Machine learning-assisted effective thermal management of rotor-stator systems

Abstract

Rotor-stator systems are widely used to facilitate highly effective energy management and process intensification but have complex design characteristics, such as skittish flow and heat transfer. Due to the difficulty in obtaining heat transfer data for thermal management of rotor-stator systems, uneconomical designs with large margins often occur. Here, we propose a comprehensive design process using machine learning guided by convective heat transfer experimental data. We organize domain knowledge involving advection and convective heat transfer using a test rig employing a naphthalene sublimation method on large surfaces. Detailed heat transfer data elucidates that local thermal-fluid characteristics are non-linear as the operating conditions change, which is supported by numerical simulation models. We develop a machine learning-based heat transfer prediction model using pure experimental data and derive the optimal algorithm model. Thermal analysis via 2D thermal circuits provides an effective design process and enables rapid quantified evaluation of critical performance considerations. An improved rotor-stator system simultaneously achieves high heat flux of 1.9 kW/m2 and low thermal resistance of 17.1 K/W to both disks. We expect our experiment-based machine learning-assisted thermal management will serve as a robust and informative indicator for improving the efficiency and feasibility of applications with multivariate thermal-fluid systems.