State model
Wind, power, rotor, generator, gearbox, bearing, pitch, yaw, and operational state share one readable runtime model.
Technical portfolio / digital twin systems
Aperture
A browser-native hybrid digital twin for utility-scale wind turbines.
Aperture turns SCADA-style telemetry into a live turbine state, interpretable model outputs, fault scenarios, and operator decisions in one static web experience.
Built by Matthew Mitchell / MMKPC Studios
4browser ML models
7reproducible fault modes
0.869 msmedian inference latency
9runtime layers
The project
From telemetry to a decision surface.
The challenge was to make a digital-twin workflow that could be inspected, demonstrated, and deployed without a backend or a specialist runtime.
Aperture combines a typed SCADA replay, physics-light state propagation, interpretable JavaScript models, scenario branching, and operator controls. The result is not presented as a billing-grade wind-farm platform. It is a transparent reference system that makes the full reasoning path visible.
/
Every signal has somewhere to go. Every alert has a reason. Every intervention can be traced back through the loop.
Runtime architecture
The operating loop
One event path connects data, models, simulation, controls, and the 3D operator view.
01SCADAtelemetry replay
02Ingesttyped state
03Featuresderived signals
04MLfour models
05Scenariofault branches
06Actuatoroperator response
073D viewvisual state
What is inside
Built as a system, not a screenshot.
Interpretable ML
Power-curve regression, gearbox forecasting, state classification, and anomaly scoring run as inspectable browser modules.
Fault scenarios
Gearbox overheat, pitch stuck, yaw misalignment, bearing fault, icing, cut-out, and normal operation can be replayed.
Operator loop
Annunciation, alarm acknowledgement, pitch, yaw, shutdown, reset, recording, replay, and export connect the twin to action.
Evidence
Measured, documented, and honest about the boundary.
The public surface includes the selected metrics, signal mapping, technical paper, evaluation notes, and figures. The raw dataset and full evaluation runner remain private development materials.
Power-curve R20.9758
Anomaly AUROC0.8601
Classifier accuracy0.9243
Gearbox forecast RMSE3.36 C
Public portfolio surface
Explore the system, then inspect the reasoning.
This public mirror is intentionally curated. The original development repository remains private; the interactive showcase and published technical artifacts are available here for review and download.