PDF - The Bartlett

 Façade’s apertures optimisation Integrating cross ventilation in Fluid Dynamics
analysis in early design stages
Chrysanthi-­‐Sandy Karagkouni MSc Adaptive Architecture and Computation Bartlett School of Graduate Studies, University College London September 2012 This dissertation is submitted in partial fulfilment of the requirements for the degree of Master of Science in Adaptive Architecture & Computation from University College London I, Chrysanthi Karagkouni, confirm that the work presented in this thesis is my own. Where in-­‐
formation has been derived from other sources, I confirm that this has been indicated in the thesis. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |2 Abstract This paper explores the potential transformations of facade openings due to cross ventilation. The aim is to provide designers a simulation model that enables the analysis of the behaviour a built form can exhibit in relation to a natural element like wind. In particular, the case studied explores the potential mutations of elastically deformable louvres in order to provide ade-­‐
quate airflow patters indoors. The paper presents a methodology that aims to integrate a nat-­‐
ural ventilation performance evaluation in the early stages of the design process. Employing Fast Fluid Dynamics (FFD), as a method to test wind-­‐induced ventilation, a performance analy-­‐
sis is conducted, whereby an array of automated feedback loops carried out by a genetic algo-­‐
rithm (GA) can produce a number of acceptable solutions. The simulation takes place in three dimensional space and enables the visual representation of the process. Towards the creation of a completed experiments set, the model incorporates the possibility of human movement through the openings of the facade as an additional factor that influences the final outcome. The experimentation conducted proves the ability of the model to yield design instances that fulfil a number of environmental criteria related to airflow and human comfort. In this light, the paper suggests that the aforementioned method can be used as an experimentation plat-­‐
form to influence the direction a designer may take when considering a design proposal. Words : 10300 | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |3 Acknowledgements I would like to express my gratitude to my supervisor Ava Fatah gen. Schieck and my technical tutor Martha Tsigkari, for their advice and guidance during our reviews. I would like to thank Angelos Chronis for his willingness to help. Also, I would like to thank Anna, Vasilis, Konstantinos and my family for their support through-­‐
out this year. This thesis has been financially supported by the State Scholarships Foundation of Greece. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |4 List of figures Figure 1. Sheikh Ahmad Jam Water Reservoir, Kashan, 2011 (source: Hensel and Menges, 2008)……….10 Figure 2. Openings in traditional buildings in Iran (source: <http://www.fotopedia.com/items/ioeBETH zKZw-­‐ApV-­‐9YE3zvE>[accessed:10 August 2012])………………………………………………………………………..…10 Figure 3. The factors that influence thermal comfort (diagram by ARUP) (source: Zola, Z., 2010)………..11 Figure 4. Wind roses application available by Autodesk Vasari (source: Autodesk,2011)……………………..12 Figure 5. The orientation of building with respect to prevailing wind direction(source:Autodesk,2011).12 Figure 6. Hostel in Southern China by Zoka Zola and Arup (source: Zola, Z., 2010)………………………………13 Figure 7. Thematic Pavilion EXPO 2012, Yeosu, Korea (source: SOMA, 2012)………………………………………13 Figure 8. Nebuta House by Molo(source: <http://molo design.com/projects/aomori-­‐nebuta-­‐house/> [accessed:3 August 2012])…………………………………………………………………………………………………………….14 Figure 9. Porous cast, Garin, G., Architectural Association, 2005 (source: Hensel and Menges, 2008)…16 Figure 10. Excel function that calculates comfort levels(source: http://lumasenseinc.com/EN/products/ thermal-­‐comfort/pmv-­‐calculation/>[accessed: 15 July 2012])………………………………………………………18 Figure 11. Wind rose for Rome in July(source: <http://www.windfinder.com/>[accessed: 15 July 2012]). The room used in this study placed on the wind rose……………………………………………………………………18 Figure 12. 3d representation of the space tested…………………………………………………………………………………19 Figure 13. The possible transformations of the louvres……………………………………………………………………….20 Figure 14. Visualisation of velocities in 3d space, CFD and Matlab graph…………………………………………….21 Figure 15. The meshing of the surfaces as it occurs due to different voxel sizes………………………………….22 Figure 16. Volume of the room and occupied zone………………………………………………………………………………22 Figure 17. Representation of the GA scheme and the possible values of the genes……………………………..23 Figure 18. Flowchart of the GA scheme………………………………………………………………………………………………24 Figure 19. Screenshot of the simulation environment………………………………………………………………………….25 Figure 20. Diagrammatic representation of the room. Experiment 1………………………………………………….26 Figure 21. Sequential pictures of a planar section during the fluid simulation. Experiment 1………………26 Figure 22. Diagrammatic representation of the room. Experiment 2…………………………………………………..27 Figure 23. Sequential pictures of a planar section during the fluid simulation. Experiment 2………………27 Figure 24. 3d graph of the velocities. Experiments 1-­‐2 ………………………………………………………………………..28 Figure 25. Fitness graph of the optimisation process for 8 louvres. Experiment 2……………………………….28 Figure 26. The optimal solution as provided by the GA. Experiment 2…………………………………………………28 Figure 27. 3d graph of the velocities. Experiments 3-­‐6………………………………………………………………………..29 Figure 28. Fitness graph of the optimisation process for 7 louvres. Experiments 3-­‐6…………………………..30 Figure 29. Comparison of the fitness of best solutions. Experiments 3-­‐6……………………………………………..30 Figure 30. The optimal solutions as provided by the GA. Experiments 3-­‐6…………………………………………..31 Figure 31. 3d graph of the velocities. Experiments 7-­‐8…………………………………………………………………………32 Figure 32. Fitness graph of the optimisation process for 6 louvres. Experiments 7-­‐8……………………………32 Figure 33. The optimal solutions as provided by the GA. Experiments 7-­‐8…………………………………………..32 Figure 34. 3d graph of the velocities. Experiments 9-­‐10……………………………………………………………………...33 Figure 35. Fitness graph of the optimisation process for 6 louvres. Experiments 9-­‐10............................33 Figure 36. The optimal solutions as provided by the GA. Experiments 9-­‐10............................................34 | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |5 Contents Abstract | 3 Acknowledgements | 4 List of figures | 5 Contents | 6 Introduction | 1 7 1.1. | Natural ventilation instead of mechanical ventilation 7 1.2. | Fast Fluid Dynamics instead of Computational Fluid Dynamics 8 1.3. | Aims and Objectives 8 1.4. | Structure of the dissertation 9 Background | 2 10 2.1. | Historical precedents 10 2.2. | Natural ventilation principles and comfort levels 11 2.3. | Recent advancements in architecture facades 13 2.4. | Simulations of Fluid Dynamics 14 2.5. | Performance-­‐driven design 16 Methodology | 3 17 3.1. | Simulation context 17 3.2. | General set up and simulation environment 18 3.3. | Space modelling and façade geometry 19 3.4. | Implementation of Fast fluid Dynamics 21 3.5. | Voxel based meshing 22 3.6. | Genetic Algorithm 23 3.7. | Multiple criteria fitness function 24 Experiments and Results | 4 26 4.1. | Experiment 1 – orthogonal openings 27 4.2. | Experiment 2 – 8 louvres free 27 4.3. | Experiment 3-­‐6 – 1 louvre open, 7 louvres free 28 4.4. | Experiment 7-­‐8 – 2 louvres in turn open, 6 louvres free 31 4.5. | Experiment 9-­‐10 – 2 louvres in a row open, 6 louvres free 33 Discussion | 5 35 5.1. | Results overview 35 5.2. | Critical assessment and further developments 36 Conclusion | 6 39 References | 41 Appendix | 44 | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |6 Introduction | 1 Performance analysis in the field of architecture allows the designer to perceive the behaviour of buildings as regard their relationship with the external environmental. Researchers focus on different ways to predict the interaction of a designed proposal with natural elements, to-­‐
wards a more comfortable built environment that achieves, in parallel, low level of energy consumption. Building engineering brings forward the issue of sustainability in order to solve everyday prob-­‐
lems, such as to provide sufficient day lighting and indoor air circulation. Solar energy and thermal analysis as well as wind engineering are considered the most popular topics that archi-­‐
tects aim to incorporate in their designs. Sustainability has become an integral issue to be re-­‐
solved while sun and wind have been widely studied. As a result, a great amount of knowledge in the territory of these natural elements is now available. It is on the designers side to inte-­‐
grate this knowledge into their plans but as a common practice it is not earlier than the last phases of design process when they engage with such issues. In the framework of this thesis, an exploration of the relationship between external environ-­‐
ment and building skin takes place. In particular narrowing down the field of environmental studies, only one parameter was tested and its possibilities in the context of performance-­‐
driven design was investigated. This study focuses on wind, since it is one of the most challeng-­‐
ing environmental factors that affect a building and has a major impact on the indoor envi-­‐
ronment provided to the occupants. Teuffel (2008) states that the correlation between the building envelope and the external forces is integral in the field of wind engineering. As wind is a dynamic phenomenon it can ex-­‐
hibit very complex behaviour in relation to the built forms, therefore the question arises how the building skin may respond to it. The efficiency of a natural ventilation can be influenced by the exterior built form, and the or-­‐
ganization of the interior spaces (Autodesk,2009). In this study an exploration is carried out towards the shaping of façade openings as they get informed about the state of wind around them. Thus, the relationship of wind and building shape is investigated on the boundary be-­‐
tween exterior and interior. In specific the experimentation will be conducted with respect to the windward and leeward façades of a low-­‐rise pavilion. The windward façade is exposed to the wind and the leeward is the opposite façade of a building. 1.1. | Natural ventilation instead of mechanical ventilation. In recent years, mechanical ventilation was preferred instead of natural ventilation as it can provide stable air conditions and resolve airflow problems triggered by inadequacies in design (Cowell, 2012). Nevertheless, heating, ventilation and air conditioning systems (HVAC) are complex and need large number of components to operate. In addition, this kind of technology | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |7 consumes great amount of energy while at the same moment does not always manage to de-­‐
liver the desired indoor climate (Kleiven, 2003). These leads architects to propose natural ven-­‐
tilation to be integrated in the design of functional built environments (Mendez, 2012). The paper follows the idea which argues that an interdisciplinary approach from the initial stages of design is mandatory for achieving successful natural ventilation concepts (Hensel and Menges, 2008; Malkawi, 2004). In this case, computational achievements like Computational Fluid Dynamics (CFD) and evolu-­‐
tionary algorithms which were investigated during the development of this thesis can be proved as potentially useful techniques to design, taking into account performance analysis. 1.2. | Fast Fluid Dynamics instead of Computational Fluid Dynamics Natural phenomena simulated in computer models require a skilful expert user. Wind engi-­‐
neers are employed for the interpretation and operation of such complex simulations (Malkawi, 2004). Moreover, powerful computer processors and a great amount of time are needed until the simulation model converges in an accurate result. This amount of power was not available in personal computers until recent years and still has not reached the level of required operating power. On the contrary, the engagement of engineers with mathematical models proved to be fruitful since new simulation models appeared. Recently, in the field of fluid dynamics a less accurate but adequately informative model was presented. This model is Fast Fluid Dynamics (FFD), firstly presented by Stam (1999), and this has been the model used for the experiments conducted in this thesis. 1.3. | Aims and Objectives The main objectives of the thesis is to investigate the architectural consequences and possibili-­‐
ties as these are formed by the integration of wind analysis in the design process. The aim is to provide optimum airflow rate and distribution in the buildings’ interior. How can a façade be designed to allow the appropriate amount of air to circulate through the internal space? In addition the study presents an exploration of natural ventilation in combination with the forced opening and closing of façade components that imply random movement of people through the building envelope. Performance-­‐oriented design has as a primary aim to introduce spaces that achieve acceptable levels of human comfort. Wind induced airflow has a significant role to the improvement of occupants’ comfort in a building. Towards a performance-­‐oriented design this thesis highlights the benefits of environmental performance analysis when incorporated during the crucial phase of conceptual design. This approach aims to give an insight to designers about the final form of the building and the behaviour it exhibits through the transformations of its compo-­‐
nents. These transformations rely on the interaction of the building shell with wind and the interaction of the facades’ components with each other. The extent to which simulation of natural airflow can potentially be a contributing parameter in the conception of performance-­‐
aware designs will be presented. For the shaping of this scenario an evolutionary algorithm | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |8 along with the FFD was employed. In the last steps of the thesis, the experimentation takes into account user requirements that imply the possibility of people circulation. The general scope of the study relies on the direct use of natural resources in the design of a building. The proposition stands on the argument that fast fluid dynamics and genetic algorithm can be proved to be contributing parameters for the shaping of a design. 1.4. | Structure of the dissertation The following section will focus on a brief description of natural ventilation as it can be found in historical buildings and its principles will be presented. Α general review of previous studies on CFD and FFD will be given. The third section of this study will address the methodology de-­‐
veloped to test the hypothesis proposed. In addition, the experiments that were conducted will be presented. The fifth section involves a detailed discussion on the results on a critical point of view and further possible developments will be proposed. Finally the conclusion in-­‐
cludes an overview of the study. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |9 Background| 2 2.1. | Historical precedents Natural ventilation is not a new phenomenon in architecture. Utilisation of the natural driving forces for the purpose of ventilation has for several decades provided the desired thermal comfort and air quality for people. In traditional architecture of south eastern countries, where the temperature is high, porous structures were widespread as a mean to provide good comfort levels for the inhabitants. The performance analysis of such buildings (figure 1) can show their efficiency to achieve high rates of air exchange due to the availability of air, passing through the perforated building envelope (RIBA, 2011). Thus, it is a common assumption that sufficient circulation of air can be achieved by openings on the building shell such as windows, louvres, grilles, wind towers and wind catchers. This method of infiltration can be traced in the wind towers and the screenwalls used in India and Arabic countries (Hensel et al., 2008; Hensel, 2008). In the case of the screenwalls, which are more related to the geometry examined in this thesis, the building envelope consists of stone (jail) (figure 2) or wooden (mashrabiaya) lattice work, that introduce a variety of aper-­‐
ture sizes. Hence, in such examples of vernacular architecture the boundary between exterior and interior space is not explicitly defined, as it happens in contemporary buildings (Hensel, 2008). Nowadays, the building envelope is usually sealed and rely on air-­‐conditioning and heating systems for sufficient air circulation. Nevertheless, the existing technology can pro-­‐
pose mechanisms that introduce a state between the past and the present. In this light, build-­‐
ing facades that operate with controllable openings in order to shift from airtight shells to per-­‐
forated structures according to the environmental conditions that change over time, can be presented (RIBA, 2011). Figure 1. CFD Analysis, Ahmad, S., Jam Water Reser-­‐ Figure 2. Openings, in traditional buildings have voir. Built around 1630, natural cross-­‐ventilation is screenwalls that enable ventilation (India). enabled by the arrangement of openings. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |10 2.2. | Natural ventilation principles and comfort levels Natural ventilation is a method to deliver air into buildings and replace the existing aged air with fresh air, using the force of wind (Cheung and Liu, 2011; RIBA, 2011). The most common types of this method is the stack effect enabled by wind towers and cross ventilation enabled by openings on opposite sides of the buildings (RIBA, 2011). The second will be the case exam-­‐
ined in this study. The wind is the driving force of the whole system. The way to allow this force move through indoor spaces and also control its distribution is to design openings that affect the circulation of air and channel it to specific directions. This will make possible to ei-­‐
ther block the air when it is not needed or introduce it indoors when it is appropriate. The pa-­‐
rameters that define the requirements of airflow depend on the comfort level standards which will be described next. It has been studied and estimated that the way people perceive their satisfaction comfort lev-­‐
els depend on external parameters as the air temperature, the mean radiant temperature, the air velocity and the relative humidity of the space they live in. In addition, two occupant-­‐
dependent variables that influence this feeling are considered to be the physical activity of each person at a specific time as well as the type of clothing someone wears (ASHRAE-­‐55, 2010; ISO, 2005) (figure 3). Moreover recent studies have shown that this model can be ex-­‐
tended to take into consideration other parameters that affect the behavioural, physiological and psychological adaptation potential of the occupants in real-­‐world settings (Nicol et al., 2009). This notion turns the comfort model into an adaptive one, that always adjusts to specif-­‐
ic situations to predict more precisely the way a person perceives the ambient quality of a space. Although such information is quite significant and should be taken into account, it is not in the scope of this thesis to work on such an intensive analysis of the comfort criteria, since the objective concerns the early stages of design. Figure 3. The factors that influence thermal comfort. Shade and air movement are the most important parameters (diagram by ARUP). Before implementing the model presented in this thesis, the local climate, existing at the place where the building will be constructed, should be studied, so as to draw some conclusions as regards the applicability of natural ventilation in the specific location. Such a study, but far more intensive than the one needed in the context of a design process, has been conducted by | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |11 previous experimental research by Carillho de Graca et al. (2002). This study examines two cit-­‐
ies in China, Bejing and Shangai, and draws conclusions about the usefulness of natural ventila-­‐
tion in each case. According to Fangers’s thermal comfort model (Fanger, 1970) and the maxi-­‐
mum allowed indoor air temperature for the warm season, the performance of passive cooling systems is evaluated. The findings describe the percentage of efficiency of natural ventilation throughout the year in both cities. Such a study can provide valuable information to architects about the appropriateness of introducing passive cooling systems in specific cities. As mentioned previously, the design of a naturally ventilated building should consider the local climate. In a real case scenario the designer, as a first step, should consult weather data from wind rose diagrams to understand the prevailing winds and their direction in the specific loca-­‐
tion where the site is located (figure 4). According to these data the orientation of the build-­‐
ing should be perpendicular to the direction of wind, during the warm periods of the year, to achieve better infiltration of wind indoors (Walker, 2010) (figure 5). This method was used also in the experiments conducted in this thesis. Figure 4. Wind roses provide information about the prevailing Figure 5. The orientation of building with direction of wind. The picture shows an application available by respect to prevailing wind direction. Autodesk Vasari, where you can overlay a wind rose diagram onto your building site. Wind, in the design-­‐process, can be considered as a factor that can influence both the exterior and the interior of buildings. The external shape of a structure can allow aerodynamic airflow, while the apertures of the external walls and the planar configuration of a space can enable efficient indoor ventilation. These factors, when resolved as one of the design problems, can improve the quality of the environment for the inhabitants and the passers-­‐by. The problem introduced when trying to optimise a building envelope, to achieve sufficient ventilation, is the difficulty to foresee the complex movement of air through the internal spaces of the building. Moreover, the losses in the airflow from the inlet opening to the outlet opening should be suf-­‐
ficiently low (Walker, 2010). Natural ventilation concept is therefore highly integrated in the building body and consequently influences the architecture, in the exterior as well as in the interior. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |12 A project that based its overall design on airflow patterns has been presented recently. The hostel designed by Zoka Zola and constructed by Arup, located in China, follows the concept of natural ventilation to provide a better environment for the inhabitants. The flows of air were thoroughly studied and shaped the final design for this building (Zola, 2010) (figure 6). Due to the optimisation of the building shell, for natural ventilation, the need for air-­‐conditioning in the specific building has been eliminated. This project highlights the possibility of performance evaluation to participate as a factor in the shaping of a design. Figure 6. Hostel in Southern China by Zoka Zola and Arup. Section of the building and airflow patterns in the space. 2.3. | Recent advancements in architecture facades The recent developments in material research are now coming to real-­‐world practical applica-­‐
tions to buildings’ facades. Soma architects and Knippers Helbig have explored the notion of materiality with the aim to construct a kinetic façade for the Thematic pavilion at Expo 2012 in Yeosu, Korea (figure 7). Based on a research about material performance inspired by nature (Knippers and Speck, 2011) they introduced a series of elastic fins, made of glass fiber rein-­‐
forced polymers, which can serve the same functionality as the common louvres. The distin-­‐
guishing characteristic in this case is the ability of each component to twist, when compressive force is applied, and thus perform deformations in a much complex way. These deformation can result in a variety in the sizing of openings. In addition, these components are enabled to move independently and the final surface as it was constructed can perform smooth move-­‐
ments. Moreover, as stated by the designers, the final surface can bear wind loads without changing its shape, which allows the use of the above presented curved geometry in a re-­‐
search which focuses on a wind driven design (SOMA, 2012). Figure 7. Thematic Pavilion EXPO 2012, Yeosu, Korea. Architect: Soma architects, Vienna. Engineer kinetic façade: Knippers Helbig | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |13 Another recent example of a façade which exhibits variations in apertures’ size can be found in the Nebuta house in Japan, realised by Molo (figure 8). In this case the façade is static and plays the role of a second outer skin. Once more, an array of fins due to their material proper-­‐
ties, in this case steel, takes advantage of their ability to bear torsion in order to create the final form. The morphology resembles moving elements that was captured in a specific mo-­‐
ment. Figure 8. Nebuta House by Molo in northern Japan. Both these projects are only two examples that depict the progress that have taken place in façade construction. The first case is a media façade that moves to create an animated built form and the second namely serves the aesthetic parameter of an architecture project. In the framework of this thesis, the question arises as regards the transformations that may occur in such morphological elements due to wind in order to enable ventilation. How would these forms change if wind analysis was incorporated into the design process? Having as a reference and starting point the above described structures, this thesis explores the possibility of a series of curved surfaces that can deform and create various sizes of openings as opposed to wind-­‐
induced ventilation. The simulation proposed, aims to give an insight to the designer either when it concerns kinematic facades about the way they could perform or when related to stat-­‐
ic configurations about their final shape as affected by wind. 2.4. | Simulations of Fluid Dynamics Computational Fluid Dynamics are models used to predict air distribution using mathematical equations to solve the flow of air in spaces (Awibi, 1989). This simulation can be either two-­‐
dimensional or three-­‐dimensional. CFD needs great amount of computational power to take place but is considered to be an accurate model of airflow (Awibi, 1989; Malkawi, 2005). In the past, several studies have used this model to simulate conditions of the real world that include the movement of air. Kato and Murakami(1985), focused on searching the distribution of ve-­‐
locities of air, in three-­‐dimensional space, in a room according to different supply openings. Awibi (1989), employed CFD to predict velocity values in a room and distribution of tempera-­‐
ture, in order to provide a helpful tool for the design process. These studies provided research material with sufficient information in order to design effective air-­‐conditioning systems but did not engage with natural wind forces. In addition, Awibi concluded that CFD needs much effort to be carried out in comparison to physical models, meaning wind-­‐tunnel simulations (Awibi, 1989). | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |14 Recently, the use of CFD has drawn the interest of a greater group of researchers and started to be considered as a way to comprehend natural ventilation problems. As mentioned previ-­‐
ously Carillho de Graca et al (2002) have implemented CFD simulation to assess the air circula-­‐
tion performance in indoor spaces in specific countries. Suga et al (2010), focused on the de-­‐
sign of windows, taking into account wind-­‐flow. Also, Cheng and Liu (2011), examined the pat-­‐
terns created by air around buildings, with the objective to optimise the void space among them, while placing blocks in proximity. These studies used the CFD model as an evaluation analysis only for orthogonal rooms or buildings, while the ones engaged with openings took into account only typical rectangular windows. Stam proposed a different way to approach the simulation of fluids. Using a semi-­‐Lagragian scheme (Stam, 1999; Chen and Zuo, 2007), he solved the equations that govern the movement of fluids using low order schemes. The FFD model was initially introduced in order to simulate “fast but visually convincing” the movement of fluids for the game industry (Stam, 2003). Chen and Zuo (2007), validated the use of this model for several situations, among which indoor ventilation can be found. In their experiments, they compared the result of FFD to data pre-­‐
sented in previous studies on airflow which was already considered valid. Their findings matched in a high degree with the given data. Thus, they propose this simulation model for further use in the field of ventilation and architecture design. In a more recent study of them, the conclusions that are drawn apart from the accuracy of the model also proves that the model presented by Stam is a much faster solution in comparison to conventional CFD models (50 times) (Chen and Zuo, 2009). Computing time is considered an important factor when talk-­‐
ing about the behaviour of the architectural forms with respect to wind during the conceptual phase of design. Although a group of researchers have been interested in the validation of this simulation and stated that the level of accuracy it can achieve is sufficient, there is a limited number of studies actually using this model in building engineering. Such an approach has been proposed by Chronis (2010) and Chronis et al (2011), as a form-­‐finding method. This study focused in the reaction of a curved surface, with a certain level of freedom in movements, to the tension in-­‐
troduced by air. His regard was not environmental but an effort to find forms which are in-­‐
formed by the wind load. Another study using FFD was conducted by Sheby (2011) where the simulation proposed, took into account pedestrian comfort in urban environment. This study was limited in two dimensional space. Hence this model to date has not been used for wind-­‐
induced indoor ventilation that can provide human comfort investigating the transformations of the apertures of a façade, as the case presented in this paper. In the studies discussed above, it is a usual practice to couple fluid simulations with genetic algorithms. In all cases this combination enables the researcher to find the best solution among a large number of possible solutions. This model have been used extensively by Malkawi, who have also noted the importance of visualisation during the optimisation process. This enables the designer to visually evaluate the good solutions carried out by the algorithmic process, providing him the opportunity to select from a number of good solutions as regards the air simulation (Malkawi et al, 2005). His studies focus only on mechanically ventilated sys-­‐
| UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |15 tems. The coupling of FFD and GA have also been recently realised and proved to be an effec-­‐
tive strategy, but none of these studies have been interested in openings optimisation for in-­‐
door natural ventilation (Chronis, 2010; Sheby, 2011). 2.5. | Performance-­‐driven design To date, the prevailing trend is the separation between the design process and evaluation due to fluid dynamics performance. A design is usually completed when ventilation criteria are in-­‐
tegrated into the proposal. Figure 9. Design, fluid analysis, fabrication. (Porous cast, Garin, G., Architectural Association, 2005-­‐06). Hensel and Menges (2008) argue that considering the performance of a design during the ear-­‐
ly stage of evaluation both the morphology of the structure and its environmental perfor-­‐
mance is important. In this light, form generation and environmental analysis become equal factors to the decision-­‐making process of an architectural proposal. The form and the behav-­‐
iour of a building become an entity that changes according to specific criteria posed by the ar-­‐
chitect with respect to a real case scenario. Thus, performance-­‐driven design can be incorpo-­‐
rated into a design methodology even from the phase of conception, which is the starting point of a project proposal. In this case, the scope of such a strategy is not to gather specific and accurate information about energy performance but rather observe and manipulate the “tendencies or patterns” (Hensel and Menges, 2008) of the design’s behaviour in accordance to external forces. This can show how sustainability can be consistent with design at the high-­‐
est level from the conception till construction. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |16 Methodology | 3 This paper proposes the implementation of a methodology that aims to become a useful strat-­‐
egy for the design of buildings that need air-­‐conditioning in order to achieve high rate of air exchange and acceptable levels of indoor airflow. Such a study can be considered useful for architectural projects located in countries with climate that exhibit high temperature and moderate or high humidity, since the airflow can provide a cooling effect and is considered as an evaporation method. Although there is not a specific brief to be fulfilled, an assumption for the parameters that would appear in such a case was made. These parameters are taken into account so that the final proposal can easily be implemented later on if a real world scenario would exist. Throughout this study, a number of experiments took place in order to prove the proposed hypothesis. In the next section, the methodology followed to enable testing is de-­‐
scribed in detail, along with some principles of ventilation that led the methodology in specific directions. 3.1. | Simulation context The methodology followed exhibits similarities with the ones that appear in the studies al-­‐
ready presented in the literature review. These similarities are the fluid simulation and main principles of the genetic algorithm (Chronis, 2010), as well as the setting of the space and the placement of openings on windward and leeward side (Malkawi et al, 2005; Cheung and Liu, 2011). In the contrary, a different approach is suggested as regards the geometry of the fa-­‐
çade’s openings, which is non-­‐orthogonal and the optimisation for natural ventilation using FFD as the evaluation method. As an initial step to set up the context for the experimentation, specific assumptions were made as regards a number of environmental and human related parameters. In this light, the model gets informed by weather data from Rome, located in the southern part of Italy, to de-­‐
termine the prevailing wind direction. Due to the hot climate in this city, it is considered a suit-­‐
able case study for a natural ventilation system. The temperature is defined to be a standard value of 25 degrees. The calculation takes place for an unfurnished room and the use of space is considered to be a communal space. Hence, the metabolic rate of the occupants is 1.2 which implies sedentary activity (ISO, 2005). The clothing insulation that influences the human com-­‐
fort levels is 0.7. Using the above values it is possible to have as an output the desired air ve-­‐
locity so as to achieve the better ambient quality in the space tested, which in this case is in a range between 0.4ms-­‐1 and 0.8ms-­‐1. A diagram of the predicted mean vote, which is the value of comfort sensation as determined by a number of people according to ASHRAE standards, can be used for this scope( figure 10). | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |17 Figure 10. Screenshot of an excel function that calculates comfort levels. The curve on the right defines the human comfort level (PMV). On the left the parameters that influence function. The coloured part of the curve gives a range of acceptable values. 3.2. | General set up and simulation environment This thesis examines a specific type of ventilation as a case to use fluid dynamics. An explora-­‐
tion has been conducted on cross ventilation, which depends on wind availability to provide airflow. It is thus difficult to foresee the way a natural element, like wind, can interact with a structure. Hence, fast fluid dynamics is proposed as a rapid way to provide some preliminary assumptions on this field. When cross ventilation is the objective, it is a common practice to place openings on the windward and leeward side of the building. Using a wind rose as a ref-­‐
erence , that depicts the directions of wind, the apertures are placed accordingly (figure 11). Therefore, the louvres, that operate as inlet and outlets, are placed on opposite sides of the room. As a result, it is anticipated that air will follow a path from the one side to the other passing through the indoor space. The louvres should force the wind to follow specific direc-­‐
tions or block it when it is considered inappropriate. This configuration of space aims to intro-­‐
duce, in the room, a natural breeze in order to provide a better environment to the occupants. As regards the rest of the structure, meaning the remaining walls, the ceiling and the floor, it has plane shape, creating an orthogonal volume of fixed area, so as to eliminate the complexi-­‐
ty of the system. Moreover, the objective of the investigation is to optimise the apertures’ form and not the overall shape of a building envelope. Figure 11. Wind rose of Rome in July. This chart provides information about the direction of wind. On the right the room used in this study is placed on the wind rose to describe the configuration of the system with respect to wind. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |18 Throughout the experiments the openings arranged on the windward façade, apart from their role to provide air in the room they are also considered as openings that allow people circula-­‐
tion. A number of experiments were conducted with each louvre totally open in order to repli-­‐
cate the use of a door in a building. Thus, the experimentation takes into consideration low-­‐
rise buildings so as to be able to combine the two parameters needed; wind and human circu-­‐
lation. In varied height buildings difference in wind force can be observed. In the case of this work, there is no need to explore high-­‐rise buildings since the movement of people through a building takes place at the ground floor. 3.3. | Space modelling and façade geometry The problem of ventilation as it can occur by various orthogonal openings on the building shell has been studied in the past and conclusions on the way windows and doors should be de-­‐
signed is now available to architects (Cheung and Liu, 2011). Though, this paper engages with curved openings optimisation for natural airflow, as they are formed by a series of louvres. The model is implemented in Processing language and allows three-­‐dimensional free-­‐form ge-­‐
ometries to be presented. The whole configuration of space is divided in two categories. The static elements and the moving elements. The static elements consist of the ceiling the floor and two lateral walls while moving elements are proposed for the two remaining sides of the space which are also the windward and leeward side. In the above described design it is sug-­‐
gested that the moving elements will replace the role of doors and windows, if it is thought as a real building. The overall volume occupied by the simulated model is characterized by its width (4.8m), length (6.15m), and height (3.9m). Figure 12. 3d representation of the space tested. On the left the openings are closed and the opposite happens on the right. The static elements are represented with transparent white colour while moving elements are solid white. The simulation of the surfaces that create the facades takes as a reference the built examples presented in section 2.3. One surface is considered as a single louvre and is placed in proximity with the adjacent surfaces so that the final outcome is an array of louvres that cover the whole width of the facades. Each louvre in the simulation environment, is a non uniform b-­‐spline sur-­‐
face (NURBS) and the algorithmic model used is based on an example created by Turner (2010). The shape of a NURBS surface is directly depended on the number of control points it consists | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |19 of, and their placement in three-­‐dimensional space. The model used three control points for each of the x and y directions. The shape of the model is also characterized by its width and its height. The unique form of a louvre is defined by its narrow width and long height. In the simu-­‐
lation every surface had 1.15m width and 3.8m height. Trying to simulate the behaviour of the elements found on the facades presented (Soma pavil-­‐
ion, Nebuta house), the louvres should perform a certain range of movements. In both pro-­‐
jects the property of twisting is prevalent. For this to be achieved, initially the upper and lower control points for each surface are considered the anchor points of the structure. The remain-­‐
ing three control points in the middle are enabled to move. This can be achieved by stimulat-­‐
ing the movement of one control point. This point is an edge point of the three placed in the middle of the overall height. Subsequently, the remaining points on x direction should move in relation to the one on the edge. The amplitude of their movement should decrease inversely proportional to the distance each of them has from the moving edge point. According to this, a Gaussian distribution was incorporated in the NURBS algorithm. Creating a Gaussian function is a simple way to achieve the expected type of relation between the adja-­‐
cent control points (figure 13). The same function was added later on to the lower three con-­‐
trol points, in order to create openings for the people circulation, as in Nebuta house, replac-­‐
ing the functionality of doors. After this addition, only the upper control points were perceived as anchor points. Figure 13. The possible transformations of the louvres. As regards the direction of movement the twisting effect observed should be replicated. The direction of displacement was constrained on the x and z axis, while the amplitude of dis-­‐
placement for both directions was coded to be equal. The Gaussian function mentioned above was implemented on both directions to affect adequately the involved control points. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |20 3.4. | Implementation of Fast fluid Dynamics FFD has already been used in a number of studies, where a detailed explanation of the scheme can be found (Stam, 1999; Chen and Zuo; 2007, Chronis, 2010). Thus, it is not the scope of the present study to explain extensively the mathematical or algorithmic process. The algorithmic model used is based on an implementation, in three-­‐dimensional space, of Stam’s model (1999) by Ash (2006), which was later on coded in Processing by Chronis (2010). The fluid solver runs iteratively to define the fluid movement that will take place in every time step. The time step used here was dt=1s and the iterations of the fluid solver was set at 8. The model is grid based which in three-­‐dimensional space consists of uniform voxels. The visualisa-­‐
tion of the whole scheme is exclusively based on a range of colours that represent the value of velocity in each voxel, using a mapping function. Furthermore, the behaviour of the system relies on the initial velocity forced into the system to stimulate the fluid. While the research focuses on natural ventilation this velocity represents the wind. Throughout the simulation, the first row of voxels on y direction is prescribed to have a specific velocity, which vary in the experiments between 1ms-­‐1 and 4ms-­‐1. This implies that the wind is blowing continuously with a specific direction and velocity. As a further step, a three-­‐dimensional vorticity confinement function was incorporated into the algorithm. This was based on a two-­‐dimensional scheme encoded in Java by Mckenzie (2004). The specific extension aims to reduce the numerical dissipation that exhibits the FFD model. Creating small swirling flows commonly found in smoke simulation (Fedkiw et al., 2001) the velocity field is influenced. The algorithm for vorticity confinement can be found in the ap-­‐
pendix. It is also worth mentioning that, a visual representation of the information that the fluid simu-­‐
lation can provide, was created. Trying to visualise the indoors condition of the airflow, a func-­‐
tion in Matlab was used to create a three-­‐dimensional graph from the given velocities (Matlab website, 2012). For every experiment a memory structure was created where the velocity val-­‐
ue of each cell located at height equal to 1.5m was kept. These values were later on parsed by a Matlab function and a three dimensional graph was created where the velocity component specified the colour and the height of the surface (figure 14) . Figure 14. Visualisation of velocities in 3d space. From left to right: FFD domain, Visualisation of velocities using the surface function in Matlab. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |21 3.5. | Voxel based meshing The fluid solver, as the majority of CFD programs, is voxel-­‐based. The voxels’ state is binary. They can be either free, where the fluid can move, or solid, which are voxels occupied by solid objects. The surfaces that represent the structure of the room and the outer bounders ,that constrain the movement of the fluid into a specific domain, represent the occupied voxels. The NURBS surfaces described in section 3.2, and the remaining elements that confine the room, was incorporated into the fluid solver, using a parametrisation technique which subdivides each surface into discrete voxels. The voxels that depend on the shape of the NURBS are ena-­‐
bled to change their state, according to the displacement of the surface, while the remaining voxels are static, and unaltered for each experiment. These voxels represent the internal boundaries of the fluid solver, redefined in every time step . The size of the fluid domain for all the experiments on the x, y and z axis was set to 76 , 150 and 50 respectively. In total 57 x105 voxels were simulated and their size was uniform. Size values between 2 and 5 were tested and 3 was chosen as the voxel size that would be used in the final experiments. Figure 15. The meshing of the surfaces as it occurs due to different voxel sizes. The simulated room occupies 34x103 voxels. Numerically the space that can possibly be occu-­‐
pied by people is taken in a distance of 0.6cm from the lateral walls and 1m of the ceiling (fig-­‐
ure 16). This region was used for the calculations that took place, in order to evaluate the per-­‐
formance of the room for natural ventilation. This is a common practice, since in the periphery of a room the condition of air does not affect considerably the human comfort (ASHRAE-­‐55, 2010; Malkawi et al., 2005). Figure 16. From left to right: the volume of the room, the room represented by voxels, representation of the occupied zone. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |22 3.6. | Genetic Algorithm An evolutionary method was implemented along with the FFD scheme to enable an automated evaluation of a large group of possible design solutions. The algorithm implemented relies on an example created by Turner (2009). In specific, this genetic algorithm (GA) is based on a standard number of population members that are evaluated by a fitness function. Each mem-­‐
ber consists a possible instance of the design. The differentiation between each instance de-­‐
pends on the genes, as they are encoded. The genes are directly connected with a specific var-­‐
iable of the model and define the final form of the design instances. In the case presented, the genes determine the displacement of the control points for each louvre on the x and z direc-­‐
tion. Theses genes, random initially, are assigned with a range of values, and the pace accord-­‐
ing to which these values can be incremented. The possible values are discrete from 0 to 5. The lower limit of this range means that the louvre is completely closed and the displacement is 0°, while the upper limit means that the louvre, at the height of 1.5m, is open at 50°. The pace of the inclination is 10° . Figure 17. Representation of the GA scheme and the possible values of the genes. The population size, after testing, was set at 64. A fitness function, related to natural ventila-­‐
tion, determines the effectiveness of its solution. Afterwards the whole population is sorted from the less to the most fulfilling option. The next step is the mutation and crossover of the possible values of the genes. Two instances are selected and a percentage of their genetic el-­‐
ements is passed to a new instance, while there is also a probability of new solutions to occur. The resultant solution is then evaluated replacing the less effective member of the existing population. A different number of generations was tested fluctuating between 500 and 1000. In the experiments presented the number of population is set at 600 since for more generation no significantly better solutions were found. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |23 Figure 18. Flowchart of the GA scheme 3.7. | Multiple criteria fitness function The fitness function elaborates the objectives of the problem into an algorithmic process. In the case presented, the airflow is adjusted due to the transformations of the facades’ ele-­‐
ments, until an appropriate airflow pattern is met. The main parameter that influences the airflow is the velocity of air in the occupied space. In this light, velocity is used to recognise the behaviour of the wind indoors. Although the velocity component was the only key element used for the evaluation scheme, it was used in a number of ways that could provide adequate information about the indoor con-­‐
ditions. Four different functions was incorporated into the fitness function creating multiple criteria for the evaluation process. These function are described in the following section. | Average Velocity The average velocity of the values found in the occupied space, was required to fall into a range as defined by the PMV (further explained in section3.1.) so as the fitness to increment. The upper limit for the average velocity was 0.4ms-­‐1 and lower 0.8ms-­‐1. If the values did not fall into this range the fitness would be influenced negatively. | Volumetric rate exchange Furthermore, since the aim is to achieve a great degree of air exchange, to replace the aged air with fresh, the calculation of volumetric range exchange was adapted. Examining the voxels located in the interior side next to inlet openings, the area they occupy was multiplied by the average velocity found in their centres. This equation is commonly found in environmental studies as regards air (Autodesk, 2011). The higher value of the volumetric rate the better the resulting fitness. Volumetric rate exchange equation: ! = ! ∗ ! 3 -­‐1
Q = airflow volumetric rate (m s ) A = area (m²) -­‐1
V = velocity (ms ) | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |24 | Standard Deviation Additionally, in order to achieve a uniform distribution of air indoors one more function was introduced. This function was formulated for eliminating the deviation from the mean velocity. Initially the variance is computed; the difference between each velocity and the mean velocity is squared and then added to a sum. According to this, standard deviation increments in a high degree when a great difference from the average value is found and respectively the fitness decreases. As stated in previous studies, it is considered as an effective way to achieve a better distribution of uniform air velocities (Malkawi et al.,2005). Standard Deviation: !
=
!
1 !!!
!
!!!(!
− !)! ! = standard deviation ! = current velocity (ms-­‐1) ! = mean velocity (ms-­‐1) n = number of items | Maximum Velocity Moreover, the maximum velocity found indoors was also integrated in the fitness function. A high number for maximum velocity affects inversely proportional the fitness value. As this function also influences the evaluation part of the GA, the solution space was further con-­‐
strained. The above values are considered integral parameters that affect the effectiveness of a natural ventilation system. Thus, it was decided to elaborate them in a way that they could equally affect the fitness of a solution. In this respect all the values as they are computed by each function was mapped to numbers raised to the power of -­‐1. Later on they were incorporated in the fitness function, to either increase or decrease the fitness value as described above. Figure 19. Screenshot of the simulation environment. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |25 Experiments and Results | 4 The experiments’ objective was the verification of the aforementioned hypothesis. The scheme modelled intents to provide information to the designer about the ventilation perfor-­‐
mance of a built form. An integral part of the experiments was the time needed for each one of them in order to converge in a possible solution. As regards the experiments that employ the GA scheme approximately 6 hours were needed for 600 generations. The first experiment consisted the ventilation performance analysis of a room with orthogonal openings so as to provide a set of measurements. These values provide the study with a performance benchmark in order to compare the effectiveness of the proposed façade, that consists of louvres, to a convetional type of façade, that consists of common windows and doors. The se-­‐
cond experiment examined the possible transformations of 8 louvres, so as to provide suffi-­‐
cient wind-­‐induced ventilation. For the rest of the experiments one or two louvres, of the windward façade, were left open throughout the simulation as if people were allowed to pass from the open louvres. 4.1. | Experiment 1 – orthogonal openings A door is placed in the middle of the width of the windward façade, along with two windows on the leeward façade (figure 20). The shape of all the openings is rectangular. This experi-­‐
ment employs only the FFD scheme and not the GA. The resultant values, as measured in the occupied zone, are: average velocity=1.71ms-­‐1, Q=0.46 m3s-­‐1, s=1.69. Consequently, the specific facades’ configuration does not achieve the critiria set. The velocity is higher than 0.8ms-­‐1, which is the upper limit of the required range, while the deviation value depicts a large differentiation among the velocity values which means that the distribution of air is not uniform. Figure 20. Diagrammatic representation of the room, a door and two windows exist on the two facades (left). The room represented by voxels, before and during the fluid simulation (middle and right respectively). Figure 21. Sequential pictures of a planar section, across the x and y axis, for z=1.5m during the fluid simulation. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |26 4.2. | Experiment 2 – 8 louvres free In this experiment the façades consist of an assembly of louvres (figure 22). The proposed de-­‐
sign is optimised by the GA according to values taken from the performance evaluation by the FFD (figure 23). During the evolutionary process, the form of each louvre is changing inde-­‐
pendently. In sum 8 louvres (four for every façade) are optimised as regards their shape, to provide the optimal sizing of opening that allows the better airflow indoors. The encoded scheme connects a control point of every louvre with one gene. As a result 8 genes regulate the overall deformation of the facades. Figure 22. Diagrammatic representation of the room, louvres are present on the two facades (left). The room represented by voxels, before and during the fluid simulation (middle and right respectively). Figure 23. Sequential pictures of a planar section, across the x and y axis, for z=1.5m during the fluid simulation for the second experiment. The fittest member of the population is evaluated for its performance. As regards the ventilation performance, it is considerably better in comparison to the condi-­‐
tions found in the first experiment (figure 24). The average velocity is measured at 0.41ms-­‐1, which is approximately the best possible result with respect to the comfort criteria set. More-­‐
over, the deviation equals 0.307, indicating a relatively homogeneous distribution of air veloci-­‐
ties. The air exchange rate equals 0.184m3s-­‐1. This value is lower than the one given from the previous experiment, however comparing the values of average velocity and Q to the previous result, the velocity has decreased 76% while Q decreased 60%. In short, the solution achieves to raise the rate of air exchange to air velocity approximately 16%. Thus, the measurements prove the benefits of such a design with respect to airflow optimisation. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |27 Figure 24. 3d graph of the velocities in the occupied zone for height=1.5m. The velocity value defines the position on the z axis and the colour. From left to right: experiment 1, experiment 2. Exp.2 exhibits better distribution of velocities The performance of the GA scheme is described in the fitness graph (figure 25). The graph in-­‐
dicates the improvement over time of the less fit, the mean and the fittest member of the population. The generated solution is depicted in figure 26. All the louvres are open in a differ-­‐
ent degree. The openings on the windward side have smaller opening inclination in compari-­‐
son to the ones on the leeward side. Figure 25. Fitness graph of the optimisation process for 8 lou-­‐
vres. Figure 26. The optimal solution as provided by the GA. 4.3. | Experiment 3-­‐6 – 1 louvre open, 7 louvres free Following the same process as in experiment two and using as a further constrain the forced opening of one louvre, the following four experiments were carried out. Experiments 3 to 6 are presented as a group since for each one of them, one of the four louvres of the windward façade was completely open during the simulation, to imply the possibility of people circula-­‐
tion through the respective opening. This happens by pressing a key during the simulation. This function restarts the genetic algorithm and allows the designer to intervene in the process. All the experiments converged into a solution that falls into the criteria set. The average veloci-­‐
ty, fluctuating from 0.493ms-­‐1 to 0.711ms-­‐1, achieved to be in the required range. The air ex-­‐
change rate is between 0.184m3s-­‐1 and 0.219m3s-­‐1, kept approximately at the same levels as in | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |28 experiment 2. The deviation, between 0.488 and 0.704, is kept in low levels in comparison to the first experiment. Figure 27. Graph of the velocities in the occupied zone for height=1.5m. From top left to bottom right: experiment 3, ex-­‐
periment 4, experiment 5, experiment 6. It is clear that exp.3 and 5 provide better solutions. In addition, the genes in this case decreased to 7. One louvre in each experiment was excluded from the optimisation process and the counterpart gene did not participate in the shaping of the outcome. The GA converged to a satisfactory solution in the same time as in the previous experiment, although the complexity of the problem to be resolved was higher, due to the high degree of wind that penetrates into the room from the completely open louvre (figure 28). It is also worth mentioning, that among the four experiments the best solution is obtained when the third louvre is open (experiment 5). The fitness graph presented in figure 29 demon-­‐
strates this comparison. In this light, the designer would be more likely to prefer this solution among the four tested. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |29 Figure 28. Improvement over generations of three population members (optimisation process). From top left to bottom right: experiment 3, experiment 4, experiment 5, experiment 6. Figure 29. Comparison of the improvement of the fittest members as taken from experiments 3-­‐6. The best solution is given by the 5th experiment where the third louvre is set to be open. The second best solution came out from the 3rd ex-­‐
periment. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |30 As regards the shaping of the optimal design instances, an interaction between the elements of the two facades is observed throughout the solution space. The louvres of the leeward side that are directly across the open louvres of the windward side are set to be closed or with a small degree of inclination (figure 30). This happens so as the air that traverse the space does not exhibit high velocity values. Additionally it enables the distribution of air in an extended part of the occupied zone. Figure 30. The final form of the louvres for each experiment. From top left to bottom right: experiment 3, experiment 4, experiment 5, experiment 6. 4.4. | Experiment 7-­‐8 – 2 louvres in turn open, 6 louvres free The following two experiments, using the same methodology as in the previous case, incorpo-­‐
rated the opening of two louvres on the windward façade. In experiment 7 the first and third louvre is constantly open and the same happens for the second and fourth louvre in experi-­‐
ment 8. The counterpart genes of these louvres are excluded from the evolution process. The remaining 6 louvres are optimised as regards their form. The average velocity, in both experi-­‐
ments is approximately 0.85ms-­‐1. This value is slightly higher than the range of velocity values required by the comfort levels. As a result, the solutions appeared in these experiments does not fulfil the criteria set. Nevertheless, in comparison to the first experiment conducted, they can still provide a better environment as regards airflow. Additionally, as appears in the fitness | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |31 graphs the fitness is kept in lower levels than the ones presented in previous experiments. Figure 31. Graph of the velocities in the occupied zone for height=1.5m. From left to right: experiment 7, experiment 8. In both cases the velocity values are high and the distribution is not uniform. Figure 32. Improvement over generations of three population members (optimisation process). From left to right: Experi-­‐
ment 7, experiment 8. The form of the optimal solutions is depicted in figure 33. The pattern appeared in the previ-­‐
ous set of experiments, where the louvre across the open ones is closed, is not met in this case. Figure 33. The final form of the louvres. From left to right: Experiment 7, experiment 8. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |32 4.5. | Experiment 9-­‐10 – 2 louvres in a row open, 6 louvres free In the remaining two experiments two louvres are left open, but in this case they are chosen to be in a row. The first and second louvre of the windward façade are constrained to be open throughout the experiment 9 and the same happens for the third and fourth louvre in experi-­‐
ment 10. Once more, the genes that participate in the optimisation process are only 6. A less anticipated result appears in this case. This happens since the average velocity takes values around 0.78ms-­‐1 and Q=0.2m3s-­‐1. These values increment the fitness of an instance and thus the optimal solutions achieve to be close to the ones presented in experiments 3-­‐6. As indicated in the fitness graphs (figure 35), this optimisation process achieved higher values of fitness. Figure 34. 3d Graph of the velocities in the occupied zone for height=1.5m. From left to right: experiment 9, experiment 10. Figure 35. Improvement over generations of three population members (optimisation process). From left to right: Experi-­‐
ment 9, experiment 10. One more contributing parameter to the accomplishment of better results in this case is the form of the winning design instances. As shown in figure 36 the openings of the free louvres in the windward façade are smaller than the respective openings in experiments 7-­‐8. This config-­‐
uration enable less air to penetrate into the room. In addition the pattern found in experi-­‐
ments 3-­‐6 appears again in this case. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |33 Figure 36. The final form of the louvres. From left to right: Experiment 9, experiment 10. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |34 Discussion | 5 This paper presents a model that investigates the different design attributes that may appear in the early stages of a design process when airflow-­‐performance criteria are taken into con-­‐
sideration. The study comprises a quantitative analysis of specific parameters, considered as influential factors in the chosen performance field. In addition, the visual representation of the potential design instances was incorporated into the scripted simulation scheme, providing the user a qualitative result for further assessment. As regards the process followed, the objective was to strike an equilibrium between accuracy and time needed for the program to run, since these are two conflicting parameters in the simulation of fluids. Specific simplifications were made, as described in the methodology section, to eliminate the issue of time and in parallel achieve an acceptable level of accuracy, thereby presenting an effective design assistance method. As regards the context of the simulated environment there are several parameters that could also be incorporated into the simulation. Nevertheless, the fundamental aspects of cross ventilation have been addressed with valuable results. This section presents a discussion with respect to the findings that came out of this study, along with some suggestions for fur-­‐
ther improvements of the proposed model. 5.1. | Results overview The objective of this study was the optimisation of the apertures’ size so that an adequate amount of air could pass through a room. The most effective way to introduce sufficient air-­‐
flow indoors is the creation of inlets that are slightly smaller than the outlets. This method en-­‐
ables a cooling effect, channelling the air through the confined inlet opening (RIBA, 2011). This design principle for cross ventilation appeared in the solutions presented in this study. As found in the second experiment, where none of the facades’ components were constrained to be open for human circulation, the optimised form of inlet openings were of smaller size in comparison to the outlet openings, proving that the solution follows the general guidelines of designing for natural ventilation. This indicates that the coupling of FFD and optimisation method, that was set up, is effective and sufficiently creates solutions that can be considered acceptable. Thus, the model generates solutions that could provide informative insight during the process of decision making, even for complex façade configurations as the one imple-­‐
mented in the present experimentation. Moreover, experiments 3 to 10 factor in human circulation and a number of louvres had openings with inclination set at 50°, enabling the remaining louvres to be optimised. Most of these experiments achieved to converge in acceptable solutions as regards human comfort levels. Among the successful experiments, a specific pattern on the configuration of louvres came up in the solutions. Firstly, once more the sizing of inlets and outlets roughly matched. Secondly, the pattern emerged, indicated an interaction between the louvres of the windward and the leeward façade. As a general overview, the louvres placed in opposite positions be-­‐
tween the two sides exhibited inversely proportional inclination creating a better distribution | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |35 of air. This shows a trend towards a specific solution to resolve the addressed problem, con-­‐
verging in an optimal form of the two opposite façades. As presented in this study the method of computing the fitness, combined four different func-­‐
tions. In this light, a system of multiple layers was created for the evaluation process. This method provided the study with a completed set of criteria with the aim to create a robust methodology that allows an accurate evaluation method. However, this method added extra complexity to the problem, which added a high level of difficulty to the optimisation algorithm towards the convergence in one solution. The four parameters taken into account were con-­‐
flicting. For example, throughout the optimisation process a satisfactory result should exhibit average velocity that falls in a specific range of low values and low maximum velocity but high volumetric rate exchange, which are contradictory attributes. Although the problem was com-­‐
plex throughout the majority of the experiments the genetic algorithm managed to converge in solutions that achieve the required standard conditions. However, the coupling of FFD and GA is a computational demanding scheme. For every design instance to be evaluated the fluid simulation should run across the whole domain, creating a time consuming performance analysis method where approximately 6 hours needed until a the GA converges in a solution. The complexity appears visually in the fitness graph of experi-­‐
ment 9 and 10 (figure 35) where the optimisation process remained steady for 300 generation in local optima until a better solution was found. 5.2. | Critical assessment and further developments This paper’s objective is to propose a way of understanding the behaviour of initial designs, in relation to natural ventilation, so that the designer can further assess them. Although the study looks at multiple criteria to evaluate the wind performance in a similar approach other factors such as sun could be resolved providing a more completed methodology. This kind of extension was not in the scope of this study as it would propose a whole new thesis subject to be resolved. As regards problem specific results, the FFD scheme proved to be an effective way to simulate the flow of air. This method was initially created in order to be used in the game industry for the simulation of smoke (Stam, 1999; Stam, 2001). In this light, it was not the primary concern to create an explicitly accurate and informative tool that can exhibit in all its extent the physics behind the movement of air. As a result, most of the information that can be extracted from this simulation relies on the velocities found in the system. This problem was tackled by the use of four different equations that take into account the velocity and volume in x, y, z coordi-­‐
nates, in order to have a more detailed image of the actual outcome of the simulation. Never-­‐
theless, this may be accurate enough for the early stages of the design process, but cannot be the driving force for the decision making at the final stage of design, which will eventually lead to the actual construction. In this case, further studies taking into account fluctuations of pres-­‐
sure due to temperature and buoyancy should be conducted. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |36 In terms of the simulation context, the simplifications made yielded a model that fulfils an ac-­‐
ceptable level of accuracy but certain improvements can be implemented in following studies. As found in environmental studies, the performance of a wind-­‐driven cooling system is directly depended on the magnitude of air velocities appeared indoors and the heat removal by air circulation through the building shell (RIBA, 2011). The model presented assumes isothermal conditions and focuses on the distribution of air velocities. The calculations that take place in this method, are based exclusively on air velocities. The fluctuation of temperatures could also be incorporated, if a more accurate CFD model had been used. In that specific study, this was avoided for a number of reasons, related to the aim of the proposal to be implemented in ear-­‐
ly stages of design and keep a low degree as regards the time needed to meet a solution. However, this could be considered as a further improvement for the fluid solver. The tempera-­‐
ture could be taken into account so as to calculate the buoyancy effect, that forces the air to move from warm to cold positions where the air is more dense. In this case, more accurate results would be extracted from the fluid simulation. Nevertheless, the methodology present-­‐
ed, where the temperature is kept in a standard value, is commonly approved, because the temperature distribution in a building with wind-­‐driven ventilation is approximately homoge-­‐
neous since in such cases the rate of air exchange is usually high (Caririllho de Graca et, al 2002). Moreover, this study highlighted the relation of ventilation performance to the building geom-­‐
etry. This relation can be considered integral for the design process. The notion followed claims that evaluating the form of a structure in parallel with its behaviour, as found by a per-­‐
formance analysis is an important issue (Hensel and Menges, 2008). However, in the present study, the model represents an isolated building in an urban area. The model uses a fixed air-­‐
flow pattern for exterior wind direction and velocity, independent of the possible differentia-­‐
tion of external conditions. The stability of this coupling, form and behaviour, can be disturbed by integrating into the model the surroundings of the building. The microclimate around a structure can affect the ventilation levels (Caririllho de Graca et al., 2002). Considering this, trees and adjacent buildings could be incorporated in the simulation environment to further study the effect of such factors in the airflow patterns and explore the sensitivity of the behav-­‐
iour of a form. In the case presented an ideal situation was simulated where the problem issued was solved in principle. As regards the optimisation algorithm, the number of variables which were needed to be optimised were relatively low. Four louvres for each of the two facades have been tested, as regards the optimum inclination they can exhibit to provide natural air movement indoors. In a real situation, a building would potentially need more louvres to be incorporated into the evolution process. This would complicate further the optimisation procedure. Already in the case study presented, the GA in some experiments tended to provide suboptimal solutions. To overcome this kind of issues, a further step could be an inquiry on the coupling of meta-­‐
heuristics models. A possible improvement could be the development of hybrid heuristics, as | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |37 the combination of a GA with a tabu search algorithm. Tabu search algorithms enable the crea-­‐
tion of a list that consists of the instances of the population which were evaluated as not ful-­‐
filling. In parallel, the process avoids the reappearance of these instances during the optimisa-­‐
tion process. This implementation could possibly eliminate the time needed for the GA to con-­‐
verge, while decreasing the possibility to yield local optima. Thus, such a method could en-­‐
hance the performance of the optimisation process and is suggested for testing in following studies. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |38 Conclusion | 6 The model presented in this study aims to approach the complex phenomenon of wind in an architectural context, so as to facilitate the design of naturally ventilated buildings. The study focused on introducing such systems during the early stages of design. The aftermath of inte-­‐
grating a technique that helps to interpret the impact of optimising the building’s shape for natural ventilation could possibly be the choice of passive cooling systems instead of mechani-­‐
cal air conditioning systems which are more popular in architectural concepts to date. The parameters that should be taken into account when designing a building are numerous, from structural issues and the emerging relationship with the surroundings to aesthetic as-­‐
pects. Shifting the interest on the behavioural qualities of a form, the paper focused on the impact of the building envelope to the internal environment created. The way the external form of the building can affect the internal ambient quality of a space was examined. The shape and size of two facades’ apertures was optimised so that the internal space would fulfil ventilation human comfort levels. The direct relation of the external appearance of a building shell to the qualities the internal space can exhibit in order to accomplish a satisfactory level of occupants sensation, was emphasized. The study focused on wind, simulating its prevalent dynamic attributes, which is a demanding computational task. Nowadays, the improvements in computational tools and the rise of scripting as a method to approach architectural problems provide novel ways to overcome is-­‐
sues that in the past relied on the designer’s expertise. Recently, new paradigms to simulate physical phenomena have emerged and their ability to influence building design has been proved. Such a paradigm is the FFD scheme employed in this study, which is a simpler fluid simulation method in comparison to commercial CFD models, but able to carry out valuable results for the initial stages of design. The methodology achieved to create a performance evaluation platform that can provide enough accurate information to the user so as to gain an insight of the behaviour of a form in relation to wind. In this light, the paper presented the ex-­‐
tent to which simulation of natural airflow can potentially be a contributing parameter in the conception of performance-­‐aware designs. In parallel, the model integrated a multiple criteria fitness function in a genetic algorithm so as to achieve optimum airflow and distribution of air in the occupied zone of a room. A specific case study, where two facades consisting of elastically deformable vertical louvres, was im-­‐
plemented to enable experimentation. A set of fulfilling design instances that have been ana-­‐
lysed for their ventilation performance have been generated. The experiments also took into consideration the potential movement of people through the facade’s openings. This intro-­‐
duced an additional constrain in the optimisation process. Discarding the complexity of the problem most of the results achieved to fulfil the standardised criteria that was set with re-­‐
spect to human comfort. Nevertheless, a refinement of the optimisation algorithm so as to integrate a hybrid-­‐ heuristic model is proposed for future investigation. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |39 The paper pointed out that building on computational achievement available to date can en-­‐
rich the decision making process. The coupling of fluid dynamics simulation and genetic algo-­‐
rithm have been proved an effective way to design taking into account ventilation. Providing a platform for experimentation, this study aims to stimulate a reciprocal conservation between performance and shape. As a general overview, this study underlined that the form of a struc-­‐
ture can be enriched by the information provided from its performance attributes. This way form and behaviour can be conceived as an entity and a structure will be able to exhibit inher-­‐
ent performative qualities from the initial stages of its creation. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |40 References American Society of Heating, Refrigerating and Air-­‐Conditioning Engineers, 2010. ANSI/ASHRAE Stand-­‐
ard 55-­‐2010 Thermal Environmental Conditions for Human Occupancy. Atlanta: ASHRAE. Ash, M., 2006. Fluid Simulation for Dummies, [online] Available at: <http://mikeash.com /pyblog/fluidsimulation-­‐for-­‐dummies.html> [Accessed 1 June 2012]. Autodesk (Autodesk Education Community), 2011. Sustainability Workshop: Wind ventilation. [online] Available at: <http://sustainabilityworkshop.autodesk.com/fundamentals/wind-­‐ventilation> [Accessed 15 June 2012 ]. Awibi, H., D., 1989. Application of Computational fluid Dynamics in Room Ventilation, Building and Envi-­‐
ronment, 24(1), pp. 73-­‐84. Chen, Q. and Zuo, W., 2009. Real time or faster-­‐than-­‐real-­‐time simulation of airflow in buildings, Indoor Air, 19 (1), pp. 33-­‐44. Chen, Q. and Zuo, W., 2007. Validation of fast fluid dynamics for room airflow, Proceedings of the 10th International IBPSA Conference Building Simulation, pp. 980-­‐983, Beijing, China. Cheung, J., and Liu, C., 2011. CFD simulations of natural ventilation behaviour in high-­‐rise buildings in regular and staggered arrangements at various spacings, Energy and Buildings, 43, pp. 1149-­‐1158. Chronis, A., 2010. Generative Fluid dynamics, Integration of fast fluid dynamics and genetic algorithms for wind loading optimization of a free form surface. Diploma thesis. MSc AAC. UCL. Chronis, A., Turner, A., Tsigkari, M., 2011. Generative Fluid Dynamics: Integration of Fast Fluid Dynamics and Genetic Algorithms for Wind-­‐Loading Optimisation of a Free Form Surface, Proceedings of the Sym-­‐
posium on Simulation for Architecture and Urban Design at the 2011 Spring Simulation Multi-­‐
conference, Boston, USA. Cowell, R., 2012. Seminar Review: The Natural Ventilation of UK School Classrooms, UCL, October 2011. Natural Ventilation News, 5, CIBSE Natural Ventilation Group Management Committee, London, pp.7-­‐8. Fanger, P. O., 1970. Thermal Comfort, New York, USA, McGraw-­‐Hill. Fedkiw, R., Stam, J., and Jensen, H., W., 2001. Visual Simulation of Smoke. In SIGGRAPH 2001 Confer-­‐
ence Proceedings, Annual Conference Series , pages 15-­‐22. Garrilho da Graca G., Chen Q., Glicksman L. R., Norford L. K., 2002. Simulation of wind-­‐driven ventilative cooling systems for an apartment building in Bejing and Shangai. Energy and Buildings, 34. pp.1-­‐11. Hensel, M., 2008. Performance oriented design, precursors and potentials. Architectural design, (78) 2, Wiley Academy, London. pp.48-­‐51. Hensel M. and Menges A., 2008. Inclusive performance: Efficiency versus effectiveness. towards a mor-­‐
pho-­‐ecological approach for design. Architectural Design, 78 (2), Wiley Academy, London. pp. 54-­‐63. Hensel, M., Hensel, D., S., Gharleghi, M., and Craig, S., 2008. Towards an architectural history of perfor-­‐
mance. Architectural Design, 78 (2), Wiley Academy, London. pp. 25-­‐37. International Standards Office, 2005. ISO 7730-­‐2 Ergonomics of the thermal environment. Analytical de-­‐
termination and interpretation of thermal comfort using calculation of the PMV and PPD indices and local thermal comfort criteria. Geneva: ISO. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |41 Kato, S. and Murakami, S., 1985. Three-­‐dimensional numerical simulation of turbulent air flow in venti-­‐
lated room by means of 2-­‐equation model. International Symposium Computation Fluid Dynamics, To-­‐
kyo. Kleiven, T., 2003. Natural ventilation in buildings. Architectural concepts, consequences and possibilities. Phd Thesis. Norwegian University of Science and Technology. Knippers J. and Speck T., 2011. Design and construction principles in nature and architecture. Bioinspira-­‐
tion and Biomimetics, 7. McKenzie, A., 2004. Fluid Solver. Java based code [online] Available at: <http://www.multires.calte-­‐
ch.edu/teaching/demos/java/stablefluids.htm> [Accessed 10 July 2012]. Malkawi A. M., 2004. Developments in environmental performance simulation. Automation in Construc-­‐
tion, 13(4). pp.437-­‐445. Malkawi A. M., Srinivasan R. S., Yi Y. K., and Choudhary R., 2005. Decision support and design evolution: integrating genetic algorithms, CFD and visualization. Automation in construction, 14. pp.33-­‐44. Malkawi A. M., and Srinivasan R. S., 2005. A new paradigm for Human-­‐Building Interaction: the use of CFD and Augmented Reality. Automation in Construction, 14(1), pp.71-­‐84. Matlab Website, 2012. Surface and mesh creation. [online] Available at: <http://www.mathworks.co.uk/ help/techdoc/ref/surf.html> [Accessed 30 July 2012]. Mendez, G., 2012, Barriers for natural ventilation in the UK. Natural Ventilation News, 5, CIBSE Natural Ventilation Group Management Committee, London, pp.3-­‐5. Menges, A., 2012. Material Resourcefulness. Activating material information in computational design. Architectural Design, 82(2), Wiley Academy, London. pp. 34-­‐43. Nicol F., J., Hacker, J., Spires, B., and Davies, H., 2009. Suggestion for new approach to overheating diag-­‐
nostics, Building Research & Information, 37 (4), pp. 348-­‐357. Royal Institute of British Architects, 2011. Sustainability Hub. Natural ventilation stuck ventilation. [online] Available at:< http://www.architecture.com/SustainabilityHub/Designstra-­‐tegies/Air/1-­‐2-­‐1-­‐2-­‐
Naturalventilation-­‐stackventilation.aspx>, London: RIBA. [Accessed 9 June 2012 ]. Sheby, N., 2011. Pedestrian Wind Comfort. Incorporation of Agent Based Model with Fast Fluid Dynamics to study pedestrian wind comfort using pedestrian behaviour. Diploma thesis. MSc AAC. UCL. Stam J., 2003. Real-­‐fluid dynamics for Games. In Proceedings of the Game Developer Conference, March 2003. Stam, J., 1999. Stable Fluids, In SIGGRAPH 99 Conference Proceedings, Annual Conference Series, pp.121-­‐128. SOMA, 2012. Thematic pavilion, EXPO 2012 Yeosu, South Korea. [online] Available at:< http://www.so ma-­‐architecture.com/index.php?page=thematic_pavilion&parent=2>. [Accessed 9 June 2012]. Suga, K., Kato, S., Hiyama, K., 2010. Structural analysis of Pareto-­‐optimal solution ets for multi-­‐objective optimization: An application to outer window problems using Multiple Objective Genetic Algorithms, Building and Environment, 45, pp. 1114-­‐1152. Teuffel P., 2008. Responsive building envelopes: Optimization for environmental impact. In Proceedings of the 6th International Conference on Computation of Shell and Spatial Structures IASS-­‐IACM: Spanning Nano to Mega, Cornell University, Ithaca, NY, USA. pp. 5-­‐9. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |42 Turner, A., 2010. Genetic Algorithm. [online] Available at:< http://www.openprocessing.org/ske-­‐
tch/3101>. [Accessed 5 July 2012 ]. Turner, A., 2010. Nurbs surface. [online] Available at:< http://www.openprocessing.org/sketch /8101>. [Accessed 10 June 2012 ]. Zola, Z., 2010. Breeze engine. Hostel, Company Retreat and Training Center in Southern China as a showcase sustainable project for the region.[online] Available at: < http://www.zokazo la.com/south_china_hostel.html>. [Accessed 3 August 2012 ]. | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |43 Appendix Processing code: In this section some integral parts of the encoded scheme are presented A. Basic main functions //setup void setup(){ size(1200, 800, OPENGL); background(255); output = createWriter(aFile); output1 = createWriter(bFile); fluid = new FLUID(numX, numY, numZ, cellSize); //initiate fluid p = new Population(popNum); //initiate the population convergenceCounter=0; //set convergence counter to zero } //draw void draw(){ background(255); lights(); smooth(); if (startFluid) { //start fluid simulation fluid.reset(); //reset the fluid solver boundariesSetup(popCounter); // mesh the surface of the current candidate convergenceCounter=0; //reset the convergence counter startFluid=false; } fluid.FluidCubeStep(); // run the fluid simulation for every frame if (convergenceCounter>60){ //check the convergence every 60 frames if (popCounter<=p.pop.length-­‐1 && !firstGen) { p.pop[popCounter].evaluate(); //evaluate each candidate until the first set of the population is completed startFluid=true; //start fluid simulation fluid.Velocities(); if (popCounter==p.pop.length-­‐1) { // if the whole population is evaluate one time popCounter=0; advancePopCounter=false; } if (advancePopCounter) { popCounter++;//advance the counter until the whole population is evaluated } if (!advancePopCounter) { p.sortPop(); // if the population is complete sort the members firstGen=true; } } else { popCounter=0; genCounter++; p.evolve(); p.pop[popCounter].evaluate(); //evaluate the current candidate p.sortPop(); //sort the population startFluid=true; } } else { convergenceCounter++; } | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |44 //setup the boundaries void boundariesSetup(int count) { for (int i=0; i<fluid.num; i++){ fluid.solid[i]=0; //reset all the cells in the domain } for (int j=0; j<numSurf; ++j) { // mesh the moving front surfaces and reset boundaries fluid.setupBoundaryMovable(p.pop[count].pheno.surf[j], 0.03, 0.03, j); // mesh the moving back surfaces and reset boundaries fluid.setupBoundaryMovable_b(p.pop[count].pheno.surf_b[j], 0.03, 0.03, j); } //mesh the permanent walls and reset boundaries fluid.setupBoundaryPermanent(); } B. Fluid Solver functions. The completed code for the fluid solver implemented can be found in Generative fluid Dynamics (Chro-­‐
nis, 2010). Here only important functions for this study are presented. 3d implementation based on 2d vorticity confinement by McKenzie (2004) <http://www.multires.calte-­‐ch.edu/teaching/demos/java/stablefluids.htm> void vorticity_confinement(float[] vc_x, float [] vc_y, float [] vc_z, float []Vx, float []Vy, float []Vz ) { float x, y, z; for (int k=1; k<N3-­‐1; k++) { for (int j=1; j<N2-­‐1; j++) { for (int i=1; i<N1-­‐1; i++) { int ind = I(i, j, k); if (solid[ind]==0) { x = (Vz[I(i, j+1, k)] -­‐ Vz[I(i, j-­‐1, k)]) * 0.5f -­‐ (Vy[I(i, j, k+1)] -­‐ Vy[I(i, j, k-­‐1)]) * 0.5f; y = (Vx[I(i, j, k+1)] -­‐ Vx[I(i, j, k-­‐1)]) * 0.5f -­‐ (Vz[I(i+1, j, k)] -­‐ Vz[I(i-­‐1, j, k)]) * 0.5f; z = (Vy[I(i+1, j, k)] -­‐ Vy[I(i-­‐1, j, k)]) * 0.5f -­‐ (Vx[I(i, j+1, k)] -­‐ Vx[I(i, j-­‐1, k)]) * 0.5f; curl[ind] = (x-­‐y-­‐z); curl[ind] =abs(curl[ind]); } } } } for (int k=2; k<N3-­‐1; k++) { for (int j=2; j<N2-­‐1; j++) { for (int i=2; i<N1-­‐1; i++) { int ind = I(i, j, k); if (solid[ind]==0) { float Nx = (curl[I(i+1, j, k)] -­‐ curl[I(i-­‐1, j, k)]) * 0.5f; float Ny = (curl[I(i, j+1, k)] -­‐ curl[I(i, j-­‐1, k)]) * 0.5f; float Nz = (curl[I(i, j, k+1)] -­‐ curl[I(i, j, k-­‐1)]) * 0.5f; float len = sqrt(sq(Nx)+sq(Ny)+sq(Nz))+0.0000001f; Nx /= len; Ny /= len; Nz /= len; vc_x[ind] += (Ny*curl[ind] -­‐ Nz*curl[ind]) ; vc_y[ind] += (Nz*curl[ind] -­‐ Nx*curl[ind]) ; vc_z[ind] += (Nx*curl[ind] -­‐ Ny*curl[ind]) ; } } } } set_bnd(1,vc_x,N1,N2,N3); set_bnd(2,vc_y,N1,N2,N3); | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |45 set_bnd(3,vc_y,N1,N2,N3); } Internal movable boundaries (louvres) // windward facade void setupBoundaryMovable(Surface surf, float du, float dv, int surfMove){ for (float u = 0.0; u <=1.0; u+= du){ for (float v = 0.0; v <=1.0; v+= dv){ PVector pt = surf.surfPos(u, v); int i = floor(pt.x/cell)+(17+(surfMove*(xWidth_louv/cell))); int j = floor(pt.y/cell)+60; int k = floor(pt.z/cell); if ( i>1 && i<N1-­‐2 && j>1 && j<N2-­‐2 && k>=1 && k<N3-­‐2){ solid[I(i, j, k)]=2; solid[I(i, j+1, k)]=2; solid[I(i+1, j, k)]=2; solid[I(i, j, k+1)]=2; } } } } // leeward facade void setupBoundaryMovable_b(Surface surf_b, float du, float dv, int surfMove){ for (float u = 0.0; u <=1.0; u+= du){ for (float v = 0.0; v <=1.0; v+= dv){ PVector pt = surf_b.surfPos(u, v); int i = floor(pt.x/cell)+(17+(surfMove*(xWidth_louv/cell))); int j = floor(pt.y/cell)+101; int k = floor(pt.z/cell); if ( i>1 && i<N1-­‐2 && j>1 && j<N2-­‐2 && k>=1 && k<N3-­‐2){ solid[I(i, j, k)]=2; solid[I(i, j-­‐1, k)]=2; solid[I(i-­‐1, j, k)]=2; solid[I(i, j, k-­‐1)]=2; } } } } Internal permanent boundaries (floor, ceiling, lateral walls) void setupBoundaryPermanent(){ for (int i=1; i<N1-­‐1; ++i){ for (int j=1; j<N2-­‐1; ++j){ for (int k=1; k<N3-­‐1; ++k) if ( i>=21 && i<=53 && j>={60 && j<=101 ){ solid[I(i, j, 29)]=2; //ceiling solid[I(i, j, 30)]=2; solid[I(i, j, 2)]=2; //floor solid[I(i, j, 3)]=2; } else if ( j>=60 && j<=101 && k>=4 &&k<=28){ //side walls solid[I(21, j, k)]=2; //left solid[I(22, j, k)]=2; solid[I(52, j, k)]=2; //right solid[I(53, j, k)]=2; } else if (i>=21 &&i<=23 && k>=4 && k<=28) { solid[I(i, 60, k)]=2; solid[I(i, 61, k)]=2; } else if (i>=51 &&i<=53 && k>=4 && k<=28) { solid[I(i, 100, k)]=2; solid[I(i, 101, k)]=2; | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |46 } } } } } Functions that influence the fitness function a. Average velocity float averageVelocity(){ float averageVel=0.0; int counter = 0; for (int i=26; i<=48; ++i){ for (int j=64; j<=97; ++j){ for (int k=5; k<=22; ++k){ int ind = I(i, j, k); if (solid[ind] == 0){ averageVel += sqrt(sq(Vx[ind])+sq(Vy[ind])+sq(Vz[ind])); counter++; } } } } averageVel /= counter; averageVel*=100; return averageVel; } b. Maximum velocity float maxVelocity() { float maxV=-­‐1000000.0; for (int i=26; i<=48; ++i){ for (int j=64; j<=97; ++j){ for (int k=5; k<=22; ++k){ int ind = I(i, j, k); if (solid[ind] == 0){ PVector Velocity=new PVector(Vx[ind], Vy[ind], Vz[ind]); float V=Velocity.mag(); if (V>maxV) { maxV=V; } } } } } maxV*=10; return maxV; } c. Standard deviation of mean velocity float standardDeviationVel(float average) { float standardDevV=0.0; float variance=0.0; int counter2 = 0; for (int i=26; i<=48; ++i){ for (int j=64; j<=97; ++j){ for (int k=5; k<=22; ++k){ int ind = I(i, j, k); if (solid[ind] == 0){ variance+= sq(average-­‐( sqrt(sq(Vx[ind])+sq(Vy[ind])+sq(Vz[ind])))); counter2++; } } | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |47 } } variance/=(counter2-­‐1); standardDevV=sqrt(variance); return standardDevV; } d. Volumetric rate axchange of air float volumetricRate(){ float avVel=0.0; float volumetricRt=0.0; float area=0; int counter = 0; //find the average velocity for the specific cells 1 cell after the openings facade for (int i=26; i<=48; ++i) { for (int k=5; k<=20; ++k){ int ind = I(i, 62, k); if (solid[ind] == 0){ avVel += sqrt(sq(Vx[ind])+sq(Vy[ind])+sq(Vz[ind])); counter++; } } } avVel /= counter; // find the area that these cells cover area=((48-­‐26)*(20-­‐5))/10; // Q=V*A volumetric rate exchange of fluids volumetricRt=area*avVel; return volumetricRt; } //save velocities to be used in Matlab surf function void Velocities(){ for (int i=26; i<48; ++i) { for (int j=64; j<97; ++j) { output1.println( i +" , "+ j +" , "+(mag(Vx[I(i, j, 15)], Vy[I(i, j, 15)], Vz[I(i, j, 15)])); output1.flush(); } } } C. Phenotype The phenotype class from the genetic algorithm that includes the fitness function class Phenotype{ Surface[] surf; Surface[] surf_b; int [] stable= { 0, 2, 4, 6, 9 }; float averVel; float maxVel; float stanDev; float volRt; Phenotype (Genotype g){ surf= new Surface[numSurf]; surf_b= new Surface[numSurf]; int counter=0; for (int j=0; j<numSurf; ++j){ surf[j]= new Surface(); | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |48 surf_b[j]= new Surface(); } //intervene in the GA with a key function if (randomValue0==true) { index= 0; randomValue0=false; } if (randomValue1==true) { index= 1; randomValue1=false; } if (randomValue2==true) { index= 2; randomValue2=false; } if (randomValue3==true) { index= 3; randomValue3=false; } for (int j=0; j<g.genes.length; ++j){ if (j%2==0 && j!=stable[index]) { //front surfaces surf[counter].ctrl_pts[0][1].x=(g.genes[j]*9)+20; surf[counter].ctrl_pts[1][1].x=(20+(xWidth_louv/N_u))-­‐((surf[counter].ctrl_pts[2][1].x-­‐ surf[counter].ctrl_pts[0][1].x)-­‐(20+(xWidth_louv/N_u)))*Gaussx(1.0, 0.0); surf[counter].ctrl_pts[0][1].y-­‐=(g.genes[j]*9); surf[counter].ctrl_pts[2][1].y=(surf[counter].ctrl_pts[0][1].y)* Gauss(1.8, 0); surf[counter].ctrl_pts[1][1].y=(surf[counter].ctrl_pts[0][1].y-­‐surf[counter].ctrl_pts[2][1].y)* Gauss(0.5, 0); } else if (j%2==0 && j==stable[index]) { //front surfaces – a louvre opens completely surf[counter].ctrl_pts[0][1].x=45+20; surf[counter].ctrl_pts[1][1].x=(20+(xWidth_louv/N_u))-­‐((surf[counter].ctrl_pts[2][1].x-­‐
surf[counter].ctrl_pts[0][1].x)-­‐(20+(xWidth_louv/N_u)))*Gaussx(1.0, 0.0); surf[counter].ctrl_pts[0][1].y=45; surf[counter].ctrl_pts[2][1].y=(surf[counter].ctrl_pts[0][1].y)* Gauss(1.8, 0); surf[counter].ctrl_pts[1][1].y=(surf[counter].ctrl_pts[0][1].y-­‐surf[counter].ctrl_pts[2][1].y)* Gauss(0.5, 0); surf[counter].ctrl_pts[0][2].x=45+20; surf[counter].ctrl_pts[1][2].x=(20+(xWidth_louv/N_u))-­‐((surf[counter].ctrl_pts[2][1].x-­‐
surf[counter].ctrl_pts[0][1].x)-­‐(20+(xWidth_louv/N_u)))*Gaussx(1.0, 0.0); surf[counter].ctrl_pts[0][2].y=45; surf[counter].ctrl_pts[2][2].y=(surf[counter].ctrl_pts[0][1].y)* Gauss(1.8, 0); surf[counter].ctrl_pts[1][2].y=(surf[counter].ctrl_pts[0][1].y-­‐surf[counter].ctrl_pts[2][1].y)* Gauss(0.5, 0); } else { // back surfaces surf_b[counter].ctrl_pts[2][1].x-­‐=g.genes[j]*9; surf_b[counter].ctrl_pts[1][1].x=(20+(xWidth_louv/N_u))+((surf_b[counter].ctrl_pts[0][1].x-­‐
surf_b[counter].ctrl_pts[2][1].x)-­‐(20+(xWidth_louv/N_u)))* Gaussx(1.0, 0.0); surf_b[counter].ctrl_pts[2][1].y=g.genes[j]*9; surf_b[counter].ctrl_pts[0][1].y=(surf_b[counter].ctrl_pts[2][1].y)*Gauss(2.9, 0); surf_b[counter].ctrl_pts[1][1].y=(surf_b[counter].ctrl_pts[2][1].y-­‐surf_b[counter].ctrl_pts[0][1].y)*Gauss(0.5, 0); counter++; } } } //Gauss float Gauss(float k, float j) { float f = exp(-­‐sq(j-­‐k)/(2*sq(1.0))); return f; } | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |49 float Gaussx(float k, float j) { float f = exp(-­‐sq(j-­‐k)/(2*sq(0.8))); return f; } void draw() { for (int i=0; i<numSurf; ++i) { pushMatrix(); translate(i*(xWidth_louv-­‐1), 100, -­‐350); surf[i].drawSurf(0.09, 0.09); popMatrix(); pushMatrix(); translate(i*(xWidth_louv-­‐1), 200, -­‐350); surf_b[i].drawSurf(0.09, 0.09); popMatrix(); } } //multiple fitness criteria float evaluate(){ averVel = fluid.averageVelocity(); maxVel = fluid.maxVelocity(); stanDev = fluid.standardDeviationVel(averVel); volRt= fluid.volumetricRate(); float fitness=0.0; if (averVel>0.4 && averVel<0.8) { //if the average velocity is in a specific range and volumetric rate of flow increases the fitness becomes better fitness+= (averVel+volRt)/2; //standard deviation and maximum velocity should minimise to optimise the fitness fitness-­‐= (stanDev+maxVel)/2; } else{ fitness+= volRt; fitness-­‐= (stanDev+maxVel+averVel)/3; } return fitness; } } | UCL MSc AAC Sept.2012 Chrysanthi-­‐Sandy Karagkouni |50