| It is crucial to detect, characterize and model events of interest in a new propulsion system. As technology advances, the amount of data being generated increases significantly with respect to time. This increase substantially strains our ability to interpret the data at an equivalent rate. It demands efficient methodologies and algorithms in the development of automated event modeling and pattern recognition to detect and characterize events of interest and correlate them to the system performance. The fact that the information required to properly evaluate system performance and health is seldom known in advance further exacerbates this issue.; Event modeling and detection is essentially a discovery problem and involves the use of techniques in the pattern classification domain, specifically the use of cluster analysis if a prior information is unknown. In this dissertation, a framework of Adaptive Learning for Event Modeling and Characterization (ALEC) system is proposed to deal with this problem. Within this framework, a wavelet-based hierarchical fuzzy clustering approach which integrates several advanced technologies and overcomes the disadvantages of traditional clustering algorithms is developed to make the implementation of the system effective and computationally efficient.; In another separate but related research, a generalized multi-dimensional Gaussian membership function is constructed and formulated to make the fuzzy classification of blade engine damage modes among a group of engines containing historical flight data after Principal Component Analysis (PCA) is applied to reduce the excessive dimensionality. This approach can be effectively used to deal with classification of patterns with overlapping structures in which some patterns fall into more than one classes or categories. |
