The new inspection system is based on a cloud/SaaS solution with distributed and load balanced worker servers, which makes it fast and easy to scale from a few users to thousands. The core system infrastructure is built for integration with other software systems with an extensive API, and designed for scalability, uptime and security. The system brings structure to data acquisition with improved quality from sensors and metadata to speed up the post processing significantly, but is also able to analyse existing data or data taken with existing acquisition methods like digital cameras or simple drones without sensors. Images are placed into geometric models to create an index to follow issues over time and analyse issues in context to placement. The analysis of acquired data can be done by computer vision and CNN (Convolutional Neural Networks). CNNs eliminate the need for manual feature extraction by extracting features directly from raw images. Deep learning algorithms can learn discriminative features directly from data such as images, text, and signals. This automated feature extraction makes CNN models highly accurate for computer vision tasks such as damage classification. For typical blade damages a probabilistic tool to predict the likelihood of and the approximate time to imminent failures will be developed. The project will also provide industry templates for inspection reporting meeting branch requirements. These reports can also be delivered as an online and interactive version with ability to zoom, rotate and measure the areas of damages so maintenance & repairs can be planned with high precision. Advanced sensors are also methodically tested for detection of structural damages with no or sparse visual footprint on the blade surface.
The project will explore use of imagery acquired from drones to inspect wind turbines and advanced computer vision techniques and deep learning will be used to automatically detect and classify turbine blade damages found from the imagery.
|Effective start/end date||01/01/2017 → 31/12/2019|