Application of machine learning is currently one of the hottest topics in the fluid mechanics field. While machine learning seems to have a great possibility, its limitations should also be clarified. In our research group, we have started a research project to construct a nonlinear feature extraction method by applying machine learning techniques to big data of fluid flow, i.e., extracting the low-dimensional nonlinear modes essential to the unsteady flow phenomena and deriving the governing equations for such low-dimensionalized dynamics. This self-review article is a focused but extended version of the keynote lecture given by the author at the 7th International Conference on Jets, Wakes and Separated Flows (ICJWSF2022). We will first introduce the use of a convolutional neural network (CNN) to learn the temporal evolution of cross-sectional velocity field in a turbulent channel flow. Subsequently, we also consider an application of CNN for extraction of low-dimensional dynamics for flow around a bluff body accompanying vortex shedding and our preliminary attempt to use the extracted low-dimensional dynamics for an advanced design of flow control.

It is now more than 100 years since the work of von Kármán (1921) on the boundary-layer flow over a rotating disk was published in the first volume of Zeitschrift für Angewandte Mathematik und Mechanik (ZAMM, Vol. 1(4), pp 233-252). Recently, there has been a large amount of work undertaken addressing the instability and transition of the boundary-layer flows over rotating disks and cones using theoretical, numerical and experimental techniques. Here we will discuss some different methods to analyze experimental data that can give insight into the instability and transition to turbulence of boundary-layer flows over rotating slender and broad cones (including the disk). At first, we discuss the pdf-method (probability density function) that allows a simple way to determine regions of instability growth, transition and fully developed turbulence. Secondly, we look at various ways to use spectral information to investigate the boundary layers giving a deeper understanding of the transition process. Finally, a method to determine the most probable flow structure leading up to fully developed turbulence is discussed. We envisage that some of these methods can be useful in analyzing instability and transition also in other flow cases.

This study identifies the coherent structure and determines its energy contribution in the upstream near-wake behind a long circular cylinder using proper orthogonal decomposition of data that is measured using particle image velocimetry. Members of the harmonic-frequency family (if they exist), which are large-scale, quasiperiodic motions, are firstly identified in the Fourier power spectrum of each mode coefficient. The other large-scaled, organized frequencies of the coherent structure in each mode coefficient are next identified by means of a condition of their peak magnitudes greater than the smallest magnitude of the identified harmonic-frequency family. The kinetic energy that is contributed by the coherent structure is then calculated. Calculation of the kinetic energy of the coherent structure in the first 39 modes, which contribute 80.04% of the total kinetic energy for all eigenmodes, is performed. An estimated 29% of the total kinetic energy is contributed by the coherent structure.

Previous studies have revealed that the change in aerodynamic drag on a test automobile under a parallel running condition is mainly because of the pressure effect, which is caused by the pressure field of the parallel running vehicle, and the effect of the change in wind direction caused by the parallel running vehicle. Apart from the drag change, which can be estimated by assuming a uniform crosswind, the change in wind direction has other effects. In this study, to clarify the other effects, the differences in the wake vortices associated with the change in drag between parallel running and uniform crosswind conditions were identified from the numerical simulation results. The surface pressure on the back excluding the pressure effect under the parallel running condition was lower on the opening side than that under the uniform crosswind condition. In addition, the wake vortices near the region of low surface pressure and their related vortices were focused. Visualizing the flows related to these vortices indicates that the non-uniformity of the flow along the upstream side and the strengthening of the entrainment at the rear end owing to the interference affect the change in the drag in the parallel running condition compared with the uniform crosswind condition. These results are useful for developing control methods to suppress the increase in drag in the parallel running condition.

In this study, a double-tube burner nozzle with a DBD-PA (dielectric barrier discharge plasma actuator) and flame holder was employed to regulate the lifted flame. The flame holder held the lifted flame and the DBD-PA regulated the lifted flame to suppress the lifted flame’s vertical oscillation. To form a lifted flame, pseudo-biogas, a mixture of CH₄ and CO₂ in the ratio of 6:4, was jetted out through an inner nozzle and air through an annular nozzle. To control the air jet and prevent the lifted flame’s vertical oscillation, the DBD-PA was installed inside an annular nozzle. The DBD-PA was driven intermittently at regular intervals to produce a tangential flow in the same direction as the jetting gas, and the flame holder was placed above the nozzle outlet on the nozzle center axis. The finding indicated that by driving the DBD-PA at the slightly lower frequency than the frequency of the naturally shedding vortices in the free jet, the lifted flame’s vertical oscillation can be reduced by collapsing the jet’s vortex ring at a steady position and with a steady cycle. By impeding the lifted flame’s vertical oscillation, it is forecasted to reduce the blowout of the lifted flame and combustion noise.

Robots that operate at a disaster site have been proposed and the dispatch of rescue robots to such site is required. However, discussion on the transport of these robots is insufficient. One method of transporting such robots is to use an airplane and the parachute. In order to safely drop these robots by parachute, it is important to investigate the aerodynamics characteristics of the robots in free fall before opening of the parachute. In the present study, the fluidic forces that acted on the robot and the wake velocities behind the robot were measured using a load cell and particle image velocimetry (PIV) in uniform flow ranges of 4 to 22 m/s. In addition, the bending angles of 30°, 45° and 60° were evaluated in terms of aerodynamic coefficients, streamlines and vorticity in the wake. As a result, changes in the fluidic forces were obtained due to the number of bending limbs. A vortex pair appeared in the wake, and the recirculation region became smaller as the number of limbs increases. In addition, the aerodynamics coefficient for the fluidic force in the z direction was a maximum at bending angle of 45° for the type-D of robot model with two limbs bent due to the vortex size and position in the bending-limb side.

Reduction of aerodynamic drag is an important topic in the development of a bicycle racing, making it extremely vital to study the flow field phenomenon on the surface of a cyclist. As the wind flows over the human body, the flow separation phenomenon will appear on the surface. The purpose of this research is to study the aerodynamic characteristics of cyclists by utilizing flow visualization method in both water channel and wind tunnel. Cyclist models used in water channel and wind tunnel testing were based on a real male cyclist and printed using a 3D printer. Flow visualization technique was employed to observe the flow phenomenon around the arms, thighs and calves of the cyclist model. It is found that there are vortices meandering along the axial direction at the arms and thighs, double separation lines appear at the lower legs. Then, the surface pressure coefficient of the model was measured on a full-scale model in wind tunnel experiment to verify the flow condition in the vicinity of the model surface. These research methods help in getting a better understanding of the structure of the flow field around the cyclist, and further achieve the purpose of drag reduction.

In order to improve the submerged cutting performance of water abrasive suspension jet an experimental investigation on the two-phase flow structure of submerged water jet with semi-coaxial air ventilation is conducted by the method of high-speed camera observation under the condition of flow dynamic similarity, and its unsteady behavior under different conditions is evaluated by analyzing of sequence images. Experiment results show that the flow pattern of air-ventilated jet depends upon the ventilation flow rate. It may be classified into bubbly flow surrounded jet, continuous air-layer covered jet and cavity-bubble flow surrounded jet. In the case of non-ventilation, high-speed water jet into water is accompanied by intensive cavitation and the flow field appears to be a narrow jet surrounded by cavitation bubble clouds. By ventilating to the cavitating jet, air-bubbles become bigger gradually and a cavity area is formed at the exit of nozzle head. When the ventilation flow ratio reaches to the level of 5 a large cavity covering the jet is formed and a continuous air-layer covered jet is then generated. Image analysis demonstrates that the continuous air-layer covering the liquid jet pulsates and sheds downstream periodically. The Strouhal number defined by the dominant frequency of jet pulsation decreases quickly with the increase of ventilation flow rate in the range of C_q ≤ 5. At a suitable ventilation flow ratio (C_q ≈ 6 ~ 8) a relative long stable continuous air-layer covering the jet is generated at high time ratio, which will dramatically enhance the processing ability of submerged water jets. Cutting tests demonstrates that the cutting ability of air ventilated jet depends on the ventilation flow ratio. The relation of processing ability with the flow structure of air ventilated jet is clarified.

90-degree elbow pipe/duct is one of the piping element which is used frequently in industrial apparatus. The separated flow which occurs at inner corner of elbow part and the secondary flow which occurs at the elbow and its downstream cause significant pressure loss especially in the case of reducing elbow. Generally the inner corner is rounded or guide vane is mounted near the corner in order to reduce the losses by controlling this separation and secondary flow, however, the authors newly propose the mounting weir-shaped obstacles on the inner wall upstream of the corner. This method is easier to introduce to the duct/pipe which has a sharp edged inner corner, because machining the corner is not needed and the shape of the object mounted is simpler than guide vane. In this study, in order to reduce the pressure loss of reducing elbow ducts by installing weir-shaped obstacle and to investigate the effect of the cross-sectional reduction ratio on the loss reduction by the obstacle, pressure measurement and flow visualization using water were carried out. Experiments by varying the position (distance from the corner) and height of the obstacle made clear that the obstacle conditions which makes the flow resistance of reducing elbow duct smaller and the loss is reduced by about 30 – 60%.

We conducted measurements of a butterfly’s motion in forward flight and numerical simulations using a computational model reflecting its motion. We measured the motion of a cabbage white butterfly (Pieris rapae), and then we constructed a computational model composed of a thorax, an abdomen, and four wings (i.e., left and right wings with fore and hind parts). Furthermore, we calculated the flow field and aerodynamic force and torque generated by the butterfly model using the immersed boundary–lattice Boltzmann method. In this simulation, we considered two types of periodic motions corresponding to slightly-descending and ascending forward flights. As a result, we found that the wing-tip and leading-edge vortices are formed on the wings and then released backward and downward in both flights. The major difference between the two flights is the flapping amplitude, indicating that the butterfly changes the flapping amplitude for each period and increases it to ascend. In addition, we considered a chimera model whose motion is based on the slightly-descending flight but partly given by the ascending flight. As a result, we found that the pitching angle and the angle of attack determine the traveling direction, but simply changing these angles does not achieve the ascending flight due to insufficient lift force. Thus, the butterfly should adjust the flapping and lead–lag angles in response to the pitching angle and the angle of attack to change the flight mode.

The vortex system on the SACCON model is studied under low Reynolds number by using the flow visualization technique and the surface pressure testing measurement in a low-speed water channel and a wind tunnel, respectively. A main topic is discussed below. The blunt leading-edge section on this SACCON model induces various flow phenomena under different angles of attack. The attached flow appearing around the upstream leading-edge region at a low angle of attack induces different flow phenomena downstream, so does the branched vortices appearing near the trailing-edge. Although this attached flow is vanished at higher angle of attack, the stability of forming an outboard vortex can be shown in the results of pressure coefficients as a function of an angle of attack. Moreover, how the blunt leading-edge contour on SACCON model and other models affecting flow field can be compared with the findings given in the previous studies. The models with only blunt leading-edge contour can prompt an outboard vortex and an inner vortex, compared to the flow fields on the models with spanwise-varying leading-edge contours. This inner vortex is formed due to the attached flow inboard passing downstream and being affected by the outboard vortex. However, this attached flow is different from the attached flow observed on SACCON model. Therefore, flow phenomena occurring on different blunt leading-edge models can be differentiated.

The cross-section of dragonfly wings has corrugated structures. In particular, the leading-edge side of the wing consists of V-shaped structures. According to previous studies, a corrugated wing may exhibit high aerodynamic performance at low Reynolds numbers (Re = O(10³)), and vortex dynamics associated with wing structure are expected to play an important role. We study the relationship between the wing structure and vortex dynamics by direct numerical simulations. It is known that, when a two-dimensional flat wing impulsively starts from a rest state, a coherent vortex called lambda vortex, which has the opposite sign to a leading-edge vortex (LEV), is generated and remains for a time interval. Our previous study suggests that, in the case of the corrugated wing, the lambda vortex collapses and is stuck inside V-shaped structures near the leading edge. The collapse of the lambda vortex was proposed as a key factor of high performance for the particular shape of the corrugated wing. In this paper, we investigate the relationship between vortex dynamics and high performance in a wider parameter range than in our previous studies. We analyse two corrugated models with different Reynolds numbers. It is revealed that the collapse of the lambda vortex is also the key to high performance for these cases.

In order to investigate the aspect of the mutual interference flow from two square cylinders, the visual water-flow observation experiment was performed. In the experiment, the pitch ratio L/d and the arrangement angle α were made into the parameter, and investigation of vortex shedding frequency or the flow pattern was performed. The pitch ratio L/d is 3 kinds, 1.13, 1.89 and 4.15, and an arrangement angle α is 7 kinds per 15 degree from 0 degrees to 90 degrees. The water flow velocity U was 0.04 m/s which correspond to the Reynolds number about 850. As the result of experiment, the complicated flow becomes clear by classification by color of tracer ink, and the aspect of the mutual interference vortex flow can understand better. Even if the pitch ratio was the same, the vortex shedding characteristics changed with arrangement angles. If an arrangement angle is smaller than a geometric duplication angle, the vortex shedding frequency from each square cylinder is the same. On the other hand, when an arrangement angle is larger than the geometric projection angle, the vortex shedding frequency from each square cylinder differs. In the smallest pitch ratio as L/d= 1.13, it is not dependent on the arrangement angle, and produce mutual interference.

In this study, two dielectric barrier discharge plasma actuators (DBD-PAs) are driven at different times to control a circular jet flow. DBD-PA was installed inside a nozzle and on a disk-shaped bluff body above the nozzle. A vortex was generated in the jet using DBD-PA in the nozzle, and the development of the vortex was controlled using the DBD-PA installed in the disk-shaped bluff body above the nozzle. The two DBD-PAs were driven with a time lag based on the time the vortex reached the bluff body. The findings reveal that the vortices did not pair with each other when the DBD-PA of the bluff body was slightly delayed from the timing of the occurrence of the vortex on the circulating flow side. When the vortices did not pair, the flow was not entrained in the circulating flow, causing the jet width to widen. The vortices paired with each other when the DBD-PA of the bluff body was driven at the same time as or long after the occurrence of the vortex on the circulating flow side. When the vortex pair with each other, they collapse quickly after pairing, and the jet width narrows, as the flow is entrained in the circulating flow. The difference in jet width downstream of the bluff body was found by driving the two DBD-PAs installed inside the nozzle and on the bluff body at different times.

Multiphase flows arise in various fields that involve complicated phenomena. Studies have shown that COVID-19 can occur via air microdroplets, and breathing jets with microdroplets turn into turbulent cloud or puffs in cases of coughing and sneezing (Bourouiba et al., 2014). Microdroplets are upturned by buoyancy in the turbulent cloud and transported without falling. Furthermore, they float in air for hours and can be transported over long distances (Mittal et al., 2020). This scenario also involves a mixed phase flow of air and droplets. To simulate these phenomena, a numerical model assuming mechanical and thermal non-equilibrium multiphase flow is required to predict the range of turbulent cloud transport. In this study, to better simulate the turbulent cloud trajectories, a viscosity term is added to a two-phase flow six-equation model (two-fluid modeling or effective-fluid modeling, EFM) developed by Liou et al. (2008). It is a development of a parameter-free, viscous multiphase flow code, based on a single-phase compressible finite-volume solver (Kitamura et al., 2013). This solver is validated in the Poiseuille flow and laminar-flat-plate problem with an isothermal wall through a comparison with the analytical solutions. A detailed simulation of coughing is performed. The location of the turbulent cloud upturned by buoyancy is compared with the data of past studies.

In the present study, the influence of the direction of the backward plasma-induced flow generated by a plasma actuator (PA) on the jet flow was investigated. In the case of the backflow PA, the centerline velocity was increased by the contraction of the main flow near the nozzle exit due to the influence of the backflow. The fluctuation of the jet boundary layer became stronger as the duty ratio was increased. Based on these factors, it is considered that the backflow by the plasma-induced flow effectively works on the diffusion of the jet structure. Furthermore, experiments were conducted by varying the Reynolds number as Re = 1.0×10³ (the effect of the induced flow is large) and Re = 3.0×10³ (the effect of the induced flow is small), and the difference in jet diffusion due to the induced flow effect was confirmed.

In the present paper the attention is focused on effects of tube pitch ratios of in-line tube banks on acoustic resonance and vortex shedding. We examined the characteristics of the acoustic resonance in the tube banks with tube pitch ratios ranging from 4.0 to 1.4 in the flow direction. Acoustic pressures were measured on the two sidewalls of the experimental apparatus. The tube surface pressure fluctuations due to the vortex shedding without the acoustic pressure were measured at the acoustic pressure nodes at which the acoustic resonance occurred. Acoustic resonance of low-order modes in the transverse direction occurred at tube pitch ratios of 4.0–2.8 in the flow direction. Alternative vortices were noticeably formed in the tube banks. As the tube pitch ratio in the flow direction of the in-line tube banks increased, the Strouhal number decreased, and as the former decreased, acoustic resonance of higher-order longitudinal modes occurred. The alternative vortices were formed in the tube banks at low gap velocities with no resonance. However, if acoustic resonance of a longitudinal mode occurred, symmetric vortices were formed in the tube banks. The onset velocity of a longitudinal mode was lower than that of a transverse mode, at small tube pitch ratios. As the number of rows of tubes increased, the acoustic resonance of transverse mode occurred at small tube pitch ratios. We discussed the prediction formula of Strouhal number at the tube pitch ratio in the flow direction of 2.0-1.4.

A typical plant that is distributed by wind is the dandelion, which possesses a complex hairy pappus that aids in flying. A pappus or group of bristly filaments found in plumed seeds is believed to have a role in drag enhancement. The pappus prolongs the descent of the seed. For wind-dispersed seeds, maintaining stability while reducing falling speed in turbulent winds may be useful for long-distance dispersal. However, it is unclear why plumed seeds chose a bristly pappus over a membrane that resembles wings and is known to increase lift during passive flight. In the current study, we analyze the dandelion pappus flight mechanism by identifying the crucial structural elements that enable its stable flight. In addition, we investigate the flow behavior around the pappus by performing float testing on several dandelion pappi in a vertical airflow pipe. The flow fields in the wake of the pappus were also measured by 2-component particle image velocimetry. The falling speed of the pappus is reliant on the posture during free-fall motion, and the aerodynamic features of the pappus are substantially affected by the posture in flight. The pappus’s posture was changed to promote vortex growth in the wake, and this vortex behavior is related to the fluid forces acting on the pappus.

Aerodynamic noise in high frequency band related in the flows around A-pillar is easily propagated to the cabin of the automobile. Understanding of the characteristics of the aerodynamic noise is needed to reduce the cabin noise. Moreover, understanding of the separated shear layer which is strongly affected by the condition of the step hight H to the boundary layer thickness δ is needed. In this study, far field sound and correlation measurement of flow velocity are measured to investigate the characteristics of aerodynamic noise and the relationship between H/δ and turbulence length scale L as eddy scale for the forward facing step. As a result, multiple gradual peaks in more than 1 kHz are observed in the aerodynamic sound spectra. Furthermore, it is found that turbulence length scale L increases for increasing of H/δ and turbulence length scale to step height for H/δ = 2.0 is almost equal to that for H/δ = 3.0. Then, it is clear that spectral peaks are strongly affected by step height. And it is confirmed that H/δ = 2.0 and 3.0 are equal in turbulence length scale to the step height.

The effect of polymer addition on transition to turbulence in a two-dimensional water-flow channel was experimentally investigated by flow visualization using reflective flakes. The flow entering the channel test section maintains a high disturbance level by expanding laterally after reaching a high Reynolds number upstream the test section. In order to obtain the intermittency factor (turbulence fraction), the visualized images were classified into non-turbulent and turbulent regions, and the streamwise scale of the streaks appearing in the non-turbulent region was estimated from the autocorrelation coefficient computed by shifting the images in the streamwise direction. The visualization results show that similar to the pure water case, intermittent flow with a patch-like distribution of turbulent and non-turbulent areas clustered by streamwise streaks is observed. The Reynolds number at which the intermittency increases shifts toward higher Reynolds numbers with increasing polymer concentration, indicating a delay of transition. The streaks appearing in the non-turbulent region elongate with increasing polymer concentration. At high concentrations, straight elongated streaks penetrate through the turbulent regions, suggesting that the polymer addition affects the stability of the streaks. These changes of the streak behavior indicate that the polymer affects not only the transition Reynolds number but also the flow structure during the transition process.

In this study, oil film interferometry (OFI) was applied to the flow around the suction surface of a two-dimensional airfoil under high lift conditions to measure the wall shear stress. However, existing OFI methods have difficulty measuring the wall shear stress in images affected by reverse and secondary flows. Therefore, the particle image velocimetry (PIV) method was applied to the Fizeau fringe images to determine the direction of progress from the calculated velocity vector. An airfoil with a wing section of NACA0012 was used, the Reynolds number was set to 8 × 10⁴, and angle of attack was set at 11°. A direct spatial domain correlation was used for the PIV analysis method. At x/c = 0.75 to 0.83, the measured local skin friction coefficient was good agreeing with the large eddy simulation calculated results reported by Miyazawa et al. (Transaction of the JSME, Series B, Vol.72, No.721 (2006)). However, a large difference between the measured and calculated local skin friction coefficients occurred at x/c = 0.17 near the re-attachment point. The wall shear stress is determined by calculating the dominant frequencies using fast Fourier transform (FFT) analysis from the matrix data obtained by mean of the pixel intensities in the analysis region in the span direction. If the Fizeau fringe image is uniform in the span direction, the program can calculate the periodic waves. When the Fizeau fringes are tilted to the analysis range, the FFT analysis of the obtained matrix data results in an error because the dominant frequency cannot be calculated. Therefore, velocity vectors were detected near the re-attachment point by adapting PIV to the Fizeau fringe images. The local skin friction coefficients were calculated by OFI measurements with the Fizeau fringe images that were rotated by the angle of the velocity vector determined by PIV.