Lace Wall

Authors | Centre for Information Technology and Architecture | CITA
Project Name | Lace Wall
Objective | Developing a machine learning optimization solution

Lace Wall explores how design integrated simulation of real world behavior of building elements and machine learning allow the design and manufacture of large scale resilient material systems. A minimal inventory of elements is used: 8mm glass-fiber reinforced plastic rods, textile cables and custom designed HDPE elements to join cables and rods together.
The project combines multiple custom-developed tools, overarching from topological exploration, geometrical form-finding to structural optimization assisted by machine learning techniques.

Object’s contribution to the project was the development of a machine learning based optimization method. So far, to find a suitable cable net an evolutionary solver was used, which was searching the solution space for the design meeting all of the buildability/performance criteria. While it is possible optimize a small sample of the wall in that way, applying the same method for the overall assembly is futile – the amount of solutions to evaluate grows exponentially with each parameter introduced into the system.

To overcome this limitation a simple heuristic is applied. First, a single unit which performs the worst is identified, after which it is optimized and reintroduced into the wall with changed cable layout. The challenge in the execution of this method is the lack of a proper metric which could indicate the worst performing unit. Rather than inventing this metric directly, a supervised machine learning method is employed for this task – backpropagation neural network. After training, it is able to indicate the units least similar to the ones which already have a solution stored in the database.

Related publications | Tamke, Martin, Mateusz Zwierzycki, Anders Holden Deleuran, Yuliya Sinke Baranovskaya, Ida Friis Tinning, Mette Ramsgaard Thomsen. 2017. “Lace Wall – Extending Design Intuition through Machine Learning.” In Proceedings of the Fabricate 2017 Conference.

About CITA | CITA is an innovative research environment exploring the intersections between architecture and digital technologies. Identifying core research questions into how space and technology can be probed, CITA investigates how the current forming of a digital culture impacts on architectural thinking and practice.

CITA examines how architecture is influenced by new digital design- and production tools as well as the digital practices that are informing our societies culturally, socially and technologically. Using design and practice based research methods, CITA works through the conceptualisation, design and realisation of working prototypes. CITA is highly collaborative with both industry and practice creating new collaborations with interdisciplinary partners from the fields of computer graphics, human computer interaction, robotics, artificial intelligence as well as the practice based fields of furniture design, fashion and textiles, industrial design, film, dance and interactive arts. (via kadk.dk/CITA)