Développement d'une méthode de caractérisation externe et interne de sprays de carburants par imagerie rapide

Abstract

Combustion quality is directly impacted by the fuel injection phase in internal combustion engines and gas turbines. The spatial and temporal distribution of the liquid needs to be managed to maximize the fuel consumed in the combustion chamber and to minimize pollutant emissions. To characterize the injection process, high-speed imaging followed by an image processing step is commonly used. Given the experimental setup (single hole injector, low fuel density, etc.), shadowgraphy with a high-speed camera is preferred in this study. Several image processing methods exist, ranging from simple image thresholding to alternative methods with contour detection followed by region segmentation. The literature review highlights a lack of information and accuracy with classical approaches. Moreover, separated methods are generally used while analyzing external (macroscopic) and internal (microscopic) spray shapes for complete spray characterization. This strategy requires complex and time-consuming calculations and may introduce a bias into the final results. The main objective of this work is to implement a single complete tool with the capability of processing spray images, leading to realistic external and internal properties. The chosen method is a hybrid method with an initial major step of image enhancing and filtering with a Bayesian multi-scale approach. This step is important regarding the poor quality of the input images. The second major step is the application of an alternative active contours detection method followed by a region segmentation to transform the contours into regions. The proposed approach is compared to well-known tools and it shows a better performance. The comparisons are mostly performed for synthetic images and natural images using visual inspection, metrics, and spectral analysis. Spray characterization is then possible by obtaining external properties (penetration length, cone angle, area and volume) and internal properties (droplet diameter distributions, mean diameters and representative diameters). The global performance of the processing algorithm allows studying the droplet morphology with a high level of confidence, owing to the preservation of contour features during the calculations (no contour extrapolation or suppression). Application on low pressure injections of isooctane shows the efficiency of the proposed solution.

Date	6 Oct 2017
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Patrice Seers (Supervisor) & Julien Weiss (Co-supervisor)