Experimental analysis of aluminum-droplet combustion in solid- propellant conditions using deep learning
Résumé
Shadowgraphy is good technique to image aluminum combustion for small solid-propellant samples. But classical detection methods struggle to provide accurate data on aluminum droplets with such different aspects. A new semantic-segmentation approach is tested using deep learning to classify every image pixel over various classes. The UNET network was trained on 45 annotated images for combustion at 1.0 and 2.0 MPa. Detection performances were compared to those of the classical Maximally-Stable Extremal Regions algorithm. The semantic-segmentation approach was able to distinguish droplets from the surrounding flame with better detection performances (Precision, Recall).
Origine | Fichiers éditeurs autorisés sur une archive ouverte |
---|
Loading...