High energy spark ignition in non-premixed flowing combustors
Sforzo, Brandon Anthony
MetadataShow full item record
In many practical combustion devices, including those used in gas turbine engines for aircraft and power generation, a high energy spark kernel is necessary to reliably ignite the turbulently flowing flammable gases. Complicating matters, the spark kernel is sometimes generated in a region where a non-flammable mixture is present, or where there is no fuel at all. This requires the spark kernel to travel to a flammable region before rapid combustion can begin in non-premixed or stratified flows. This transit time allows for chemical reactions to take place within the kernel as well as mixing with surrounding gases. Despite these demanding conditions, the majority of research in ignition has been for low energy sparks and premixed conditions, not resembling those found in many combustion devices. Similarly, there is little work addressing this issue of spark kernel evolution in the non-premixed flowing environment, and none available that control the time allowed for transit. The goal of this thesis is to understand the development of a spark kernel issued into a non-premixed flow and the sensitivities of the ignition process. To this effect, a stratified flow facility for ignition experiments has been fabricated utilizing a high speed schlieren and emission imaging system for visualizing the kernel motion and ignition success. Additionally, OH chemiluminescence and CH PLIF were used to track chemical species during the ignition process. This facility is also used to control the important variables regarding the flow and spark kernel interaction to quantify the influence on ignition probability. A reduced order model employing a perfectly stirred reactor (PSR) has also been developed based on experimental observations of the entrainment of fluid into the evolving kernel. The simulations provide additional insight to the chemical development in the kernel under different input conditions. This model was enhanced by introducing random perturbations to the input variables, mimicking a practical situation. A computationally efficient support vector machine was trained to replicate the numerical model outputs and predict ignition probabilities for nominal input conditions, providing comparison to experimental results. Experimental and numerical results show that initial mixing with non-flammable fluid quickly reduces the ability for the kernel to ignite the flammable flow, resulting in a strong influence of the inlet temperature and the kernel transit time on the probability of ignition. Once the kernel reaches the flammable mixture, entrainment of this flow occurs, which requires on the order of a vortex turn-over time before chemistry can begin. Initial chemical reactions include endothermic fuel decomposition, further reducing the kernel temperature prior to heat release, creating a competition between the cooling effect of additional mass entrainment and the delayed heat release reactions. CH PLIF results show that flame chemistry is initially confined to a thin region that corresponds to the interface layer where the flammable gases mix with the hot kernel fluid from the vortex entrainment of ambient gas. The dependence of the ignition probability to variations in flow conditions is captured reasonably well by the reduced order model, validating the PSR approach and the probability prediction tool. The development of this reduced order model is a major contribution of this work with the ability to predict the effects of the important physical ignition processes, which can be used when considering an ignition system's feasibility. This work will provide knowledge to guide the use and design practices in industry, as well as a simple model to test ignition feasibility based on mixing, entrainment, and chemical reactions. Furthermore, the flow facility is well characterized, and a database has been developed that can provide validation points for future computational simulations. Future modeling will be important to further understand fluid dynamic effects that are difficult to measure experimentally, and study a broader range of conditions.