Intelligent and quantitative ligament breakup event analysis in 65 kHz off-axis holographic video of swirl spray

Abstract

The mechanism of ligament breakup events in sprays significantly impacts both daily life and industry, particularly with regard to swirl sprays in aero-engines. Specifically, three critical challenges must be addressed: tracking the trajectories of ligaments across frames, elucidating the relationships between parent and daughter ligaments, and capturing the spatial and temporal information of breakup events. This study introduces an intelligent pipeline for the quantitative analysis of ligament breakup events. The integration of Segment Anything Model 2 (SAM 2) enables automatic segmentation, producing ligament masks from low signal-to-noise ratio holographic videos. To effectively provide prompt information, four techniques are proposed: YOLO detection for initial bounding box prompts, box post-processing, video clip generation, and prompt giving. Based on the preliminary masks generated by SAM 2, the grayscale values of the masked pixels are utilized for refined segmentation through clustering of K-means, accurately delineating the internal structures and edges of the ligaments. A manually curated validation dataset is established, and three evaluation metrics (precision, recall and F1 score) are used to assess the segmentation performance of seven different methods, demonstrating the superior efficacy of the proposed approach. The influences of frame rate and image quality are also discussed. Subsequently, the centroids and areas of connected regions within the segmentation results are extracted. A light flow field is utilized to initiate ligament trajectory matching. To determine the relationship between parent and daughter ligaments during the breakup process, the stability of relative horizontal positions between connected regions before and after the breakup is analyzed, significantly reducing the calculation time. Finally, breakup lengths, ligament lengths, and breakup times are calculated to represent the statistical characteristics of breakup events.