2015年9月4日 星期五

K9K | Holistic Architective Agency

Motivation

The theories of architecture, building technology and the simulation of the building systems are always attractive to me. I imagine that, in the future, I can establish an R&D lab, K9K, in which I can push the theories and building technology forward, meanwhile the theories can be implemented in some real projects.

K9K basically is working on two fields of study, design theory and fabrication. The research of design theories is to establish the mathematical models of architecture design, digital fabrication, interactive architecture, etc. Fabrication research is focusing on discovering new architectural materials, studying automatic fabrication, developing building technologies, etc. One field is in virtual world and the other is in a physical world. Although the goal of K9K is to connect these two world, I still think we should approach this goal in two different ways.

Multidisciplinary Research

Because of my EE (electrical engineering) and architectural background, I am looking for the opportunity to combine these two alternative fields of study I've involved in. I believe that some technologies from EE can be used in architectural field, for instance, machine learning, digital communication, recognition technology and digital IC design. Digital IC design is one of my professional skills, especially in digital communication systems. Hence, I keep trying to bring some concepts and methodologies into architecture design.

Not only electrical engineering, but also other fields of study are welcome in K9K. I believe that we should re-think the design in a more holistic way. Hence, multidisciplinary study will be the principle in K9K. Everything is relative to one another, and all the theories may be the elixirs in different fields. By integrating different fields of research, we may find more creative solutions for architects.
  
Business Mode

The aim of K9K is to keep generating pioneering technologies and accumulating the knowledge for architects. The income of this lab may be from exporting technologies and transferring patents. In fact, it works like most consultant companies, K9K will cooperate with architects, contractors or manufacturers. Also, K9K can be an educational institute for those students who are just graduated from school. K9K can also be viewed as a post-research institute and help students complete the first practice work in which they still can test some new ideas. Furthermore, K9K offers not only the consultant services but also the advanced products. Hence, the short term goal of K9K is to be a professional consultant institute and solution provider for architects. For the goal of long term, K9K will be able to produce some advanced products and patents, which may be the main financial source of K9K in the future.

Structure of K9K, T.C. Kurt Hong, Summer 2015
Archive

As a research institute, accumulating knowledge is a critical work for K9K. K9K will keep generating technologies, meanwhile, K9K will keep documenting all the experiments including the process and the result as the references for architects and educators. The research data should be archived and well-organized for the next generation. K9K attempts to be the library which keeps updating knowledge and a platform for architecture professionals.

RESEARCH | Mathematical Models

Mathematical model for digital fabrication

While I was making the models with the machines and the digital tools, an idea occurred to me that some concepts from digital communication could be used in the theories of digital fabrication. For example, the behavior of the digital fabrication may be described as a equation as below:

Y = HX + N ;

In this equation, "X" is the digital model and "Y" is the physical result, "H" is the feature of the machines or the tools, "N" means the noise while manufacturing. Actually, it is a fundamental representation of digital communication. I think this equation is suitable for digital fabrication because the whole process is similar to communication. The digital model can be viewed as the message we want to transmit, and the physical result is the message we actually receive, the manufacturing process is like the channel which has some effect on the transmitted message, and, of course, we have some inaccuracy which is like the noise. Assume that we use this model to describe the process of fabrication, we can observe that the H and the N are not welcome in fabrication since they cause the distortions. Therefore, we may need some methods to remove them. Once we can remove the effects and the noise from the manufacturing, we may connect virtual world and physical world seamlessly.

However, most architects work without the mathematical models and hence they can hardly analyze their design process systematically. If there is a systematic methodology which can analyze the process of design for architects, they can check where the errors are and avoid them in advance. The equation above may not be the best way to describe the process, but we can keep looking for other mathematical models for architectural design. Furthermore, I proposed a method to analyze the systems in architecture in my graduation thesis[1]. Although it is still primitive, I believe that we can use this method to establish the mathematical models.

Equalizer  in Digital Fabrication

In the previous section, the equation "Y=HX+N" is proposed as the basic mathematical model for digital fabrication. This equation will be pushed further to eliminate the gap between digital model and physical result in this section.

Assume that we use Rhino (or any other software) to draw a solid cube on the computer, and we use a CNC router drilling the foam to implement it. However, the physical cube might be a little different from the cube in the computer, including the tool marks, the texture of the material or other distortions. The entire process can be viewed as a type of communication, the virtual model (the cube on Rhino) is the message we want to transmit, the physical model (the foam cube) is the message received, and the CNC router is the channel which the message pass through. In fact, the difference between virtual model and physical model is similar to the channel effect in digital communication. Typically, the engineers would use some methods to eliminate the effect to make the original signal (virtual model) approximate to the received signal (physical model).

To eliminate the distortion, we can feed the information of the distortion back to the digital design tools, and make prevention in advance. This method is called equalization which is similar to the prestressed design in civil engineering, and actually we may find many cases in architectural and industrial design. For example, a famous Japanese architect, Jun'ya Ishigami, designed a table with this concept. In this design, he fed the gravity factors back to the virtual model to obtain a prestressed structure.

Table, Junya Ishigami, 2005-2006

The mathematical model of equalization can be described as below:

Y = HX +N ;
E is defined as Y/X ;
E = H + N/X ;
H = E - N/X ;
H approximates to E ; ( Ignore N if N is small enough or X is large enough )
The goal is to find X' to achieve Y=X;
X = HX' + N ;
X' = (X-N)/H = X/H -N/H ;
X' = X/E - N/E ; (Ignore N if N is small enough )
X' = X/E ; (Equalized Models)
Y' = H(X/E) + N ;
Y' = H(X/(H+N/X)) +N ;
Y' = HX/(H+N/X) + N ;
Y' = HX(H-N/X)/(H^2 - (N/X)^2) + N ;
Y' = H^2X -HN/ H^2 +N ; ( (N/X)^2 is small enough )
Y' = X- N/H + N ;

Y' = X + N ; (N/H +N can be viewed as N )
The error = N ; The manufacturing effect is equalized ;

According to the equation above, the fabrication process can be equalized to obtain a result approximating to the original design. The error left is the noise during manufacturing, which can be reduced by other methods.

In Sean Ahlquist's lecture, we had a conversation about the simulation and physical models. It is hard to analyze the behavior of fabrics because the forces from fabrics are so complicated. However, we may use Kinect to capture the deformation and establish a new digital model. Then, we may use this new digital model to obtain the equalization factor (E), and we can reduce the distortion in the next work. We don't need to analyze the force from each fabric, all we need is the factor, which can be calculated after the experiment. Furthermore, if we take advantage of machine learning mechanism, the digital tools may be trained well after several feedback iterations. Equalization is just one of the possible theories in digital communication which can be adopted in digital fabrication, I believe that more theories can be developed for digital fabrication.

Architecture Description Language (ADL)

Comparing to other fileds of study, architectural design is basically on modeling-based tools such as Rhino, Revit, AutoCad and SketchUp. However, we may just describe the functions, the systems or the spacial relationships, in other words, the logic of the design, and let computer generate the forms for us.

In IC design industry, engineers adopted scripting-based design in 1984, even earlier [2]. In the beginning, IC designers drew the electrical layouts by hands, and they tried to arrange those transistors and wires to obtain optimized results, for example, smallest area or highest speed. But, quickly they gave up the hand drawing and turned the process into scripts. The engineers use computers to generate the optimized results, all they need to do is to describe the behaviors (logic) of the circuits. Hence, the IC design industry emerged in a extremely pace, because the designers have more time and energy to design the circuits in a higher level, and therefore they can create high complex chips (imagine that if the engineers have to draw 1.4 billion transistors in Intel CPU by hands). It turns out, human beings started to create the product with intelligence such as the computers, laptops, tablets or smart phones.

Although the architects nowadays adopt some scripts to design such as Grasshopper, Processing or CityEngine, most architects still use modeling-based tools, some even draw by hands. In other words, we keep using the limited calculation power (human brains) to optimize the entire architecture. Once we use the description language to design a building, one of the advantages is "Design Rule Check (DRC)". While designing a building, there are many rules and limitations, thus the architect may need an additional person to check all the rules for architect. Sometimes some rules may be ignored and hence it may cause architect and client extra time and cost to fix it. DRC helps architect check all the rules in advance, and architects can adopt different libraries according to different country or environments. This concept is from IC design, this tool is important because the manufacturing cost of IC is extremely high, engineers have to exclude all the possible bugs before tape-out. The feature, high manufacturing cost, is similar to architecture, thus I think we may introduce this idea to architecture.

The second advantage of ADL is that architects can use computer to optimize the design. Once the architects can script the logic of the architecture, the computer can generate all the possible models. And, architects can input different constrains such as minimum area, minimum budget or maximum capacity to obtain alternative optimized results. It is like the parametric design, the difference is that architects describe the logic rather than the forms while designing. Moreover, this language may be the foundation of automatic design, which will be elaborated in the next section.

The structure of ADL can be described as below:

Flow Chart of ADL, T.C. Kurt Hong, Summer 2015

RESEARCH | Machine Learning

One of the concepts that I am eager to apply to architectural design is "Machine Learning". Also, machine learning may be applied on both software development and physical fabrication. Moreover, I think machine learning may be the key of the building technology evolution.

Pixelwise Context-aware Semantic Segmentation

Meta-Design

In physical aspect, robotic fabrication is the focus in building technology. However, the robots can be more intelligent and thus help architects deal with the issue about construction. For example, the automatic vacuum robot, iRobot, is equipped with some simple algorithm to learn how to optimize the path while it is cleaning a house. The robot can collect the data from environment and keep calculating the best solution in its brain. Hence, if we equip a mechanism on robots which allow robots to collect data through fabrication every time, the robots can operate more efficiently and gradually learn how to fabricate by itself.

However, automatic fabrication is the first level application of machine learning. In software aspect, machine learning can be the core of the smart software for architects, furthermore, it may also be the kernel of the automatic design. In the future, the robots may design architectures just like what architects do now. Not only doing optimization, evaluation, analysis for architects, but also creating architectures. That means architects have more energy and time to design in a higher level and lead architecture to the next phase.

"A city is not a tree" [3], a famous article of Christopher Alexander, in which a concept is depicted that there are enormous systems in our circumstances. Architects may easily ignore the systems hidden in the environment and make some terrible decisions. There are too many systems to analyze, therefore, architects can hardly optimize the design without a powerful analysis tool. Hence, the parametric design showed up in the recent years. Architects tried to convert the modeling design to script design in order to obtain more information from environment to optimize the design. With scripting, architects can try a spectrum of parameters and simulate the result in virtual environment. However, the parametric design might be upgraded to a higher level, automatic design. Automatic design can collect all the coherent information and calculate a optimized result for client. The design can be generated by an AI system, especially the design is an optimized design which is considering all the systems and factors. Or, the AI system can calculate all possible designs and pick up the best one for architects.

For instance, Google and Amazon are trying to build up a system which can analyze all the data collected from the world. Amazon is working on the the prediction of the shopping behavior of human beings, and Google is working on IoT technology which can connect all the objects in a house and make the house smarter. Furthermore, Google is also working on big data analysis, which is the kernel of development of the AI system. In fact, the fundamental technology of those applications such as AI, IoT, big data and prediction of human beings, is machine learning. Therefore, I believe that the architecture design will get a great leap by adopting this technology. On the other hand, in the future, architects' job is to design the algorithm of automatic progress, and that is the second level of design.

System Description, Tzu-Chieh Hong, 2014

Intelligent & Interactive Design

In my third year in architectural college, our assignment is to design an interactive installation in the studio. From that time, I keep researching the interface between human and architecture. Typically, we treat architecture as a container and human being is the content inside. Or, we treat the architecture as a sculpture and human being is the viewer outside. It is arbitrary that I use only two relationships (actually many architects have designed a lot of architectures which are aimed to blur the boundary between inside and outside). However, most of the relationships between architectures and human beings are single directional. The relationships are static and monotonous. Recently, architects are trying to endow architecture with more ability to response human beings. Thus, the interactive design turns into a popular topic in architecture study.

Boundary Functions, Scott Snibbe, 1999

The first level of engagement: Respond like a mathematical function.

The interactive architecture nowadays contains a simple algorithm, it responses back to people according the inputs that people give to it. In fact, it is like a vendor machine or an automatic door, it responses people in a simple way. There are a lot of applications of this kind of interaction. 

The second level of engagement: Respond like a Siri.

In the second level of engagement, the interactive architecture can collect data from users and adjust the response modes. According to the behavior of the subject, the architecture can learn from the subject and train itself to create a dialogue with the subject. Meanwhile, the architecture can connect to the internet and communicate with other objects to gather more information to build up the specific response mode. 

The third level of engagement: Respond like a person.

The next phase of interactive design is to create a brain for an architecture. The brain can not only think but also create something. In the book, "What Technology Wants" by Kevin Kelly, the author depicts that everything will have a small brain, which is a microchip, and those brains are connected to one another as a big network just like the concept of IoT [4]. Therefore, the network is like neuron network in human's brain, it can think, even create. In the future, the architecture will be equipped with this ability to think like a person, even smarter, and it can create some wonderful things for human beings. For example, a computer in an old movie, Electric Dreams (1984) [5], composed a song for the girl it loved. The music is beautiful and full of emotion. From this perspective, the ultimate goal of interactive design is to create a "human-like" machine which has the ability to create. However, I think the ultimate goal is not to create a human-like machine but create an AI system better than human. The machine can think more holistically than human beings do, it won't ignore any information and simplify the problem.

RESEARCH | Fabrication

Material Research

Discovering new architecture materials is also a critical research topic in K9K. Ten decades ago, architects adopt new materials such as glass, concrete and steel as the new material in architecture. After a century, we are still using glass, concrete and steel as the main materials in architecture. In automobile industry, plastic or carbon fabric is very common. Apple adopt sapphire crystal as the new material of the screen of iPhone, now the sapphire crystal is very common in smart phone design. Before that, Apple used uni-aluminum piece and CNC router to manufacture MacBook. Television industry replaced CRT with LCD to reduce the size, recently the television industries adopted OLED and ePaper to obtain a banding screen and save more energy.

New material can always push an industry forward. Even though the architecture industry is too complex to catch up the pace of other industries, we can still try some advanced material in architecture. For example, the ePaper may be used as the wall paper in a house to display the information for people. Transparent solar panel may be the substitute of glass, it can be a good insulation material and generate power for people. Furthermore, the new material should be energy productive, intelligence embedded, and environmental friendly. One of the jobs in K9K is working on the research of new material and make prototypes for architects. Architects may purchase the simulation results from K9K before they spend lots of money on an unknown material. K9K can cooperate with material suppliers as an institute to test and verify the product from architect's perspective.

Energy Productive Brick, Tzu-Chieh Hong, 2014

Automatic Fabrication

Besides the limited materials, today we still use a lot of human power to build a house, even though we use pre-fab system. Why can't we just let house grow by itself? I imagine that we can place some materials around the site and let several robots build a house automatically. Those robots can work like human workers, they can detect the weather and communicate with each other. The building work load is distributed to each robots automatically, and the working schedule can be optimized. Furthermore, the robots can adjust the schedule and fabrication process according to the unexpected factors they detect.

Robotic Fabrication, University of Michigan

Skylar Tibbits, a professor in MIT, has a lab which is focusing on self-assembly systems in architecture. His research is to develop the objects which can aggregate themselves to form a building (or a structure). However, we may approach this goal in a different way. Instead of producing lots of special components, we can let robots to build our house with normal materials. Nevertheless, the robots today just follow a fixed plan to operate. Actually they can't think and response to un-expected factors. What if we equipped those robots with the ability of machine learning? They may be more efficient, smart and robust. Once the robots can train themselves and train other machines to build a house, we may make building technology a great leap. And the key of this evolution is machine learning.

Furthermore those robots can arrange the whole schedule of construction by themselves according to budget, supply chain, design complexity, etc. They will connect to the entire supply chain and estimate the delivery time of each material, and thus optimize the storage and the time.

Conclusion

The multidisciplinary research is always what I am interested in, it requires the cooperation among various people with various backgrounds. K9K is an institute which offer a platform and resources to connect them together and implement the ideas. The machine learning, AI system or the technologies mentioned in previous sections are just a small part of the possible technologies which can be applied to architecture design, there are still a lot of possibilities for K9K. Therefore, keeping introducing other knowledge into architecture design is the main task of K9K. D-School in Stanford University, ISD in University of Michigan and TDIS in National Chiao Tung University are similar models for K9K, even though their multidisciplinary research is based on educational purpose. To sum up, even though it is hard to run a firm like K9K, I will still be dedicated to multidisciplinary research.

Reference

Footnotes
[1] Sharing Components - Multi-Systems & Uni-Solution, Tzu-Chieh Hong, September 2014.
[2] Verilog, the description language for IC design, 1984
[3] A City Is Not A Tree, Christopher Alexander, 1965

[4] IoT, Internet of Things:
IoT is another developing concept in architecture. All the objects can communicate with each other, ,and thus the human behaviors can be predicted. Furthermore, the entire house can operate according to the people living inside. Also, the house can detect the circumstances to adjust the energy output to achieve a more efficient mode. However, the IoT technology requires many skills from computer science and electrical engineering, the architects can hardly realize the theory behind the operations. To eliminate the gap, K9K can be the bridge and develop the IoT kit for architects and simplify the interface. 

[5] Electric Dreams, Virgin Films, 1984

Bibliography
1. Roger L. Martin, The Design of Business: Why Design Thinking is the Next Competitive Advantage (Harvard Business Review Press; Third Edition edition, October 13, 2009).
2. Kevin Kelly, What Technology Wants (Penguin Books, September 27, 2011).
3. Marty Neumeier, Metaskills: Five Talents for Robotic Age (New Riders; 1 edition, December 1, 2012).
4.Tim Ingold, Making: Anthropology, Archaeology, Art and Architecture (Routledge; 1 edition, May 5, 2013).
5. Christopher Alexander, A City is not a Tree (London: Design, 1966).
6. Christopher Alexander, Notes On The Synthesis of Form (Massachusetts: Harvard University Press, 1964), 28-54.
7. Aldo Rossi, The Architecture of the City (Massachusetts: The MIT Press, 1984), 15-28.
8. Gary T. Moore, Emerging Methods in Environmental Design and Planing (Massachusetts: The MIT Press, 1968), 21-37
9. John Habraken et al., Housing of the Millions (Belgium: NAi Publishers, 2000), 213-284.
10. Foreign Office Architects, Phylogenesis (Barcelona: Actar, 2003), 30-35.
11. Reiser + Umemoto, Atlas of Novel Tectonics (New York: Princeton Architectural Press, 2005), 196-205.
12. Scott Marble, Digital Workflows in Architecture, Design-Assembly-Industry (Basel: Birkhauser, 2012).
13. Stephen Friedberg, Linear Algebra, 4th Edition (Chicago: Pearson, 2002).
14. Shu, Lin, William Ryan, Error Control Coding (New Jersey: Prentice Hall, 2004).

沒有留言:

張貼留言