design with computer

The Problem with Parametric Modelling & Optimisation

Over the last couple of years I have been involved on a number of projects whereby a parametric model has its variables optimised to given, sometimes multi-objective, performance criteria. The key word here is ‘objective’ in that the subjective aspects of the design have already been decided upon (and constrained) by the architect, although they may oversee the optimisation process and make sure they’re still happy with the outcome.

Although I am trained as an architect myself, working for an engineering firm means that by the stage I get a look in on most projects the crucial decisions have already been made. Still, we dive in like good engineers and do our problem solving within our cosy fixed boundaries, publish papers on using genetic algorithms or simulated annealing - but still the designs are actually sub-optimal often resulting in cost overruns and the requirement to do additional work – something the parametric model was initially meant to prevent. Why is this?

Although the structure of the parametric model can be changed by the design team, as the complexity of the model increases the dependencies become difficult to adjust and design freedom actually decreases. As a result, the parametric relationships tend to get locked-in. A similar experience to ours was found by Holzer et al. (2008) with their stadium roof project investigation:

“At one point in the setup of the parametric model schema, changes required by the design team were of such a disruptive nature that the parametric model schema could not cope with them.”

I found many examples where this aspect of parametric modelling occurred - the software itself was determining an inflexible method of working incapable of responding to changes outside of its system. Instead going back to square one, hacking away inefficiently at the parametric topology is the only option available due to time constraints.

“If reformulation is selected (this is common for architectural problems) the analytical stage of the design must be revisited and parameters and constraints adjusted.” Hudson R (2008).

In light of this it is clear that there must be a certain skill required to set up a parametric model that is flexible enough to cope with changes in an unknowable and wicked future. This is a sound approach if you can predict the future, believe solely in cause and effect and have never heard about a certain story with regards a butterfly flapping its wings. But this is real life. Things change during the design process, and not only that; when we start building our models we usually don’t know what the problem is itself. Is parametric modelling software the right tool to be using at all?

Parametric models at present are no different to CAD or indeed BIM in some aspects in that designers usually only deal with one building typology which they assemble in a symbolic graph structure. This graph structure is viewed top-down and has to be comprehensible by a single human brain in order to be tweaked, something that is also done in a linear fashion (explicit history, transactions, etc). Evolutionary solvers tagged onto parametric models (in order to enable cyclic graphs) do not address this inflexibility at the meta-level, that is, the topology of the symbolic graph itself, instead reinforcing the belief that because something has been ‘optimised’, it must be a good design solution – this is a fallacy.

I therefore propose that we look again at what we are doing when we create a parametric model and optimise sliders, and how we are constraining ourselves to building typologies before we have really explored the problem domain itself. Perhaps we need to turn instead at ideas in complexity science for how nature handles environmental complexity during its design process. The use of complex adaptive systems is an attractive one because without a strict discrete structure like a parametric model, bottom up rules give rise to emergent structures with no locked-in ‘explicit history’. Our tools could adapt to changing objectives & subjectives (aesthetics, sociology, iconography, etc) that are still measured globally by tweaking rules at a lower level and letting new buildings emerge without pre-determined typologies – something crucial at the conceptual stages of design. Interestingly, we could potentially go beyond the limitations of top-down graph structures that are comprehensible by the individual and really put our computers to good use in the design of complex buildings that are coupled with a complex environment.

References:

Holzer D., Hough R. & Burry M., 2008, “Parametric Design & Optimisation for Early Design Exploration”, in International Journal of Architectural Computing, 04:05, p.638.

Hudson R., 2008, “Frameworks for Practical Frameworks for Design in Architecture”, in 26th eCAADe Conference Proceedings, p.850.