Thermal performance analysis of photovoltaic/thermal hybrid systems with artificial roughness using machine learning models

• Hybrid solar panel achieved max TE of 28.2% at 0.0115 kg/s • SVR+DT model outperformed ANN and CatBoost with R² = 98.74% • Type 3 roughness design improved TE by ∼20% over smooth plate • ML predictions matched experiments, reducing need for repeated trials. The ability of hybrid solar panels to simultaneously generate thermal and electrical energy has garnered significant attention as the demand for non-traditional and renewable energy sources increases. A flat plate hybrid solar system with various artificial roughness configurations was experimentally evaluated for thermal performance and efficiency supported by machine learning (ML) techniques for prediction. Three synthetic roughness configuration was tested: Type 1 (no fins or baffles), Type 2 (nine fins), and Type 3 (four fins and 24 baffles). These configurations were selected to systematically assess the effect of increasing turbulence—where fins and baffles are expected to disrupt the laminar boundary layer, enhance fluid mixing, and improve heat transfer compared to a smooth duct. Two distinct air mass flow rates (MFR) of 0.0115 k g / s and 0.0082 k g / s were used in the tests. Thermal efficiency (TE) was selected as the target variable determined based on the first law of thermodynamics through energy balance equations. TE was modeled and predicted using two supervised ML techniques: Support Vector Regression (SVR) and Decision Tree (DT). Experimental data was used for both training and validation to assess the models' accuracy and consistency. Both models demonstrated strong predictive power with respective root mean square error values of 0.8692 and 0.8956. The nonlinear interactions between flow conditions, roughness geometry, and TE were represented by both models. By combining ML approaches, notably SVR and DT, the research demonstrates that hybrid prediction frameworks can reliably anticipate TE trends in hybrid solar systems, reducing reliance on exhaustive experimentation.