Akademik Veri Yönetim Sistemi
Applied Thermal Engineering, cilt.294, 2026 (SCI-Expanded, Scopus)
This study presents Artificial Intelligence (AI) driven and physics guided machine learning workflows for advancing the thermofluid design of Split Air Conditioners (SACs) by using Stereo Particle Image Velocimetry (SPIV) derived outlet line velocity profiles as targets for fast, data efficient airflow surrogate modeling. Using 186 SPIV measured configurations spanning volute curvature, tongue angle, and vortex wall geometry, two predictive models were developed and evaluated independently: an Extreme Gradient Boosting (XGBoost) regressor and a Physics Informed Neural Network (PINN). Both models demonstrated strong performance across low and high velocity regions, while the PINN promoted stable and physically plausible outlet-profile predictions through physics-guided regularization. In an independent stratified hold out evaluation, XGBoost achieved 11.5% Mean Absolute Percentage Error (MAPE) and coefficient of determination R2 = 0.92 over the full outlet profile. Consequently, the proposed workflow can reduce computational cost and experimental effort compared to conventional Computational Fluid Dynamics (CFD) based analyses and repeated SPIV prototyping/measurement cycles, enabling faster design iterations. This study proposes an approach for multi output prediction of the 44 point outlet line velocity profile, utilizing SPIV measured profiles as ground truth targets. Additionally, it provides a design oriented assessment that balances prediction accuracy, physical plausibility, and computational efficiency, with specific evaluation of the high velocity outlet region.