Abstract
Abstract
This research presents a finite element enhanced physics-informed neural network (FE-PINN) framework which is more versatile than traditional PINNs and can be applied to various civil engineering problems. Unlike traditional PINNs, a finite element discretization scheme is used to discretize the space so that the governing equation is reduced to an ordinary differential equation (ODE) as opposed to a partial differential equation (PDE) used in traditional PINNs. FE-PINNs are hypothesized to be more efficient, and easier to use than traditional PINNs, especially in civil engineering problems.This study applies these FE-PINNs to a one story building which is idealized as a spring mass system to inversely estimate its stiffness parameters. Results obtained from this pilot study provides optimism for application of FE-PINNs to more complex systems.
Abstract
This research presents a finite element enhanced physics-informed neural network (FE-PINN) framework which is more versatile than traditional PINNs and can be applied to various civil engineering problems. Unlike traditional PINNs, a finite element discretization scheme is used to discretize the space so that the governing equation is reduced to an ordinary differential equation (ODE) as opposed to a partial differential equation (PDE) used in traditional PINNs. FE-PINNs are hypothesized to be more efficient, and easier to use than traditional PINNs, especially in civil engineering problems.This study applies these FE-PINNs to a one story building which is idealized as a spring mass system to inversely estimate its stiffness parameters. Results obtained from this pilot study provides optimism for application of FE-PINNs to more complex systems.