torquescope


cold-start fault detection for wind turbines. No new sensors, no historical failures, no ML team required.

Status
active · validated across three farms
Domain
wind · diagnostics
Period
2026 to present
Role
founder · principal

Torquescope reads the SCADA data wind farms already produce and tells the operator which turbines are quietly going wrong. It needs no new sensors and no library of past failures to learn from. It is, in effect, the Ab Astris methodology turned inward, pointed at a drivetrain rather than a star, asked to flag the early warning before the gearbox does.

Criticality counter · the alarm trips at 72.FIG. I72alarm02550100fault flaggedDAY 1DAY 7DAY 14DAY 21DAY 28SCADA WINDOW · KELMARSH T03 · GEARBOX HS BEARING
Fig. 1The criticality counter accumulates anomalies and trips the alarm at 72: typically days to weeks before failure.

A modern wind farm generates more SCADA data than its operators have time to look at. The faults that matter most (gearbox bearings, generator windings, pitch-system drift) declare themselves weeks before they become outages, but only to a reader who knows where to look. Most fault-detection products require either bespoke vibration sensors or a dataset of historical failures the operator has spent years not having.

Torquescope was built for the operator without either. It reads the same SCADA stream every turbine already produces and reasons about the drivetrain from temperature, power, and ambient conditions alone.


A hybrid v5 pipeline combines Lomb-Scargle periodic analysis with a Normal Behaviour Model (binned power × ambient → expected temperature) and a criticality counter that accumulates anomalies and resets on normal operation. Alarm threshold is set at 72. The CARE benchmark scores the result on coverage, accuracy, reliability, and earliness.

The same modules, unchanged, were then pointed at three open-data commercial wind farms. They worked.


On the CARE benchmark, torquescope scores 0.588 against the 0.66 autoencoder baseline. Close enough to be useful, and reached without the autoencoder's appetite for training data. Across Kelmarsh, Penmanshiel, and Hill of Towie, the same pipeline identified turbines with elevated anomaly rates and surfaced dual generator-bearing drift on the Siemens fleet without any retraining.


Live demonstration available at torquescope.com, with the dashboard running across Demo, Fleet Health, and Resource Intelligence modes. We are in conversation with operators, OEMs, and validation partners (including ORE Catapult's Levenmouth turbine) about extended field trials.