Machine learning driven metaheuristics to optimize the design-operational parameters for enhanced pool boiling CHF subject to a range of surface-liquid combinations

Critical heat flux (CHF) is crucial for heat-flux controlled boiling applications. Designing heat sinks with larger CHF limits is essential for two-phase thermal systems. However, identifying optimal operational and heater surface morphological parameters remains challenging. Most empirical correlations focus on CHF prediction rather than inverse design, showing large errors and limited applicability. We previously developed a deep neural network (DNN) for CHF using a database of 13,000 data points from 42 pool boiling studies, incorporating 29 working fluids across various surface textures and operating conditions. The DNN model achieved a R² score of 0.97. The present study addresses the inverse design problem: identifying optimal design and operational parameters for targeted CHF values. An optimized Genetic Algorithm (GA) model is proposed to predict these parameters. The analysis includes HFE-7000, Water, HFE-7200, and FC-72, evaluated under typical CHF benchmarks for substrates with thermal conductivities from 205 to 385 W/m.K. For HFE-7000, at CHFs of 120 kW/m² and 252 kW/m² on a substrate with 385 W/m·K conductivity, inclination angle values decrease as CHF increases, due to vapor removal difficulties. For HFE-7200 at CHFs of 41.11 kW/m² and 93.8 kW/m² on a 205 W/m·K substrate, parameter values increase with CHF, except temperature and surface roughness. These models serve as tools for predicting CHF and identifying optimal parameters for heating surfaces to meet desired CHF targets.