The Explainable AI lens: revealing the turbulence dynamics

This study explores the potential of Explainable Artificial Intelligence (XAI) algorithms, with a particular focus on the SHapley Additive Explanations algorithm (SHAP), for providing novel insights into the complex dynamics of turbulence. Turbulence is a highly non-linear and chaotic phenomenon, which presents significant challenges in terms of prediction and modelling. It is often the case that traditional methods are unable to adequately capture the intricate details of turbulent flows, thereby necessitating the exploration of advanced computational techniques. In this study, we utilise convolutional neural networks (CNNs) in conjunction with the SHAP algorithm to analyse and predict turbulent flow patterns in an axisymmetric jet. The CNNs are trained on experimental data with the objective of identifying and learning the underlying patterns of turbulence. The objective of integrating the SHAP algorithm is to enhance the interpretability of the neural network outputs, thereby facilitating a more comprehensive understanding of the physical phenomena that govern turbulence. The findings of the study demonstrate that the SHAP algorithm is an effective method for identifying the most influential features that contribute to the neural network's predictions. This provides valuable insights into the mechanics of turbulent flows. This enhanced interpretability not only facilitates the validation of the model's predictions but also contributes to the field of fluid dynamics by uncovering so far obscured aspects of turbulence.