| description abstract | Abstract. The recently proposed system identification via validation and adaptation (SIVA) method allows system identification, uncertainty quantification (UQ), and model validation directly from data. Inspired by generative modeling, SIVA employs a neural network (NN) that converts random noise to physically meaningful parameters. The known equation of motion utilizes these parameters to generate fake accelerations, which are compared to real training data using a mean square error (MSE) loss. For concurrent parameter validation, independent datasets are passed through the model, and the resulting signals are classified as real or fake by a discriminator network, which guides the parameter-generator network. In this work, we apply SIVA to simulated vibration data from a cantilever beam that contains a lumped mass and a nonlinear end attachment, demonstrating accurate parameter estimation and model updating on complex, highly nonlinear systems. We demonstrate that the SIVA method is applicable to cases where an existing linear model is available and one only wishes to add onto that model, and for cases where a sustained load is applied to the system, such that the external forces must be included in the SIVA method. The results demonstrate that SIVA can accurately perform model updating and provides a foundation for future research on its use for continuous model improvement to capture parameters changes as a system ages. | |
