Active Learning and Conditional Autoencoders

Innovative Approaches for Navigating Nonlinear Design Spaces of Multi-Story Structural Systems

Master thesis

This master thesis develops AI-driven methods for design space exploration of multi-story frame structures, focusing on geometrically and material nonlinear behavior. The research implements active learning strategies and a conditional autoencoder tailored to structural engineering applications. Motivated by the need to efficiently navigate complex design spaces, the study considers nonlinearities, costs, and utilization rates in multi-story structures. By combining active learning for strategic sampling with a conditional autoencoder for dimensionality reduction and generative design, the thesis aims to create an intelligent framework for optimizing structural configurations. The findings are expected to provide insights into AI-driven design exploration in structural engineering, potentially transforming approaches to complex, nonlinear structural problems.

Prerequisites

We are searching for a civil engineering student with experience in non-linear FEM modelling. Some first insights into Machine and Deep Learning (e.g., PyTorch, TensorFlow) or Python programming are a plus. The project will be conducted in close cooperation with Hilti AG in Liechtenstein and is equipped with a student employment opportunity. The preferred language for this thesis project is English.