A failure found on the day it happens is an unplanned outage, an unscheduled mobilization, and whatever the wind was doing that week. The same failure found six weeks earlier is a task on a plan. Condition monitoring exists to move the date.
The approach to a limit is recorded in full, even though nothing was exceeded and nothing needed to stop.
Gearbox, generator, main bearing, and pitch system degradation appear as drift in temperature, power, and rotor speed relationships well before a threshold alarm is raised. The models learn the normal signature of each individual machine and report departures from it.
A signal is only useful if it arrives before the decision it should inform. Every alert states the period over which the drift developed and how far it has progressed, so a planner can judge whether it fits the next campaign or requires a visit of its own.
An alert that sends a crew to a healthy machine spends the same money as a real one, and it spends the team's confidence as well. Thresholds are set against the cost of the intervention, which is not the same offshore as onshore.
The analysis runs on the channels the turbine already exposes. Where a dedicated vibration-based condition monitoring system is installed, its data is read alongside the SCADA record rather than replaced by it.
A turbine's alarm system exists to stop the machine when an operating limit is crossed, and it does that reliably. What it is not designed to do is describe the weeks during which a component approached that limit, because during those weeks nothing was exceeded and nothing needed to stop.
The approach is nonetheless recorded. A bearing that is failing runs warmer than comparable bearings at the same load, ambient temperature, and rotor speed, and it does so long before it runs warm enough to trip. The information is in the ten-minute record. What is missing is a reference for what warm means on that specific machine in those specific conditions.
That reference is what the models supply. A per-turbine, condition-matched expectation turns an absolute temperature into a residual, and a residual that grows steadily across six weeks is a different object from a reading that is merely high on a warm afternoon at high load.
The work a failing component requires does not change with the date it is discovered. What changes is the mobilization, the production lost while the machine waits for parts nobody ordered, and whether the failure was given time to damage anything else.
Onshore that difference is an unplanned crane mobilization set against a planned one. Offshore it is larger, because access is weather-dependent and a vessel is chartered rather than dispatched, so work that can be consolidated into a single planned campaign is work that does not require a mobilization of its own.
It is also the difference between a component exchange and a consequential failure. A gearbox bearing replaced on schedule is a serviceable event. The same bearing left to run to failure can take the gearbox with it, which converts a planned task into a major component replacement and a long outage.
An alert's position in the list combines the probability that the degradation continues, the production at risk while the machine is out, and the cost of the intervention itself. A minor drift on a high-yield turbine entering a strong wind season can legitimately outrank a more advanced signal on a machine that is already curtailed for much of the period.
Each alert shows the components of its score rather than only the result. An operations manager who disagrees with the weighting can see the effect of a different one, which is a more productive conversation than disputing a number whose derivation is not visible.
Lead time, not detection, is the measure that matters. An alert that is correct on the day of failure has told the team nothing it was not about to discover.
See how OpenTurbine can help your team understand performance, anticipate operational issues, and make better decisions across your wind assets.