This presentation reviews the historic use of machine learning (ML) in NDT and the current state of the art. Some of the ways to work around the traditional lack of training data are introduced and evaluated. The performance and validation of modern deep learning networks is presented and compared to human inspectors.
Attendees will come away with an understanding of:
- recent developments and applications of machine learning in NDT
- the large amount of flawed and unflawed data sets provided by ML
- performance and reliability requirements of NDT