Contribution
Frequency-Domain-Based Parameter Identification of a Finite Element Vehicle Body in Vibroacoustics
* Presenting author
Abstract:
Uncertainty consideration in the early development of complex systems like vehicle bodies prevents expensive modifications close to the production start. In previous work, the authors have identified sensitive parameters in a NVH body-in-white finite element model, reducing the uncertainty parameter space for further investigations by nearly 90%. However, finding reliable distributions of these sensitive input parameters, like material parameters and geometry variations, is challenging, especially for large-scale vibroacoustic models with numerous parameters. This work solves this challenge using neural networks based on frequency domain data like the admittance curve. Due to the curse of dimensionality, directly determining the parameter group of 170 parameters with a fully connected neural network is complicated. Hence, the authors analyze the impact of the number of parameters on the performance of the neural networks. In addition, they use both a fully connected feed-forward neural network and a one-dimensional convolutional neural network. The latter demonstrates enhanced performance compared to the fully connected model. Lastly, the authors present a method to ascertain distributions of the analyzed parameters based on synthetic measurement information, substantially enhancing these finite element simulations and offering an approach to address the obstacle of identifying uncertain variables in large-scale finite element frameworks utilizing frequency-domain-based data in vibroacoustics.