Electromechanical actuator bearing fault detection using empirically extracted features

Show full item record

Redirect: RIT Scholars content from RIT Digital Media Library has moved from http://ritdml.rit.edu/handle/1850/15376 to RIT Scholar Works http://scholarworks.rit.edu/theses/5930, please update your feeds & links!
Title: Electromechanical actuator bearing fault detection using empirically extracted features
Author: Sridhar, Rahulram
Abstract: Model parameter estimation when coupled with Principal Component Analysis (PCA) and Bayesian classification techniques form a potentially effective fault detection scheme for Electromechanical Actuators (EMAs). This work uses parameter estimation algorithms based on linear system identification methods, derives a novel feature extraction algorithm based on PCA and analyzes its performance through simulations and experiments. A Bayesian classifier is used to create well defined EMA health classes from the extracted features. Research contributions on fault detection in EMAs are significant because EMA faults and their detection are not yet well understood. Potential future applications - such as in primary flight control actuation in aircraft - require that quality fault detection systems be in place. Therefore, fault detection of EMAs is a vast area of ongoing research where highly capable solutions are gradually becoming available. Prior work in parameter estimation methods for feature extraction in DC motor drives - which includes EMAs - are amongst those available. While PCA is a popular feature extraction solution in a number of frequency-based fault detection approaches, the use of PCA for feature extraction from model parameters for detecting bearing faults in EMAs has not been previously reported. In this work, a linear difference model is applied to the EMA system data such that fault information is distributed amongst the estimated model parameters. A direct comparison of the parameter estimates from healthy and degraded systems offers little insight into health conditions because of the weak effects of faults on the signal data. However, the application of PCA to uncorrelate the linearly correlated model parameters while minimizing the loss of variance information from the data effectively brings out fault information. The present algorithm is successfully applied to data collected from a Moog MaxForce EMA. The results are consistent and display effective fault detection characteristics, making the developed approach a suitable starting point for future work.
Record URI: http://hdl.handle.net/1850/15376
Date: 2012-08

Files in this item

Files Size Format View
RSridharThesis8-2012.pdf 4.324Mb PDF View/Open

The following license files are associated with this item:

This item appears in the following Collection(s)

Show full item record

Search RIT DML

Advanced Search