Modeling and Monitoring of Functional Computer Experiments

Computer experiments play a very important role in engineering and science research and have received a lot of attention from industries (GE, GM, IBM etc.) and research institutes (Los Alamos National Laboratory, Oak Ridge National Labs, SAMSI, etc.). By utilizing complex computer models, use of computer experiments has opened a new avenue towards a better understanding of systems or phenomena that are difficult to carry out physically. Recently, there has been significant research in developing Bayesian approaches for modeling computer experiments and validating computer models. While the developed methods and results have been shown to be very powerful and promising in many applications, they are largely limited to computer models with single (uni-variate) output. Much less research has been developed on modeling and monitoring functional computer experiments. Functional data have arisen in many engineering applications with dynamic computer models, e.g. vehicle crashworthiness test, dynamic stress analysis, and response histories study. The difference between functional data and traditional data is that the output responses or input variables are of functional forms or curves (e.g., acceleration versus time). Although many methods for analyzing computer experiments are available, the literature on functional responses is rather limited.The objective of this research is to develop a systematic modeling approach that helps computer model developer and users (designers and scientists) (i) understand the impact and importance of validating the adequacy of functional computer models, (ii) model and monitor functional computer experiments. In particular, we will developinterpretable and accurate meta-models that combine functional physical experiments and computer model output together.computer model validation metrics and procedures that allow model develops to assess the prediction capability and allow designers to optimize engineering designs.a surveillance strategy for monitoring the adequacy and accuracy of functional computer models dynamically.practical solutions for model developers, scientists, and engineering designers.The complexity of the proposed research requires expertise and experience in engineering design and technical skill in statistical experimental design, modeling, and analysis. The PI has previously received NSF funding on modeling and validating computer experiments and has worked on many engineering applications extensively. He has also done a lot of research in monitoring multivariate and correlated data as well as written papers in modeling computer experiments with functional responses.