Generative Design and How to implement it in Architecture
Case Study: Free-Form Structures based on Iranian-Traditional-Geometries
The spur intro
While we designers are heading to have every element intelligent by data manipulation, there is a significant consideration on how to deal with this evolution through Architecture, Engineering, and Construction (AEC) processes that have been correlated with technology. It has almost been a decade now since “Integrated Design” has become traveling in interdisciplinary projects, and we could go beyond the talk and see the picturesque operations of digital futures from the framework of Artificial Intelligence (AI). AI could let us exploring more profound layers of cognition in paradigms of elements’ behavior (Functional Paradigms) through every complex process gets/sets impact from/to AEC. However, each of these functional paradigms should have been elicited from tons of iterations and data gathering; they could be managed by high qualified algorithms that have been approachable by Machine Learning. There is always a “but” in corroborating such paths when getting out of theory to getting on practice, and it is how designers can take advantage of the related cutting edge technologies mentioned by the terms in this paragraph. This is the whole scenario this article tries to discuss within instances, definitions, and hands-on projects.
Hold on a sec! That was a complicated paragraph filled with technical terms a bit vague to follow, which had also been written comprehensively too. OK! I’m not going to remain the tone from A-Z like that, so we would end up having a complicated text or whatever instead; let’s cook our meal a bit soothed and under a step-by-step recipe.
First and foremost, if you are new to the topics of machine learning and artificial intelligence, you will find tons of fantastic staff in the field from top researchers by just google it, especially on Edx and Coursera, even though I suggest taking a look at Lynda.com where you can find some of the best courses available for grasping the whole concept of AI and ML, taught by Doug Rose which I myself back-packed by. As this article tries to push the limits of understanding new techs for designers, you may also find “Intelligent Architecture towards integration” from my blog as a key to commence. We are not going to discuss the basics of these terms, which takes a lot to know before any step forward but, if you are on the second base of deliberation on how data can flow to the design process already know the context of such technologies, that would probably ensure a higher corresponding to this article. So, by all mentioned above in facilitating your way through, let’s jump in the question we already flagged:
“How designers can take advantage of the cutting edge technologies and where to start?”
Returning such questions would not necessarily have an identified response; therefore, we need to review what we are heading to in design in advance. We will have the subsequent subtitles with a described version of digital design followed by one of the trends in using technology known as Generative design and end it up with a research project titled: Free-Form Structures based on Iranian-Traditional-Geometries. The advanced version of the answer we might consider as a proper response to the main question will be published in further articles where Machine Learning takes on the rest of digital design progressions as evolving technologies.
1. What we are heading to in digital design?
Architecture is an excellent field for interdisciplinary projects no matter what jokes say: “Architects are engineers do not know math”. The more complexity we have in a project’s demands as qualifications, the more we can have integrated design principles’ impressions palpable. In fact, by integrating data from different fields, using engineering and design according to what I mention a lot about functional paradigms, and using these data cumulatively in real-time, we will have a more dynamic ambiance to solve our problems affirmatively. This is what every sustainable idea looks for. The question would be considerably bold here: how can AEC get closer to getting done intrinsically with technology? Any idea?!
Recently we have seen outstanding practices from world leaders in the field from Harvard, MIT, ETH, ITKE as integrated design projects. Hence, I highly recommend you check out how they determine the processes of manipulating data within AEC processes. In 2019, Stanislas Chaillou presented new approaches to analyzing architectural planning by historical layouts of data sets using AI models for stylizing the procedure. Autodesk generative design also defines a new broader space for designers to get to the most advantageous level of their design. Many other firms have started developing techniques and can spot Siemens by Solid Edge, a bunch of well-known grasshopper add-ons like Ameba, Millipede, Monolith, and brand new evolutionary solvers like Wallacei. Design principles using such cutting-edge techs by AI and Machine Learning such as TOPOS (using Cuda by Nvidia to have GPUs for calculations), Owl, Crow, LunchBoxML, DODO, where we use data sets for having more layers of information connected fully or categorized by specific features and weight disciplines towards data-driven design. Nathan Miller and his brilliant team have always been an excessive motivation for me myself, and I invite you to check their awesome blog posts and their achievements at Proving Ground. Design is not the only field where AI and MLs are being used. Engineering is also under the sage of these new strategies. We can find Digital Twin models where sensors and data come together for better solutions over simulations correlated based on reality, monitoring the conditions and optimized decision making. The process is assigned to control and make decisions based on predicting situations, categorizing different outputs within specific features the engineering process defined for. I have to mention remarkable works done by CITA and particular ones by the Innochane project where scholars join professionals as a corroborating bound to bring ideas to the real world from these aspects. There are many projects there demonstrating how implementing data gets to more complex but performative outcomes.
2. Generative design! Or/and what?
Designing powered by AI is going to be more and more fluent in practice, and as computational design enthusiasts, we might get to know how different algorithms work to come up with the idea of having data-driven design by functional paradigms. Such conditional parameters from various factors like structural analysis, energy performance, architectural planning, etc., are implemented proportionately by algorithms towards integrated ideas. Consequently, designing each architecture element, each process in engineering, and each step of construction would be more optimized and reliable to be responsive to complex problems.
Design iterations could be generatively directed from diverse parameters and factors to have results functionally adapted to project demands. The number of outputs we would have from different generations expedites the optimization process while setting a discrepancy in load cases and other meaningful active impact features. The process has been explained, presents the “Generative Design” approach, which is a high qualified trend in engineering design these days. Just consider we have an algorithm using data sets’ paradigms to have features we want more calibrated and adequate in design generations to have a better commensurate between qualifications we want and the computed output.
Accordingly, Generative Design would be simplified in defining a high qualified algorithm mainly powered by AI for a particular functional paradigm within designing an element. Whether the activated factor in the design process is bearing loads and force flows or other features, we as designers might be concerning about like, occupants’ flows in a space according to an office plan or how solar incidents on a specific form would be distributed, it is a data-driven design optimizer according to a particular feature. You can accelerate gaining the knowledge behind it and dive into the Autodesk Generative Design page’s research. But for now, we want to go through a generative design process taking on data manipulation using Python and Grasshopper3D for free-form structural design to experience how data can get into a live link from design to generating 3d models.
3. Free-Form Structures based on Iranian-Traditional-Geometries
One of the most exciting challenges behind this research project’s scene through engineering designs is free from structures with a great potential to have complexity besides performance. Lots of progressions have happened in recent years from world leaders of such beautiful structures. One of the popular impressive teams is Block Research Group, well-known as BRG from ETH Zurich. However, when it comes to complexity, there would be no boundary to qualify design alternatives, but we can develop it by variant geometry and emergent responses as long as we have an idea to reference. This could be set by the geometry selected for every element in 3d format, where we can use the term: “strength through geometry” (By prof. Block). This is the main idea that has been emphasized in this project. We will review the process by demonstrating an image processing scenario to generate models from Iranian Girih Geometries’ images. This is just an initial idea about traditional Iranian traditional master’s deliberations on geometry they had frequently used back then and how it would be correlated with a free-form structure as well. The project is still undergoing, while the next steps would be structural analysis simulation to fulfill how the outputs work beyond aesthetics.
There are many applications in a high range of Girih geometries in certain physical elements of traditional Iranian architecture. They have been used on windows, planning, and bricks arrangement with massive and large span domes, minarets, porticos, and kar-bandi, which devotes to line spans of arches defines the force flows under gigantic structures. The question is how to set our input data as phase one corresponding to Figure 2? Check the image below as an explanatory visual procedure of this issue.
While the input is ready for analysis, the process would be done back and forth until we have a great pile of layers to start the paradigm recognition part using the Machine Learning Algorithm. This part has not been completed yet in this project, but it is under development to have the most data as input we need. At this stage, the project uses RhinoCommon, Anaconda, and GHPythonRemote for loading NumPy, Pillow, in further steps OpenCV and Matplotlib. This is to define an algorithm to read an image pixel by RGB codes and track them with Alpha mode toward a surface subdivision setting. The later steps are dedicated to having these data as mesh points controllable by number sliders to change the meshing parameters and have the process toward a 3d structure on free form surface as structural elements.
4. Conclusion
Consequently, the procedure from gathering data from pixelate images of Girih geometry to imbibe the modeling algorithm based on the paradigm of which pixel has the attribute we define as the feature we want is the explicit recognition we want. This is the exact part; the high qualified algorithm can calculate the images to generate a database. The next step of developing this project is to have horizontal and vertical section planes of structural analysis on each 3d model generated by different Girih geometry, other free-form models, and also variant 3d meshing, which defines the structural elements. There is a lot to consider within various fields, potentially to have a new edge in design, engineering, and fabrication based on such functional technology, and this project is just a straightforward one to comprehend the processes while you are using Machine Learning algorithms. Every machine learning project has its own input, definition, and output procedure, and that is why we need to know how to code to program the algorithms according to project demands. In the article, we did not start to explain inside the algorithms or how they actually work, to perceive the whole concept while there are not many materials from an architectural perspective. We may discuss it further in details about different Machine Learning Algorithms, how they are implemented, what the related terms are, and how we should use contributive tools which make each procedure more convenient and simplified for us as designers.
