Structural health monitoring of a satellite antenna: A benchmark between the Modal Method, iFEM coupled with Modal Strain Expansion, and a Physics Informed Neural Network

Shape sensing techniques for real-time deformation reconstruction are an increasingly important topic for Structural Health Monitoring (SHM), particularly in the context of digitalisation and digital twins. Artificial intelligence and artificial neural networks are showing promising results for such tasks. Physics-Informed Neural Networks (PINN) have recently gained interest due to less training data-dependent predictions by incorporating physics laws into the training. To benchmark such a PINN for solving inverse problems in shape sensing, referred to as iPINN, this paper presents a study on its performance against the Modal Method (MM) as an established technique on the example of a composite space antenna. Moreover, a combination of the two mainstream methods for shape sensing is tested on the same space structure. Starting from a few strain measurements, the strain field is first expanded using a variation of the MM. Then, displacements are reconstructed using the inverse Finite Element Method (iFEM). This study shows that both the iPINN and the combination of MM and iFEM can produce predictions of deformed shapes, although further investigations are needed to improve their accuracy.