ModalPINN: an extension of Physics-Informed Neural Networks with enforced truncated Fourier decomposition for periodic flow reconstruction using a limited number of imperfect sensors

Abstract

We propose a new architecture for Physics-Informed Neural Networks (PINN) specialised for oscillatory phenomena.• The hard-coded truncated Fourier decomposition allows a better precision than the classical approach at an equivalent number of degrees of freedom and computing time.• The proposed format shows great convergence for field reconstruction with data from a limited number of sensors and can overcome problems of time synchronisation and noise.