Combustion Simulations with reduced mechanisms using hybrid optimization
(2018) FYSK02 20181Department of Physics
Combustion Physics
- Abstract
- In order to simulate complex combustion systems, the kinetic mechanisms describing the chemical processes have to be reduced. To minimize the error introduced by the reduction, coefficients of the reaction rates included in the mechanisms can be adjusted in ways so that simulations using the reduced mechanism behave like simulations that use detailed mechanisms. One can then run simulations with different sets of reaction coefficients and compare the result to that of a detailed mechanism thus knowing if the new set of coefficients was better
than the old or not. This can be done using an optimization algorithm that in smart ways picks sets of coefficients and uses them in simulations. In this project, simulated annealing and a genetic... (More) - In order to simulate complex combustion systems, the kinetic mechanisms describing the chemical processes have to be reduced. To minimize the error introduced by the reduction, coefficients of the reaction rates included in the mechanisms can be adjusted in ways so that simulations using the reduced mechanism behave like simulations that use detailed mechanisms. One can then run simulations with different sets of reaction coefficients and compare the result to that of a detailed mechanism thus knowing if the new set of coefficients was better
than the old or not. This can be done using an optimization algorithm that in smart ways picks sets of coefficients and uses them in simulations. In this project, simulated annealing and a genetic algorithm is used together to create a hybrid algorithm in order to mitigate each other’s weaknesses to ultimately find better coefficients. (Less) - Popular Abstract
- Optimization is a subject that has been relevant ever since math was invented, and it basically means finding the best element out of a set of elements. This can be applied to many things but is usually about finding the highest or lowest value of a mathematical function known as maximum and minimum respectively. This tool for finding the best value has lately become very popular in Machine Learning and Neural Networks, where computer programs, without specifically programming them to do a certain task, can learn by themselves. They only need training data and something that tells them how good they're doing, and this is where optimization algorithms are important. These optimization methods can also be used in computer simulations of... (More)
- Optimization is a subject that has been relevant ever since math was invented, and it basically means finding the best element out of a set of elements. This can be applied to many things but is usually about finding the highest or lowest value of a mathematical function known as maximum and minimum respectively. This tool for finding the best value has lately become very popular in Machine Learning and Neural Networks, where computer programs, without specifically programming them to do a certain task, can learn by themselves. They only need training data and something that tells them how good they're doing, and this is where optimization algorithms are important. These optimization methods can also be used in computer simulations of physical processes such as combustion of fuel. Optimizing properly will result in well behaved, realistic simulations that will ultimately help scientists in all fields where optimization is relevant.
So how does this have anything to do with a blind man climbing a mountain using Darwinism? There are many different kinds of optimization algorithms, often efficient at different tasks. One algorithm that is commonly used is called Simulated Annealing which is very analogous to the blind man in question. Imagine that the blind man can take a step in a random direction with a certain probability. If the next step he's taking is higher than the last, the probability is higher. The reason the probability to take a "good" step isn't 100\%, is because if the blind man reaches a local peak, he would think that he has reached the top of the mountain, while he might in reality be no where near it. Giving him the opportunity to take a step in the downwards direction but making steps in the upwards direction more likely, makes for a higher chance to make it the very top than just always taking the steepest climb.
In order increase the climbers chance to get to the peak, one can introduce another optimization algorithm. Genetic Algorithm is a type of Evolutionary Algorithm that picks many different points of the set, or the "mountain" and does a fitness test that can depends on many things, but usually just the height of the points. It then kills of the bad points, and keeps the good points to breed them with each other. It saves the offspring and the parents, and adds new random points until one has the desired population size. It then does a new fitness test for all the points, and the process is repeated until the user seems fit. This way, one spreads the search of good points over the entire mountain, and hopefully the best point will eventually be the peak.
The idea is then to use these algorithms in tandem in a way the lets them make up for each others weaknesses. For example, the Genetic algorithm has high precision, but if it runs into a local maximum, it can have a hard time getting unstuck. One can then use Simulated Annealing to "walk down" from the local maximum and continue the search for the global maximum. The steps Simulated Annealing takes are often in the vicinity of the original point, so it can also serve as tool to "scout" around points of interest. There are many details in both algorithms that can be used together to strengthen the overall performance of the hybrid algorithm, and endless combinations of other kinds of algorithms. Finding the best combination and using them together intelligently can greatly improve the quality of whatever is using the optimization methods.
This is being worked on in the department of Combustion Physics in Lund, where one simulates combustion of fuel using Chemkin. Because real events of combustion has way too many processes that play very little part in the overall reaction, one simply ignores them. This means that one has to make up for them by increasing the amount of key processes by very precise amounts. There are still roughly 30 processes being accounted for, so one needs to adjust a combination of 30 parameters for the process to look realistic. This is where the optimization algorithm is highly efficient compared to doing it manually. It uses the techniques described earlier to basically find the best parameters to get the best simulation, or in other words; find the peak of the mountain. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/8943116
- author
- Burman Ingeberg, Marius ^{LU}
- supervisor
- organization
- course
- FYSK02 20181
- year
- 2018
- type
- M2 - Bachelor Degree
- subject
- keywords
- Optimization, Reduced mechanism, Hybrid optimization, simulated annealing, Genetic algorithm
- language
- English
- id
- 8943116
- date added to LUP
- 2018-05-31 17:03:48
- date last changed
- 2018-05-31 17:03:48
@misc{8943116, abstract = {{In order to simulate complex combustion systems, the kinetic mechanisms describing the chemical processes have to be reduced. To minimize the error introduced by the reduction, coefficients of the reaction rates included in the mechanisms can be adjusted in ways so that simulations using the reduced mechanism behave like simulations that use detailed mechanisms. One can then run simulations with different sets of reaction coefficients and compare the result to that of a detailed mechanism thus knowing if the new set of coefficients was better than the old or not. This can be done using an optimization algorithm that in smart ways picks sets of coefficients and uses them in simulations. In this project, simulated annealing and a genetic algorithm is used together to create a hybrid algorithm in order to mitigate each other’s weaknesses to ultimately find better coefficients.}}, author = {{Burman Ingeberg, Marius}}, language = {{eng}}, note = {{Student Paper}}, title = {{Combustion Simulations with reduced mechanisms using hybrid optimization}}, year = {{2018}}, }