abstract
- This work addresses the challenge of active vibration attenuation in flexible structures by proposing an improved control architecture based on Finite Element Differential Neural Networks (FEDNN). Unlike traditional black-box identifiers, the FEDNN incorporates the spatial topology of the Finite Element Method, enabling a physics-consistent approximation of spatiotemporal dynamics. The network is identified offline using experimental data acquired from a cantilever beam, which serves as a canonical distributed parameter system. Based on this data-driven model, two distinct control strategies are synthesized and compared: a nodal Model Reference Adaptive Control (MRAC) and a distributed Neural-Sliding Mode Control (NSMC). Numerical results demonstrate that while both approaches successfully reject persistent non-vanishing disturbances, they exhibit distinct operational characteristics. The study concludes by analyzing the trade-offs between the robustness of the distributed sliding mode approach and the smooth, energy-efficient performance of the proposed nodal MRAC scheme.