(215d) Novel Techniques for Surrogate Based Global Sensitivity Analysis on Low Pressure Fluidized Bed Encapsulation

This paper presents a novel approach to quantify sensitivity indices for the input parameters used in a low pressure dynamic fluidized bed encapsulation system by surrogate model based global sensitivity analysis (GSA). The physics- based model involves different kind of input variables which can be both spatial and temporal in nature. The input variables are related to fluidization media properties, atomization air and coating solution properties, size and scalability of the bed etc, whereas the output variables are coating thickness of the final product and the coating efficiency. A two-fold scheme can be developed for GSA, where the effect of input parameters can be tested on the dimensionless numbers associated with fluid dynamics, heat and mass transfer, and later those can be mapped further to the final output variables on the basis of coating performance. The physics model-based GSA involves rigorous computational cost due to repetitive model evaluations for sample generation using Monte Carlo simulation, especially when the model has a large number of input parameters. To compensate for that, a data driven approach is required in the form of surrogate modeling to simplify the physics-based model. Three different surrogate modeling techniques have been discussed namely: polynomial chaos expansion (PCE), support vector regression (SVR) and neural network based radial basis function (RBF). The accuracy and efficiency of each of these methods have been compared in the scope of the fluidized bed encapsulation model on the basis of Sobol indices.