航空学报 > 2014, Vol. 35 Issue (9): 2513-2521   doi: 10.7527/S1000-6893.2014.0097

基于贝叶斯因子和二阶概率方法的圣地亚热传导模型确认挑战问题

赵亮, 杨战平   

  1. 中国工程物理研究院 电子工程研究所, 四川 绵阳 621900
  • 收稿日期:2013-12-02 修回日期:2014-05-12 出版日期:2014-09-25 发布日期:2014-05-28
  • 通讯作者: 杨战平,Tel.:0816-2494743 E-mail:zhanpingy@aliyun.com E-mail:zhanpingy@aliyun.com
  • 作者简介:赵亮 男,博士研究生。主要研究方向:模型验证与确认、系统可靠性评估。Tel:0816-2493265 E-mail:swjtu_zhaoliang@126.com;杨战平 男,博士,研究员,博士生导师。主要研究方向:复杂系统仿真和评估。Tel:0816-2494743 E-mail:zhanpingy@aliyun.com
  • 基金资助:

    国防预研项目(426010401);中国工程物理研究院科学技术发展基金(2012B0403058)

Sandia Model Validation Thermal Challenge Problem Based on Bayes Factor and Second-order Probability Method

ZHAO Liang, YANG Zhanping   

  1. Institute of Electronic Engineering, China Academy of Engineering Physics, Mianyang 621900, China
  • Received:2013-12-02 Revised:2014-05-12 Online:2014-09-25 Published:2014-05-28
  • Supported by:

    National Defense Pre-research Foundation (426010401); Science and Technology Development Foundation of China Academy of Engineering Physics (2012B0403058)

摘要:

针对美国圣地亚国家实验室为促进模型确认方法的发展所提出的热传导挑战问题,根据现代模型确认思想总结了该挑战问题解答中应包含的核心内容。在确认度量环节采用贝叶斯因子考察了实验数据是否支持所给模型,在此基础上通过二阶概率方法得到了模型预测的分布,以此计算出模型预测结论的置信度。该过程中考虑了模型参数的随机不确定性和认知不确定性,最后通过灵敏度分析辨识了模型参数的不确定性对模型预测的影响。研究表明该挑战问题中的实验数据支持所给模型,模型预测受导热系数的不确定性影响最大,模型预测材料在调控条件下的失效概率不满足调控要求,该结论的置信度为99.97%。

关键词: 模型确认挑战问题, 贝叶斯因子, 确认度量, 二阶概率方法, 模型预测, 灵敏度分析

Abstract:

The objective of the thermal model validation challenge problem presented at American Sandia National Laboratory is to develop methodologies associated with model validation. The article summarizes the key content to answer this problem based on modern model validation idea. Bayes factor is employed as a validation metric to find whether the experimental data supports the model. The distribution of model prediction is obtained by the second-order probability method which propagates both aleatory uncertainty and epistemic uncertainty through the model, and the confidence of the model prediction is calculated from the distribution. Finally the article makes a sensitivity analysis with the model parameters to identify their effects on the prediction. The study indicates that the experimental data supports the given model, the thermal conductivity contributes most to the uncertainty in the model prediction, and with 99.97% confidence we can conclude that the failure probability of the material under regulatory condition predicted by the model doesn't meet the regulatory criterion.

Key words: model validation challenge problem, Bayes factor, validation metric, second-order probability method, model prediction, sensitivity analysis

中图分类号: