| Description |
x, 79 leaves : illustrations ; 28 cm |
| Summary |
"In this thesis, the detection and prediction of faults in rotating machinery is undertaken and presented in two papers. In the first paper, Principal Component Analysis (PCA), a well known data-driven dimension reduction technique, is applied to data for normal operation and four fault conditions from a one-half horsepower centrifugal water pump. Fault isolation in this scheme is done by observing the location of the data points in the Principal Component domain, and the time to failure (TTF) is calculated by applying statistical regression on the resulting PC scores. The application of the proposed scheme demonstrated that PCA was able to detect and isolate all four faults. Additionally, the TTF calculation for the impeller failure was found to yield satisfactory results. On the other hand, in the second paper, the fault detection and failure prediction are done by using a model based approach which utilizes a nonlinear observer consisting of an online approximator in discrete-time (OLAD) and a robust adaptive term. Once a fault has been detected, both the OLAD and the robust adaptive term are initiated and the OLAD then utilizes its update law to learn the unknown dynamics of the encountered fault. While in similar applications it is common to use neural networks to be used for the OLAD, in this paper an Artificial Immune System (AIS) is used for the OLAD. The proposed approach was verified through implementation on data from an axial piston pump. The scheme was able to satisfactorily detect and learn both an incipient piston wear fault and an abrupt sensor failure"--Abstract, leaf iv. |
|
