[3M2-GS-10u] AI application

11:30 AM - 11:45 AM JST(2:30 AM - 2:45 AM UTC)

[3M2-GS-10u-03] Convolutional Model Order Reduction for Lagrangian Mechanics

〇Taiki Umetsu¹, Takashi Matsubara^1,2 (1. Hokkaido University, 2. RIKEN-AIP)

Keywords:

Lagrangian Mechanics,Model Order Reduction,Convolutional Neural Network

Lagrangian Neural Networks (LNNs) are a method for modeling physical systems by learning the system's Lagrangian, thereby enforcing conservation of energy. Applying LNNs to large-scale systems is computationally challenging, so researchers have explored combining LNNs with model order reduction techniques. However, previous approaches used fully connected autoencoders for model order reduction, which can ignore spatial structure and locality. In this study, we propose to use convolutional autoencoders for model order reduction. Experimental results show that the proposed method substantially improves reconstruction performance for object shapes and demonstrates high stability in long-term prediction.