Session: 04-35: Combustion Modeling III
Paper Number: 81756
81756 - Automatized Experimental Combustor Development Using Adaptive Surrogate Model-Based Optimization
Lean premixed combustion is the state of the art technology to achieve ultra low emissions in stationary gas turbines. However, lean premixed flames are susceptible to thermoacoustic instabilities, lean blow-off and flashback. In order to increase the stability of the flames, an increased equivalence ratio and thus increased production of nitrogen oxides must be accepted, at least locally.
Many combustion systems use multiple fuel injection lines to allow for a variable fuel distribution in the combustion chamber. The design of such an injection system is always related to the balancing between the levels of emissions and flame stability within the gas turbine's range of operation. In practice, this development process involves computationally and financially costly simulations and experiments and is based to a large extent on the experience and intuition of the engineers.
Machine learning and the adaptation of models through artificial intelligence have experienced a surge in development in the past years, mainly caused by the increased availability of computational capacity. Data-driven optimization methods can avoid cumbersome theoretical studies and might be able to efficiently find solutions for complex problems. These methods are very generic and have been applied to a wide range of scientific fields. The goal of this study is to show the potential of these optimization methods to experimental burner development in order to accelerate the design process.
To modify the mixture field close to the flame, a special pilot burner is installed into a series premix swirl combustor. The SLM printed pilot features 61 different positions of fuel injection. Each of the injector lines is equipped with an individual valve, such that the distribution of fuel-air mixture upstream of the flame can be modified variously.
A data-driven optimization method is used to find an optimal subset of injection locations by automated experiments. The optimizer controls the fuel valves and uses live measurements to find a distribution that generates minimal NOx emissions, while ensuring flame stability. The applied method is based on a surrogate model that is constantly updated during the measurements. It combines techniques, commonly used in the field of machine-learning and was developed in order to reduce the amount of required measurement points to find a minimum.
The method is applied to various optimization tasks and different operational points of the industrial gas turbine combustor. The results are analyzed and advantages and limitations of the proposed method are discussed. The work shows the potential of the proposed approach for accelerated and efficient combustor design.
Presenting Author: Johann Moritz Reumschuessel Technische Universitaet Berlin
Presenting Author Biography: Moritz Reumschüssel is a Resercher and phD-Student at the Chair of Fluid Dynamics under supervision of Prof. C.O. Paschereit. He is working on the application of data-driven methods and machine lerning in the field of combustion and thermoacoustics.
Authors:
Johann Moritz Reumschuessel Technische Universitaet BerlinPhilipp Maximilian Zur Nedden Technische Universitaet Berlin
Jakob G. R. von Saldern Technische Universitaet Berlin
Thoralf G. Reichel Technische Universitaet Berlin
Bernhard Cosic MAN Energy Solutions SE
Christian Oliver Paschereit Technische Universitaet Berlin
Automatized Experimental Combustor Development Using Adaptive Surrogate Model-Based Optimization
Paper Type
Technical Paper Publication