doi: 10.1016/j.procs.2015.05.476

CFD post-processing in Unity3D

Matthias Berger1,2 and Verina Cristie2 1ETH Zürich, Zurich, Switzerland.

2Future Cities Lab, SEC, Singapore. {mberger, verina}@arch.ethz.ch

Abstract In architecture and urban design, urban climate on is a strong design criterion for outdoor thermal comfort and building’s energy performance. Evaluating the effect of buildings on the local climate and vice versa is done by computational fluid dynamics (CFD) methods. The results from CFD are typical-ly visualized through post-processing software closely related to pre-processing and simulation soft-ware. The built-in functions are made for engineers and thus, it lacks user-friendliness for real-time exploration of results for architects. To bridge the gap between architect and engineer we propose vis-ualizations based on game engine technology. This paper demonstrates the implementation of CFD to Unity3D conversion and weather data visualization. Keywords: CFD post-processing, Unity3D, urban climate, urban designs, visualization

1 Introduction Urban planning is nowadays an interdisciplinary endeavor, as shown by ETH Zürich’s Future Cit-

ies Laboratory*, the Centre for Liveable Cities in Singapore†, or the British Future Cities Catapult‡. Visualizations and models have been the channel of choice for communication in architecture. With the interdisciplinary input from citizens, engineers, politicians and other experts, relying on a visually oriented language as the common denominator is becoming crucial for the effectiveness of the inter-disciplinary approach (Grêt-Regamey et al., 2013; Pettit et al., 2012).

One field of input to urban planning is coming from joint efforts in computational science, engi-neering and urban climate studies: The advancement of computational power in combination with re-duced operating costs in the last two decades enabled computational fluid dynamics (CFD) methods to enter urban climate studies on meso scale, with resolution of 500m – 25km, and micro scale, with res-olution of 10cm – 10m (Ashie and Kono, 2011; Mochida et al., 1997). There is still an upper limit of available resources for meso scale simulations, which often prefers isothermal Lattice Boltzmann

* www.futurecities.ethz.ch † www.clc.gov.sg ‡ futurecities.catapult.org.uk

Procedia Computer Science

Volume 51, 2015, Pages 2913–2922

ICCS 2015 International Conference On Computational Science

Selection and peer-review under responsibility of the Scientific Programme Committee of ICCS 2015c© The Authors. Published by Elsevier B.V.

2913

methods with focus on wind flow over more complex turbulence models due to significant less com-putational time (Miao and Shen, 2014).

Several researchers have already pointed out the difference in requirements of engineering visuali-zations for engineers only compared to interdisciplinary (or non-expert) stakeholders (Stone et al., 2014). Existing engineering visualization tools such as ParaView§, Tecplot 360**, or ANSYS Fluent†† are lacking of user-friendliness for those outside the domain. Suggested was a user-centric focus by implementing game-like environments (Greenwood et al., 2009), or directly employing game engines like Unity3D (Indraprastha and Shinozaki, 2008, 2009; Stojanovic et al., 2014). The object of visuali-zation is multi-dimensional data, and as Donalek et al. highlight, it is not about raw (or big) data, “it is about discovery and understanding of meaningful patterns hidden in the data” (2014). A different ap-proach on interactive visualizations can be described as computer aided drafting (Pauwels, 2014), where interoperability is the challenge. Tools for stakeholders can help stakeholders to understand complex scientific aspects (Berger, 2012; Grêt-Regamey et al., 2013), and furthermore help research-ers to study the stakeholder (Prendinger et al., 2013).

Our goal was to provide a visualization tool called CFDtoUnity3D based on CFD simulation re-sults from professional engineering software. This tool is to be differentiated from visualization tools for architects based on visual correlation of epiphenomena, like in the ‘Specialist Modelling Group’ of Foster + Partners or the space syntax framework by Hillier (Hillier and Hanson, 1984). Unity3D has several benefits over proprietary CFD post-processing software: 1) it does not require a license to exe-cute, 2) it is platform independent, 3) can be used web-based or on tablets, 4) is optimized for visual performance, and 5) does not require expert knowledge and skills to run. It is then open to all kinds of stakeholders on a multitude of devices consequently. On the downside, the features for analyzing CFD results in established post-processing software have to be adapted to Unity3D and coded from scratch. The results in Unity3D are far less accurate than established post-processing software, but an architect, urban designer or planner will get much more qualitative performance insight of a specific design. One do not need to worry about minimizing uncertainty of parameters (wind speed, temperature, etc.) for any given point in the simulation.

A true design cycle including CFD would be out of sync for a real-time visualization due to exten-sive computational needs of CFD. However, the CFDtoUnity3D tool shall enable a rapid analysis cy-cle as in (Birch et al., 2013). The post-processing commonly done in CFD can be as complex as the actually numerical simulation, especially for visualizing the streamlines of wind flow (McLoughlin et al., 2010). In their excellent review paper McLoughlin et al. emphasize the problems resulting from computational time and irregular grids on the data side as well as visual clutter and occlusion when visualizing in 3D.

2 Methodology

2.1 Visualization workflow Our input consists of 3D building models and its CFD simulation. Buildings were modeled with

3dsMax or Rhino and imported to Unity3D using the .obj or .fbx format. CFD simulation results were exported to text format (.txt or .csv) from ANSYS Fluent. Temperature, humidity, and pressure data was preprocessed into 3D spatial data structures so that it will be easier to be used during interactive exploration. Unity3D finally will do the rendering to be shown on display(s) and the user will be able

§ http://www.paraview.org ** http://www.tecplot.com/products/tecplot-360/ †† http://www.ansys.com/Products/Simulation+Technology/Fluid+Dynamics/Fluid+Dynamics+Products/ANSYS+Fluent

CFD post-processing in Unity3D Matthias Berger and Verina Cristie

2914

to interact and navigate through the environment by using a controller. Figure 1 provides a flowchart showing the implementation of our tool.

2.2 Pre-processing and rendering Pre-processing

ANSYS Fluent produces mesh and nodes in the simulation space with irregular distances from each other. The number of nodes nearer the building geometry compared to the number of nodes in the spaces where there is no geometry like sky or corners of the simulation space is higher. In our study case we use existing CFD simulations (Papadopoulou et al., 2015; Vernay et al., 2015) and 3D-models generated by Lidar and aerial photogrammetry (Gruen, 2013) of the CREATE Tower within Universi-ty Town campus in Singapore (1°18'14.5"N 103°46'25.1"E). The simulation space from CFD extends to 2.0×1.2×0.3 km with approximately 14 million nodes. It employs hexahedral mesh with minimum cell size of 5cm. Steady RANS equation in combination with the Realizable k-epsilon model is used to describe the turbulent flow. Four simulations are executed with specific wind directions occurring in Singapore. A logarithmic wind speed profile is used at the inlet with a wind speed of 3.5 m/s at 16m heights.

Figure 1: Flowchart of the CFD data visualization. On the input side in red there is the scenery and in green

the numerical results of the CFD simulation. Blue represents the processing around the Unity3D game engine and orange is the communication between visualization and user.

Pre-processing is done to divide nodes into spatial data structures such that accessing information of temperature, pressure, and humidity given a position in the space becomes easier. The most straightforward method is to divide the space into a regular grid (in k×l×m). The grid resolution is de-cided such that the size of each cell is (space length/k × space width/l × space height/m). Given a spe-cific position in which we want to know what the temperature, pressure, or humidity value is, we

CFD post-processing in Unity3D Matthias Berger and Verina Cristie

2915

could just divide it by the grid resolution to find its index in each X, Y, and Z-axis. The grid resolution therefore corresponds to the level of detail. If the grid resolution is too big, there will be lack of details – each cell encompasses a too big space. Conversely, if the grid resolution is too small, we will have a lot of details yet it will consume much computer memory and reduce the interactivity frame rate of the program. As some of the cells will be empty during conversion from mesh to grid due to unequal dis-tribution of the nodes in the space, we apply linear averaging from surrounding data points to fill up empty cells. The grid is saved in the three-dimensional array structure of C#.

We decided to introduce a second data structure – octree (Meagher, 1980). Octree data structures allow subdivision of space to into eight regular children spaces. This means that we are able to use bigger octrees for space where there are not many nodes, and use smaller octrees (subdivided octree) on the space near the buildings. This will allow more level of details near the building. If there is any empty node, the values could be interpolated based on the other children octrees on the same space. In our implementation, the .xml format is used to store the octree after pre-processing due to natural tree structure the .xml format has, making the file easily readable. The user has to define the smallest octree size, which is the opposite of the normal procedure. Each level we go up will increase the cell size by factor two (or cell volume by factor eight). We are to find the maximum node size such that it can en-compass the whole simulation space. If needed, empty space is added to the dimensions of the simu-lated space from CFD to return integer values of space. Once done, we can start to construct our oc-tree using the given data. Each of the processed octree cells contains information of center, (octree-) level, temperature, pressure, and humidity.

Rendering & Lightmapping

Like other game engines, Unity3D is also able to support lighting and rendering systems. Unity3D provides graphics classes and renderer classes with its own line and mesh classes for visualization. We use the GL-class that provides a low-level graphics library capable to issue commands similar to OpenGL. Given the raw text data then we construct triangles and planes for visualization. Unity3D provides the Update() function where each point will be updated after certain frame rate for animation. Since the building is static, lighting are precomputed and saved into Unity’s Lightmap so that it will not consume computer resource during runtime and a frame rate supporting interactivity is achievable. We explicitly decided not to use surface textures of buildings and terrain unlike others (Piga et al., 2013; Wang et al., 2010), to preserve a clear view on the simulation results.

2.3 GUI and interaction Navigation

To enable interactive exploration, we allow the user to have two navigation modes: Walkthrough (see Fig. 5 right) and Helicopter (see Fig. 5 left) mode. In Walkthrough mode, we have four degree of freedom for exploration: right-left movement, forward-backward movement, turning left-right (yaw), and tilting upward-downward (pitch). The Helicopter mode allows one more degree of freedom, up-ward-downward movement.

Map Functionality

We also provide map functionality in which the user can have an overview of the whole building landscape (as in Fig. 2). User’s position is marked by a red arrow on the map. A small map is shown on the top right of the screen to give user awareness of the immediate surroundings. The zooming lev-el of surrounding map can be adjusted from the Settings panel. To have an overall view, users can choose to maximize the map. Top down view of the whole area can be shown on the full screen (see Fig. 2 right). On this screen, user is able to perform two operations: teleport (1) – such that user can immediately access place that he/she wants to focus out of the entire landscape, and measurement (2) – such that the user can correlate what they see with actual measurement (in meters). In addition, we also provide cursor positions at the center of screen such that user is able to measure actual height of

CFD post-processing in Unity3D Matthias Berger and Verina Cristie

2916

building in meters. This is done by using the ray casting from the center of camera to center of screen to know where the ray hit the building, and hence the height could be calculated.

Figure 2: CFDtoUnity3D GUI (left). Map functionality in maximized zoom mode (right). Settings

The user is also allowed to choose certain options or visualization modes that he/she wants to see in the Settings panel (see Fig 2 left):

- Scale Panel to give the actual data value corresponding to the color shown in visualization. - Positional Data Label to give information of wind speed, temperature, humidity, and pressure

value of the current position of user. - Visualization Option to choose what data visuals to be shown (wind streamlines/ temperature/

humidity/ pressure). Only one data type can be seen at one time. Wind animation can be played concurrently with other visualization options. Density level of static streamlines visualization is also adjustable.

- Navigation and Control Panel to provide user with navigational keys and controls information. Once users are familiar with the keys, they can disable this panel.

- Data Culling Mask to show data visualization on the top-down view map panel. - Global Movement activates same movement for all available screens while using multiple screens. - Global Settings activates same settings for all available screens.

3 Results Wind Streamlines Tube

Both wind streamlines tube and wind animation are based on wind streamlines data from CFD (streamline ID, Position, Time, and Speed as in Fig. 1). The tube is static (see Fig. 3 left) and time in-formation is not used. A procedural mesh generation method is being used to generate 3D tube like mesh from each line. The radius of the tube should not be too small avoiding streamlines to look like fine hair lines, neither it should be too big that lines will start overlapping and occlude each other. To create tube, we create circle which center position is the streamline position and is perpendicular to the streamline direction. Once the circles were generated at each point of streamlines, outer point of neighboring circles are connected to form triangles and thus the tube surface is created.

Wind Streamline Animation

The wind position needs to be calculated dynamically on every time step in the animation based on the given discrete data (see Fig. 3 right). Taking account of the curve nature of wind, we interpolated points in between data points to get intermediate wind position. Catmull-Rom Splines are used to pro-vide cubic spline with smooth tangent (Catmull and Rom, 1974). This resulted in streamline animation that represents wind speed and direction at each point of time.

CFD post-processing in Unity3D Matthias Berger and Verina Cristie

2917

Figure 3: Wind streamlines tunnel (left). Wind streamline animation (right).

Users are also allowed to set the number of waves to be visualized. Waves are produced at time in-terval, meaning that at each new wave, animation of streamline is run from the starting point. Setting a higher number of waves allows better understanding of the wind movement in the overall simulation area. The wind trail is generated using the LineRenderer class of Unity3D. Blue transparent texture is passed to LineRenderer settings to give the effect of moving wind.

Wind Wavefront Animation

Wind wavefront animation (see Fig. 4) used the same points as generated by wind streamline ani-mation. However, instead of drawing trails we are drawing a meshed plane. Initial points of the wind streamlines are coming from a wind outlet at one end of the simulation area and will travel through the area to the other opposite end. While the streamlines come from the same initial plane, the given wind ID from ANSYS does not come in regular sequences. Thus, we need to triangulate the initial points. Delaunay (Delaunay, 1934) triangulation is applied to find the initial arrangement of the ID that makes up the triangles. The subsequence animation steps will then follow the same triangle sequence. As the wind moves, however, points will move possibly too far from each other and triangle size be-comes too big. We stopped drawing these artifact triangles, as the focus is the wave front of the wind. This is a compromise between understandability and accurateness. In many other simulation cases the tail of a turbulence or vortex would be of higher interest and hence our methodology not applied. The color of the points will change according to the speed information extracted from ANSYS data. Using this wave front animation, one can clearly see how the wind plane transformed when the wind first touched the building surface and subsequently as it travels through the buildings.

Figure 4: Wind wavefront animation movement at different time steps.

Pressure & Humidity Plane To overcome depth perception, visibility and occlusion issue, we allow dynamic exploration such

that temperature, pressure, or humidity data are presented in a plane that is always facing the user (see Fig. 5). Given the user’s position in 3D space, the forward vector is calculated. Users are allowed to scroll in and out to adjust the distance to the plane. We set the plane width to span across the building. Once we have calculated the orthogonal line on the ground (base of the plane), we then sample points

CFD post-processing in Unity3D Matthias Berger and Verina Cristie

2918

along the line to find temperature and other information of this point from our spatial data structure. By expanding it to the Y-direction, we can then get the points needed to draw the plane.

Figure 5: Pressure plane viewed from Helicopter mode (left). Temperature plane viewed from Walkthrough mode (right).

4 Discussion Multiple Screens with Controller

We presented our early prototype to a group of architects and researchers and one of the valuable feedbacks we received was to have different images to compare so that user will be able to infer mean-ingful insight from the data. This resulted in multiple screens implementation of CFDtoUnity3D. Pos-sible combinations that can be done on multiple screens: (1) Same building with same data on all screens, with different positioning/ different settings on each screens; (2) Same building with different data (example: same building with different wind directions); (3) Different building with different data (example: buildings that are different in their geometries – we can see the difference it will cause to the wind direction). Additional functionality could be exploited by using multi-touch surface screens (Chen and Schnabel, 2011).

In our initial testing with our 35 megapixel video wall (8k in 4x4 2k screens) driven by an Intel®

Xeon® CPU E5-2687w @3.10GHz dual processor, 32GB RAM, four Nvidia Quadro K5000, an inter-active frame rate at >10fps is achievable when we created less than ten visualization windows. The multiple screen mechanism is implemented using Client-Server architecture on the same computer so that each client can run the data independently. A default dialogue is initialized on program start on the server screen. When our program is launched, the default dialogue (see Fig. 6 left) will prompt the user to enter the number of screens which will be the clients. The user has to set data sources for every screen. Once all the screens are launched, the user can select any of the screens to be controlled. Glob-al movement and global settings can then be activated.

Figure 6: Default dialogue (left). User exploration of multiple screens using the Wii U Pro Controller (right).

CFD post-processing in Unity3D Matthias Berger and Verina Cristie

2919

Lastly, as we want to enable the exploration in the most user-friendly way, we implemented the Nintendo Wii U Pro Controller (see Fig. 6 right) as the navigational control in the environment. The Settings panel also can be activated from the controller. We used JoyToKey‡‡, a shareware, to override button controls of the controller, which is connected to PC using a USB adapter.

Since architect’s and urban planner’s understanding, behavior and feedback while exploring our

CFDtoUnity3D visualization tool should strongly influence the implementation, we plan extensive user studies throughout 2015. Visual effects, appearance and features are subject to change and further development.

5 Conclusion In agreement with Chen and Schnabel (2011) we see the future of urban planning to be most likely

successful by opening up towards democratic and participatory design (Kunze et al., 2012). This can be achieved through breaking the monopoly on information and knowledge, in our case held by archi-tects and engineers, by an alternative approach on simulation and visualization. As we demonstrated with the CFDtoUnity3D visualization tool exploration, navigation, and understanding can be enhanced by game-like environments. This setup is empowering the user (or stakeholder of a design/ planning process, which usually excludes engineers) to make informed decisions based on accurate CFD simu-lations by means of a representative and qualitative visualization.

Special attention has been put on the interface or communication channel, to use Shannon’s (1948) words. Game controllers and multi-touch screens have been implemented as well as displays beyond 4k resolution in single and multiple screen view mode. Optimization of the computational perfor-mance by minimizing needs (e.g. memory) is ongoing, but on the other hand depending on the visuali-zation platform. By using Unity3D we permit a wide range of devices to visualize the simulation re-sults, and no commercial software license is required to run the application.

As the research is ongoing, user studies are done consecutively and expected to refine the func-tionality and visual appearance.

References [1] Ashie, Y., Kono, T., 2011. Urban scale CFD analysis in support of a climate sensitive

design for the Tokyo Bay area. International Journal of Climatology 31, 174-188.

[2] Berger, M., 2012. Urban heat-balling - A review of measures on reducing heat in tropical and subtropical cities, in: Sheng, T.K., Hing, P. (Eds.). Sustainable Future Energy 2012 & 10th Sustainable Energy and Environment, Brunei Darussalam pp. 445-451.

[3] Birch, D., Kelly, P.H., Field, A.J., Simondetti, A., 2013. Computationally unifying urban masterplanning. Proceedings of the ACM International Conference on Computing Frontiers. ACM, p. 32.

[4] Catmull, E., Rom, R., 1974. A class of local interpolating splines. Computer aided geometric design 74, 317-326.

[5] Chen, I.R., Schnabel, M.A., 2011. Multi-Touch: The Future of Design Interaction. CAADFutures 2011 : Designing Together, Liège, pp. 557-571.

‡‡ http://joytokey.net/en/

CFD post-processing in Unity3D Matthias Berger and Verina Cristie

2920

[6] Delaunay, B., 1934. Sur la sphère vide. A la mémoire de Georges Voronoï. Bulletin de l'Académie des Sciences de l'URSS, Classe des sciences mathématiques et naturelles, No. 6: 793–800.

[7] Donalek, C., Djorgovski, S., Davidoff, S., Cioc, A., Wang, A., Longo, G., Norris, J.S., Zhang, J., Lawler, E., Yeh, S., 2014. Immersive and Collaborative Data Visualization Using Virtual Reality Platforms. arXiv preprint arXiv:1410.7670.

[8] Greenwood, P., Sago, J., Richmond, S., Chau, V., 2009. Using game engine technology to create real-time interactive environments to assist in planning and visual assessment for infrastructure. 18th World IMACS/MODSIM Congress, Cairns, pp. 2229-2235.

[9] Grêt-Regamey, A., Celio, E., Klein, T.M., Wissen Hayek, U., 2013. Understanding ecosystem services trade-offs with interactive procedural modeling for sustainable urban planning. Landscape and Urban Planning 109, 107-116.

[10] Gruen, A., 2013. Generating a 3D Campus Model. GEOInformatics 16, 30-36.

[11] Hillier, B., Hanson, J., 1984. The social logic of space. Cambridge University Press, Cambridge.

[12] Indraprastha, A., Shinozaki, M., 2008. Constructing virtual urban environment using game technology. 26th eCAADe Conference: Architecture in Computro Antwerpen pp. 359-366.

[13] Indraprastha, A., Shinozaki, M., 2009. The investigation on using Unity3D game engine in urban design study. Journal of ICT Research and Applications 3, 1-18.

[14] Kunze, A., Burkhard, R., Gebhardt, S., Tuncer, B., 2012. Visualization and Decision Support Tools in Urban Planning. Digital Urban Modeling and Simulation. Springer, pp. 279-298.

[15] McLoughlin, T., Laramee, R.S., Peikert, R., Post, F.H., Chen, M., 2010. Over Two Decades of Integration Based, Geometric Flow Visualization. Computer Graphics Forum. Wiley Online Library, pp. 1807-1829.

[16] Meagher, D., 1980. Octree Encoding: A New Technique for the Representation, Manipulation and Display of Arbitrary 3-D Objects by Computer. Rensselaer Polytechnic Institute (Technical Report IPL-TR-80-111)

[17] Miao, H., Shen, J., 2014. Developing Three-Dimensional CFD City Model Based on Public Accessible Information for Street-Level Wind Risk Assessment. 19th Australasian Fluid Mechanics Conference, Melbourne.

[18] Mochida, A., Murakami, S., Ojima, T., Kim, S., Ooka, R., Sugiyama, H., 1997. CFD analysis of mesoscale climate in the Greater Tokyo area. Journal of Wind Engineering and Industrial Aerodynamics 67, 459-477.

[19] Papadopoulou, M., Raphael, B., Smith, I.F.C., Sekhar, C., 2015. Optimal sensor placement for time-dependent systems: application to wind studies around buildings. Journal of Computing in Civil Engineering [under review].

[20] Pauwels, P., 2014. Supporting Decision-Making in the Building Life-Cycle Using Linked Building Data. Buildings 4, 549-579.

[21] Pettit, C., Widjaja, I., Russo, P., Sinnott, R., Stimson, R., Tomko, M., 2012. Visualisation support for exploring urban space and place. XXII ISPRS Congress, Melbourne.

[22] Piga, B.E., Cibien, L., Legnani, A., Mungo, S., Zappa, D., Signorelli, V., Morello, E., 2013. A sense-archive of urban space. EAEA-11 conference 2013, Milano, pp. 251-258.

CFD post-processing in Unity3D Matthias Berger and Verina Cristie

2921

[23] Prendinger, H., Gajananan, K., Zaky, A., Fares, A., Molenaar, R., Urbano, D., van Lint, H., Gomaa, W., 2013. Tokyo Virtual Living Lab: Designing Smart Cities Based on the 3D Internet. IEEE Internet Computing 17, 30-38.

[24] Shannon, C. E., (1948). A Mathematical Theory of Communication. The Bell System Technical Journal. p. 55.

[25] Stojanovic, V., Falconer, R., Blackwood, D., Paterson, G., Fleming, M., Bell, S., 2014. Interactive visualisation of heat loss and gain for early-stage energy appraisal of the built environment. Procs 30th Annual ARCOM Conference, Portsmouth, pp. 743-752.

[26] Stone, R., Guest, R., Pahl, S., Boomsma, C., 2014. Exploiting gaming technologies to visualise dynamic thermal qualities of a domestic dwelling. BEHAVE2014 – Behavior and Energy Efficiency Conference.

[27] Vernay, D.G., Raphael, B., Smith, I.F.C., 2015. A Model-based Data-interpretation Framework for Improving Wind Predictions Around Buildings. Journal of Wind Engineering & Industrial Aerodynamics [under review].

[28] Wang, S., Mao, Z., Zeng, C., Gong, H., Li, S., Chen, B., 2010. A new method of virtual reality based on Unity3D, Geoinformatics. 18th International Conference on. IEEE, pp. 1-5.

CFD post-processing in Unity3D Matthias Berger and Verina Cristie

2922