Abstract: Aiming at the non-linear characteristics of vibration signals, a method of fault diagnosis of turbine rotor based on isometric feature mapping (ISOMAP) and support vector machine (SVM) is proposed. For the advantages of ISOMAP that it can maintain the inner feature while projecting the high dimension data onto low dimension data space. The vectors form low dimension data space are extracted, which produced by ISOMAP from high dimension space of vibration signal, and consider them as feature vectors, adopting a new kernel function SVM classifier to diagnosis fault types. The proposed ISOMAP-SVM approach is applied to the rotor fault diagnosis. The results show that this method has higher fault diagnosis accuracy, it can make a visual outcome of fault diagnosis. Compared with other kernel functions, the new kernel function SVM has better diagnosis effect.

Key words: Fault diagnosis, Isometric feature mapping, Manifold learning, Vibration signal