Motivation:

In real-world applications, damage detection using non-destructive evaluation (NDE) techniques has become an important task for which data sets can be generated by experiments or by numerical simulations. The presence of such data sets enables the use of data-driven science and data assimilation techniques for the identification of damages in composite materials such as Fibre Metal Laminates (FMLs). In this subproject, methods which are (i) model-free, (ii) model-based and (iii) model-assisted will be examined in order to accomplish this objective.

Aims:

• Use ideas from parametric model order reduction and data assimilation for the identification of damages in FMLs.
• Identify damage parameters in FMLs by Physics Informed Neural Networks (PINNs).
• Detect, segment and characterise damages in FMLs using artificial intelligence algorithms.

Approach:

• Construct reduced order models tailored for damage detection combined with data assimilation techniques for the identification of unknown damage parameters.
• Incorporate the damage detection process into the learning process of PINNs and learn the damages alongside with the solution.
• Collect experimental and/or synthetic CT/GUW  data to train a machine learning model. Use the trained model to detect, segment, and characterize damages in FMLs.