Visual Analytics for Generative Design Exploration: An interactive 3D data environment for a computational design system facilitating the performance-driven design process of a nearly Zero-Energy sports hall
Year 2018Â
Author(s)     Jamal van KastelÂ
ML Tags
Decision Trees
Clustering
Self-Organizing Maps
Topic Tags
Generative Design
Analytic Hierarchal Process (AHP)
> Software & Plug-ins UsedÂ
Rhinoceros and Grasshopper for generating building models which will be used for the performance simulationsÂ
Grasshopper plug-ins for building performance simulations: Â
DIVA, Radiance, DAYSIM for illuminance Â
Ladybug for PV energy valuesÂ
Honeybee, EnergyPlus, OpenStudio for operative temperaturesÂ
Octopus GH plug-in for multi-objective optimization with an evolutionary algorithmÂ
ModeFRONTIER, a stand-alone design optimization software, for clustering data using a SOMÂ
Microsoft Excel & Unreal Engine for processing building information Â
Unreal Engine for visual analyticsÂ
> Workflow
Thesis Report Figure 2.2: Scheme depicting the interconnectedness of software used in this thesis
Final presentation figure: Workflow of the Computational Design System. Feedback loops related to the user (e.g. changes in the design concept because of findings in the visual analytics tool) are not shown in this workflow
> Summary
Image by Jamal van Kastel
This thesis' main contribution is a visual analytics tool that allows for the visualization of data and design geometries in a highly dynamic, game-like 3D data environment, allowing for the exploration of both quantified and non-quantified performances of a collection of design options.Â
This thesis' Computational Design System aids in the design of a nearly Zero-Energy sports hall for a sports facility in Overhoeks, Amsterdam.Â
The Computational Design System (CDS) consists of two different parts of a workflow, each addressing specific steps. The first half of the workflow is used to generate a dataset of design options. The latter half is used to evaluate this dataset. Â
The Iterative Design System is used as follows:Â
Design alternatives are generated using parametric models developed in Rhino & Grasshopper. Multiple Grasshopper scripts are used to generate design options that encompass multiple design concepts.  LIMITATIONS:  generating random geometry does not necessarily result in a single optimal solution therefore optimizations are completedÂ
Evolutionary solver Octopus is used to automatically generate design options, iteratively converging towards optimized design solutions.Â
Then the Data Analytics system is used as follows:Â
The data is processed and clustered to aid the designer in decision making. Machine learning is used for some of the techniques. The data in processed in multiple different ways to provide different levels of information:Â
Self Organizing Maps are used to reduce the dimensionality of the data, creating a two-dimensional representation of the high-dimensional design space. The Self-Organizing Map creates ‘soft clusters’ of similar design options and augments the dataset with approximations for non-simulated areas of the design space.  LIMITATIONS: not easy-to-read overview of the cells’ performancesÂ
A decision tree is used to cluster the data based on the user-defined criteria Â
Hierarchical clustering is used to cluster the data based on the user-defined criteriaÂ
Annual Performance Data is visualized using a method best described as a combination of a stacked bar graph and an analytic hierarchal process, these are methods used to differentiate between non-empirical values of optionsÂ
Pictograms are used to visualize the designs with best or worst performances, this is to aid the designer in quickly identifying these characteristicsÂ
A correlation matrix can be used to further indicate interrelationships between design aspects Â
The above noted different methods of clustering and visualizing the data are combined in the Unreal Engine to create a highly interactive environment for visual analytics. The visual analytics tool positions all buildings in a ‘data landscape’, a visual representation of the various data analytics techniques. Features of the landscape give insight in the buildings’ performances. By integrating multiple data analytics methods and facilitating access through game-like means of interactivity, users can intuitively explore the design space to make informed design decisions.Â
> Additional Visuals
Thesis Report Figure 2.1: Classification of subsystems in the Computational Design System
Thesis Report Figure 9.2: Current feedback loops (t.) and recommendations for additional feedback loops (b.)
Image by Jamal van Kastel: Canyons are used as a visual metaphorto visualize the SOM U-matrix
Image by Jamal van Kastel: Users may control a character to walk through the landscape
Image by Jamal van Kastel: Menu showing annual performance values. Tile # 145 with dsign alternative 222271 is selected. For the cells that are hovered over (currently cell # 583), relative performances to that cell are calculated
Image by Jamal van Kastel: Menu showing 6 design alternatives bookmarked by the user. Relative performances compared to the tile the user hovers over (currently tile # 607) are automatically calculated
Additional information provided by Jamal van Kastel:
> Possible Applications
For ideas on how to implement some of the above mentioned techniques, please see Â
‘Possible applications for students to try with SOM’Â
 ‘Possible applications for students to try with Clustering’Â
 ‘Possible applications for students to try with Surrogate Models’Â