The SimuStruct dataset contains 1000 cases of 2D rectangular plates with holes under load along with measurements of von Mises stress.
In this 3rd and final part of the Heat series, we delve into the idea of enhancing generalizational power in Neural Networks so they can learn more complex aspects. We exemplify these ideas by running Physics Informed Neural Networks (PINNs) on a custom-designed domain and boundary condition.
In this 2nd part of the series, we show that Neural Networks can learn how to solve Partial Differential Equations! In particular, we use a PINN (Physics-Informed Neural Network) architecture to obtain the results we obtained with classical algorithms in Heat #1.