AI is increasingly used in industry and has found a new outlet in inspection routines. Siemens Gamesa partnered with Fujitsu to finalize an inspection solution that involves AI to identify potential failures of huge wind blades.
The fibreglass blades used by modern wind turbines are massive structures that can be up to 75 meters long But the extremely arduous conditions in which they operate mean that even tiny flaws in their construction can result in a catastrophic failure. Intense inspection of completed blades is therefore necessary – a task that when carried out purely visually by a human operator can take up to six hours per blade.
But at one of Europe’s leading manufacturers of such blades Siemens Gamesa, that approach has been replaced by one involving automated inspection techniques linked to an artificial intelligence database.
As a result, inspection routines for each new blade can now be completed in as little as 90 minutes. Human eyesight is reserved for close-up examination of possible flaws identified by the automated system, which uses non-destructive ultrasonic scanning technology.
According to Antonio de la Torre, CTO for Siemens Gamesa, this application of AI is just the most recent example of the company’s exploration of its potential:
AI projects have been done at Siemens Gamesa for a couple of years in applications like remote diagnostics of wind turbines or site-specific wind controller optimization. The use of AI technology for blade inspection as quality assurance is relatively new.
The technique is now in use for the Siemens Gamesa production sites at Aalborg in Denmark and Hull in the United Kingdom. This new application will not be the last, however, as the company is now convinced of its potential for further exploitation.
De la Torre confirms the company is currently testing AI technology as part of blade integrity management within the wind service business unit.
Wind farms can be inspected with automated drones that capture images from all rotor blades on-site. The images are analyzed by intelligent software that indicates and assesses all irregularity in the high-resolution images of the blades. This application will be launched as a product within the next year.
The application at Siemens Gamesa uses a basic AI “engine” supplied by the Japanese company Fujitsu which has been customized to the user-specific requirements.
This usually requires large amounts of labelled data and some form of machine learning, for instance using deep neural networks. Many AI use cases are so application-specific that even closely aligned AI engines need to be tuned or adapted to customer needs, but Fujitsu also has a number of generic AI application APIs including translation and object recognition.
Siemens Gamesa’s use of the technology also exemplifies the way that AI can be applied to the greatest effect within industry.
Dr Snelling states,
The most important areas in industry for AI are with complex processes and decision-making where much of the work is tedious or repetitive, but still requiring a significant level of judgement. The result is usually a partnership between a human and the AI, where the final call is still made by an engineer.
For him, the ideal areas for industrial AI implementation, are “non-destructive testing, predictive maintenance and process optimization.”
Moreover despite the sophisticated nature of the technology, both the necessary training and implementation can be quite straightforward.
There are two parts to an AI application – training and operations. In the Siemens Gamesa case the training required moderate levels of computing power – just a small cluster with multiple GPUs. For operational needs a normal PC or laptop is adequate.
Actually getting an implementation up and running can also be quite speedy.
We have had experience with proof of technology requiring only a week to verify the technical approach. For usable deployment a few week to a few months is sufficient.
As such he is confident that AI applications in industry will continue to expand:
This is definitely a growth time both in terms of achieving operational efficiency and exploiting whole new business strategies based on AI. Industries that do not embrace AI soon will fall behind the pack.