الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 119 |  |  |
| | | from | 119 | | |
Abstract
Physics Informed Neural Networks (PiNNs) is a new field of research that combines the power of deep learning with the physical laws governing a system. PiNNs leverage differential equations –which most physics problem based on it- as a regularization for a learning task, allowing for the incorporation of physical constraints into the neural network architecture. This approach has the potential to overcome the limitations of traditional numerical methods in deep learning, such as increased time and the need to modify equations for each method. In order to develop this knowledge, the proposed methodology was implemented and applied through two case studies of one of the building blocks of optical waveguides which is slab waveguide.
In this work, an enhanced perspective for analyzing optical waveguides is provided using PiNNs which may overcome the drawbacks of traditional methods. A solver for a 1D dielectric slab waveguide with accuracy reaching 99% and a mean square error of 〖10〗^(-5) was made and the proposed model could analyze slab waveguide with very small change (Δn=〖10〗^(-5)) in refractive index between its regions with relative error of 〖10〗^(-6).
The results of this thesis were compared with previously published results, and a high degree of agreement was observed between them and the results of this model. This work is important because it provides a new perspective for analyzing optical waveguides using physics-informed neural networks, which may overcome the drawbacks of traditional methods and lead to improved accuracy and efficiency in analyzing optical waveguides.