Assessment on bearing performance degradation using improved discriminant LPP

Improved Discriminant Local Preserving Projection (IDLPP) method is investigated to get the knowledge of degradation from testing data for the bearings of momentum wheels. By the means of adjusting the neighbor parameter according to its sample density, IDLPP method is less sensitive to neighbor parameters than LPP method. Furthermore, IDLPP method makes great improvement on feature extraction by introducing a supervision for category information of data and combining maximum identification criterion into traditional LPP method. With these advantages IDLPP method achieves its application on degradation feature extracting from vibration data. Simulation results demonstrate that IDLPP has a satisfactory performance on the dimensionality reduction and it also improves the manifold distortion.