Generative Design of Mass-Timber Floors

Institut

Lehrstuhl für Akustik mobiler Systeme (TUM-ED)

Typ

Semesterarbeit / Masterarbeit /

Inhalt

theoretisch /  

Beschreibung

Motivation

While timber construction is gaining popularity, developing building components that meet specified acoustic performance standards remains a significant challenge. Wooden floor systems typically consist of multiple layers, whose interactions yield complex sound insulation properties. In particular, impact sound represents a major source of annoyance in residential buildings, motivating the development of accurate and robust prediction methods. At present, reliable assessment often requires costly experimental validation. To address this limitation, a data driven prediction tool is being developed that employs machine learning techniques to estimate the sound insulation characteristics of wooden floor systems based on laboratory measurement data.

Your Task

A comprehensive set of measurement data has been collected and integrated into a MySQL database. Using this dataset, feed-forward neural networks and tree-ensemble models were trained in an initial comparative study to predict airborne sound insulation. The next step inverts that prediction process to create a design aid that proposes floor assemblies that meet specified acoustic targets. In this thesis, you will train a variational autoencoder (VAE) to learn compact latent representations from the compiled dataset. A predictor trained jointly with the VAE will map points in latent space to sound-insulation spectra. Candidate assemblies will be found by optimizing in latent space toward desired acoustic performance; the decoder will then reconstruct those candidates into full floor con f igurations. The objective is to produce physically plausible, acoustically appropriate floor-assembly designs.
Your contribution will play a crucial role in refining the existing predictive models and enabling a reliable assessment of the sound insulation of mass-timber floor constructions. By joining our team, you have the opportunity to apply your theoretical knowledge in a real-world context, gaining valuable insights into the development and deployment of machine learning solutions.

Voraussetzungen

• Knowledge of Python/PyTorch
• High motivation and a structured working style
• Experience in probabilistic modeling and hyperparameter optimization

Application

Please submit your application, including a brief statement of interest, your CV, and a transcript of records via email.

Möglicher Beginn

jetzt

Kontakt

Markus Mörwald, M.Sc.
Raum: MW1535
Tel.: 08928955136
markus.moerwaldtum.de

Ausschreibung