A hyper-optimisation method based on a physics-informed machine learning and point clouds for a flat plate solar collector

Abstract This paper presents a new way to hyper-optimise flat plate solar collector using combination of regenerated point clouds, constructal theory, and physics-informed machine learning (PIML). The behaviour the is studied as radiation ambient temperature change, both precise numerical simulation PIML. novel hyper-optimisation method integrates these two approaches improve performance collector. In this method, volume fluid solid structure (FPSC) transformed into clouds based on theory. are then continuous uniform 3D geometry optimised parameters. To put modified version practice, computational used generate training data set for learning, specifically neural networks. After thoroughly verifying results, PIM trained generated set. study marked first time that regular replaced with PIML output reduce cost prediction. second layer calculation, deep network make predictions outputs by Seven independent parameters predict heat transfer efficiency over time, including flux, mass flow rate, inlet temperature, number pairs clusters, fraction nanofluid, while 16 hidden layers 63 learnable neurons engaged in prediction layer. matrix redefined theory principles series iteration loops (CTFPSC). results indicated could calculation costs 18.31% compared method. addition, reveal maximum outlet temperatures possible when N c > 3 p 5.