关键词: 局部平衡判别投影, 降维, 故障诊断, 旋转机械

Abstract: Aiming at the problems of classification difficulty caused by multi-class and high-dimensional complex characteristics of rotor fault data, a LBDP dimensionality reduction algorithm was proposed. First of all，the mixed features of the rotor vibration signals were extracted from multiple angles in time domain, frequency domain and time-frequency domain，and the high-dimensional feature sets were constructed. The original feature sets were fused by LBDP algorithm, and the low-dimensional sensitive feature subsets which might best reflect the intrinsic information of the faults were selected. Then the low-dimensional feature subsets were input into K-nearest neighbor(KNN) classifier for training and fault classification. The effectiveness of the proposed method was verified by the vibration signal sets of a double-span rotor systems, and it is proved that the method may extract the local discriminant information comprehensively and make the difference among fault categories clearer. 

Key words: locality-balanced discriminant projection(LBDP), dimensionality reduction, fault diagnosis, rotating machinery