HYDRA Distributed Multi-Objective Optimization for Designers

Architectural design problems can be quite involved, as there is a plethora of – usually conflicting – criteria that one has to address in order to find an optimal, performative solution. Multi-Objective Optimization (MOO) techniques can thus prove very useful, as they provide solution spaces which can traverse the different trade-offs of convoluted design options. Nevertheless, they are not widely used as (a) they are computationally expensive and (b) the resulting solution space can be proven difficult to visualize and navigate, particularly when dealing with higher dimensional spaces. This paper will present a system, which merges bespoke multi-objective optimization with a parametric CAD system, enhanced by supercomputing, into a single, coherent workflow, in order to address the above issues. The system architecture ensures optimal use of existing compute resources and enables massive performance speed-up, allowing for fast review and delivery cycles. The application aims to provide architects, designers and engineers with a better understanding of the design space, aiding the decision-making process by procuring tangible data from different objectives and finally providing fit (and sometimes unforeseen) solutions to a design problem. This is primarily achieved by a graphical interface of easy to navigate solution spaces of design options, derived from their respective Pareto fronts, in the form of a web-based interactive dashboard. Since understanding high-dimensionality data is a difficult task, multivariate analysis techniques were implemented to post-process the data before displaying it to end users. Visual Data Mining (VDM) and Machine Learning (ML) techniques were incorporated to facilitate knowledge discovery and exploration of large sets of design options at an early design stage. The system is demonstrated and assessed on an applied design case study of a master-planning project, where the benefits of the process are more evident, especially due to its complexity and size.