Project Case Study

Developed an LSTM machine learning workflow to predict wind turbine power from SCADA data, reconstruct drivetrain torque, and evaluate gearbox tooth loading. The predicted torque was used with AGMA methods to calculate tangential loads, bending stress, contact stress, and safety factors for a wind turbine gearbox.

Start Case Study Back to Projects View PDF Report
Tools Python / TensorFlow / SCADA / AGMA
Result R² ≈ 0.88 / MAE ≈ 79.5 kW / Gear safety factors > 1
Wind turbine nacelle cutaway showing drivetrain and gearbox components
01

Objective / Problem

What needed to be solved

Predict gearbox loading from real wind turbine operation

Wind turbine gearbox loads are not constant because wind speed, rotor speed, yaw angle, and power output change continuously during operation. The goal of this project was to use real SCADA data to predict turbine power, convert that power into drivetrain torque, and use the reconstructed torque history to evaluate gear tooth loading.

This matters because gearbox failures are one of the most expensive wind turbine problems, and stress histories such as tangential load, bending stress, and contact stress help explain how the gearbox behaves under real operating conditions.

Project snapshot

Quick details

Role Machine Learning / Gearbox Stress Analysis
Timeline 2025
Industry Renewable Energy / Wind Turbine Drivetrain
Deliverable LSTM model, torque reconstruction, AGMA stress analysis, final report and presentation
02

Approach and Tools Used

Method

LSTM torque reconstruction and AGMA stress workflow

A stacked LSTM model was trained on SCADA time-series data using wind speed, wind direction, rotor speed, yaw angle, theoretical energy, and power. The predicted power was converted into drivetrain torque, then propagated through a three-stage planetary-helical gearbox model. AGMA methods were used to calculate tangential tooth loads, bending stress, contact stress, and safety factors for the helical gear stages.

Tool stack

Python, deep learning, SCADA, and gear stress methods

Tool stack used for the machine learning model, drivetrain torque reconstruction, and gearbox stress calculation workflow.

Python TensorFlow LSTM SCADA Data AGMA Excel Gearbox Design
04

Results / Outcome

LSTM Accuracy

R² ≈ 0.88

SCADA power prediction

LSTM model predicted wind turbine power from SCADA data with stable test performance, reaching about 79.5 kW MAE and 126.5 kW RMSE.

Gearbox Loading

Torque → Stress

Mechanical load reconstruction

Predicted power was converted into drivetrain torque and used to calculate tangential tooth load, bending stress, and contact stress through AGMA gear rating methods.

Engineering Value

Physics + ML

Data connected to stress analysis

This project proves that machine learning predictions can be connected to mechanical engineering analysis, turning real turbine operation data into gearbox loading and gear stress histories.