// nuwright
● live

torquescope

Flags wind-turbine drivetrain faults from existing SCADA data, with no new sensors.

open torquescope.com ↗
job
early fault warning
domain
energy · diagnostics
runs
hosted · API
class
signal pipeline
scale
tested on 95 benchmark datasets, 5.2 million SCADA rows
since
2026

One call

  1. 01BaselineA periodic baseline for each temperature sensor
  2. 02HeuristicRise, volatility and trend over a seven-day window
  3. 03NormalExpected temperature for the power and weather conditions
  4. 04HybridHigh only when both agree; one alone is damped
  5. 05AlarmA running count of anomalies; the alarm needs a sustained run

In · out

inputs
  • existing 10-minute SCADA data
  • temperature sensors
  • power output
  • ambient temperature
outputs
  • anomaly score per turbine
  • running anomaly count
  • alarm

Where the person is

A high score needs two independent methods to agree, and an alarm needs sustained evidence, so one noisy reading never trips it.

code decides● yes · the final say is deterministic

What is measured

Against: The CARE wind-turbine fault benchmark: 95 datasets from three wind farms.

CARE score 0.588, against 0.66 for the autoencoder baseline it aims to beatmeasured
Warned 29.5 to 59.3 hours ahead on four selected fault eventsmeasured
Applied unchanged to three more wind farms (Kelmarsh, Penmanshiel, Hill of Towie; 41 turbines)stated

Stack · links

PythonFastAPISupabase (Postgres)Three.jsRenderVercel
last updated 2026-07-01 · from source, not marketingclaims carry their own status