Machine learning ยท Digital twin
Predictive Maintenance of Solar Panels
Solar farms lose output when a panel fails and nobody notices. This project predicts panel defects with machine learning and shows them on a 3D replica of the farm, so less of the checking has to be done by hand.
The farm
A sun, a field of panels
Large solar farms are checked manually, which is slow and leaves faulty panels running for longer than they should. The project goal is simple: reduce that manual labour and the downtime that comes with it.
The panel
Every panel is a data source
The system is built from sensors, a Raspberry Pi and a Flask API. Each panel reports on its own condition, so problems can be spotted from data instead of walking the rows.
Green energy
More of the power you can make
A panel that is quietly failing wastes clean energy. Catching defects earlier means a farm produces closer to what it was built to produce.
Digital twin
A copy of the farm in 3D
The panels are replicated in Unreal Engine as a digital twin. When the model predicts a defect, it appears on the replica, so a person can see which panel needs attention without leaving the screen.
The model
A CNN predicts defects
A convolutional neural network (CNN) predicts defects, and the results feed the 3D replicas. Demo footage, the architecture and the repository link will be added here.
Placeholder: demo video or screenshots