Doctoral Dissertations

Date of Award

5-1999

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Major

Mechanical Engineering

Major Professor

Roy J. Shulz

Committee Members

Firouz Shahrokhi, Dinesh Mehta, Roger Crawford

Abstract

Hundreds of individual sensors produce an enormous amount of data during developmental turbine engine testing. The challenge is to ensure the validity of the data and to identify data and engine anomalies in a timely manner. An automated data validation, engine condition monitoring, and fault identification process that emulates typical engineering techniques has been developed for developmental engine testing.An automated data validation and fault identification approach employing enginecycle-matching principles is described. Engine cycle-matching is automated by using an adaptive nonlinear component-level computer model capable of simulating both steady state and transient engine operation. Automated steady-state, transient, and real-time model calibration processes are also described. The model enables automation of traditional data validation, engine condition monitoring, and fault identification procedures. A distributed parallel computing approach enables the entire process to operate in real-time.The result is a capability to detect data and engine anomalies in real-time during developmental engine testing. The approach is shown to be successful in detecting and identifying sensor anomalies as they occur and distinguishing these anomalies from variations in component and overall engine aerothermodynamic performance. The component-level model-based engine performance and fault identification technique of the present research is capable of: identifying measurement errors on the order of 0.5 percent (e.g., sensor bias, drift,level shift, noise, or poor response) in facility fuel flow, airflow, and thrust measurements; identifying measurement errors in engine aerothermodynamic measurements (rotorspeeds, gas path pressures and temperatures); identifying measurement errors in engine control sensors (e.g., leaking/biased pressure sensor, slowly responding pressure measurement) and variable geometry rigging (e.g., misset guide vanes or nozzle area) that would invalidate a test or series of tests; identifying abrupt faults (e.g., faults due to domestic object damage, foreign object damage, and control anomalies); identifying slow faults (e.g., component or overall engine degradation, and sensor drift). Specifically, the technique is capable of identifying small changes in compressor (or fan) performance on the order of 0.5 percent; and being easily extended to diagnose secondary failure modes and to verify any modeling assumptions that may arise for developmental engine tests (e.g., increase in turbine flow capacity, inaccurate measurement of facility bleed flows, horsepower extraction, etc.).The component-level model-based engine performance and fault identification method developed in the present work brings together features which individually and collectively advance the state-of-the-art. These features are separated into three categories: advancements to effectively quantify off-nominal behavior, advancements to provide a fault detection capability that is practical from the viewpoint of the analysis,implementation, tuning, and design, and advancements to provide a real-time fault detection capability that is reliable and efficient.

Download
Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS