Design Space Exploration and Explanation via Conditional Variational Autoencoders in Meta-model-based Conceptual Design of Pedestrian Bridges

TL;DR

This work addresses the challenge of exploring high-dimensional conceptual design spaces in the AEC domain by introducing a conditional variational autoencoder (CVAE) that serves as both forward performance predictor and inverse design generator conditioned on performance requests. The framework uses a two-stage synthetic data pipeline, combining Latin Hypercube Sampling with analytical and finite-element performance simulations to build a dataset of $\mathbf{x} \in \mathcal{D} \subset \mathbb{R}^d$ and $\mathbf{y}=\mathcal{P}(\mathbf{x})$, and trains a CVAE with a latent space of dimension $d_z=2$ to map between designs and performances. A decorrelated loss term and sensitivity analysis enable explainability, while a Revit/Dynamo interface demonstrates an interactive inverse-design workflow, including Pareto visualization of performance objectives. The results show accurate forward predictions and meaningful design sensitivities, illustrating the potential of the approach to act as a co-pilot for conceptual bridge design and beyond, with practical implications for BIM-integrated, data-driven design exploration and rapid iteration.

Abstract

For conceptual design, engineers rely on conventional iterative (often manual) techniques. Emerging parametric models facilitate design space exploration based on quantifiable performance metrics, yet remain time-consuming and computationally expensive. Pure optimisation methods, however, ignore qualitative aspects (e.g. aesthetics or construction methods). This paper provides a performance-driven design exploration framework to augment the human designer through a Conditional Variational Autoencoder (CVAE), which serves as forward performance predictor for given design features as well as an inverse design feature predictor conditioned on a set of performance requests. The CVAE is trained on 18'000 synthetically generated instances of a pedestrian bridge in Switzerland. Sensitivity analysis is employed for explainability and informing designers about (i) relations of the model between features and/or performances and (ii) structural improvements under user-defined objectives. A case study proved our framework's potential to serve as a future co-pilot for conceptual design studies of pedestrian bridges and beyond.

Figures (7)

• Figure 1: Project site, M 1:750, from Woodtli et al. woodtli_svec_2021
• Figure 2: Overview of design features and performances for the bridge design task for synthetic data generation
• Figure 3: Architecture of our CVAE network acting both as surrogate as well as generative model.
• Figure 4: Model performance on four different performance metrics predicted by the encoder. Optimally, all points should lie on the diagonal.
• Figure 5: Sensitivities of the costs w.r.t. the input features (design variables) for: (left) a single user-chosen bridge, (right) a batch of 100 randomly generated bridges around the user-chosen bridge.