Search Results

During the last decades, big steps have been taken towards a realistic simulation of NVH (Noise Vibration Harshness) behavior of vehicles using the Finite Element (FE) method. The quality of these computation models has been substantially increased and the accessible frequency range has been widened. Nevertheless, to perform a reliable prediction of the vehicle vibroacoustic behavior, the consideration of uncertainties is crucial. With this approach there are many challenges on the way to valid and useful simulation models and they can be divided into three areas: the input uncertainties, the propagation of uncertainties through the FE model and finally the statistical output quantities. Each of them must be investigated to choose sufficient methods for a valid and fast prediction of vehicle body vibroacoustics. It can be shown by rough estimation that dimensionality of the corresponding random space for different types of uncertainty is tremendously high.

In order to predict reality as accurately as possible leads to the fact that numerical models in automotive vibroacoustic problems become increasingly high dimensional. This makes applications with a large number of model evaluations, e.g. optimization tasks or uncertainty quantification hard to solve, as they become computationally very expensive. Engineers are thus faced with the challenge of making decisions based on a limited number of model evaluations, which increases the need for data-efficient methods and reduced order models. In this contribution, variational autoencoders (VAEs) are used to reduce the dimensionality of the vibroacoustic model of a vehicle body and to find a low-dimensional latent representation of the system.

Transfer path analyses of vehicle bodies are widely considered as an important tool in the noise, vibration and harshness design process, as they enable the identification of the dominating transfer paths in vibration problems. It is highly beneficial to model uncertain parameters in early development stages in order to account for possible variations on the final component design. Therefore, parameter studies are conducted in order to account for the sensitivities of the transfer paths with respect to the varying input parameters of the chassis components. To date, these studies are mainly conducted by performing sampling-based finite element simulations. In the scope of a sensitivity analysis or parameter studies, however, a large amount of large-scale finite element simulations is required, which leads to extremely high computational costs and time expenses. This contribution presents a method to drastically reduce the computational burden of typical sampling-based simulations.

Uncertainties play a major role in vibroacoustics - especially in car body design in the preliminary development because of the overall spread in the production that should be covered with one simulation model. Therefore, we use uncertain input parameters to determine the stochastically distributed admittance of the car body before each part of the car is fully designed. To gain a stochastic result - the stochastically distributed admittance curve - we calculate a deterministic finite element simulation several times with sets of stochastically distributed input parameter values. To reduce simulation time and cost of the car model with many million degrees of freedom we focus on the uncertain parameters that show a significant influence on the admittance curve. It is therefore necessary to be able to accurately estimate for each parameter if its influence on the admittance of the car body plays a major role for the noise vibration harshness simulation.

In vehicle NVH development, vibroacoustic simulations with Finite Element (FE) models are a common technique. The computational costs for these calculations are steadily rising due to more detailed modelling and higher frequency ranges. At the same time, the need for multiple evaluations of the same model with different input parameters, e.g., for uncertainty quantification, optimization, or robustness investigations, is also increasing. Therefore, it is crucial to reduce the computational costs in these cases. A common technique is to use surrogate models that replace the computationally intensive FE model to perform repeated evaluations. Several different methods in this area are well established, but with the continuous advancements in the field of machine learning, interesting new methods like the Gaussian Process (GP) regression arises as a promising approach.