Off-campus UMass Amherst users: To download campus access dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.

Non-UMass Amherst users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

Dissertations that have an embargo placed on them will not be available to anyone until the embargo expires.

Document Type

Open Access Dissertation

Degree Name

Doctor of Philosophy (PhD)

Degree Program

Mechanical Engineering

Year Degree Awarded

2014

First Advisor

Dr. Kourosh Danai

Second Advisor

Dr. Yossi Chait

Third Advisor

Dr. Christopher V. Hollot

Subject Categories

Mathematics | Mechanical Engineering | Other Mechanical Engineering

Abstract

This dissertation addresses health monitoring of aircraft engines. Two methods are offered for engine degradation assessment: (1) a direct method to isolate degradation of engine components in-flight, and (2) an inverse method to quantify the level of degradations post-flight. The noted feature of the degradation isolation method is its independence from training, which makes it suitable for on-board implementation. The degradation quantification method, on the other hand, is a multi-output method of parameter estimation with the advantage of leveraging the shape attributes of model outputs. The representation of the shape attributes of the various time series, by continuous wavelet transforms (CWTs), is the salient feature of both the direct and inverse methods developed in this research. It enables the isolation of regions in the time-scale plane, called “signatures,” wherein the wavelet coefficients of a given transform time series dominate the wavelet coefficients of the others. These methods of engine degradation isolation and quantification have been validated numerically using the transient outputs of a high-bypass turbo-fan engine model provided by Pratt & Whitney Company.

In the direct method, residuals will continually be formed in-flight to represent the difference between individual outputs and their baseline. These residuals will then be contrasted with each other to reveal “degradation signatures,” denoting the effect of the present degradation on individual residuals. To perform degradation isolation, the observed degradation effect will be compared with the pre-established effect of individual components’ degradation on the outputs according to the engine model. These pre-established effects are defined according to the sensitivity of outputs to component parameters, denoted as “output/parameter signatures,” and to combined component parameters, denoted as “output/component signatures.” The effectiveness of the proposed method is evaluated in engine simulations. The results indicate that with the suite of outputs currently available on-board 70% to 96% of the degraded components simulated can be isolated for new and older engines.

In the inverse method, parameter signatures are extracted to denote the regions of the time-scale domain wherein individual output parameter sensitivities are dominant. Justified by this dominance, the prediction error can be attributed in these regions to the error of the corresponding model parameter. This enables parameter estimation to be performed on a small set of wavelet coefficients. These isolated regions of the timescale plane also reveal numerous transparencies and degrees of freedom to be exploited for parameter estimation. The transparencies include the quality measures of the parameter signatures. The degrees of freedom entail the various shape attributes of outputs that can be included through different wavelet transforms, selectiveness of regions of the parameter signatures that are closest to the edge points (modulus maxima), among others. It is shown that by taking advantage of these transparencies and degrees of freedom, the robustness of parameter estimation can be improved. The results also indicate the potential for improved precision and faster convergence of the parameter estimates when shape attributes are used in place of the magnitude. Although the inverse method has proven effective in several platforms, it is found to be less effective than nonlinear least squares in application to the engine model, due to the lack of distinction between its output sensitivities.

Share

COinS