Data Clustering of Hybrid Rocket Combustion Flame

Abstract

Hybrid rocket propulsion is a promising technology for many applications, because it allows for cost reductions, still being able to deliver similar performance as solid and liquid propulsion. However, the complex fluid dynamic and combustion phenomena taking place in the hybrid rocket combustion chamber are not fully understood yet and still a matter of ongoing research. In the last years, many combustion tests with different paraffin-based fuels were performed at the German Aerospace Center (DLR) with an optically accessible combustion chamber. For gaining a better insight into the combustion process, the tests were captured with a high-speed video camera. This led to a huge amount of data images for each test, which needed to be analysed in detail. In the framework of this study, the combustion data set was clustered with a K-means++ algorithm, in order to get insights into the complex flow phenomena. Since the algorithm might converge to local optimal solutions, many repetitions were performed to ensure that a global optimal solution is reached. Moreover, an analysis was necessary to estimate the number of relevant clusters in each experiment. From the results it is possible to observe that different clusters, corresponding to different combustion phases, were found by the algorithm. Moreover, the combustion flame behaviour is changing depending on the oxidizer mass flow and fuel.