第１部門①

[11001-05-03] Construction of real-time structural vibration analysis using PINN (Proceedings of Symposium on Applied Mechanics)

*Toko Okuda¹, Mayuko Nishio¹ (1. University of Tsukuba)

Keywords:PINN, vibration analysis, machine learning, real-time simulation

In this study, a real-time dynamic analysis of structural members was conducted using Physics-informed neural networks (PINN). Currently, many studies are working on making structural simulations interactive and real-time. However, the high computational cost of structural dynamic simulation limits its application. In this study, PINN was used to conduct the real-time vibration analysis of a cantilever beam as a basic investigation. Prior to real-time simulation, a PINN model was constructed to solve the undamped free vibration problem. To improve the prediction accuracy of PINN, the number of automatic derivatives was reduced and weights were added to the loss function. Sequential learning and prediction for real-time simulation were then performed by PINN at fine time steps.