Surrogate Model and Optimisation. With the data from the DoE the optimisation process is commenced. First, the response surface model is used to create a surrogate model that correlates the various performance parameters in pump mode and turbine mode to the design parameters [8]. In this study, a design exploration tool by Dassault Systemes called ISIGHT is used to perform the optimization, as well as Design of Experiment and response surface model. Input parameters and output responses associated with all the design points (95 designs from DOE) are arranged in a table format. Initially a Surrogate Model approximation is performed on the data using linear regression response surface model. Then optimization method based on a Non-dominated Sorting Genetic Algorithm – II (NSGA-II) [9], with 100 X 100 (population size X generations) is used to explore the design space, using the surrogate model only. A few best design candidates from optimization are chosen from among Xxxxxx points and are simulated in CFD to assess performance improvements.
