Magma Works extends capabilities of column design app

February 2021
Our Magma Works team has been working on extending the capabilities of the ACE (Advanced Column Engineering) app. Latest developments focus on batch design and optimisation. Using clustering techniques and evolutionary algorithms, new features enable you to import critical sets of columns, and optimise them using a limited and predefined number of different designs.

After finite element analysis has been processed, columns are imported simultaneously into ACE. Using the clustering techniques — such as k-means or spectral clustering — commonly used in machine learning applications, columns are grouped based on their moments and axial load response. Each cluster is associated with an independent column design. An evolutionary algorithm called simulated annealing is used to optimise each design. The optimisation process covers all usual design checks, including fire design. During the process, the interaction diagram of the optimised column is displayed by the app and evolves to best fit the cluster loads, becoming narrower and narrower around the loads.

The video above shows an example case with 1,160 columns, grouped in six clusters. Optimisation results in a solution with 18% less concrete than the input design. This kind of batch column optimisation should be included in a global iterative process — new column sizes must be updated in the finite element model and new internal forces computed. After a couple of iterations of these two steps, the column clustering should result in the same groups and the optimisation process should converge to a stable output.

The development of this process is part of Whitby Wood’s digital R&D programme. We are looking at further improvements — for example, clustering techniques need to be investigated, compared and improved. Current methods can result in the creation of non-optimal column groups in relation to the next step of design optimisation. The optimisation stage could also be improved. Evolutionary algorithms are efficient for many applications but can suffer from poor time performance. Using more-recent machine learning techniques to get a first-guess design, based on the input loads, could help in reducing optimisation time.

Note … the currently available version of ACE via download does not yet include the above capabilities. Sign up to our newsletter to be kept informed of updates. To download the app, see

To share our work, we have created a new platform — Magma Works on Github. The apps and digital tools create as part of Magma Works help us design better buildings, and they will cover every stage of the design process. All our digital tools are open source. For more information, see

  news index