PhD dissertation
Transcript: Wavelengths: 476, 543, 623 nm Smoke particles: soot Universidad Politécnica de Madrid “To pave the way for future devices relying on remote sensing to present the users with artificially generated images that increase their situation awareness and thus their safety while conducting exposed tasks in smoke, turbid water or other particle-filled atmospheres”. pencil Metamaterial and sub-wavelength reconstruction Related Work Explore possible application of Compressed Sensing techniques: Grow, but still in the pm-nm range ...how can we process it? (in an easy and meaningful way) Implementation constraints Activated pixels with power @time interval Frame burst: 240 fps Object Injection & suspension of particles using methanol Molecule & Particle Dynamics Smoke particles: soot Concept setup Media Validation overview Temperature: 25ºC Simulation 2. Aggregation Molecules and soot particles follow: Kinetic theory [<100m/s] 83% (where is a random variable following the scattering profile and generated by the inverse CDF method. Simulation results representing received pixels below sensibility threshold Concept: "synthetic smoke" MonteCarlo Calibrate Different output modes Creation of a visualization tool Generalization of the model to other particle-filled atmospheres Performance study: UVM vs. manual optimizations Pipe length: 0.4 m UVM (CUDA > 6.0) Receiver: 320x240 pixels @10x10 um Results and discussion (I) Emitter: pinhole, 10 um Visualization tool of "synthetic smoke" Arbitrary particle types* Measurements MFP: 50000, 100000, 200000 um Arbitrary environment conditions Self-consistency checks Power represented in false color scale Universidad Politécnica de Madrid Parametric design 3D printing Polystyrene calibrated homogeneous microspheres José María Nadal Serrano optim. 3D print Propagation [sub]model Simulations Concept Golden rule: Partition and balance load across multiprocessors Maximize concurrent active threads Optical [sub]model The model has been designed with performance in mind: it has been implemented in a highly parallel architecture to minimize runtimes. A number of studies have been conducted on parallel code performance optimizations and practical conclusions have been drawn. Setups: 1, 2, 3, 6 um + mix 1-2-6 Advisor: Marisa López Vallejo Dept. Ingeniería Electrónica & We have bridged the gap between microscopic and macroscopic worlds in some sense: making use of the computational power of GPUs, we could create a radiation propagation model at optical and infrared frequencies able to provide with macroscopic data aggregating individual particle-radiation interactions. This poses a truly challenging environment in the limit between Quantum physics, Electrodynamics and Thermodynamics. The model has been designed as a framework model: its modularity has made it possible to generalize it from smoke to host a wide scope of particle-filled atmospheres. Conclusions: Follow the golden rule Start with UVM Then go for "ninja optimizations"... ...but watch your cost optimization function Specific test data: Wavelength = 543 nm MFP = 200000 um Particle size = 1 um Temp = 25 ºC MFP: 50000, 100000, 200000 um UVM + "Ninja opts." They define the particular fire conditions for the simulation: Temperature, incident wavelength, mean free path, fuel Using the data above: Particle size distributions are obtained Scattering distributions are obtained for each particle type Other "physical data" is also fed into the model for simulation: environment, medium, emitter, sensor types and geometries... Nadal-Serrano JM, Lopez-Vallejo M: A Survey on Theoretical and Practical Aspects of Imaging Aids for Artificial Vision in Professional Environments. IEEE Sensors Journal 2015, 15(5):2719 – 2731. Nadal-Serrano JM, Lopez-Vallejo M: A time-resolved Monte Carlo smoke model for use at optical and infrared frequencies. Fire Safety Journal 2015, 71:299 – 309. Nadal-Serrano JM, Lopez-Vallejo M: A Performance Study of CUDA UVM versus Manual Optimizations in a Real-World Setup: Application to a Monte Carlo Wave-Particle Event-Based Interaction Model. IEEE Transactions on Parallel and Distributed Systems 2016, 27(6):1579 – 1588. [Two additional publications are currently under preparation] Big computational load Many concurrent "processes" (threads) No interaction between the threads (no interaction photon-photon) Model overview: conceptual blocks Reasonable Mie approximation using "equivalent spheres" under certain circumstances [P Hull et al.] Our model: Edge between micro- and macroscopic worlds q-static, short distance (tens of meters) Incorporates ideas from other models Conclusions III Registers < Shared < Global memory An in-depth multidisciplinary, top to bottom analysis of the problem of temporary loss of vision in harsh environments and its consequences on spatial perception has been conducted, leading to a subsequent analysis of the fundamental variables that take part on that process. A basic (low-level) study of the
Prezi Present
Prezi Video
Prezi Design
