Project Details
Projekt
Design to Acoustics through Deep Learning
Subject Area
Acoustics
Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
Term
since 2022
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 501927736
Due to the negative environmental impact of high noise levels (health, discomfort), great efforts are being made to reduce sound pressure levels in product and vehicle design. We plan to reduce noise emitted from mechanical structures via deep learning-driven design optimization. The considered mechanical structures and models are aimed at potential real world applicability. We base our approach on the successful method of the first project phase involving a surrogate model that guides the generative model. While we applied grid-based methods in the first project phase that are primarily suitable for 2D design domains, in this project phase we will develop methods that operate on complex 3d geometries and capture interactions with fluids while also enabling the combination with models for other mechanical properties (multi-physics). Furthermore, we will also consider the manufacturability of designs. We start by a problem definition from an engineering point of view inspired by a real application that represents the typical complexity of vibroacoustic systems. Simultaneously we will explore machine learning techniques for processing 3D data of mechanical structures. Later we will develop techniques that facilitate design in practice, including design optimization, active learning and improving sample efficiency to address expensive simulations. To incorporate the acoustic properties of a product in the early design stage process, the design engineers rely on rapid feedback of the acoustic characteristics of the system. Our design assistant is designed further to address this need: It shall enable engineers to change the design and get immediate response of acoustic target quantities. Often during design, acoustics is not the only aspect that needs to be considered. Therefore, we will also investigate techniques on how to combine multiple surrogate models from different modalities, potentially involving other SPP projects. This way, a design engineer can test different designs in an interactive manner and discuss it with customers or experts from other disciplines.
DFG Programme
Priority Programmes
