The AES Corporation is a Fortune 500 global power company generating and distributing electric power in 15 countries. As part of AES’s renewable energy generation portfolio, they own and operate windfarms in multiple geographical locations.
The Challenge: Automating Visual Aerial Inspections
Due to the difficult operating environment of a wind turbine, AES partnered with a 3rd party, Measure, to perform inspections using drone technology. Previously, images captured through the drone inspections were reviewed by a human workforce to identify equipment readiness and maintenance issues. AES was actively working to develop a computer vision solution to reduce the amount of images requiring human review. AES needed help developing a platform to manage the labeling effort, model training and prediction process for their wind turbine and drone visual inspection solution.
“Working with Google, we identified ClearObject to be our local GCP partner to help us architect and develop our platform using the latest thinking in cloud and serverless tools available from Google. ClearObject has been a great partner and worked to quickly develop this platform for us”
-- Nicholas Osborn, Global Machine Learning PMO, The AES Corporation
The Solution: Vision Aerial Intelligence Platform
ClearObject assisted AES in the development of the computer vision solution by instrumenting a reliable
data processing pipeline within Google Cloud Platform (GCP), enabling the AES data science team to focus on model development and training. Once appropriate intervals were reached in the computer vision model, it was operationalized within the data pipeline, resulting in a reduction in the amount of images requiring human review to identify equipment failure.
Leveraging object detection with Google’s AutoML Vision technology, ClearObject was able to develop a minimal viable machine learning model (MVM) to detect gelcoat and lightning damage. Initial results showed our MVM could detect 95% of gelcoat and lightning damage while reducing the number of images reviewed by 55%. With further
iteration, this model will be further trained to detect more damage types, increase confidence, and reduce the number of images needing reviewed leading to increase cost savings and process improvement. The same technology and approach can also be leveraged to incorporate other use cases such as inspections for solar arrays, T&D lines, hydro power structures, boiler structures, safety checks and more.
more effective and efficient
inspections in the field.
of images eliminated from needing human review after initial round of model training. This is expected to improve to 90% as the model continues to learn and improve overtime.