Data-driven dimensional analysis of critical heat flux in subcooled vertical flow: A two-stage machine learning approach

Abstract

This study presents a novel two-stage machine learning algorithm that identifies dominant dimensionless numbers for critical heat flux (CHF) prediction, circumventing the limitations of traditional dimensional analysis by integrating the Buckingham Pi theorem with advanced computational methods, including the active subspace method and machine learning. When applied to extensive datasets, including 1438 data points from subcooled departure from nucleate boiling (DNB) conditions and validated against a non-overlapping 24,284-point lookup table dataset, our algorithm effectively discerns and validates the dimensionless numbers critical to the CHF phenomena. Notably, it identifies a significant negative correlation between a newly discovered dominant dimensionless number and the boiling number Bochf = q″w,chf / (G hfg), and introduces a new simplified scaling law for water under earth's gravitational conditions, outperforming existing models with substantial predictive accuracy. This scaling law and the identified dimensionless numbers, derived from an initial dataset of 1438 data points, exhibit exceptional generalization capabilities when validated against a significantly larger dataset of 24,284 points. These findings also emphasize the nonlinear effects of inlet quality on the boiling number Bochf. The contribution of this research lies in its potential to deepen the understanding of CHF based on data-driven fashion, with the potential to enhance thermal management and design optimization across various applications.