Skip to Main content Skip to Navigation
Theses

Outils pour l'étude conjointe par simulation et traitement d'images expérimentales de la combustion de particules d'aluminium utilisées dans les propergols solides

Abstract : The addition of aluminum particles in the solid propellant loading improves propulsive performance, but can also lead to various adverse phenomena, including pressure oscillations. Research has been carried out for many years to improve the understanding of these phenomena, particularly through the use of numerical simulation. However, the input data of the numerical simulation, especially the size and the initial velocity of the aluminum particles in the flow, are often difficult to obtain for real rocket motors. ONERA has been developing a shadowgraphy set-up for several years to visualize aluminum particles near the surface of propellant samples in combustion. The present study deals with the development of tools to analyze the experimental images of the shadowgraphy set-up and to improve the interaction with the two-phase digital simulation. A first part concerns propellant samples containing inert particles, which interest is to make it possible to validate the measurement methods on relatively simple images and with reference data. The implemented tools concern the detection and the tracking of particles in image sequences, as well as the location of the surface of the propellant. Good correspondence of size distributions was obtained with reference distributions. The velocity of particles leaving the surface has been confronted with a simplified model of particle transport in a constant flow. The use of this model has made it possible to emphasize the importance of the population of detected tracks in order to make good use of an average velocity profile, particularly in terms of average diameter. A two-phase flow simulation was then carried out for the shadowgraphy experiment. Different parameters were studied (type and size of mesh, thermodynamic parameters ...) in order to obtain a simulated stationary field for propellant flow. The movement of the simulated inert particles could be compared to the experimental profiles for different injection strategies, either using a mean diameter or using a lognormal distribution. The other part of the study is devoted to the analysis of experimental images of the combustion of aluminum particles. The complexity of the images under these conditions has led to the use of a deep learning semantic segmentation approach, aiming to classify all the pixels of the image into different classes, in particular aluminum droplet and flame. The learning was conducted with a restricted base of annotated images using the U-Net neural network, with various adaptations on the processing of the experimental images were studied. The results are compared to a reference technique based on MSER object detection. They show a clear gain in the use of neural techniques for the segregation of aluminum drops of the flame. This first demonstration of the use of convolutional neuronal network on propellant shadowgraphy images is very promising. Finally, we draw perspectives on experimental image analysis and numerical simulation to improve the joint use of these two tools in the study of solid propellants.
Complete list of metadatas

Cited literature [121 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-02421172
Contributor : Abes Star :  Contact
Submitted on : Friday, December 20, 2019 - 12:09:08 PM
Last modification on : Friday, October 23, 2020 - 4:52:25 PM
Long-term archiving on: : Saturday, March 21, 2020 - 6:12:05 PM

File

76217_NUGUE_2019_diffusion.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-02421172, version 1

Collections

Citation

Matthieu Nugue. Outils pour l'étude conjointe par simulation et traitement d'images expérimentales de la combustion de particules d'aluminium utilisées dans les propergols solides. Mécanique des fluides [physics.class-ph]. Université Paris-Saclay, 2019. Français. ⟨NNT : 2019SACLS229⟩. ⟨tel-02421172⟩

Share

Metrics

Record views

221

Files downloads

151