论文部分内容阅读
现有的测试性验证试验方案在一定的风险约束下都需要较大的故障样本量。针对这一问题,以故障检测率(fault detection rate,FDR)为验证指标,提出利用研制阶段试验数据和专家信息制定测试性验证试验方案的贝叶斯方法。该方法首先利用研制阶段试验数据建立了产品的FDR增长模型,以此描述FDR在研制阶段的变化趋势,然后利用专家信息确定模型中的超参数,进而得到FDR的验前分布,最后依据贝叶斯最大后验风险准则制定了新的测试性验证试验方案。通过实例的对比分析表明,与经典试验方案相比,新方案样本量减少效果明显,在样本量保持不变的条件下双方风险大大降低。
The existing test verification scheme requires a large amount of fault samples under certain risk constraints. Aiming at this problem, a Bayesian method based on fault detection rate (FDR) is proposed to develop a test verification scheme based on experimental data and expert information. The method first uses the experimental data to establish the FDR growth model of the product to describe the trend of the FDR in the development stage and then uses the expert information to determine the hyperparameters in the model to obtain the pre-test distribution of the FDR. Finally, Sri Lanka’s largest a posteriori risk guidelines developed a new test verification test program. The comparative analysis of the examples shows that compared with the classical experimental scheme, the effect of reducing the sample size of the new scheme is obvious, and the risk of both sides is greatly reduced under the condition that the sample size remains unchanged.