Programmable Architecture
-Towards Human Interactive, Cybernetic Architecture-

Kensuke Hotta (B.Eng, M.Eng, Msc),
Architectural Association School of Architecture

Chapter 5 Data and Analytical Methods

5-1 . Evaluating Performance in ‘Intelligent Systems’

While chapter 4 elaborates the proposal of PA architecture with its physical and virtual models as well as multiple hierarchies, chapter 5 addresses the issue of how to quantify this new kind of relationship with the environment focusing particularly on light levels.

The proposed approach involves first, performing some baseline case studies. These are correlated using a software interface which ‘metabolizes’ the results. This ability will lead to a diversity of applications executable from a small architecture scale up to larger urban patches with multiple intelligent units. After the software is ‘trained’, it should be evaluated in a functional environment since it may not always know how to respond usefully to its complex real world environment. Following this PA will be evaluated to see how it can work effectively for human society and the environment.

If architecture is truly ‘programmable’, its metabolic interface can be adjusted based on the results of the ongoing evaluative process. Moreover, as the software is ‘intelligent’, it will have a self-developed system such as a neutral network. When the system works well, it should be able to record its history of ‘training’ and ‘learning’ and prove that it is possible to create adaptive software-metabolisms without any change to the hardware.

第5章 データおよび分析方法

5-1 . 知能システムの性能評価




5-2 . Details of Methodology in Previous Experiments

Prior to this thesis, the author proposed the concept of “Programmable-Kinetic-Fabric for Architecture'' at the Bartlett in the UCL as a M.Sc. master’s degree thesis, in which several methods were utilized to control architectural machines. Three types of architecture were identified within an experiment that focused on active light shading in order to develop kinetic architecture. In addition a number of tools were used in the experiment, in particular Arduino (hardware) and the Genetic Algorithm (software). This case study related to both hardware and software systems. (fig 5-2,1)

Two key challenges (potentials) came out of this project. The first challenge (potential) was the hardware; architectural hardware should be capable of receiving a signal, and acting accordingly. If we look at cars or any other intelligent object, we can easily find that some systems are allowed to change, not only morphologically but also functionally. Another challenge (potential) was the software; there is no single software system that can control all aspects of an architecture system. The exceptions to this rule would be an environmental or energy management system, but this is only a part of complex architectural design.

Fig 5-2,1: Illumination Experiment, on the thesis at UCL, 2008 (by Author and A.Hotta)

5-3 . Initial Physical Experiments

5-3-1. A Building Envelop Experiment

During the 2010~2011 period, several experiments were performed. The first experiment involved a comparison of the amount of sunlight within different building envelopes containing the same footprint that were controlled by a ‘Selfish System’ or a ‘Concession System’. The Selfish System’s control principle is that each envelope object tries to get maximum sunlight without considering others. In contrast the Concession System‘s control principle is that each envelope is allowed to work as part of a group cooperating for group efficiency.

Fig 5-3-1,1: Building Envelope Experiment: Setup (by Author)

The Rope on the diagram shows the outline of the building, here called the ‘envelope’. The footprint is fixed at the bottom point with pins. Within this limitation the rope (envelope) can move and change its shape freely. The pin in the middle of the rope represents the light detector. The detector’s efficiency depends on the angle towards the light source (Sun). The light calculation is a function (cosine) of the angle from the perpendicular of the sun’s rays at that point along the rope.

The experimental esults indicated that the ‘Concession System‘ got a higher score (1190 points) than the ‘Selfish System’ (1003 points)(fig 5-3-1,3)(fig 5-3-1,4). This means, if an urban patch [note: is ‘urban patch’ a technical term or your term? If it's your term you should define it or use a different word like “neighbourhood” or “area”. ] operates using the Concession System they can achieve higher efficiency. Based on these results a number of issues were identified for further development. Firstly the precise logic of the ‘concession’ used to achieve a higher score needs to be precisely defined; secondly the issue of a central or distributed controller system needs to be addressed and finally the location and nature of the evaluator needs to be addressed (the score was calculated by hand in the experiment).

Fig 5-3-1,2: Building Envelope Experiment: Initial Explanation (K.Hotta)

There are 3 envelopes in this field (patch). The light source comes from the upper right position making shadows on the opposite side. The way of calculating illumination at the various points is the same as the previous diagram. The actual functions are shown below.

Amount of Sunlight building1

= Σ {(S1-1)+(S1-2)+(S1-3)}

= Σ {(x*cos0°)+(x*cos(-50°))+(x*cos(-89)°)}

Amount of Sunlight building2

= Σ {(S2-1)+(S2-2)+....(S2-13)}

= Σ {(x*cos62 °)+(x*cos53 °)+(x*cos60 °)+(x*cos53 °)+(x*cos34 °)+(x*cos11 °)+(x*cos(-48

°))+(x*cos(-86 °)+(x*cos(-119°)}

Amount of Sunlight building3

= Σ {(S3-1)+(S3-2).....+(S3-8)}

= Σ {(x*cos0°)+(x*cos(-50°))+(x*cos(-89)°)}

The total score of this field (Urban patch) is simply the sum of these 3 scores.

Fig 5-3-1,3: Building Envelope Experiment : Selfish Envelopes

The total score was 1003 points (Building1: 250points; Building 2: 700 points; Building3: 53 points)

Fig 5-3-1,4: Building Envelope Experiment: Concession Envelopes

The total score was 1190 points (Building1: 200points; Building 2: 680 points; Building3: 310 points)

For these questions, it is worth referring to animal physiology especially the reflex system. (fig 5-3-3) Humans have two different reflexes - spinal reflexes which are autonomic, and the cerebrum based reflexes which are a more intelligent and voluntary reflex from within the brain. Both have their merits, the first providing speedy reactions, the second providing ‘smart’ reactions. Animal sensing systems use a combination of local reflexes and central reflexes to achieve dynamic reactions. How might this work in a machine system?

Fig 5-3-1,5: Hybrid Reflex System drawn by Author, Referring from Dr.Joseph Mcnairm,MDC

5-3-2. Kinetic Robot Experiments

The second experiment which was carried out used a different reflex system based on the hypothesis that the combination of different reflexes would result in more light in the experimental space and thus achieve higher scores. The objective of this experiment was the same as the above where building envelopes change their shape to maximize the amount of sunlight using real time sensing. There were two models used, each having different control principles. One is a centrally controlled system where the central reflex is a coded reaction using electrical signals (fig 5-3-4). The other is a distributed system, where the distributed-local reflex is material-based using Bio-metal(R), which is a type of shape memory alloy (fig 5-3-5). These robotic structures were operational.

Fig 5-3-1,4: Building Envelope Experiment: Concession Envelopes

This is a screenshot from the video of the experiment. The machine was controlled using Arduino, and a light responsive envelope. The machine is made of a flexible belt with 3 small servos and sensors attached. These servos are connected to the belt with threads, so if a sensor detects light the whole envelope will change its shape in real time. The control system is a 'centered' type consisting of a component made up of a servo and a sensor.

Fig 5-3-2, 2 : Tensegrity Arch (Material based responsive system)

This is a screenshot from the video of the experiment. The model is made with bamboo lodes (brown), elastic strings (red), and shape memory alloy (Thin black wire). When warmed up with a dryer, the whole structure starts changing its shape instantly. This system is a material based responsive system so there is no central brain (processor).

5-4 . What is Going to Be Examined

The word ‘Programmable’ within this thesis refers to the quality of optimising a person’s activity within a specific environmental condition. In this experiment, illumination levels were used to evaluate the architecture, in particular the roof’s adaptability to varying levels of illumination. This particular value was chosen as optimum illumination levels have already been established in many environmental institutes, such as IES; Illuminating Engineering Society of North America, or MS1525; Malaysian standard, or Panduan Teknik JKR or JISZ9110; Japan industrial Standard. Here, IES’s standard illumination level is used for the setup. Since there is no established index to show the level of adaptability of a structure to environmental changes, the author attempted to develop a process for measuring ‘correspondence to required environment’, which would identify the extent to which an architectural machine can follow its target or ‘objective’ function when dealing with the environmental change around the machine (fig5-4, 1).

fig 5-4,1: Experiment’s Diagram

5-5 . Examine, Evaluate and Compare a Fixed and Kinetic Roof

The diagram below (fig.5-5,1) (fig.5-5,2) compares the performance of a fixed roof and a kinetic roof in terms of illumination performance. The two columns on the left show the relationships between an activity, a room or space and its required illumination level based on the IES-The Illuminating Engineering Society of North America-which is the recognized technical authority on illumination. The red line shows the available environmental (solar) illumination over 24 hours referred to as the Sun Light, SL(t), while the black line shows the necessary levels of light within the structure over 24 hours referred to here as the Objective Function, OF(t). On top of this the performance of the Fixed Roof, FR(t), and the Kinetic Roof, KR(t) are overlaid. What can be seen here is that the kinetic roof closely matches the objective function over the 24 hour period and is therefore more efficient and thus is able to encourage specific activities with more accurate amounts of light, than the fixed roof.

fig 5-5, 1: The relation between Room illumination level and architectural functions (K.Hotta)

The graph indicates the correspondence between illumination level and Activity, and also Activity and Architectural functions. The data is referred from IES ; the Illuminating Engineering Society of North America. (

fig 5-5, 2: The Estimated relation between required room illumination level and time-based architectural performance (K.Hotta).

The graph indicates the estimated result that corresponds between required illumination level for certain architectural functions and architectural performance as illumination level. The black line shows required illumination level (=objective function), Blue line shows estimated illumination level (architectural performance) under fixed roof, lastly Green line shows as well but under kinetic roof. The coloured hatches (S0,S1,S2) show the difference between objective function and estimated performance on each, it indicates the performance of the each roof. Smaller is better in here.

Here sunlight refers to general illuminations. These values are set on the assumption that the place is Japan in the summer and the weather is fine. The maximum illumination level might be 100,000 (lx) at peak time, which is generally around noon.

SL(t); Sun Light


fig 5 5-3: Sun Illumination (K.Hotta)

fig 5 5-4: Sun Illumination (K.Hotta)

fig 5-5-5: Logarithmic calculations (By Auther)

The resulting calculations yielded an S1 value of 46 and an S2 of 30. These values were calculated using a logarithm based 1.01. The logarithm calculation is useful when the values are not easily comparable and where the area (s) potentially has no limit with the possibility of the value diverging to infinity.

• Defined Indicator

Using the logarithmic calculations the correspondence to the required levels of illumination of the Objective Function is measured using new indicators (x; x1, x2). This indicator’s (x’s) range is from 0 to 1 where 1 represents absolute correspondence to the required light levels. For the value to approach 1 there is less tolerance for error, in other words, this roof system must follow more precisely the Objective Function.

• The result of this demonstration

The Kinetic Roof’s ‘correspondence to required environment’ is x2=0.76, whereas the Fixed Roof’s ‘correspondence to required environment’ is x1=0.66. From this we can say that the Kinetic Roof meets the illumination needs of the building programme more effectively. In addition if the Kinetic Roof has programmability built into it, it would adapt over longer periods of time and under different weather conditions as well.

5-6 . The Contribution

With the development of ‘smart’, ’metabolic’ systems like kinetic structures they needed a new way of evaluating their conditions as well as a way of being evaluated. The development of the ‘correspondence to required environment’ index provides a way of doing this. This index could be expanded to cover a wide range of environmental factors. As the proposed architectural machine took a higher score in this indicator, it means that the system has more resistance against the environment, thus making it more adaptable to a wider range of conditions. When the roof system is applied to extreme climatic zones such as hot deserts, or the frigid Arctic, it will work more effectively to maintain the required inner conditions.

fig 5 5-6: Sun Illumination (K.Hotta)

The left hand diagram shows high time-responsiveness, which means the output function quickly constricts to match the objective function. The right hand diagram has a lower time-responsiveness. The system takes time to reduce the difference between actual output and objective function.

System will adapt the output following the objective function. So the system can be judged the difference between objective function and actual output function. But also system can be evaluated with time. Dynamic system is laid on the timeline, so the adaptability can also be shown like the diagram. In the idea of homeostasis, the quicker response is the better. However the adaptability (=the difference between objective function and system output ) in the certain time duration, are not always correspondance with time-responsiveness. Because high convergence will bring smarter answers in the stable objective function, on the other hand it could be a lack of adaptability in the case the objective function is fluctuating.