Deep reinforcement learning for the control of conjugate heat transfer

This research gauges the ability of deep reinforcement learning (DRL) techniques to assist control conjugate heat transfer systems governed by coupled Navier–Stokes and equations. It uses a novel, “degenerate” version proximal policy optimization (PPO) algorithm, intended for situations where optimal be learnt neural network does not depend on state, as is notably case in open-loop problems. The numerical reward fed computed with an in-house stabilized finite elements environment combining variational multi-scale (VMS) modeling governing equations, immerse volume method, multi-component anisotropic mesh adaptation. Several test cases natural forced convection two three dimensions are used testbed developing methodology. approach successfully alleviates induced enhancement two-dimensional, differentially heated square cavity controlled piece-wise constant fluctuations sidewall temperature. also proves capable improving homogeneity temperature across surface three-dimensional hot workpieces under impingement cooling. Various tackled, which position multiple cold air injectors optimized relative fixed workpiece position. flexibility framework makes it tractable solve inverse problem, i.e., optimize injector distribution. obtained results showcase potential method black-box practically meaningful computational fluid dynamics (CFD) systems. More significantly, they stress how DRL can reveal unanticipated solutions or parameter relations (as symmetrical actuation turns offset from symmetry axis), addition being tool optimizing searches large spaces.