Acoustic emission (AE) technology has recently found its way in condition monitoring of rotary equipment due to its advantage of earlier detection of defects and anoma- lies in comparison to vibration analysis. However, there has been very little industrial application of AE signals for condition monitoring of safety-critical equipment if any, partly due to the diffculty in processing, interpreting and classifying the acquired data in a highly reliable fashion. The motivation in this thesis was to develop a methodol- ogy for inferring health related information in a gas turbine without intruding the engine.
Our work has targeted a broad class of rotary equipment known as cyclostationary processes, therefore, instead of analyzing particular AE samples of gas turbines we have tried to build a mathematical framework that would suit any arbitrary machine comply- ing certain conditions. The result of our work mainly encompasses a feature extraction technique that eliminates the random e↵ects associated with a gas turbine AE signal, and a hypothetical testing method for classification of AE signals with any desirable level of certainty, subject to a set of assumptions and conditions.
We have validated our methodologies and derivations using actual real-life gas turbine AE signals, and compared our solutions with some of the techniques published in the literature.