Abstract: Support vector machines can avoid over-fitting and have better generalization ability as compared with neural networks. In this paper, the support vector machines based fault diagnosis algorithms are developed for aero-engines. The influence of kernel function is discussed and the chunking algorithms are analyzed. The methods are presented to select normalized parameter and kernel parameter. With the proposed algorithms, the support vector machines can give correct fault diagnosis results for the gas path components of an aero-engine. The results show that the fault diagnosis algorithms are also able to meet the application requirements and can keep robust when the measurement inputs are disturbed by noises.

Key words: aero-engine, support vector machines, fault diagnosis, kernel, generalization