Sensitivity method for extreme-based engineering problems
提出一种简单方法，用于分析工程模型中输入因素对极值（最大值或最小值）结果的影响，无需定义概率分布，计算成本低，适合可靠性设计。
<p>Engineering models are becoming increasingly complex, involving a larger number of variables. As a result, engineers struggle to deeply understand the models and therefore validate and properly use them. Sensitivity analyses (SA) are usually conducted to help with this task. Nonetheless, the knowledge and computational efforts required to conduct SA increases with the complexity of the model. This paper presents a conceptually simple method to conduct the SA of extreme-based engineering problems, that is, those problems whose outcome of interest is their maximum or minimum values. This type of problems is of relevance for reliability-based design. The method requires to iteratively maximize or minimize the model by fixing one input factor at each time. Then, the variation of the optimal surface is evaluated. The method permits the visualization and measure of the input factors’ influence on the extreme model response, with no need of defining the probability distribution of the input factors. Both the easiness of application and interpretation, along with a low computational cost for problems dealing with extreme values, make the proposed method very convenient to practitioners.</p>