Visualization of multi-channel sensor data from aero jet engines for condition monitoring and novelty detection
The work presented in this paper seeks to determine if it is possible to determine if an aero gas jet engine is behaving normally by learning the expected variation of its Feature Detector (FD) scores and comparing subsequent flight data against a model of the expected scores. We want to be able to...
| Main Author: | |
|---|---|
| Format: | Conference item |
| Language: | English |
| Published: |
2007
|
| Subjects: |
| Summary: | The work presented in this paper seeks to determine if it is possible to determine if an aero gas jet engine is behaving normally by learning the expected variation of its Feature Detector (FD) scores and comparing subsequent flight data against a model of the expected scores. We want to be able to characterize the engine by monitoring combinations of FD scores so that it is possible to quickly and intuitively ascertain if an engine is behaving normally according to the previously learned model. This paper demonstrates that it is possible to fuse vibration data from multiple channels into ‘score vectors’ and then determine an optimal mapping that can represent these high-dimensional features in 2-D for visualization. We show that this low-dimensional representation of the score data can adequately capture differences between sets of flight data that allow instances of abnormal engine behaviour to be identified. This paper proposes a number of different models, each associated with a sub-set of engine scores that represent the condition of a particular engine shaft in a three-shaft gas aero jet engine and demonstrates the proposed method on data from the intermediate pressure (IP) shaft of an engine. |
|---|