Physics-Embedded Deep Learning to Predict Real-Time Flow Parameters in Complex Thermodynamic Machinery

Abstract

Flow through complex thermodynamic machinery is intricate, incorporating turbulence, compressibility effects, combustion, and solid–fluid interactions, posing a challenge to classical physics. For example, it is not currently possible to simulate a three-dimensional full-field gas flow through the propulsion of an aircraft. In this study, a new approach is presented for predicting the real-time fluid properties of complex flow. This perspective is obtained from deep learning, but it is significant in that the physical context is embedded within the deep learning architecture. Cases of excessive working states are analyzed to validate the effectiveness of the given architecture, and the results align with the experimental data. This study introduces a new and appealing method for predicting real-time fluid properties using complex thermomechanical systems.