Adaptive Hessian LLE in mechanical fault feature extraction

In response to mechanical fault in high dimensional feature extraction problem, this paper presents an adaptive choosing neighborhood of manifold learning algorithm. The algorithm based on manifold local curvature estimation of tangent space, can make of all sample points adaptive selection of neighborhood. Adaptive selection of neighborhood algorithm applied to the Hessian locally linear embedding (HLLE), the improved HLLE in the neighborhood graph is constructed to ensure the local linearity, thus ensuring the Hessian LLE reduction performance. Eventually the adaptive HLLE is applied to the rolling bearing of four kinds of fault feature extraction, extracted from the sample of low-dimensional feature and recognition accuracy results show, adaptive HLLE algorithm in neighborhood selection on parameter selection has a stronger robustness, extracting mechanical fault low-dimensional feature more accurate.