Predicting the heat transfer enhancement in solar flat plate collectors using physics informed neural networks

This study aims to enhance the overall thermal efficiency of the flat plate solar water collector system (FPSWCS) by utilising a novel absorber tube design that increases the heat transfer rate. To achieve this goal, both a conventional and a modified version of the FPSWCS with increased contact surface length is designed and constructed. Comparative experiments were carried out on the roof top of the solar thermal engineering laboratory, SRM Institute of Science and Technology (10.7560°N, 78.6513°E), Tiruchirappalli, India. The experimental results revealed that the overall performance of the modified FPSWCS, with its extended line of contact length surpassed the conventional rig by 14.08% (in Level – 1) and 17.61% (in Level – 2). Additionally, this study explores the use of the physics-informed neural network (PINN), a machine-learning technique to predict the thermal performance recorded by both the conventional and modified systems. The PINN proposed herein incorporates four features (three inputs and one output). Real-time data were employed to assess the model's predictive accuracy, demonstrating that the results closely align with real-time values. The proposed PINN model achieves an impressive 99.92% accuracy in predicting the thermal power of a flat plate collector.ABBREVIATIONS: SECD: solar energy conversion device; FPSWCS: flat plate solar water collector system; C-FPSWCS: conventional flat plate solar water collector system; N-FPSWCS: novel flat plate solar water collector system; D-F: direct flow; W-F: web flow; SPI-F: spiral flow; P-F: parallel flow; O-F: oscillatory flow; SER-F: serpentine flow; MFR: mass flow rate (kg/s); AT: ambient temperature (°C); FIT: fluid inlet temperature (°C); FOT: fluid outlet temperature (°C); APT: average plate temperature (°C); G: irradiance (w/m2); Nu: Nusselt number; L: length (m); W: width (m); H: height (m); RH: relative humidity (%); WS: wind speed (m/s); ΔP: pressure drop; PEC: performance evaluation coefficient; ηT: thermal efficiency (%); ηE: exergy efficiency (%); Re: Reynolds number; nv: number of vents; Lat: latitude; Level 1: first phase of the experiment; Level 2: second phase of the experiment; PINN: Physics informed neural network; NN: neural network; L: Overall loss; Ldata : data driven loss; Lphysics: physics informed loss; Lreg: regularisation loss; PDE: partial differential equation; δ: partial differential; ET-C: thermal power output recorded by the C-FPSWCS (Watt); ET-N: thermal power output recorded by the N-FPSWCS (Watt); ΔT-C: temperature difference recorded by the C-FPSWCS (°C); ΔT-N: temperature difference recorded by the N-FPSWCS (°C); AO: accuracy (%); pred.: predicted value; exp.: experimental value