Development and Application of Aero-engine Experimental Data Mining Algorithm Library

摘要

This paper presents the application of several anomaly detection algorithms in experiment data from engine test bed. Several anomaly detection algorithms are programmed in Python language and integrated into an algorithm library named PyPEFD (Python Package for Engine Fault Detection). The algorithm library includes Gaussian Mixture Model, Feature Weighted Fuzzy Compactness and Separation (WFCS), Sequential Probability Ratio Test (SPRT), Variational Autoencoder, Dynamic Time Warping, Mahalanobis Distance, Singular Value Thresholding, Random Forest and Multivariate State Estimation Technique. These algorithms can analyze the structure and characteristics of the engine test data, and mine the hidden fault information in the data, so as to detect the fault or fault trend of aero-engine test data. This paper also presents a preview of the algorithm library.