Begell House Inc.
Journal of Flow Visualization and Image Processing
JFV
1065-3090
28
1
2021
MEASURING FLUID INTERFACES, CORNERS, AND ANGLES FROM HIGH-SPEED DIGITAL IMAGES OF IMPACTING DROPS
1-19
10.1615/JFlowVisImageProc.2020032697
D.
Biolè
Laboratory of Technical Physics, University of Liverpool, Liverpool, L69 3GH,
UK
Volfango
Bertola
Laboratory of Technical Physics, University of Liverpool, Liverpool, United
Kingdom, L69 3GH
drop impact
contact line
contact angle
image processing
Modern high-speed digital cameras enable the investigation of fluid flows with unprecedented resolution in both space and time. A remarkable example is given by the study of drop impact and dynamic wetting phenomena, which occur on time scales of few milliseconds and exhibit features, such as the contact angle, which are often treated as geometric singularities. Whilst a good resolution of the camera sensor and a high-quality optics are obvious prerequisites to obtain accurate measurements, the extraction of quantitative information from digital images always requires some kind of processing algorithm. In particular, the accurate measurement of the position and velocity of fluid interfaces, and that of contact angles, require the identification of image features such as contours, edges, and corners, which represents a nontrivial problem of digital image processing with fundamental applications in machine vision systems. This work illustrates systematic analytical procedures to identify fluid interfaces, corners, and angles in high-speed digital images of fluid flows, with focus on their application to the study of drop impact and dynamic wetting phenomena.
LIQUID AND VAPOR PHASE BEHAVIOR OF HIGH PRESSURE n-HEXANE SPRAY
21-44
10.1615/JFlowVisImageProc.2020034163
Manas Kumar
Pal
School of Mechanical Engineering, VIT-AP University, Amaravati, Andhra Pradesh, 522237, India; Internal Combustion Engines Laboratory, Department of Mechanical Engineering, IIT Madras, Tamilnadu, 600036, India
n-hexane spray
liquid penetration
vapor penetration
maximum liquid phase temperature
Injection strategy determines the air-fuel mixture inside the combustion chamber and hence controls the combustion and emission in direct injection (DI) engines. Higher injection pressure compared to the present-day scenario can be used to improve the in-cylinder air-fuel mixing for gasoline direct injection (GDI) engines. This study presents previously unavailable high-pressure spray characteristics data of n-hexane at nonevaporating and evaporating conditions. Various spray parameters like spray penetration, cone angle, and maximum liquid phase penetration are studied with high injection pressures. Existing correlations of cone angle and liquid length are tuned with the experimental results. The maximum liquid phase temperatures are calculated for evaporating conditions. It is observed that, for an evaporating spray, the maximum liquid phase temperature of the fuel is increased with increase in surrounding gas density for a constant surrounding gas temperature. Air-fuel mixing ratios are also calculated, and the results show that for a constant surrounding gas density, higher injection pressures increase the air-fuel mixing quality at a certain time from the start of injection.
FLOW IMAGE GENERATION ALGORITHMS FOR IMPROVING GAN
45-59
10.1615/JFlowVisImageProc.2020034486
Xianhong
Zhang
Heilongjiang Institute of Technology
Shusen
Li
College of Mechanical and Electrical Engineering, Northeast Forestry University,
China
image generation
GAN
activation function
optimizer
Generative adversarial network (GAN) is widely utilized for image synthesis, but the original GAN has many drawbacks in image generation, such as unstable training, poor image quality, and insufficient diversity. This paper attempts to solve these problems by establishing a simple and effective GAN to enhance the quality of output images. This GAN is optimized using a scaled exponential linear unit (SELU) activation function in the generator, which prevents the gradient from disappearing. In addition, the loss function is adjusted to further increase the diversity of images. Furthermore, mini-batch gradient descent (MBGD) is incorporated into Adam's algorithm to enhance the robustness of the network training process. The experimental results from the MNIST data set and the flow images show that the improved GAN is effective.
EDGE DETECTION ALGORITHM FOR COLOR IMAGES BASED ON THE REACTION-DIFFUSION EQUATION AND THE CELLULAR NEURAL NETWORK MODEL
61-80
10.1615/JFlowVisImageProc.2020033379
Xianhong
Zhang
Heilongjiang Institute of Technology
Shusen
Li
College of Mechanical and Electrical Engineering, Northeast Forestry University,
China
color image edge detection
cellular neural network
reaction-diffusion equation
adaptive threshold
This paper presents a cellular neural network (CNN) based on the reaction-diffusion equation to directly increase the accuracy of color image edge detection and solve the problem of color image edge adaptive threshold detection. It analyzes the dynamic properties of the FitzHugh-Nagumo (FHN) reaction-diffusion model and optimizes the diffusion coefficient of the reaction-diffusion equation to make the threshold more adaptive. Finally, the cell neural network improved by the FHN model is applied to the edge detection of color images in the HSV space. This paper's numerical experiments show the effectiveness of its approach, which quantitatively and qualitatively outperforms the commonly used methods.
EXTRACTING, PROCESSING, REPRESENTING, AND VISUALIZING DIFFERENCE FIELDS FOR EFFECTIVE ANALYSIS OF CFD DATA UNCERTAINTY
81-103
10.1615/JFlowVisImageProc.2020035313
Zhanping
Liu
Department of Computational Modeling and Simulation Engineering, Old Dominion University, ODU-CMSE 1300 ECSB, 4700 Elkhorn Avenue, Norfolk, VA 23529, USA
David
Kao
data visualization
uncertainty visualization
CFD data
color map
data processing
feature extraction
data analysis
line integral convolution
This paper presents a suite of techniques for extraction, processing, representation, and visualization of the uncertainty embedded in computational fluid dynamics (CFD) simulation data by addressing a field of difference (FoD) that is derived from any two cases of an ensemble of 2D scalar data, without assuming any statistical or simulation model, without any input from domain experts, and without any user interaction. To cope with a sheer range, the data values of an FoD are exponentially divided into several bands, for which 10 or an adaptively identified real number is utilized as the base. Exponential data banding (EDB) discretizes the continuous signal to bring quantization cues in color mapping. By integrating multiple color maps that are visually segmented and significantly contrasted in hue, hybrid color mapping (HyMap) offers both qualitative visualization and high-level quantitative visualization to attain a focus + context representation. Incorporated with HyMap are Relief Image Synthesis (RIS) and Differencing by line integral convolution (Diff-LIC). RIS takes an FoD as height data and local gradients as normal vectors, employing an illumination model to produce an image of "terrain". Diff-LIC extends LIC, a texture-based approach intended for flow data, to uncertainty visualization by creating an artificial special 2D vector field. Furthermore, velocity vector data associated with the source scalar data of the FoD, if available, can be visualized simultaneously to provide a contextual layer over which an image of FoD is superimposed to achieve fluid data fusion (FDF). These proposed techniques serve as primitive elements on which composite algorithms may be developed to visualize the uncertainty exhibited by an ensemble of results that a numerical simulation runs for a large number of times to yield. EDB, HyMap, RIS, Diff-LIC, and FDF are among the components of our framework for CFD data uncertainty visualization. Preliminary tests demonstrate that they are well suited for visual analysis of difference fields.