|
| A combined active learning Kriging model and sequential importance sampling for hybrid reliability analysis with random and interval variables |
| Received:June 01, 2021 Revised:June 15, 2021 |
| View Full Text View/Add Comment Download reader |
| DOI：10.7511/jslx20210601414 |
| KeyWord:inverse problem multi-source uncertainties manifold learning polygonal convex set λ-PDF |
| Author | Institution |
| 韩旭 |
河北工业大学 机械工程学院, 天津 ;湖南大学 机械与运载工程学院, 长沙 |
| 刘杰 |
湖南大学 机械与运载工程学院, 长沙 |
| 陈金龙 |
湖南大学 机械与运载工程学院, 长沙 |
|
| Hits: 542 |
| Download times: 256 |
| Abstract: |
| In order to effectively solve the multi-source uncertain inverse problem coupling response uncertainties and model uncertainties, the manifold learning method with mixed uncertainty quantification is proposed to realize the identification of uncertainties of the unknown structural parameters. The uncertainties from the measured responses are quantified by using the probability model, and the uncertainties from the modeling parameters are quantified by using the polygon convex set model. By establishing a manifold mapping model between model parameters and the cumulative distribution function (CDF) of unknown parameters, the proposed method realizes the decoupling of the uncertainties from responses and model, and transforms the multi-source uncertain inverse problem into the inverse problem with uncertainty from the measured responses. The manifold learning method can transform the high-dimensional CDF into the characteristic parameters in the low dimensional manifold space. Through establishing the mapping relationship between model parameters and characteristic parameters, the CDF of the inversed parameters under the specific model parameters can be rapidly predicted. Further more, the derivative λ-PDF and dimension reduction integral method is proposed to realize the transformation from the inverse problem considering uncertain measured responses into a deterministic inverse problem. The P-box model is adopted to quantify the uncertainty of the inverse parameters. The proposed method can not only realize the effective identification of the unknown parameters, but also accurately quantify the comprehensive impact of the response uncertainties and model uncertainties on the inverse results. |
|
|
|