Anomaly Detection Techniques With Real Test Data From a Spinning Turbine Engine-Like Rotor
- Created on Saturday, 01 September 2012
Online detection techniques to monitor the health of rotating engine components are becoming increasingly attractive to aircraft engine manufacturers in order to increase safety of operation and lower maintenance costs. Health monitoring remains a challenge to easily implement, especially in the presence of scattered loading conditions, crack size, component geometry, and materials properties. The current trend, however, is to utilize noninvasive types of health monitoring or nondestructive techniques to detect hidden flaws and mini-cracks before any catastrophic event occurs. These techniques go further to evaluate material discontinuities and other anomalies that have grown to the level of critical defects that can lead to failure. Generally, health monitoring is highly dependent on sensor systems capable of performing in various engine environmental conditions and able to transmit a signal upon a predetermined crack length, while acting in a neutral form upon the overall performance of the engine system.Spin simulation tests were conducted on a turbine engine-like rotor with and without an artificially induced notch at different rotational loading speed levels. Health monitoring verification was performed by integrating three different advanced machine-learning algorithms for anomaly detection in continuous data streams from spinning tests of a subscale turbine engine-like rotor disk up to a speed of 10,000 rpm.
This study compares an outlier detection algorithm (Orca), one-class support vector machines (OCSVM), and the Inductive Monitoring System (IMS) for anomaly detection on the data streams. These techniques were used to inspect the experimental data under the same operating conditions employed in the tests, and using the measured vibration response (blade tip clearance) as a key input to check the viability of these techniques on detecting the disk anomalies and to evaluate the performance of each methodology. The performance of the algorithm is measured with respect to the detection horizon for situations where fault information is available. Further, this work presents a select evaluation of an online health monitoring scheme of a rotating disk using a combination of high-caliber sensor technology, high-precision in-house spin test system facilities, and unprecedented data-driven fault detection methodologies.
The methodologies applied in this study can be considered as a modelbased reasoning approach to engine health monitoring. Typical modelbased reasoning techniques compare a system model or simulation with system sensor data to detect deviations between values predicted by the model and those produced by the actual system. In fact, a model-based reasoner uses the collected system parameter values as input to a simulation and determines if a particular set of input values is consistent with the simulation model. When the values are not consistent with the model, a “conflict” occurs, indicating that the system operation is off nominal. The results obtained showed that the detection algorithms are capable of predicting anomalies in the rotor disk with very good accuracy. Each detection scheme performed differently under the same experimental conditions, and each delivered a different level of precision in terms of detecting a fault in the rotor. Overall rating showed that both the Orca and OCVSM performed better than the IMS technique.
This work was done by Ali Abdul-Aziz,
Mark R. Woike, Nikunj C. Oza, and Bryan
L. Matthews of Glenn Research Center.
Inquiries concerning rights for the commercial use of this invention should be addressed to NASA Glenn Research Center, Innovative Partnerships Office, Attn: Steven Fedor, Mail Stop 4–8, 21000 Brookpark Road, Cleveland, Ohio 44135. LEW-18758-1