Wind Turbine Blade Centroid Detection

Problem Statement

Measure blade geometry and centroid movement within structured grids using computer vision so that the mechanical performance of wind turbine blades can be assessed without manual measurement.

Dataset Description

• blade_cg_data.xlsx, blade_cg_comparison.xlsx, and blade_cg_comparison.jpg store the measurement records and visual comparisons.
• The images_v2/ folder is referenced in the notebook but not included here because the raw imagery is proprietary. Provide your own high-resolution blade photos following the filename pattern images_v2/<index>.jpg for the notebook to run.
• Supporting auxiliary exports live under cg_visuals/ and cg_visuals_v2/ for reference, but they are not replayed automatically.

Methodology

1. Resize the imported image set to a consistent scale (default 10% of original resolution).
2. Calibrate the metric conversion by clicking two points whose real-world distance is known and computing millimeters per pixel.
3. Apply HSV masking to isolate blade edges and run contour detection to estimate blade outlines.
4. Compute centroids per contour, export coordinates to Excel (openpyxl), and log the results along with scale-adjusted measurements.

Key Findings

• The notebook writes centroid coordinates (X, Y) and summary statistics back to Excel while logging any missing segments.
• Metric comparisons between blades and the generated comparison image help visualize alignment over time.

Technologies Used

• Python 3.9+
• OpenCV (opencv-python)
• NumPy
• pandas
• openpyxl

How to Run

1. Install dependencies via pip install -r requirements.txt.
2. Populate images_v2/ with .jpg blade photos that match the notebook’s naming pattern.
3. Run Centroid.ipynb interactively in a Jupyter environment, clicking the required calibration points when prompted.
4. Check blade_cg_data.xlsx for exported centroid records and use the included comparison sheet for quick verification.

Future Improvements

• Automate the calibration step by reading a ruler present in the image rather than manual clicks.
• Add edge-case handling for partial blade occlusion in the frame.
• Export results directly into a SQLite or CSV log for versioning.