Flow field reconstruction from spray imaging: A hybrid physics-based and machine learning approach based on two-phase fluorescence particle image velocimetry measurements

Abstract

The interaction between liquid spray and the surrounding air is crucial in fluid research, especially in the study of fuel spray and combustion. However, the fuel spray–air interaction is a complex process influenced by multiple factors, including fuel type, fuel injection pressure, and fuel temperature. These factors are coupled together, making it challenging and time-consuming to accurately capture the spray–air data using traditional experimental methods alone. The current study proposes a hybrid physics-based and machine learning model for utilizing spray images to reconstruct ambient flow fields. The novelty of this work lies in leveraging the spatial characteristics of spray and airflow data to optimize feature extraction and reduce unnecessary nonlinearity in the model. Consequently, the model offers complementary advantages, improving model interpretability and reducing its reliance on massive data. The training dataset is collected using a combined diagnostic approach, utilizing Mie-scattering imaging and fluorescence particle image velocimetry. The liquid spray and the ambient air velocity field are measured simultaneously under a wide range of experimental conditions, including different fuel types, fuel injection pressures, and fuel temperatures. The reconstruction results are validated against unseen experimental data. In general, the reconstruction results indicate that the model is accurate, fast, and robust for different fuel conditions and injector types. It provides an innovative way to reconstruct airflow fields based on spray images (spray density distribution). These findings highlight the potential of integrating physics-based and machine learning methods for multiphase flow diagnostics, paving the way for broader data-driven applications in fluid research.