| DOI | Resolve DOI: https://doi.org/10.25046/aj030602 |
|---|
| Author | Search for: Cheung, Catherine1; Search for: Kilfoyle, Nicolle; Search for: Valdés, Julio2; Search for: Sehgal, Srishti1; Search for: Chavez, Richard Salas1 |
|---|
| Affiliation | - National Research Council of Canada. Aerospace
- National Research Council of Canada. Digital Technologies
|
|---|
| Format | Text, Article |
|---|
| Subject | condition indicators; failure prediction; intrinsic dimension; low-dimensional spaces |
|---|
| Abstract | Advances in technology have enabled the installation of an increasing number of sensors in various mechanical systems allowing for more detailed equipment health monitoring capabilities. It is hoped the sensor data will enable development of predictive tools to prevent system failures. This work describes continued analysis of sensor data surrounding a seizure of a turbocharger within a propulsion system. The objective of the analysis was to characterize and distinguish healthy and failed states of the turbocharger. The analysis approach included mapping of multi-dimensional sensor data to a low-dimensional space using various linear and nonlinear techniques in order to highlight and visualize the underlying structure of the information. To provide some physical insight into the structure of the low-dimensional representation, the transformation plots were analyzed from the perspective of several engine signals. By overlaying operating ranges of individual sensor signals, certain regions of the mappings could be associated with distinct operational states of the engine, and several anomalies could be related to various points in the turbocharger seizure. Although the failed points did not map to an obvious outlier location in the transformations, incorporating expert domain knowledge with the data mining tools significantly enhanced the insight derived from the sensor data. |
|---|
| Publication date | 2018-11-01 |
|---|
| Publisher | ASTES Publishers |
|---|
| In | |
|---|
| Language | English |
|---|
| Peer reviewed | Yes |
|---|
| Export citation | Export as RIS |
|---|
| Report a correction | Report a correction (opens in a new tab) |
|---|
| Record identifier | b988de27-c86d-4b94-a375-8d4189fbb50a |
|---|
| Record created | 2019-04-23 |
|---|
| Record modified | 2021-09-17 |
|---|
