Use of machine learning (ML) is in its early stages within the domain of phase-change heat and mass transfers, yet ML holds transformative potential for both research and practical applications. Droplet evaporation, a fundamental phenomenon in phase-change processes, is pivotal in diverse industries such as inkjet printing, microelectronics cooling, and chemical processing. Conventional experimental and numerical methods, while informative, often fall short in fully capturing the complex multiscale, multivariate dependencies, and nonlinear interactions that govern evaporation dynamics, including vapor shielding and selective evaporation. ML offers a robust set of techniques for advanced high-dimensional data analysis, predictive modeling, and real-time monitoring, effectively addressing these limitations. This review explores the integration of ML into droplet evaporation studies, highlighting key challenges and opportunities in optimizing experimental setups, enhancing spatiotemporal analysis, improving model accuracy, and leveraging image processing to interpret complex, nonlinear features embedded in visual data. By concentrating on droplet evaporation, this work aims to establish a foundation for broader advancements in phase-change heat transfer. Furthermore, the methodologies and insights presented herein are expected to extend naturally to related areas such as condensation and other evaporation-driven processes, fostering new directions in thermal management and fluid dynamics research.
This work presents a comprehensive approach to modeling, recognition, and analysis of solid particles of various sizes moving in a gaseous medium using computer vision and machine learning techniques. The study is based on laboratory experiments conducted in a shock tube facility with flow velocities up to 900 m/s. Visualization of particle motion was carried out using high-speed shadowgraph imaging. Custom image processing algorithms were developed to detect and track particles with sizes ranging from 10 μm to 30 mm. The processing pipeline includes grayscale conversion, adaptive thresholding, morphological operations, contour detection, and centroid tracking. In addition, deep learning models based on the you only look once (YOLO) architecture were trained on a labeled dataset of particle images to enable real-time detection and localization of individual particles within complex flow fields. The analysis also included automatic estimation of particle brightness and background intensity, which provide an additional metric for particle identification and classification. Quantitative results were obtained through tracking-based reconstruction of particle trajectories, x-t diagrams, and velocity-time profiles. These tracking procedures are conceptually related to particle tracking velocimetry (PTV) methods, as they rely on the precise localization and temporal correlation of individual particle positions for velocity and trajectory analysis. The proposed methods demonstrate high efficiency in the detection and analysis of fast-moving solid particles in gas flows and are applicable to a range of problems, including laboratory modeling of multiphase flows.
The article explores the possibility of using structural image features, generated by the machine vision system of mobile robots, to identify an object of interest. Two approaches to solving the problem of object identification using structural image features are considered. The first approach is based on the direct construction of a segmented image using fractal dimension as a characteristic of the structural properties of the original image. The determination of fractal dimension values is proposed to be carried out based on the presence of "tails" in the fractal dimension histogram. The conditions and the range of fractal dimension values at which the desired object is identified in images with high object density have been determined. It has been established that the range of fractal dimension values, at which signs of object identification appear in the brightness distribution histogram, falls within a certain interval 2.945 ≤ D ≤ 2.951. At the same time, the appearance of object identification signs occurs when the original image is subjected to Gaussian noise with a certain variance, σN = 44. The second approach is based on the preliminary segmentation of images by contrast, followed by the construction of a segmented image using a structural feature. It is shown that this approach allows for an unambiguous determination of the fractal dimension value and does not lead to an increase in the number of computational operations.
During a volcanic eruption, fine ash particles collide with each other and form larger aggregates. Ash aggregation plays an important role in volcanic plume modeling, as larger aggregates fall closer to the vent exit, while smaller particles can travel far before falling out. Preferential concentration can promote ash aggregation by increasing the likelihood of collisions among the particles. This process is challenging to model due to the complexities associated with turbulence and particle tracking. To improve these models, we investigate the effects of several properties on the clustering of particles in a jet flow, including the Reynolds number, Stokes number, mass loading, and relative humidity. We vertically eject compressed air embedded with particles into a quiescent chamber at Reynolds numbers between 5 x 10(3) and 10 x 10(3). We study hollow glass and solid nickel particles, each similar to 13 mu m in diameter, resulting in Stokes numbers between 1.0 and 9.4. Particle mass loadings range from one to three percent, while relative humidity varies from 40 to 75%. Particles are illuminated using a laser sheet, and images are captured with a high-resolution camera. Clusters are identified by examining the densities of particles in the series of instantaneous images. At Stokes numbers near one, the number of clusters is more than an order of magnitude higher, with nearly twice as many clusters downstream. In addition, the average area and perimeter are larger than at high Stokes numbers. However, when normalized, the distribution of clusters is largely independent of Reynolds number, Stokes number, mass loading, and relative humidity.
Experimental work was performed by means of particle image velocimetry (PIV) to investigate axisymmetric turbulent jets impinging on a semi-cylindrical convex surface. Measurements were performed for one surface curvature ( D/ d approximate to 8 . 2 ) and four L/d (pipe jet exit to surface) ratios at a Reynolds number R e approximate to 30000 . The study focused on the free jet region before impingement and the wall jet formed after impingement, as well as the developing boundary layer over the convex surface, with the aim of specifying the vortex structures and high turbulent intensity regions. The jet impingement behavior was characterized by large radial and tangential velocities as well as increasing rms velocities along different radial lines (crossing the convex surface). The linear stochastic estimation (LSE) method was employed to identify the dominant vortical structures in the flow field and also to correlate them to other relevant parameters, such as rms velocities of the wall jet. The results show an entrainment of the flow carried about by the curved surface geometry, especially for large L/d ratios. They also show high turbulence fluctuations in the tangential velocity when the distance from the stagnation point was reduced and even distribution of dominant vortical structures for L/d = 2. Both the peak rms tangential velocity and half velocity width decreased for all L/d ratios when increasing the distance from the stagnation point over the convex surface. These results can be used for the interpretation of heat and mass transfer aspects with convex surfaces and could serve as a benchmark for future numerical studies.
The field of skin cancer detection has experienced significant advancements, largely due to the incorporation of artificial intelligence (AI) techniques. This review traces the evolution of skin cancer detection methods, beginning with visual inspections and progressing through classical machine learning, deep learning, and collaborative initiatives. The essential contributions of the International Skin Imaging Collaboration (ISIC) in promoting data sharing and establishing benchmarks are emphasized. Insights gained from comparing the diagnostic abilities of human experts and AI algorithms underscore the unique strengths and challenges of each, fostering the development of synergistic human-machine collaborations. The review also delves into the enhanced accuracy and robustness provided by deep learning-based ensemble models. The rising significance of explainable AI is addressed, highlighting its role in enhancing the transparency and trustworthiness of AI systems. Additionally, the recent advent of vision transformers, which utilize attention mechanisms for superior image analysis and detection, is explored. This comprehensive overview offers a detailed understanding of the advancements in skin cancer detection, emphasizing its potential for early diagnosis and improved patient care, and outlines future research directions.
Cavitation occurs in fluid flows in areas of the low pressure. It negatively affects the operation of hydraulic engineering units because it causes corrosion, which leads to the destruction of machine parts. Cavitation can occur not only in the main stream but also in the slits available in any mechanism. In this paper, computer simulation of cavitation flow in a slit channel arising behind a cylindrical body of flow is considered. The length of the slit channel is 145 mm, width is 120 mm, and height is 1.2 mm. Cavitation occurs behind a cylindrical body with a smooth surface with the forming circle center 50 mm away from the beginning of the channel and 60 mm from the side wall. The dynamics of the cavity development behind the cylinder is described. The frequency of occurrence of cavities and their relative time of their existence are calculated.
This work presents high-speed shadowgraph imaging of a high-speed flow visualized through windows of a rectangular shock tube test section. Images of striped structures in the supersonic gas boundary layer on the side glass walls were first obtained, and their spatial-temporal characteristics were studied using image processing. The laminar-to-turbulent transition in the boundary layer behind the shock wave was visualized. Computer vision methods, including deep learning algorithms, were applied to analyze the visual data from the experiments. Techniques such as contour detection, Fourier and wavelet analysis, and neural network-based classification (supervised learning) were employed to automatically recognize and categorize the flow stages, enhancing the understanding and analysis of flow structures.
Since 1960, the Siberian Thermophysical Seminar has traditionally been held in the academic town of Novosibirsk in Russia by the Kutateladze Institute of Thermophysics SB RAS. In 2024, the seminar was dedicated to the 110th anniversary of the birth of the academician Samson Semyonovich Kutateladze, who the institute is named after, and the 300th anniversary of the Russian Academy of Sciences.
The design of a corrugated eight-lobed convergent-divergent (CD) nozzle with three different corrugation area ratios (CAR = 0.08, 0.15, and 0.022) was performed, and the numerical simulations were carried out with a nozzle pressure ratio (NPR) of 5. Further, the predictions were compared with the experimental data of a circular CD nozzle with the same geometric parameters. A comparative analysis of velocity contour, Mach number, pressure, velocity, and Reynolds stresses was conducted with all four nozzles. It was evident that a moderately underexpanded jet with 'X'-shaped shock cell structures was formed in all the cases. Moreover, the strength, position, and length of the first shock cell structure were similar for all the nozzles. However, as the flow proceeds downstream, the nozzle with the higher CAR exhibits a larger deviation from the circular CD nozzle. It has characteristics with the early appearance of shock cells, early dissipation of turbulent kinetic energy, and a high rate of turbulent momentum transfer. These findings suggest that corrugated CD nozzles with optimized CAR values can effectively promote fluid mixing and noise reduction, offering potential applications in aerospace and industrial processes.
