Zustandsanalyse von Brennstoffzellensystemen mittels Maschinellem Lernen

Abstract

Airplanes with fuel cell systems (FCS) as energy suppliers represent a promising alternative to conventional combustion engines. This innovative technology still requires research in areas such as design, control, and efficiency. Due to the complex influences of environmental conditions and a variety of resulting operating states, the processes in FCS cannot be calculated with physical models. This bachelor thesis addresses this issue and attempts to determine the state of an FCS using machine learning (ML). The goal is to determine the state of an FCS in operation based on a concise output of software in real-time. To investigate this, datasets were generated in a test laboratory in Hamburg and analyzed with ML algorithms. The result after applying unsupervised learning techniques to the dataset is a set of plausible clusters that can be interpreted as states. In a detailed analysis, appropriate scalers, dimension reduction (DR) methods, and clustering algorithms (CA) were developed to analyze datasets with univariate parameters (UP) in steady states. The end result is a structured ML pipeline with serial processing of the DS using ML algorithms (MLA), enabling visualization of meaningful state analysis (ZA).