Physically Informed and Data Driven Direct Models for Lamb Waves based SHM: Advantages and Drawbacks of Existing Approaches

Abstract

Validation and testing of Lamb wave based SHM algorithms requires numerous simulations that require themselves qualitatively and quantitatively consistent models representative of the physical behavior of the monitored structures and that are in agreement with experimental data. Finite elements models appear as an interesting solution to achieve this goal but are associated with large computational costs and low generalization abilities. On the other hand, data driven machine learning approaches are computationally very efficient and can predict fine details but at the cost of low physically interpretability. Original approaches trying to build physically informed models balancing the advantages and drawbacks of physics-based approaches and of machine learning approaches also exist and will be discussed in the context of Lamb waves based SHM of aeronautical structures.