A hydrodynamic prediction model of throttle orifice plate using space filling and adaptive sampling method

The hydrodynamic prediction model improved design efficiency of the flow control application. A global sampling process with space filling method, adaptive sampling method, and neural network is proposed to generate a hydrodynamic prediction model for the flow control device. The optimized Latin hypercube sampling method is applied to create a uniform set of initial sample points for analysis. An automatic computational fluid dynamics analysis process is developed to provide data for the hydrodynamic objective of the multi-stage throttle orifice plate. By self-learning from previous sample points, the new sample points placed in the region of interest are collected by the adaptive sampling method in several iterations. Two numerical examples and the flow rate modeling for the throttle orifice plate are provided to demonstrate the approximation capability of the proposed process. Ten prediction model cases are established by a back-propagation feed-forward neural network, and the test data show that the model constructed by optimized Latin hypercube sampling and adaptive sampling has 1% mean relative error and 4.85% maximum relative error, which is nearly 25% more accurate than the optimized Latin hypercube sampling model. The proposed global sampling process is efficient to build the hydrodynamic prediction model and shows potential in the flow control design area.