Physics-Informed Proper Orthogonal Decomposition for Accurate and Superfast Prediction of Thermal Field

Abstract Thermal field prediction has garnered ever-increasing attention as an urgent and vital issue in broad applications ranging from thermal management, performance prognosis, lifetime evaluation, and safety assessment, to energy conversion and carbon neutrality. Suffering from the huge amounts of data and iterative iterations, traditional full-order prediction methods are overstretched for rapid predictions and analysis of complex physical fields. In contrast, reduced-order methods, like proper orthogonal decomposition, can tackle such issues with accelerated computational efficiency but predictions and design may be physically inconsistent or implausible. Here we develop a physics-informed proper orthogonal decomposition for the acceleration of thermal field prediction. By introducing a unified index matrix to reduce the amount of processed data and to uniform the physical equations with the reduced-order equations, we achieve accurate and superfast predictions of thermal fields for unstructured grid, validated by typical complicated spray cooling experiments. The amount of data to be processed achieved a reduction of ten million times, with a maximum computational speedup of 101 times. The physics-informed proper orthogonal decomposition framework is demonstrated to be highly efficient and accurate and can be extended to address a wide range of scientific and technological applications beyond thermal field predictions.