In this project, a methodology for encoding geometry was proposed, that provides users a solution to create a geometry without knowledge of physics and how to use specific software for form finding. 

For the case study, numerical parameters which provide information of canopies made by a form finding software were used to create a dataset of 2001 geometries. 

Neuron network was trained by the dataset to output numerical values to replicate catenary curves, and eventually canopy. The network is composed of a sequential model with 4 dense layers. The number of total params is 637. In the end of 1000th epoch, trained model showed converged loss function and 89% accuracy.

The result indicates that predicted values are able to replicate original geometry to some extent, while there are more improvement spaces in the process of encoding and replication. In further research, concepts of Kangaroo and structural metrics could be involved in the process of data encoding and assessment of results.


MLed Kangaroo is a project of IAAC, Institute for Advanced Architecture of Catalonia developed at MaCAD (Masters in Advanced Computation for Architecture & Design) in 2022 by Students: Takeaki Sakakibara and Alexander Lopez. faculty: Gabriella Rossi and assistant faculty: Hesham Shawqy