A grid combination method, based on a modified hyperbolic grid solver, is developed to generate tw+ dimensional grid systems in regions with prescribed grid distribution along all of the boundaries. The hyperbolic grid solver uses a predictor-corrector scheme and is a modified version of the hyperbolic grid generation method of Tai, Yih, and Soong, which applied the Roe first order upwind splitting to the Steger and Chaussee grid generation equations. Four grid sys-0 intersection angle between grid lines terns are first marched from each boundary toward its opposite boundary and then combined to give the a,o functions defined in As arc length, Eq.(12) A t , . . . = t i -ti-1,. . . user specified constant, Eq.(9) eigenvalue of C scaling factor, Eq.(ll) coordinates on computational domain t , 7
SUMMARYThis study investigates how high-pressure water-mist system discharge methodologies influence the fire extinction performance for pan pool fires and the corresponding mechanisms of restraining fire. The fire source is a pool-fire burner. Fine water spray is injected using a portable device. The additive in the watermist is neither toxic nor corrosive. All the tests are regarded as fuel controlled. The fire test parameters are fuel type, nozzle discharge angle, and additive solution volume. The fuels used are heptane, gasoline, and diesel. Nozzle discharge angles are 30, 45, and 60 • with respect to the ground. Additive solution volumes are 0% (pure water), 3, 6, and 10%. Test results indicate that the nozzle discharge angle and additive solution volume in a water-mist fire extinction system play a significant role. Fire extinguishing efficiency is influenced by mist effects and the additive. Furthermore, the water-mist system can reduce radiation and can provide good protection for operators using portable fire extinguishing equipment.
This study applied the numerical simulator tool FDS (fire dynamics simulator), Version 5.53, and focused on the simulation of the natural smoke flow ventilation design system, an innovative ventilation design using the parallel processing technology MPI (message passing interface). The design was then compared with the exhaust efficiency of a typical natural smoke vent. The natural smoke flow ventilation design system was located at the top of the factory, where smoke streams effectively converged. Therefore, the source of fire was designed to be 2 MW, which has a better exhaust efficiency than typical natural smoke vent with same area. The simulation discovered that the exhaust efficiency of the natural smoke ventilation design systems is higher than that of typical natural smoke vent with 2 times the opening area and that was not affected by external wind speed. Instead, external wind speed can help to enhance the exhaust efficiency. Smoke exhaust of typical natural smoke vents was affected by external wind speed, even leading them to become air inlets which would disturb the flow of air indoors, leading to smoke accumulation within the factory.
o O 1.2 1.0 0.6 0.4 0.2 0.0 Minimum Length Nozzle Wall Contours ---MOC Prediction --Neural Net Prediction -Me = 5.0 -Me = 3.0 _ 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 Scaled X-Coordinate Fig. 3 Comparison of nozzle wall shapes.Since the relation between exit Mach number and minimum length is a simple one-dimensional map, a polynomial approximation was used as the first network. The second network had a single neuron input layer, two hidden layers with four neurons each, and an output layer with four neurons. The traditional MLN problem can be solved using method of characteristics (MOC) procedures and a program to that end was developed by the authors and used to generate training data. This phase involved randomly generating 200 exit Mach numbers in the range [1.05, 5]. The MOC program was then run for each Mach number and a fourth-order polynomial of the presented form was used to fit the shape output from the MOC program. The polynomial coefficients along with the corresponding exit Mach number and minimum length formed the training sample set. Network training was accomplished through repeated presentation of the training samples and backpropagation of error. Training both networks took 6 h on an IBM RS/6000 workstation.The networks were then verified by randomly generating 40 Mach numbers in the range [1.05, 5] and running both the MOC code and the neural networks for each exit Mach number and obtaining the corresponding shapes. Errors were computed using 50 (4) where {jc/}f ( l i are points along the length of each nozzle. Typical results can be seen in Fig. 3. Good agreement was obtained between the neural network predicted shapes and the MOC predictions over the range of Mach numbers considered. The average error in shape was 3.9% and the average error in exit Mach number was never more than 3.1%. It must be noted that the 40 nozzle shapes were predicted by the neural networks in only 2.4 s of CPU time on an IBM RS/6000 workstation. ConclusionsExperiments with two design problems have demonstrated the ability of neural networks to accurately and rapidly solve inverse aerodynamic problems. Once trained, a neural network is far superior computationally than any standard technique. In situations where inverse problems must be solved in real time, neural networks are certainly beneficial. Practical difficulties with this approach are associated with training data generation, sizing of the network, and choice of training algorithms. Problems more complex than those discussed here may require large amounts of computer time to solve the forward problem and, thus, it would be difficult to generate sufficient training data. Some complex problems may not have an analytic forward solution available and would need to resort to expensive experimentally generated training data. Robust neural networks designed with minimal training are needed and is an area suggested for further research. AcknowledgmentsWe are grateful to Rob Markin, Karen Harwell, and Dedra Caddell, undergraduate students in the Department...
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.