Building Upon Last Week's Exploration of Generative Design

In our previous session, we delved into the realm of generative design, examining how tools like Generative Design Dynamo can facilitate the exploration of complex design problems through algorithms. We explored a variety of practical examples, from optimizing construction material costs versus operating performance to analyzing building forms for solar exposure. The key takeaway was the inherent power of generative design to efficiently navigate the design space, exploring multiple possibilities and identifying optimal solutions.

Why Integrating Machine Learning with Generative Design

Integrating building performance analysis results in the early stages of conceptual design supports effective design exploration and decision-making. However, such integration requires domain expertise and can be technically challenging, time-consuming, and computationally expensive. In this context, machine learning demonstrates great potential for building seamless workflows that provide faster analysis feedback. Yet, one of the critical challenges in the architecture domain is the lack of diverse datasets needed to train effective prediction models.

This session will provide a brief overview of the potential of machine learning in building analysis and prediction models, and it will demonstrate how we can capitalize on generative design to compensate for the lack of data. We will present a workflow where generative design features in Revit, Dynamo, and Insight software are utilized to generate a synthetic dataset for training a machine learning model focused on cost-analysis prediction.

Through this process, we aim to illustrate not only how machine learning can enhance our design workflows but also how it can be integrated effectively with generative design to overcome data limitations and improve architectural decision-making processes.

Key Learnings

• Learn how to design workflows with Generative Design in Revit and Dynamo for building synthetic data sets to be used in training machine-learning models.
• Discover the diversity of mass model geometry required to represent a comprehensive set of possible building types.
• Learn how to represent your data to be used in training machine-learning models.
• Discover potential uses of machine-learning models toward achieving faster analysis in early conceptual design stages.