Predicting Motion of Engine-Ingested Particles Using Deep Neural Networks

Abstract

The ultimate goal of this work is to facilitate the design of gas turbine engine particle separators by reducing the computational expense to accurately simulate the fluid flow and particle motion inside the separator. It has been well-documented that particle ingestion yields many detrimental impacts for gas turbine engines. This ingestion is of concern for operation in environments where dust, ash, or ice persist. The consequences of ice particle ingestion can range from surface-wear abrasion to engine power loss. Ice particles are chosen for this study because of their relevance to civil aviation. It is known that sufficiently small particles, characterized by small particle response times (τp), closely follow the fluid trajectory whereas large particles deviate from the streamlines. The behavior of small particles hints at a method for larger particle trajectories because the higher order terms (HOT) in the asymptotic particle acceleration solution can be shown to be O(τp). By explicitly considering τp, these HOT can be derived. Rather than manually deriving these terms, this work chooses to implicitly derive them using machine learning (ML). Inertial particle separators are devices designed to remove particles from the engine intake flow. Particle separators contribute to both elongating the lifespan and promoting safer operation of aviation gas turbine engines. Complex flows, such as flow through a particle separator, naturally have rotation and strain present throughout the flow field. This study attempts to understand if the motion of particles within rotational and strained canonical flows can be accurately predicted using supervised ML. This report suggests that preprocessing the ML training data to the fluid streamline coordinates can improve model training. Furthermore, this work provides some guidelines for applying ML, particularly deep feed-forward neural networks, with physics driven multiphase flow data. Additionally, the ML model is able to predict the particle accelerations in the fully rotational and irrotational canonical laminar flows quite well. For combined flows with rotation and strain, however, the model struggles to predict the particle accelerations.