Dispersion Simulation and Visualization for Urban Security Authors: F. Qiu, Y. Zhao, et al. Visualization II Instructor: Jessica Crouch Presenter: Mike.

Dispersion Simulation and Visualization for Urban Security

Authors: F. Qiu, Y. Zhao, et al.

Visualization II

Instructor: Jessica Crouch

Presenter: Mike Jones

Problem

• Dispersion simulation is computationally intensive, particularly for the complex geometries found in urban areas.

• How do you accurately simulate the dispersion of airborne contaminants display them in near real time?

Motivation:Civil Defense/Emergency Response

Planning and Training

• Nuclear, Biological, Chemical (NBC) Attack

• Hazardous Material Spill

“The ability to construct accurate, easy-to-understand analysis of dangerous contaminant release incidents is an absolutely crucial component of civil defense planning and execution. When decisions have to be made during an actual crisis, essentially infinite speed is required of the predictions and yet the analysis must be performed with high accuracy…”

http://www.kentucky.com/multimedia/kentucky/raildocs/washrelease.pdf

http://www.kentucky.com/multimedia/kentucky/raildocs/washrelease.pdf

Methods

• Lattice Boltzmann Model (LBM)

• Multiple Relaxation Time LBM

• Sensor Feedback to improve accuracy

• GPU for acceleration

• Visualization of buildings

• Visualization of smoke

Previous Work:Dispersion Nomographs

• Not mentioned in the article, but common in practice.

http://www.kentucky.com/multimedia/kentucky/raildocs/washrelease.pdf

Previous Work:URBAN & VTMX Experiments

• Provided empirical data.

• URBAN– Conducted in Salt Lake City in 2000.– Focused on resolving interaction between scales.

• VTMX- Held in Salt Lake Valley in 2000.

- Studied vertical transport and mixing.

Previous Work: Modeling• QUIC

– QUIC-URB: Empirical algorithms for wind fields around buildings.

– QUIC-PLUME: Computes dispersion using random walk based on results of QUIC-URB.

• Multi-scale simulations sharing data and results:– COAMPS: Meteorological effects, including wind

fields, at the urban scale.– HIGRAD: Computes wind fields and transport around

buildings.– FEM3MP: Wind fields around individual buildings.

Previous Work: Lattice Boltzmann Model• Capabilities:

– Micro-level model for fluid dynamics.– The summation of the micro-level calculations yields

accurate macro-level simulation.

• Advantages:– Easy to code.– Naturally parallelizable.– Models complex boundaries and thermal effects.

• Description to follow…

Previous (and follow-on) Work: GPU acceleration- “Practical use of LBM usually requires parallel supercomputers…”

- Commodity graphics hardware speed doubles approximately every six months…

-GPUs are designed to be parallel to accommodate individual RGBA channels.

-GPU acceleration of the LBM has shown an increase in speed by a factor of 8.

Previous Work: Texture data reduction

• Geometry and texture data can be recorded together or separately, but result in large data files… Not practical for this purpose.

• Reduce file size by:– Creating a grammar reflecting common textures

to be repeated.– Add texture to landmarks, leave the rest without

texture.

Previous Work: Smoke Rendering

• Volumetric ray tracing.

• Photon Mapping.

• Both are computationally intensive…

CURRENT WORK!

Lattice Boltzmann Model• Goal: Avoid computationally intensive

Navier-Stokes equations.• Overview:

– Divide area into small cubes.– Within each cube, define representative

velocity vectors.– Define distribution function for the velocities.– Two step, discrete-time, process:

• Transport along velocity vecors.• Resolve collisions.

Lattice Geometry:

http://www.cs.sunysb.edu/~vislab/projects/amorphous/yezhaoweb/melting.pdf

Lattice Boltzmann Model(Single Relaxation Time)

http://www.cs.sunysb.edu/~vislab/papers/hardwareLBM.pdf

Lattice Boltzmann Model(Single Relaxation Time)

http://www.cs.sunysb.edu/~vislab/papers/hardwareLBM.pdf

Lattice Boltzmann Model(Multiple Relaxation Time)

Sensor Feedback

• Simulation is limited:– Single density Single contaminant??

• Sources of error:– Rounding Error– Discretization Error

• Solution: Use sensors to provide feedback

Sensor Feedback

• Report a weighted average between the sensor value and the simulation value with the weighting decreasing with distance from the sensor…. Does not provide adequate results.

Sensor Feedback

• Report a weighted average …

• Modify boundary conditions by adding external body force to account for the difference.

Hardware Acceleration

• Layout data in texture memory…

• Convert LBM operations into fragment programs which can be executed in a rendering pass…

Visualization

Essential to allow users, especially in casualty response situations, to rapidly digest the vast amount of information and make decisions with it!

DATA INFORMATION INTELLIGENCE

BUILDINGS: Textures

• Texture memory is already allocated … to the LBM portion of the simulation.

• Implemented using a small number of high resolution texture images repeated to cover the building.

• Use shading to break up the pattern and mask the repeatability.

BUILDINGS: Textures

BUILDINGS: Texture Coordinates

• Match stored texture with the original building floor height and window width.

• For each building, select the texture which most closely matches a multiple of these dimensions.

Smoke:

• Smoke particles’ postitions and velocities are determined by the LBM simulation.

• Each particle is rendered as a textured splat using the half-angle slicing technique.

Smoke: Half-angle determination

Smoke: half-angle projection

Evaluation

• They show dramatic improvements in processing power, but do not compare accuracy… Would their method work with the URBAN data they mention?

Evaluation

• How long would it take for the end user to understand, and trust, the results? They provide good visualization, but no method for evaluating the accuracy. How big is the added boundary value? What is the standard deviation in the sensor readings?

Evaluation

• I do not see any attempt to quantify the threat. What is the lethal dose and how is that displayed? How can the user tell if it is safe to evacuate through this cloud or not?

Evaluation

• I do not see any attempt to quantify the threat. What is the lethal dose and how is that displayed? How can the user tell if it is safe to evacuate through this cloud or not?

Conclusion

• Bottom Line: This looks like a great enhancement for a video game, but I am having a hard time seeing the usefulness for the stated purpose.

Conclusion• They continued the work in several papers..

– Implementing Lattice Boltmann Computation on Graphics Hardware

– Handheld version– This is a work in progress

Discussion Questions

Would you make the “Shelter-in-place versus evacuate” decision based on this product? What would the authors have to do to allow you to make that decision?

Discussion Questions

How accurate does the flow around individual buildings need to be? How would you conduct V&V based on that need?

Discussion Questions

• How much computation time is taken up by the smoke rendering? (24%) Is it well spent? Does this step increase the usefulness or merely the appearance?