The paper proposes a refined stochastic fault classifier for jet engine vibration diagnostic based on advanced stochastic concepts. The statistical data to be analyzed are the spectrum profiles of vibration measured at different locations in the engine. The statistical spectrum profiles are idealized by non-homogeneous stochastic fields with non-Gaussian probability distributions. The proposed stochastic classifier is based on the decomposition of the statistical correlation matrix of spectrum profiles using a Karhunen-Loeve (KL) expansion. Two stochastic classifiers are proposed, namely a “global” and a “specific” fault classifier. The “global” KL classifier, which is a scalar quantity, is an efficient anomaly/novelty detection tool for identifying incipient fault diagnosis with small amplitude fluctuations. The “specific” KL classifier, which is a vector quantity, is a refined diagnostic tool for identifying the engine malfunction causes. An illustrative example of a turbofan engine is included.
