Fault Detection in Wind Turbines

AI powered Wind Turbine Inspection

Given the safety hazards and time taken to carry out wind turbine inspections, it is not surprising that the use of drones for this purpose is on the rise. Not only do they help carry out regular inspections but also help identify and gather data to support warranty claims.

The conventional manual inspections using ropes and platforms can at best inspect 2-3 wind turbines a day while a semi-automatic drone could inspect 12-15 turbines in the same time. A 2 man team consisting of a drone pilot and sensor operator can maneuver and click high definition images of different parts of the wind turbine, typically inspecting each blade in under 10 minutes of flight.

Aerial wind turbine inspections using drones consists of 3 phases

Use of drones instead of humans can cut down the time required for the first phase by a factor of 6-8 times and cost by 8-10 times. At the same time, they help reduce the hired manpower, safety hazards and costs.

The second phase of “Analysing” the collected images can either be done manually or automatically using cutting edge machine learning models. The task of manually looking at hundreds of high definition images is tedious and mundane which leaves it vulnerable to human errors. This also consumes significant time and requires hiring skilled labour who are experienced in this domain. On the other hand, use of AI for the second phase, suffers from no such issues resulting in highly accurate analysis in a cost effective way.

Automating the second phase is where Nanonets fits in. We build and deploy custom inspection models which identify and categorize the different kinds of faults with high accuracy and speed. Each inspection model is trained and fine tuned to each use case. Depending on the size, each wind turbine can be covered in 10-20 high definition images. Our models are capable of analysing these images at the rate of 2-5 wind turbines a minutes, covering wind farms within hours. The same analysis would take days if done manually.