SACE: Statistics-Aided Computer Experiments

随着计算的进步，用大型计算机代码实现的复杂数学模型已经成为描述物理过程和真实系统的有效工具。许多行业已经广泛使用计算机模型来取代物理实验，并在缩短开发周期和成本方面看到了好处。在这项提案中，将结合统计学和工程学的思想，为计算机实验的规划、建模和分析开发新的方法和技术。将开发一种贝叶斯方法来整合来自两个来源的数据：高精度但昂贵的实验和低精度但昂贵的实验。由于系统或产品的性能可能会受到其噪声变化的严重影响，因此在进行优化之前需要考虑系统的稳健性。将开发稳健设计和优化方面的新技术，以处理具有许多噪声因素的问题，并解决多种性能衡量标准。为了说明和测试所提出的方法，将考虑在蜂窝合金材料设计和数据中心热分布方面的具体应用。更广泛的应用包括制造、材料、环境监测、水文学和影响动力学。

With advances in computing, complex mathematical models implemented in large computer codes have become effective in describing physical processes and real systems. Many industries have made extensive use of computer models to replace physical experiments and seen the benefits in reduced development cycle time and cost. In this proposal a combination of statistical and engineering ideas will be used to develop novel methodologies and techniques for the planning of computer experiments, their modeling and analysis. A Bayesian approach will be developed to integrate data from two sources: high-accuracy but expensive experiments and low-accuracy but cheaper experiments. Since the performance of a system or product can be severely affected by its noise variations, system robustness needs to be considered before performing optimization. New techniques in robust design and optimization will be developed to handle problems with many noise factors and to address multiple performance measures. Specific applications to cellular alloy material design and data center thermal distribution will be considered for illustrating and testing the proposed methodology. Broader applications include manufacturing, material, environmental monitoring, hydrology, and impact dynamics.