Deep learning‐based reduced order models for the real‐time simulation of the nonlinear dynamics of microstructures

We propose a non-intrusive deep learning-based reduced order model (DL-ROM) capable of capturing the complex dynamics mechanical systems showing inertia and geometric nonlinearities. In first phase, limited number high fidelity snapshots are used to generate POD-Galerkin ROM which is subsequently exploited data, covering whole parameter range, in training phase DL-ROM. A convolutional autoencoder employed map system response onto low-dimensional representation and, parallel, nonlinear trial manifold. The on manifold described by means feedforward neural network that trained together with autoencoder. strategy benchmarked against solutions clamped-clamped beam real micromirror softening multiplicity solutions. By comparing different computational costs, we discuss impressive gain performance show DL-ROM truly represents real-time tool can be profitably efficiently system-level simulation procedures for design optimization purposes.