Reliability analysis of gear vibration based on dimensionality reduction visualization and Kriging

To solve the problems of large computation and low precision during gear vibration reliability analysis, a reliability analysis method based on dimensionality reduction visualization and Kriging was proposed. Sample points were generated by Monte Carlo method. These points were transformed into two-dimensional pole feature space, and then Kriging model was used to predict the dividing line of safe and failure regions. When predicting the dividing line, an active learning approach of selecting points was introduced to establish Kriging model so that the utilization rate of sample points was improved dramatically, thanks to the properties of nonlinear prediction and error estimation of Kriging. Through gear vibration reliability analysis, and by comparing with traditional dimensionality reduction visualization technique, it is shown that the number of calls to the performance function changes from 975 numbers to 149 numbers, and calculation time changes from 12 400 s to 1 810 s. What's more, the result of this method is consistent with that of 100 000 Monte Carlo simulation, so the efficiency and correctness is validated. © 2016, Editorial Department of Journal of Aerospace Power. All right reserved.