Abstract: Data mining is one of the important means for engine fault diagnosis, the training data of fault diagnostic model always comes from fault data or experimental data. In practical operation, it is difficult to obtain a large number of different fault data, and the same kind of faults may have different parameters with various failure degrees, which makes limited generalization of the diagnostic model. In lack of fault data, how to obtain a better diagnostic model becomes significant. Fingerprints identifies the small deviation of the main performance parameters of engine faults. Figures from fingerprints can be used as basic data of different faults with data minning. According to the fingerprint chart of fault diagnosis principle, unit vector, ratio coefficient and correlation coefficient are used to extend data, then training data are obtained by applying different methods of noise addition, helping the diagnostic model getting more generalized.

Key words: data mining, fault diagnosis, fingerprint, unit vector, ratio coefficient, correlation coefficient