A site average conceals the machines producing below the fleet around them. Performance intelligence establishes an expected output for every turbine from the conditions it actually experienced, then reports the shortfall in annual energy and in revenue.
Four measurements, all of them derived from the ten-minute record a wind fleet already writes.
Models are fitted per turbine from matched wind speed, direction, air density, and turbulence rather than from a manufacturer's datasheet power curve. The expectation describes the asset as installed and as it currently operates.
A month of low production may be a low wind resource or a degrading asset. Comparing each turbine against units operating in the same conditions at the same time is what makes the difference attributable rather than assumed.
Static yaw misalignment, pitch calibration drift, and wake interaction reduce output continuously without raising a fault. All three are visible in the operating record, and most are correctable without an intervention at height.
Each deviation is reported in megawatt-hours per year and in its revenue equivalent, so a work list can be sequenced by the return on the intervention rather than by whichever asset was escalated most recently.
Operational reporting is built around availability and total production, and both are honest measures of different things. Availability counts the hours a turbine was able to operate. Production counts the energy it delivered. Neither answers the question that governs annual energy production, which is whether a machine that was available and running produced what the conditions allowed.
Answering that requires a reference. The reference most often used is the manufacturer's power curve, which describes a new machine in standardized air on a test site rather than a twelve-year-old asset in complex terrain operating behind two rows of neighbors. Measured against it, almost every turbine in a mature fleet appears to underperform, and a finding that applies to every asset directs attention to none of them.
The reference that carries information is the fleet itself. When a turbine is compared against comparable units operating in the same wind at the same time, the residual belongs to the machine, and a residual that persists across seasons belongs to something that can be found and corrected.
The inputs are the ten-minute SCADA channels a turbine already records: active power, nacelle wind speed and direction, nacelle position, rotor speed, pitch angle, ambient and component temperatures, and operating state. Twelve months of history is sufficient for a first assessment, because a shorter window contains a single season of wind resource and cannot distinguish a machine that is degrading from one that is simply being measured in an unrepresentative quarter.
The record is validated before it is modeled. Curtailment, forced outage, and periods of manual intervention are identified and held out of the fit, because a model trained through a curtailed period learns curtailment as normal behavior and then reports its absence as a gain. Sensor faults are treated the same way. A nacelle anemometer that has drifted is one of the most common sources of an apparent loss that is not there, and identifying it is part of the analysis rather than a precondition for it.
Nothing is installed at the turbine, nothing changes in the control system, and no certification work with the original equipment manufacturer is required. That constraint is deliberate: anything that touches the machine converts a data question into a warranty conversation, and moves the decision from an operations team to a legal one.
Each finding identifies the turbine, the loss mechanism, the period over which the deviation was observed, the annual energy at stake, and the evidence supporting the estimate. It also carries a confidence, because a production deficit is an inference from a model rather than a counted quantity, and a figure presented without that distinction will not survive the meeting it was produced for.
The intended reader is the person accountable for the fleet's output. A finding is written to be acted on by an operations team and defended to an investment committee without being translated between the two, which is the point at which most performance analysis loses its authority.
Establishing whether a machine is genuinely underperforming is the work the models do. On most fleets that is the step an engineer completes before diagnosis can begin.
See how OpenTurbine can help your team understand performance, anticipate operational issues, and make better decisions across your wind assets.