Sensor Error Compensation in Engine Performance Diagnostics

The pursuit of reliable engine performance methods is as old as the gas turbine itself. Contemporary methods of engine fault isolation and assessment have centered around the use of modem estimation algorithms similar in nature to the Kalman filter and its derivatives. Those who are familar with such programs will offer the following quip, (apparently known to many but attributed to none), that the Kalman filter works well when you already know the answer, in other words, when the a-priori estimates are sufficiently close to the true values such as to produce small measurement residuals. Unfortunately, the ubiquity of measurement error ensures that this is generally not the case. A pivotal requirement for successful diagnostics becomes the ability to detect and assess the measurement error existing in raw engine data in an effort to mitigate its effect on potential engine fault mis-interpretation. The purpose of this paper is to outline the mathematical basis for the techniques in present use and to offer a generalization to a large measurement error compensation method which has been utilized effectively for over a decade.Copyright © 1994 by ASME