Abstract
This paper presents a new way to hyper-optimise a flat plate solar collector using a combination of regenerated point clouds, constructal theory, and physics-informed machine learning (PIML). The behaviour of the flat plate solar collector is studied as solar radiation and ambient temperature change, using both precise numerical simulation and PIML. The novel hyper-optimisation method integrates these two approaches to improve the performance of the solar collector. In this method, the volume of fluid and solid structure of the flat plate solar collector (FPSC) is transformed into point clouds based on constructal theory. The point clouds are then regenerated into a continuous and uniform 3D geometry using optimised parameters. To put the modified version of the flat plate solar collector (FPSC) into practice, a computational method is used to generate a training data set for machine learning, specifically for neural networks. After thoroughly verifying the computational results, the PIM is trained using the generated training data set. This study marked the first time that a regular computational method is replaced with PIML output to reduce the computational cost of prediction. In the second layer of calculation, a deep neural network is used to make predictions based on the outputs generated by PIML. Seven independent parameters are used to predict heat transfer and efficiency over time, including time, heat flux, mass flow rate, inlet temperature, number of pairs and clusters, and volume fraction of nanofluid, while 16 hidden layers and 63 learnable neurons are engaged in this prediction layer. The geometry matrix is redefined by constructal theory principles in a series of iteration loops to generate the first flat plate solar collector based on constructal theory (CTFPSC). The results indicated that the hyper-optimisation method could reduce calculation costs by 18.31% compared with the regular computational method. In addition, the results reveal that maximum outlet temperatures are possible when N c > 3 and N p> 5. ? 2023, The Author(s).
