Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Physics informed neural networks with focus on the solution of inverse problems arising in vibrations of rods

Nikolou Kalliopi

Full record

---

Summary

Physics-Informed Neural Networks (PINNs) offer a transformative approach to solving complex problems governed by partial differential equations (PDEs). This thesis investigates the application of PINNs for bothforward and inverse problems in the context of a vibrating rod system. The forward problem focuses on approximating the rod’s displacement over time, given the governing PDE, initial conditions, and boundary constraints. The inverse problem centres on identifying unknown parameters, such as material properties, directly from observed data. PINNs eliminate traditional numerical methods’ reliance on mesh generation and efficiently handle challenges posed by high-dimensional systems, noisy data, and irregular geometries. By leveraging their ability to integrate physical laws with observational data, PINNs achieve accurate solutions while reducing computational complexity. This thesis further explores adaptive optimization techniques to improve convergence and accuracy, particularly for inverse problems where initial parameter estimates are far from their true values. The results demonstrate that PINNs effectively model the physical behaviour of the vibrating rod and accurately recover system parameters, showcasing their potential as a robust alternative to classical numerical methods. This work highlights the versatility and adaptability of PINNs, paving the way for future research into their application in more complex, real-world systems.

Available Files

Services