Design of electric powertrains to achieve NVH performance using autoencoders and a physical meaningful latent space

One of the fundamental differences in the perception of electric vehicles is how their radiated noise is perceived with respect to classic internal combustion engines. Even though electric vehicles are usually quieter, the tonal content of the radiated noise can be more annoying. Considering this, the work presented in this paper proposes a novel approach that starts from the assumed radiated noise spectrum profile as input to a neural network that can produce parameters to design the powertrain that would generate that specific noise profile. The proposed network acts as an autoencoder where the latent space is forced to have a physical meaning, being the powertrain parameters. Since a multiple combination of parameters can result in a similar noise profile, a variational autoencoder approach is also presented. The network predictions are validated against results of a 3D CAE model. Overall, the mean absolute error is around 5 dBA for this feasibility study, which aims to demonstrate the concept. This work reverts the logic on optimisation problems as it starts from the quantity that interacts with the user of the system and then it predicts the parameters to fulfil that. While only microgeometry changes and bearing preloads are considered in this study, other parameters could be incorporated, according to the requirements.