Evaluation of the Effect of Image Processing Methodology on Vapor Length and Spray Angle in a Constant Volume Combustion Chamber

<div class="section abstract"><div class="htmlview paragraph">In direct injected engines the spray formation is important for both combustion performance and emission formation. Thus, being able to compare how the spray formation is affected by changes in nozzle design, injection pressure or fuel formulation is an important area of research for all engine sizes. This becomes especially important for the introduction of new sustainable fuels, or for fuel injection optimization to increase efficiencies and minimize the formation of emissions such as particles. High-speed imaging of the fuel spray using the schlieren technique is well established for this purpose, and the Engine Combustion Network (ECN) has developed multiple guidelines to ensure that a similar experimental approach is used in different laboratories around the world. For the initial image processing, the ECN provides a procedure based on an image-temporal-derivative approach. Many researchers however rely on intensity-based thresholding, preceded by contrast adjustment, background removal and filtering, to detect spray boundaries. In setups representing smaller engines (bore diameter of approx. 85 mm), the impact of image processing might give less of an impact due to each pixel representing a small distance. However, for larger combustion chambers, such as the one used in this study with a window size of 150 mm, a difference of one or a few pixels may cause a larger difference in the obtained spray penetration length and spray angle. Therefore, this paper studies the sensitivity of different image processing choices on the final images and obtained spray parameters. The investigation shows that the resulting spray parameters are relatively robust to changes in the image processing parameters. The main difference is seen when adjustments are made to the image binarization threshold. The image contrast can be increased to ease visual interpretation, without compromising measurement accuracy. Both the Gaussian and median filter proved useful to improve the robustness of edge detection. A filter size of 10-15 pixels, representing 2.2-3.4% of the maximum penetration depth, appears optimal.</div></div>