A Hybrid Stochastic-Neuro-Fuzzy Model-Based System for In-Flight Gas Turbine Engine Diagnostics
Abstract
One key aspect when developing a real-time in-flight risk-based health management system for jet engines is the development of accurate and robust fault classifiers. Regardless of the complex uncertainty propagation in the data fusion process, the selection of fault classifiers is the critical aspect of a health management system. The paper illustrates the application of a hybrid Stochastic-Fuzzy-Inference Model-Based System (StoFIS) to fault diagnostics and prognostics for both the engine performance. The random fluctuations of jet engine performance parameters during flight missions are modeled using multivariate stochastic models. The fault diagnostic and prognostic risks are computed using a stochastic model-based deviation (using a gas-path analysis model) approach. At any time the engine operation for the future is approached as a conditional reliability problem where the conditional data are represented by the past operational history monitored on-line by the engine health management (ERM) system. To capture the complex functional relationships between different engine performance parameters during flight fast an adaptive network-based fuzzy inference system is employed. This increases significantly the robustness of the ERM system during highly transient in-flight conditions. Both the monitored and fault data uncertainties are considered in a multidimensional parameter space, with two probabilistic-based safety margins employed for fault detection and diagnostic, as follows: (i) Anomaly Detection Margin (ADM) and (ii) Fault Detection Margin (FDM).
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 05, 2001
- Accession Number
- ADP013480
Entities
People
- Dan M. Ghiocel
- Joshua Altmann