| CPC G06F 30/10 (2020.01) [G06F 30/27 (2020.01); G06F 30/28 (2020.01)] | 20 Claims |
|
1. A computer-implemented method for heat exchanger design in the manufacture of a heat exchanger, the heat exchanger design comprising boundary representation of one or more shapes that form the manufactured heat exchanger, the method comprising:
receiving initialized control points that set a curve or composite curve, the curve or composite curve define a geometric shape for the one or more shapes of the heat exchanger, the control points located within a design space;
performing iteratively:
parametrizing the design space with a parametric function using the curve or composite curve set by the control points to define surface boundary conditions;
determining heat transfer and pressure drop of the heat exchanger design defined by the control points using the parameterized design space;
determining whether the heat transfer, pressure drop, or both, meet one or more predetermined conditions;
where the one or more predetermined conditions are met, performing no further iterations;
where the one or more predetermined conditions are not met, determining an updated policy that maximizes a cumulative reward toward maximized heat transfer and minimized pressure drop, the updated policy determined by performing deep reinforcement learning, an agent of the deep reinforcement learning taking the control points and the determined heat transfer and pressure drop as input, and determining updated control points using the updated policy;
determining the one or more shapes, from the updated control points, that form the heat exchanger design; and
outputting the heat exchanger design to manufacture the one or more shapes comprising the heat exchanger.
|