UT automatic diagnostic using machine learning and numerical databases
S Le Berre, D Roue, X Artusi, R Miorelli and C Reboud
Machine learning techniques (ML) offer highly promising possibilities to efficiently solve inversion problems and, therefore, to provide reliable and automatic diagnostics. Two important challenges related to the use of ML in this context are, first, the availability of data for the training stage and, second, the identification of relevant descriptors or so-called features to be learnt. Working with efficient simulation tools proves very useful in addressing both of these challenges. In this communication, the authors discuss this strategy and present corresponding developments recently achieved in the CIVA software platform. The authors illustrate the benefits of these numerical tools on a typical phased array ultrasonic testing application case. In particular, the authors demonstrate how optimisation and quantitative assessment of the diagnostic performance is carried out by means of sensitivity studies with respect to influential parameters.
