COMPLAS 2027

IS045 - Identification of Nonlinear Nonsmooth Differential Equations Under Inequality Constraints Using Artificial Intelligence: Applications to Industrial Contact Mechanics Problems

Organized by:
Keywords: Data-driven modelling
Repeated impacts with friction occur in many industrial systems subjected to vibrations and mechanical clearances, and lead to irreversible phenomena such as wear, damage accumulation and surface degradation. Predicting their long-term effects remains challenging, because the underlying impact–friction laws are difficult to identify from observed mechanical responses. Identifying wear laws directly from experimental data is a very difficult task. We therefore start with a better-understood phenomenon, impact combined with friction, and aim to identify the impact–friction law from the trajectories of the system, as a first step toward the identification of wear. The study case is a viscoelastic Timoshenko beam impacting two rigid obstacles with friction. The data are not experimental: they are generated by numerical simulations driven by a known impact–friction law, so that the identified law can be compared with the reference one. Since this inverse problem is highly nonlinear and ill-posed, we restrict attention to impact–friction models deriving from a dissipation potential, which we seek to recover in an explicit and interpretable form. In doing so, we aim to extend to this nonsmooth setting data-driven identification methods developed for smooth dynamical systems, such as SINDy.