Design of Computer Experiments: A Review
- Modern DoE techniques are comprehensively reviewed.
- A detailed classification and chronological evolution of modern DoE research is presented.
- Our numerical and visual analyses revealed the excellent high dimensional performance of SOB3.
- Rapidly growing class of adaptive DoE is critically discussed.
- Several potential opportunities for future research in modern DoE are discussed.
In this article, we present a detailed overview of the literature on design of computer experiments. We classify the existing literature broadly into two categories viz. static and adaptive design of experiments (DoE). We discuss the abundant literature available on static DoE, their chronological evolution, and their pros and cons. Our numerical and visual analyses reveal the excellent performance of Sobol sampling based on recent work of Joe and Kuo (SOB3) at higher dimensions while showing that Hammersley (HAM) and Halton (HAL) sampling are suited for lower dimensions. Our investigation of these techniques highlight the vital challenges that are dealt by adaptive DoE techniques, an upcoming class of modern DoE. They employ intelligent and iterative strategies that combine system knowledge and space-filling for sample placement. Adaptive DoE literature is critically analyzed based on the
Keywords: Adaptive sampling, Computer experiments, Design of experiments, Space-filling, Surrogate development,
Associated Project: Numerics