Physics-informed neural network for vibration analysis of large membranes

Abstract Dynamical systems are mathematical models often represented by differential equations (DEs), which can be linear or nonlinear. The developments in machine learning (ML) applications have paved the way to solve dynamical systems with ease. However, the major hindrances involved in all those algorithms are their inability to predict noisy and sparse datasets as well as to incorporate the prior physical information. Most of the implementation of neural networks is to perform nonlinear transformations from input to output and also data-driven. The new scientific computing paradigms viz. physics-informed neural networks (PINNs) have revolutionized traditional utilization of ML algorithms. It represents a promising set of algorithms that integrate the training process with the known physical properties governed by DEs. Here, we have implemented PINNs, to solve the vibration equation of large membranes. Finally, the obtained results are verified with the ground truths through simulations for tracking the performance of proposed algorithm.