Paper No. 80
W Nash1,2, A Lockwood3, M Brameld3, L Holloway2,4, N Birbilis1,2 – 1Woodside Innovation Centre, Clayton, Australia, 2Department of Materials Science and Engineering, Monash University, Clayton, Australia, 3Woodside Energy, Perth, Australia, 4MEnD Consulting Pty Ltd, West Perth, Australia
The confluence of aging infrastructure and commodity prices has created a scenario where maintenance costs are often prohibitive and inspection has become the primary means of ensuring the integrity of critical infrastructure. This inspection comes at a considerable cost due to the need provide access to equipment that is often elevated and operating in an oil and gas plant setting. Deep learning and artificial intelligence computer vision systems are revolutionising many fields, and condition assessment can take advantage of this progress to provide expert opinion systems on site, and direct inspection resources to areas of high risk. In this work, the capability of semi-supervised deep learning models to conduct rapid condition assessment of complex steel structures is presented. Currently this technology is applied in post-processing of site photos, although this provides for some time saving, a roadmap is provided to deploying this software for on-site assessment in real time, for example with cameras mounted on drones, or augmented reality headsets, and explore the challenges remaining for field deployment. The discussion will continue to the pitfalls of artificial intelligence systems and how to avoid them. Finally, a discussion of the extensibility of these systems will be presented to illustrate how the same approach can be used to tackle problems in other areas.