Adaptive Virtual Support Vector Machine for the Reliability Analysis of High-Dimensional Problems
Abstract
In this study, an efficient classification methodology is developed for reliability analysis that maintains an accuracy level similar to or better than existing response surface methods. Sampling-based reliability analysis requires only classification information, a success or a failure, but response surface methods provide real function values as their output, which requires more computational effort. The situation is more challenging with high-dimensional problems due to the curse of dimensionality. In the newly proposed virtual support vector machine (VSVM), virtual samples are generated near the limit state function by using linear or Kriging-based approximations. The exact function values are used for approximations of virtual samples to improve the accuracy of the resulting VSVM decision function. By introducing the virtual samples, VSVM can overcome the deficiency in existing classification methods where only classified function values are used as their input. The universal Kriging method is used to obtain virtual samples to improve the accuracy of the decision function for highly nonlinear problems. A sequential sampling strategy that chooses a new sample near the true limit state function is integrated with VSVM to maximize accuracy. Examples show that the proposed adaptive VSVM yields better efficiency in terms of modeling time and the number of required samples while maintaining similar or better accuracy, especially for high-dimensional problems.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 2011
- Accession Number
- ADA558429
Entities
People
- David Lamb
- Hyeongjin Song
- Ikjin Lee
- Kyung K. Choi
- Liang Zhao
Organizations
- University of Iowa