Turbulent flow prediction: Lagrangian Particle Tracking-Deep Learning (LPT-DL) based models

Turbulent flow is a common occurrence in various natural and artificial processes. Predicting and modeling turbulent flow poses a significant challenge, as traditional numerical methods are limited by computational costs and constraints. Furthermore, conducting experimental studies on turbulent flow is constrained by factors like scale and expenses.

In recent times, the integration of deep learning with High-performance computing has emerged as a promising approach to simulate and predict fluid flow behavior. In this study, we propose a deep learning methodology that combines long short-term variants and Transformer models to forecast the velocity of a strained incompressible turbulent flow. The prediction is based on experimental datasets based on Lagrangian particle tracking technique, specifically focusing on Taylor microscale Reynolds numbers, Re_λ ranging from 100 to 500, and vertical mean strain rates 2S of 4 and 8 s^-1.

The results obtained from this approach demonstrate remarkable achievements. Nonetheless, further investigations are essential to determine the capability of these models in predicting the duration of flow periods and the range of Reynolds numbers they can accurately handle. By addressing these aspects, we can enhance the reliability and utility of deep learning techniques in turbulent flow prediction.