It learns what healthy looks like for that specific valve
Over a commissioning period the unit builds a baseline from the valve as installed — its own hysteresis, its own response timing, its own noise floor. Every later judgement is made against that valve's history rather than a generic specification, which is what makes the same device workable across valve families and duty cycles.
It condenses what it sees into hourly measurements
Raw high-rate signals are reduced on the device into a set of hourly engineering measurements — hysteresis, response threshold, null behaviour, tracking error, current per unit command, band-limited vibration energy, acoustic emission rate, and others. Processing happens locally. The monitor does not depend on a cloud service to form a judgement.
A proprietary AI engine projects each measurement forward
This is the part that makes CANARY a prognostic system rather than a condition monitor. A purpose-built model tracks every measurement against its own limit, estimating level, trend and acceleration together, and answers a forecasting question: will this cross its limit within the warning horizon, and when. It requires corroboration across several measurements before escalating, which is what holds the false-alarm rate down. The estimator is deliberately biased toward the earlier bound — in our 75-day pilot, every single time-to-failure estimate fell before the actual failure rather than after it.
It reports in two levels, with a reason
WATCH means a measurement is drifting and worth keeping an eye on. WARNING means a failure limit is forecast within the horizon, and it names the failure mode and the measurements behind the call. A hard-wired healthy contact and a Modbus register are available for plants that want the signal in their existing system.
Architecture
One monitor per valve, one central unit per site
Valve units
One per monitored valve, fitted inline. Each unit measures, learns and forecasts on its own. It keeps its own log on local storage and serves its own web interface, so a technician can reach it directly at the machine without site infrastructure.
Self-organising network
Valve units pass their results back to a central unit over a wireless mesh, relaying for each other so a unit out of direct range reaches the central through its neighbours. Adding a valve to the network does not require new cable runs across the plant.
Central unit
Shows the whole fleet in one view, holds configuration for every valve unit, and distributes firmware updates out to them with automatic rollback if an update fails. This is also where site-wide settings and the administration account live.
Integration. Results are available over Modbus for a plant SCADA or PLC, on a hard-wired healthy contact for systems that want a dry contact, and through the web interface each unit serves. The monitor fits alongside what the plant already runs rather than asking it to be rebuilt.
Why inline matters
Measuring at the component, not inferring from the process
Plant-level monitoring sees the symptom
A SCADA trend showing cycle time creeping up tells you something in the machine is degrading. It rarely tells you which of the forty valves on that machine is responsible, and it only moves once the degradation is large enough to affect the process.
Component-level monitoring sees the cause
The signals that move first are at the valve: coil current waveform, spool response timing, the acoustic signature of bubble collapse at the metering edges. Measuring there gives both an earlier signal and an unambiguous answer about which component to service.