Akademik Veri Yönetim Sistemi
INTERNATIONAL COMMUNICATIONS IN HEAT AND MASS TRANSFER, cilt.172, 2026 (SCI-Expanded, Scopus)
Flow boiling in microscale is known for its complex nature and high performance; however, by using machine learning (ML) techniques, predictive analyses can be performed for complex flow nature by eliminating long and costly experimental processes. Here, ML models were applied for first time to datasets got from flow boiling experiments performed by cross-collector/distributor heat sink with pin-fins under transient and variable thermal loads. Four different algorithms-Artificial Neural Networks (ANN), Ensemble of Trees (ET), Support Vector Machines (SVM), and Linear Regression (LR)-were applied and compared to estimate characteristics of micro-flow-boiling via input parameters, namely, mass flux (G = 155, 220, 285 kgm-2 s-1), inclination angle (IA = 0,10,25,40,55 degrees), and effective heat flux (q(y)eff=179.7 to 382.6 kWm-2). A high-speed-visualization was performed. It resulted that increasing inclination angle had a beneficial impact on thermal performance at relatively low mass flux (G = 155 kgm-2 s-1). For higher mass fluxes (G = 285 kgm-2 s-1), lowest thermal performance was observed at highest angle (IA = 55 degrees). Pressure drop (Delta P) increased with applied heating power (qap), and influence of IA on Delta P was negligible. The Ensemble of Trees (ET) model provided most accurate and reliable predictions across all target variables, while Linear Regression (LR) model showed least accurate predictions among the four models.