Predictive maintenance of wind turbines using drone imagery

Overview

Drones are used to inspect wind power stations because they are cheaper, faster, and safer. Our client was a drone pilot agency servicing an energy utility company to monitor and inspect their wind turbines. The wind turbines have many huge interconnected parts and it becomes imperative to monitor the health of the parts for both safe and seamless operations. The drones were being used to take images and videos of the wind turbines periodically. Our client wanted to devise an efficient predictive maintenance system that would assess the health of the turbines and intercept failures. Taking action after a major breakdown would amount to huge costs and losses due to work shutdown. The main purpose was to anticipate failures, avoid production shutdowns and accidents. Historical data collected by the drone could provide useful insights to detect the signs and symptoms of machine failure before critical damage.

Solution

We developed a Deep Neural Network model to recognize the health and degradation of the Wind Turbine across various parameters including rust, dents, dust, etc. We first collected the drone images from our client to generate a customized data pertaining to the good health and each deterioration class of the Wind Turbine. We then trained a Deep Neural Network to inspect the Wind Turbine to spot any anomalies on the surface of the Wind Turbine.

Impact

The predictive maintenance system was able to detect the failures ahead of 6 months with almost 85% accuracy. The model was able to pinpoint the parts which are likely to fail including the type of failure and their life. The model lead better planning of maintenance and work orders, lowering costs, and reducing risk to workers onsite by almost 40%.

Technology stack