Detection coverage
Seven failure modes, and an honest account of each
Prediction only works for failures that develop gradually and change something measurable on the way. Below is every mode CANARY watches, what physically goes wrong, and how the proprietary prognostic engine performed against each one in our 75-day pilot programme. The failures it could not predict are on this page too.
Spool & sleeve wear
Mechanism. Wear of the spool and sleeve, rising friction, a weakening centring spring.
What drifts first. Hysteresis widens, the response threshold grows, the null offset drifts, tracking error rises and the energy needed per stroke climbs. The valve still goes where it is told — it just works harder to get there.
Silting (contamination)
Mechanism. Fine particles settling into radial clearances while the spool rests, and in the pilot stage.
What drifts first. A rested spool sticks. The monitor compares how much extra command a spool needs to break away after resting against how much it needs while already moving. That difference is silt lock, and it grows as contamination builds.
Erosion & internal leakage
Mechanism. High-velocity flow and particles rounding the metering edges; internal seals beginning to bypass.
What drifts first. Pressure gain at the null falls and the hydraulic null moves. This mode needs pressure transmitters on ports A and B — without them the monitor reports it as not monitored rather than guessing.
Cavitation & aeration
Mechanism. Bubble collapse at the metering edges from high pressure drop, low back pressure, air in the oil or high temperature.
What drifts first. Collapse is impulsive and loud in a band well above what the machine itself produces. The monitor watches high-frequency vibration energy, how impulsive that energy is, and acoustic emission level and hit rate.
Solenoid coil degradation
Mechanism. Shorted turns and insulation ageing from heat, continuous energisation and voltage surges.
What drifts first. Current per unit of command rises as a weakening coil needs more current to hold the same spool position. With the coil's thermal data entered, an estimated coil temperature is tracked against its class limit.
Valve electronics
Mechanism. Ageing electrolytic capacitors, a drifting power stage, position-sensor electronics, connector or solder fatigue.
What drifts first. Supply-current ripple and standing current climb, the reported actual value gets noisier and starts to glitch, and the enable-to-acknowledge handshake slows.
Wiring & supply
Mechanism. Corroding or loosening supply pins, a weak plant supply, surges on the plant side.
What drifts first. Source resistance rises, measured as the voltage droop per unit of current drawn. Supply sag under load deepens and over-voltage surges are counted.
Pilot programme results
How far ahead the warning came
Two hundred runs per failure scenario against a modelled plant with two-shift operation, seasonal oil-temperature swing and realistic measurement noise. The column that matters operationally is the first one: how often a correct warning arrived at least eight days before the valve reached its failure limit.
| Failure mode | Warned 8+ days ahead | Median lead | Confidence |
|---|---|---|---|
| Spool & sleeve wear | 98% | 12.5 days | Strong |
| Silting (contamination) | 90% | 12.0 days | Strong |
| Erosion & internal leakage | 98% | 14.0 days | Strong |
| Cavitation & aeration | 98% | 18.4 days | Strong |
| Solenoid coil degradation | 72% | 9.9 days | Moderate |
| Valve electronics | 74% | 9.7 days | Moderate |
| Wiring & supply | 88% | 11.0 days | Strong |
Every scenario produced a warning. In all seven modes the engine warned in 100% of runs; the column above measures how many of those warnings arrived with at least eight days of usable lead. Where the figure is lower, the warning still came — a median nine to ten days out — but roughly one run in three or four gave only five to eight days.
Limitations
What CANARY cannot do
A monitor that only advertises its strengths is hard to plan around. These are the boundaries as they stand.
Failures with no precursor
- A sudden mechanical break in the spring or pilot stage
- A single large particle jamming the spool — gradual silting is predictable, one particle is not
- A cable that is cut
- Insulation breakdown with no thermal or electrical warning
What no sensor on the unit sees
- External leakage (weeping). Nothing on the monitor detects oil on the outside of the valve. It needs a separate oil-presence sensor on an auxiliary input.
- Erosion without pressure transmitters. Edge wear changes what flows, not where the spool sits. Without pressure at ports A and B it is effectively invisible.
Where the engine is weakest
Coil degradation and valve electronics are the lowest-scoring modes because both were modelled as fast, strongly accelerating failures — most of the change arrives at the end. Both still warned in every run, a median nine to ten days ahead.
Conditions and assumptions
- The leakage measurement assumes a critically lapped spool. A valve with large spool overlap has no pressure gain at the null to measure.
- Silt-lock detection needs the spool to rest periodically. A valve that never rests is watched for silting only through response and tracking.
- Under harsh plant disturbance the false-warning rate rises from 0.10 to roughly 1.6 per valve-year.
On the status of these numbers. They come from simulation of the diagnostic engine against modelled degradation, using engineering estimates for failure limits, degradation shapes and noise. The device has since been tested in a field environment on valves in service, and the data collected correlates with these simulated results. We are continuing to expand that field data set and will restate the figures against it as the evidence base grows. We publish the simulation basis so you can weigh it rather than take it on trust.