High-speed imaging in fluids

Versluis, Michel

doi:10.1007/s00348-013-1458-x

High-speed imaging in fluids

Review Article
Published: 14 February 2013

Volume 54, article number 1458, (2013)
Cite this article

Download PDF

Access provided by Autonomous University of Puebla

Experiments in Fluids Aims and scope Submit manuscript

High-speed imaging in fluids

Download PDF

Michel Versluis¹

7084 Accesses
122 Citations
Explore all metrics

Abstract

High-speed imaging is in popular demand for a broad range of experiments in fluids. It allows for a detailed visualization of the event under study by acquiring a series of image frames captured at high temporal and spatial resolution. This review covers high-speed imaging basics, by defining criteria for high-speed imaging experiments in fluids and to give rule of thumbs for a series of cases. It also considers stroboscopic imaging, triggering and illumination, and scaling issues. It provides guidelines for testing and calibration. Ultra-high-speed imaging at frame rates exceeding 1 million frames per second is reviewed, and the combination of conventional experiments in fluid techniques with high-speed imaging techniques is discussed. The review is concluded with a high-speed imaging chart, which summarizes criteria for temporal scale and spatial scale and which facilitates the selection of a high-speed imaging system for the application.

Brandaris Ultra High-Speed Imaging Facility

High-Speed Imaging of Shock Waves and Their Flow Fields

Measurement in opaque flows: a review of measurement techniques for dispersed multiphase flows

Article Open access 13 May 2020

Use our pre-submission checklist

Avoid common mistakes on your manuscript.

1 Introduction

The beauty of slow-motion movies captured with high-speed imaging has traditionally been described along the words “making the invisible visible,” “seeing is believing,” “seeing the unseen,” “making flow motion into slow motion,” “science or art,” or “capturing the moment.” In recent years high-speed imaging has taken a groundbreaking step forward in our scientific world of flow research, and improved high-speed imaging technology has pushed new and exciting insights into the physical mechanisms underlying flow and flow-related phenomena. Typical applications of high-speed imaging include car crash testing, air bag deployment, machine vision technology for packing and sorting, high-speed impact and material testing, sport science, ballistics and (nuclear) detonation and explosions. In fluid dynamics, applications of high-speed imaging include propulsion and cavitation, combustion, turbines and supersonic flows, sprays and jets, and shock waves. Emerging applications in microfluidics, biomedicine, and biomechanics require top-of-the-line high-speed imaging systems, that is, high frame rates at high spatial resolution at a high number of frames. Recent examples of these applications include animal locomotion, such as that of the water striders (Hu et al. 2003), of insect flight (Fry et al. 2003), of the Common Basilisk (Basiliscus basiliscus) or Jezus Christ lizard that runs on the surface of water (Glasheen and McMahon 1996), and that of the snapping shrimp (Versluis et al. 2000; Lohse et al. 2001) and Mantis shrimp (Patek et al. 2004). In microfluidics we find applications in pinch-off phenomena in a dripping faucet (Shi et al. 1994; Cohen et al. 1999), splashing (Thoroddsen and Takehara 2000; Thoroddsen 2002; Xu et al. 2005), jet instabilities and jet breakup (Anna et al. 2003; van Hoeve et al. 2010), and ink jet printing (Chen and Basaran 2001; de Gans et al. 2004). In medicine, time-resolved dynamics of ultrasound contrast agent microbubbles insonified at MHz frequencies (de Jong et al. 2000; Chomas et al. 2000; Bouakaz et al. 2005; van der Meer et al. 2007), shock wave lithotripsy for controlled kidney stone fracturing (Pishchalnikov et al. 2003), and cell membrane permeabilization through acoustic streaming and sonoporation (Marmottant and Hilgenfeldt 2003; Ohl et al. 2006; Wamel et al. 2006; Kudo et al. 2009) are all to be visualized at nanoseconds timescale.

Probably, the most striking icon in high-speed flow visualization is the crown-shaped splash made by a milk droplet, Fig. 1, as it hits the surface of a liquid captured by Harold Edgerton in 1936 using flash photography, Fig. 1b. Edgerton, who was the inventor of the high-speed strobe flash light, was and is a master in high-speed still photography. He produced pictures of hummingbirds in flight (Edgerton and Killian 1954), a water jet from a faucet (Edgerton et al. 1987), and the vortical air movement near spinning fan blades (Edgerton and Killian 1954) by freezing the motion of the object using his microseconds flash technique. Edgerton was not the first to discover the crown-shaped splash, as it was Worthington (Worthington and Cole 1897, 1900), who worked on splashes for all of his career in studies of impacting solids and fluids, and who made the first studies on the crown-shaped splash by carefully drawing out the contours of the impact phenomenon after a 3-μs spark plug illumination in 1877, see Fig. 1a, after earlier work in high-speed photography using spark discharges by Fox Talbot (1852), Cranz and Glatzel (1912), and Mach and Mach (1889). The Worthington jet, which arises when a sphere or a droplet plunges into a pool of water, creating a cavity which collapses under the water pressure, forming an up-shooting jet, is named after him. His remarkably detailed piece of work [A Study of Splashes (Worthington 1908)] is an inspiring masterpiece which is still very attractive and actual as we speak. And indeed, the use of high-speed flash photography is still highly appropriate today. In fact, still photography is the preferred option for high-speed flow visualization. It is inexpensive, is flexible, has superior resolution and dynamic range and provided that the illumination pulse is short enough, and has excellent motion arrest. Flash photography should in fact come first to mind when studying high-speed flow phenomena (Thoroddsen et al. 2006). We have investigated the milk coronet formed after the impact of a red-colored milk droplet as it hits the surface, see Fig. 1c. The image was taken with a 12.3 Mpixel Nikon D300 SLR digital photo camera connected to a programmable trigger/delay unit and a high-speed flash source. The photograph reveals that only the inside of the crown is coated with the fluid of the impacting droplet, which is most prominently visible at the level of the satellite droplets, formed after pinch-off of the crown jets, which are half-covered with a splash of red dye. Experimentally, the situation becomes much more complex when the flow phenomenon under study is dynamic. We can have high-speed events and transients, and when the phenomenon is non-reproducible, non-repetitive, or non-localized in time, high-speed flash photography will be of little scientific value. And as we need now to capture the series of flow phenomena at the timescale of the process, here is where high-speed imaging comes into play.

High-speed imaging in flow visualization is primarily aimed at obtaining precise information about the position and dimensions of the fluid flow at a series of instants in time, that is, to resolve to the best possible extent the spatial and temporal scales. High-speed imaging has been key to a series of discoveries, and to put the recent high-speed imaging technologies in historic perspective, we give a brief overview of the history of high-speed imaging. It all started with the pioneering work of Muybridge and Mozley (1887) and Etienne-Jules Marey (1868, 1894) who built the first framing cameras in the late nineteenth century (1860–1880). This was long before motion picture film was patented in 1894. Muybridge’s challenge was to settle a long-standing debate: “Is there a moment in a horse’s gait when all four hooves are off the ground at once?” Muybridge was a pioneer of scientific research in motion analysis. He devised a camera system consisting of 12 individual negative film cameras, each triggered milliseconds after each other, to collect a series of motion pictures, see Fig. 2. Muybridge’s multi-camera concept was also applied in the Manhattan project. Berlyn Brixner (1992) at Los Alamos used a set of 37 synchronized cameras in a row, all recording at the conventional 30 frames per second, to record the first nuclear explosion at 1-ms time interval. With the need for ever increasing recording speeds, rotating mirror technology was introduced for ultra-high-speed imaging for thermonuclear weapons research in the 1950s. The first ultra-high-speed rotating mirror camera was constructed, capable of recording at 1 million frames per second (Mfps). The incoming image was spinned around by rotating prisms or mirrors along the inside of a 1-m-radius arc covered with a static negative film rail. Using a very clever optical configuration, known as the Miller principle (Miller 1949), the experiments were photographed, and the pictures revealed timing errors in detonation which neither the ballistics engineers nor the physicists had anticipated. This was the last major technical hurdle of this history-changing project, solved by the ultra-high-speed camera. A series of rotating prisms, rotating mirror, streak cameras, and rotating drum cameras were developed in the years to follow, see overviews in Igel and Kristiansen (1997) andParker and Roberts (2002). All systems in those days relied on negative film, and for many reasons, these have now been replaced, first by VHS and Video8 in the 1980s, followed by digital CCD and CMOS units after the turn of the millennium. In the race for the need for speed, the mechanically driven cameras were superseded by electro-optical camera systems, see the overview in Honour (2002), notably the image converter cameras. These cameras are a modified version of the Gen I image intensifier tubes where the beam of photoelectrons, formed at a high-gain photocathode, was rapidly deflected such that multiple images could be projected separately onto a phosphor screen, where it was photographed and in later years imaged onto a CCD imager. Image converter cameras were commercialized as Imacon cameras by Hadland, and similar models were offered by Hamamatsu and Cordin.

Digital high-speed imaging systems are available today from 200 frames per second to 200 million frames per second. Systems vary from compact handheld consumer-type cameras to systems filling an entire laboratory. Two classes of imaging systems are common: first, systems with a single CCD and CMOS chip. These systems are capable of acquiring 1,000–5,000 frames per second at up to 1080p HD-TV resolution, which also makes them highly popular for sports and nature and science documentaries. Many of these systems can achieve higher frame rates by rapidly shifting the charge to neighboring cells in the chip at the expense of a considerable reduction in resolution, for example, 128 × 48 pixels at 500 kfps for a typical high-end camera. This obviously limits the use of these types of cameras for frame rates exceeding one million frames per second; one ends up with only a couple of pixels. That is why the second class of imaging systems makes use of the old concepts of Muybridge and Brixner, that is, if one cannot solve the problem with one camera, one simply uses more. The number of frames for a sequential recording is then dictated by the number of cameras available in the camera system, for example, 128 frames for the Brandaris camera (Chin et al. 2003) running at 25 Mfps and eight frames for the highest imaging frame rates of 200 Mfps for the image converter cameras (Honour 1994), such as the Imacon-200. But the limitations of high-speed imaging systems are not only in the frame rate and the number of frames. Selection criteria for the use or the choice of a particular high-speed imaging system also include a detailed assessment of the system’s sensitivity, actual (!) resolution, pixel sizes, trigger options, shutter speeds, the size or weight, and of course cost.

A typical recording with a digital high-speed camera is shown in Fig. 3. It shows how a droplet impacts on a pool of water, forming a cavity that collapses. A thin microjet is formed on the axis of the collapsing cavity, pointing downward, and a microbubble with a radius of the order 100 μm pinches off from the jet. From synchronous hydrophone measurements, we identify that the familiar sound of dripping water is directly associated with the formation of this bubble, not with the contact of the droplet with the water surface (Prosperetti et al. 1989). The sound is harmonic and is related to the size of the microbubble formed through its eigenfrequency f ₀ = 3.3 mmkHz/R ₀ (Minnaert 1933). And since the droplet impact formation changes from droplet to droplet, also the bubble formation process changes from droplet to droplet, leading to the characteristic “plik, plok, pluk” sound (audio online).

This particular series of events was recorded at 1,000 frames per second, with an exposure time setting of 150 μs. But what dictates the frame rate and what dictates the exposure time? What is it in this dripping water problem, and in flow visualization in general, that we are after? What is it that we want to investigate? Is it the impact of the droplet or the formation and inertial dynamics of the cavity? Is it the microjet formation, the microbubble pinch-off, or the microbubble oscillations leading to sound formation? The answer is that we need to know the relevant timescale of the problem; in other words, the flow problem dictates how the high-speed imaging needs to be done. And therefore, the purpose of this review is to provide insight into high-speed imaging technology and to get acquainted with the various high-speed imaging techniques. It will offer basic assessment of the high-speed imaging protocols: first flow problem-oriented and then imaging hardware-oriented. We offer rule of thumbs and we discuss common pitfalls. We will discuss ultra-high-speed imaging, at timescales shorter than a microsecond, important for microfluidic applications. We discuss high-brightness illumination techniques and offer ways to calibrate and test high-speed imaging systems. And finally, we conclude with unique combinations of traditional experimental imaging methods in fluid mechanics (Schlieren, PTV and PIV, fluorescence) with high-speed imaging techniques, opening up a wealth of new opportunities for experiments in fluids.

2 High-speed imaging basics

2.1 Definitions

With reference to Fig. 4, we start by defining a set of keywords associated with high-speed imaging. First of all, the frame rate or recording speed is defined as the number of frames taken per second. The units are given in frames per second and are abbreviated with fps. Engineering notation is used to further shorten the units; 1 kfps represents 1,000 frames per second; 1 Mfps represents 1 million frames per second. Normal video-rate CCD cameras record at a frame rate of 25 fps (PAL in Europe) or 30 fps (NTSC in US and Japan), and typical commercially available high-speed cameras record at a frame rate of 1–5 kfps. Each frame of a regular video-rate camera consists of an interlaced image composed of odd and even fields, a technology dating back from the television standard and an important technique of improving the picture quality of a video signal on CRT devices without consuming extra bandwidth. The odd fields fill the odd rows, the even fields fill the even rows, and each field is taken sequentially at a time interval of 20 ms for PAL and 16.7 ms for NTSC. For high-speed events, interlacing causes motion artifacts and aliasing, as the frame is composed of two fields taken at different times. Most high-speed imaging systems therefore use progressive scanning, where each field is a complete frame.

The interframe time is defined as the time in seconds between two successive frames, and the interframe time is therefore the reciprocal function of the frame rate. An ultra-high-speed camera accommodating a set of intensified CCD cameras can have an interframe time of 5 ns; hence, it achieves a frame rate of 200 Mfps.

The exposure time is the duration in seconds that the frame is exposed to the object. The exposure time is typically controlled by a mechanical or electronic shutter. For optimum temporal resolution, the exposure time should be equal or smaller than the interframe time. To reduce motion blur, the exposure time is typically chosen to be shorter than the interframe time. Some camera systems do not allow to set the exposure time independently from the interframe time, so for these systems the only other way to control the exposure is to control the pulse length of the illumination source, for example, by using a pulsed Xenon flash or a light-emitting diode (LED). In the following we will refer to it as the illumination time, see also Fig. 4b.

Now, the important questions we have to ask ourselves before we do the high-speed imaging experiment are as follows:

(a)
What frame rate is required?
(b)
What exposure time is required?
(c)
How do these relate to the magnification?

We should thereby keep in mind that the criterion is quite different for a pure visualization and for illustrative purposes rather than for a quantitative evaluation of the high-speed event.

2.2 Frame rate

To answer the first question, we will need to find out the relevant timescale of the high-speed event. In some applications we look at a cyclic event of which we know the (driving) frequency. Then according to the Nyquist sampling theorem, the required sampling rate or frame rate should be more than twice, but conveniently at ten times the event frequency. In many applications we have little clue of the relevant timescale. One may be inclined to rely on the typical velocity as a guide of the relevant timescale; however, you will easily be misled as the timescale also scales with the inverse length scale.

The optimum frame rate f can be estimated from:

$$f = \frac{N\cdot u}{l} $$

(1)

where N represents the required number of samples, as discussed above a minimum value of 2, but a typical value of 5–10 would be more appropriate, u a typical velocity, and l a typical length scale. Eq. (1) reveals the importance of high-speed imaging in microfluidic applications, where l is small and consequently f is large. Similarly, one should realize that a velocity of 1 m/s corresponds to 1 μm/μs, which means that looking at μm length scales requires μs interframe times at these velocities, corresponding to Mfps frame rates.

To exemplify the selection of the proper frame rate, we first refer to Fig. 5 showing the Kaye effect (Kaye 1963; Collyer and Fisher 1976). Here a thin stream of shampoo is poured from a height of 20 cm. The shampoo forms a viscous heap, and once in a while, a thin jet of shampoo leaps from the heap. The timescale of this problem is governed by the length scale l of the incoming jet, which is of the order of 2 cm, and by its velocity. The velocity u can be estimated from a balance between the potential energy $\dot{m}gh$ and the kinetic energy $\frac{1}{2}\dot{m}u^2,$ leading to $u=\sqrt{2gh}.$ With a height of 20 cm, we obtain an estimated velocity u = 2 m/s, independent of the mass flow rate $\dot{m}$. Using Eq. (1) we then find that the leaping shampoo experiment zoomed in at the heap (field of view of 20 mm) can be visualized with a frame rate of order 1 kfps. Tracking of the microbubbles entrained in the incoming stream confirms the impact velocity. Tracking of the very same microbubbles in the outgoing stream, see Fig. 5c, reveals the energy dissipation in the dimple and was key to our understanding of this problem (Versluis et al. 2006): At some instant, due to a favorable geometry, the incoming jet will slip away from the heap, and while for a viscous Newtonian fluid such a slip would only lead to a small disturbance in the wrinkling or coiling pattern, in the shear-thinning shampoo the resulting high shear rate forms a low-viscosity interface leading to an expelled jet at low inclination. Meanwhile, the incoming jet will exert a vertical force on the viscous heap forming a dimple. The dimple deepens because of the sustained force exerted onto it by the incoming jet, thereby erecting the outward-going jet. The inclination of the streamer steepens until it hits the incoming jet and disturbs or even interrupts the inflow, thereby halting the Kaye effect.

As a second example, we now look at oscillating microbubbles, see Fig. 6, used as contrast bubbles in medical ultrasound imaging. The bubbles have a typical size of 1–5 μm, they are coated by a stabilizing phospholipid shell, and the bubbles are driven near resonance (Minnaert 1933) by an ultrasound pulse with a typical frequency of 1–5 MHz. The bubbles generate an echo which is picked up in the far-field in pulse-echo mode by a transducer. While the surrounding tissue reflects the ultrasound linearly, the nonlinear bubble dynamics leads to a harmonic contribution, which by picking up the harmonics leads to the contrast with the tissue. And as blood is a very poor ultrasound scatterer, the injection of these microbubbles facilitates the real-time perfusion imaging of organs. From potential flow arguments, the nonlinear bubble echo pressure P _s at a distance r can be calculated directly from the bubble oscillations: $P_{\rm s}=\rho R({R\ddot{R}+2\dot{R}^2})/{r^2},$ with R the radius of the bubble, $\dot{R}$ the velocity of the bubble wall, and $\ddot{R}$ the acceleration of the bubble wall. To understand and to optimize contrast-enhanced medical ultrasound imaging, it is therefore fundamental to record the bubble dynamics with sufficiently high accuracy as to be able to calculate its time derivatives, that is, velocity and acceleration, for nonlinear echo pressure evaluation. Here the system is governed by the driving frequency of 1 MHz, and thus, the sampling frame rate must be of the order of 5 Mfps for the velocity, and another factor 5 higher for the acceleration, 25 Mfps. It will be shown later how such a high imaging frame rate can be achieved. Note that in this particular example of vibrating microbubbles, the relative amplitude of oscillation ${\Updelta R}/{R_0}$ is of the order of 10 % of the ambient radius R. With $R = R_0 + \Updelta R {\rm sin} \omega t$ we find that the velocity of the bubble wall $\dot{R} \simeq \omega \Updelta R{\rm cos}\omega t$ is of order 1 m/s, the same as for the leaping shampoo. So again, velocity is not the governing factor for the frame rate determination, but must rather be combined with the length scale of the problem.

2.3 Exposure time

To answer the second question (on the exposure time), let us first make the remark that for many applications we do not need a high frame rate per se; however, for delineation purposes and for sufficient spatial resolution for scientific analysis, it is advantageous to reduce the exposure time to freeze the motion. Figure 7, for example, shows the collapse of a cavitation bubble produced by the snapping shrimp (Versluis et al. 2000). The shrimp closes its snapper claw with lightning speed, and water is squeezed out from the claws. It forms a jet with a velocity of 25 m/s, fast enough to create a cavitation bubble due to the reduced pressure of the jet, which is below the vapor pressure of water. The cavitation bubble grows to a size of several millimeters, followed by an inertial collapse, see Fig. 7b. At a frame rate of 2 kfps, the interframe time is 500 μs and the exposure time here is set at 25 μs. The cavitation bubble exists only for 300 μs or so; if the exposure time would have been set to the interframe time, the bubble dynamics would be completely smeared out and the bubble would probably not even be visible. With the current recording settings, very small details of the collapsed bubble cloud are captured, see again Fig. 7b. If we were to resolve the full dynamics of the bubble collapse, then we would have to record at a reference timescale of the order of the exposure time used here (as this freezes the motion). A 25 μs interframe time then corresponds to a frame rate of 40 kfps, see the details in Versluis et al. (2000).

In a similar way we image, in a granular fluid, the impact of a steel ball into loose very fine sand, see Fig. 8. It creates a splash, followed by an intense jet with a height that exceeds the release height of the ball (Lohse et al. 2004). The frame rate was 2,000 frames per second, and the exposure time was set much smaller to capture the fine details of the grainy splash, the jet, and the clustering of particles within the jet.

A shorter exposure time always goes at the expense of the signal level. It decreases linearly with the exposure time setting; therefore, it must be compensated by an increased illumination level. Going from video-rate imaging to high-speed imaging sometimes requires an increased illumination level of four orders of magnitude, which is not easily overcome; therefore, illumination considerations are included in detail in this review. The exposure time can be as short as 1 μs for a CCD camera using an electronic shutter. Specialized double-frame particle imaging velocimetry (PIV) cameras have an exposure time of 200 ns, while an intensified CCD camera can have an exposure time as short as 250 ps owing to the fast electronic gating of the image intensifier.

The exposure time can be set in several ways. The simplest way to control the exposure time is by using a mechanical shutter. The fastest mechanical shutters open and close in just under 1 ms. A shorter exposure time can be achieved by controlling the opening and closing phase in separate shutters; for instance, the start of the exposure can be initiated by the perforation of a thin metal foil, while the end of the exposure time can be controlled by redirecting or removing a mirror. In some systems we find liquid crystal shutters. They can be switched down to milliseconds timescales, but the transmission can be poor for an open shutter (50 %), and the faster shutters lack sufficient contrast ratio and light bleeds through in the “closed” position.

In CCD and CMOS cameras, the exposure time is set by an electronic shutter that can be set manually (by DIP switches), electronically, or by software. Shutter times can be as short as 1 μs. Even shorter exposure times can be accomplished by using an image intensifier, which converts photons imaged onto a photocathode to electrons. The electrons are accelerated onto a multi-channel glass plate (MCP), which amplifies the electrons by 4–5 orders of magnitude. Double-MCP configurations are used to achieve even higher gains. The amplified electrons then impinge on a phosphor screen to convert them back into photons. Image transfer is conserved through the image intensifier at the expense of loss of resolution and the addition of noise, in particular for higher gain settings. A fiber-optic taper couples the phosphor screen onto a CCD chip for recording, see Fig. 9a. The advantage of using the image intensifier is threefold. First, as the name implies, it intensifies the image. Secondly, and most important for our interest here, is the ability to gate the exposure time, both the timing and the duration. By switching the voltage between the photocathode and the MCP, the electrons can be either accelerated or repelled from entering the MCP; switching of the 250 V can be accomplished in 5 ns. And finally, as the phosphorescence of the phosphor screen decays slowly (order milliseconds for a P-43 phosphor), it gives us the ability to capture fast events at nanoseconds timescale with a CCD with a relatively long video-rate readout time of 16.7 or 20 ms. For some fluorescence applications, the image intensifier is also a convenient tool to transform ultraviolet photons, which cannot be detected with a silicon-based chip, to photons in the visible, blue or green, depending on the phosphor type.

2.4 Magnification

Now to answer the quantitative measure of the exposure time setting, we remind us that the application of high-speed imaging in flow visualization is primarily aimed at obtaining precise information about the position and dimensions of the fluid flow at a series of instants in time. It is therefore of prime importance to capture the smallest details in both temporal and spatial resolution at high contrast. With the previous sections in mind, we can now quantify the criteria to achieve optimum high-speed flow visualization. These are

(1)
to fulfill the spatial Nyquist criterion,
(2)
to maximize the signal-to-noise ratio at the sensor,
(3)
to minimize motion blur.

Assuming that an imaging objective lens with the highest possible numerical aperture is chosen for the experiment, the effective optical magnification M must be sufficient to avoid undersampling of the image with respect to the spatial resolution, R _s, as defined by the Rayleigh criterion (Davidson and Abramowitz 2002). The spatial Nyquist criterion requires that at least 2 pixels lay within $M \cdot R_{\rm s}$. Hence, it follows that the magnification should be such that:

$$M>2r_{\rm p}/R_{\rm s} $$

(2)

where r _p is the pixel size of the sensor located in the image plane. Larger magnifications allow denser spatial sampling (known as oversampling), but also result in a smaller field of view and lower image brightness. The magnification is therefore limited by the optimum size of the object image and the available illumination intensity.

To ensure maximum image contrast, the intensity of the illumination should be adjusted to cover the full dynamic range of the camera sensor. The extent to which the intensity can be varied is, however, limited due to the reciprocal relationship between illumination intensity and exposure time. To clarify, an image with a certain fixed brightness can be obtained with either low light conditions and longer exposure times, or, vice versa, a bright flash and short exposure. In the situation where an object is motionless, both settings result in identical images. However, if the object of interest is moving, the image becomes susceptible to motion blur. This undesired effect causes a smeared appearance of the image of the object due to its displacement during the time the image is recorded. Minimizing motion blur constitutes the third criterion that would be required to accurately capture a single high-resolution image of a moving object, and this can be achieved by adjusting the temporal resolution of the imaging system. As discussed before, this temporal resolution is determined by the camera exposure time τ _c or by the illumination time τ _p. Evidently, the actual temporal resolution τ will be the shortest of both durations; hence, τ = min (τ _p, τ _c). If we now define $\epsilon$ as being the displacement in pixels of the object moving with velocity u during time $\tau: \epsilon = \tau M u/r_{\rm p}$ motion blur will be minimal if $\epsilon \le 1$. In most experiments the object velocity, the magnification, and the pixel size are predetermined so motion blur is avoided by choosing an exposure time or illumination time such that:

$$\tau\le r_{\rm p}/M u $$

(3)

Drop formation in inkjet printing is an almost perfectly reproducible process, and single-flash photography is frequently applied for its visualization (Dong et al. 2006; Hutchings et al. 2007), see Fig. 10a. Here we demonstrate the use of two different flash illumination sources to capture the front end (or head) of an inkjet microdroplet of 100 pL in volume and with a velocity u of about 20 m/s. The printhead is an experimental prototype developed by Océ Technologies B.V., similar to the printheads used in references (de Jong et al. 2006; Wijshoff 2010). The printhead ejects droplets with a diameter of 30 μm at a typical velocity of 20 m/s during drop formation. Some 35 μs later, the tip of the droplet reaches a terminal velocity close to 6 m/s. We use a high-intensity Luxeon LED or iLIF (van der Bos et al. 2011) for illumination with pulse durations of τ _p = 400 ns and τ _p = 7 ns, respectively. The incident light is focused through a collimating lens onto the region of interest. The collimation lens was chosen such that the numerical aperture (NA) matches with that of the microscope objective ensuring maximal optical resolution (R _s≈ 2 μm) and intensity (Davidson and Abramowitz 2002). The microscope (Olympus BX-30MF) is equipped with a long working distance (WD) 20× objective (Olympus, WD 25 mm/NA = 0.25). The camera is a Lumenera LM165 with a sensitive Sony EXview HAD CCD sensor with a pixel size of 6.5 × 6.5 μm² and a resolution of 1,392 × 1,040 pixels. The scale factor for the camera and the 20× microscope objective combination is 320 nm/pixel. Figure 10b, c shows the results of the microdroplet head after it has just exited the nozzle and has slowed down to its terminal velocity. The practical advantages of the combination of high-intensity and short pulses of light for imaging of fast phenomena become apparent from a detailed comparison of the images in Fig. 10. Using Eq. (3) for the indicated magnifications, we find that the droplet can only be imaged without motion blur if the illumination times are smaller than τ = 16 ns. The droplet displacement during exposure $\epsilon$ is small enough for the iLIF, Fig. 10c; however, the pulse length of the LED is too long, resulting in motion blur, as can readily be seen in Fig. 10b.

2.5 Scale and scaling issues

As a final example, we now present two cases that relate to the problem of scaling. These include an example of a finite-time singularity and a typical example of how scaling affects high-speed imaging analysis.

Figure 11a shows the discharge of a liquid from a microscopic nozzle at a sufficiently large velocity which leads to a continuous jet that breaks up into droplets, so-called Rayleigh breakup, due to the interfacial capillary forces (Plateau 1873; Rayleigh 1879). Here we can estimate the required frame rate for the formation of microdroplets directly from the capillary breakup time $\tau_{\rm b} = \sqrt{\rho r_0^3 / \gamma}$. With a jet radius r ₀ of 20 μm, a density ρ of 1 × 10³ kg/m³, and a surface tension γ of 0.072 N/m, we find τ _b = 10 μs, which translates to a frame rate of 500 kfps. The diminutive Rayleigh jet of Fig. 11b has a radius of 1.25 μm, and the breakup timescale is reduced to approximately 150 ns and would therefore require imaging at 50 million frames/s. The frame rate with which Fig. 11b was taken was 13.7 Mfps (interframe time of 73 ns) and just barely fulfills the Nyquist theorem.

Figure 12 shows the collapse of a cylindrical cavity formed after a small disk-shaped plunger is pulled into the water. The collapse is driven by the pressure of the water; near the water surface, it is not so high and the collapse speed is relatively low. Near the plunger the water pressure is high, but cavity formation has just started. Consequently, the cavity collapses somewhere in the middle where two jets are formed upon collapse, one upward and one downward. An estimate of the typical length scales and velocities, for example, those of the plunger: l = 10 cm and u = 1 m/s, using Eq. (1) gives us a frame rate of 100 fps. If we want to study better the moment of collapse, we zoom in, which we know from Eq. (1) inevitably leads to higher frame rates. However, the radius of the neck, r ₀, goes to zero following a universal scaling behavior with r ₀ ∝ (t − t _c)^α, where t _c is the time of collapse, and (t − t _c) the time until collapse, and where α represents the power law scaling exponent. A simple power law behavior can be predicted based on a purely liquid inertia-driven collapse giving rise to a 1/2 scaling exponent (Longuet-Higgins et al. 1991; Oğuz and Prosperetti 1993), although in reality the problem is much more complicated than can be described here (Gordillo et al. 2005; Bergmann et al. 2006; Thoroddsen et al. 2007). The scaling of the neck radius leads to an increasing velocity closer to final collapse, until it reaches infinity, an example of a singularity in finite time, when $t \rightarrow t_{\rm c}$. And while 100 fps was sufficient to capture the dynamics of the cavity collapse at a scale of 10 cm, at a scale of 1 cm we will need a frame rate of 10 kfps, while at a scale of 1 mm we will need ultra-high-speed imaging at a frame rate of 1 Mfps. Thus, zooming in leads to a dramatic increase in the high-speed imaging effort.

3 High-speed imaging systems and techniques

3.1 High-speed photography

High-speed flow phenomena and small-scale flow phenomena that are typically encountered in microfluidics require both a high spatial resolution and a high temporal resolution. To freeze the motion during exposure, we have identified in Sect. 2.4 given above that we can employ either a camera with a short exposure time or a light source capable of emitting flashes of short duration. The first option generally requires a fast high-speed camera system, which is often limited in its pixel resolution (Thoroddsen et al. 2006), see Sect. 3.3.4. For this reason, flash photography is the preferred method for experiments where high-resolution single images are required, since the method does not require a fast shutter and can in principle be performed with any type of camera. Furthermore, the pulse duration of flash illumination sources can easily be shorter than the shortest exposure time of a high-speed camera.

3.2 Stroboscopic imaging

We have also identified that once we are interested in real-time flow dynamics, we need to resort to imaging at the timescale of the event (which can become as small as nanoseconds in certain cases), with the exception of reproducible and repetitive events. In the latter cases, we can use stroboscopic imaging without recourse to the use of an (expensive) ultra-high-speed camera system. In stroboscopic imaging we apply the high-speed flash photography technique, and we delay all consecutive recordings by a fixed time spacing δt, where δt is a fraction of the period, T, of the event. δt is tuned such that it spans the interframe time of the camera, which as mentioned before can in principle be any type of camera, for example, a CCD video camera or a high-resolution digital SLR photo camera. Playing the captured frames in a video sequence then provides us with a stroboscopic movie of the event, which in many cases is indiscernible from a real-time high-speed movie, but with a dramatically improved spatial image resolution.

Figure 13 shows the stroboscopic recording of inkjet microdroplets, see the details previously discussed in Sect. 2.4. A short flash was created by exciting a laser dye solution with a pulsed laser (Zijlstra and Ohl 2008; van der Bos et al. 2011). The resulting fluorescent light has a very high intensity, while the undesired coherence of the laser light, which would lead to interference and speckle, is removed. The illumination time of the light source is only 7 ns. To record the images, a sensitive CCD camera (Lumenera LM165) was used. As an example to estimate the effect of motion blur, based on Eqs. (2) and (3), we list Table 1 where we compare the stroboscopic flash technique and ultra-high-speed Brandaris camera. It shows the minimum imaging timescales, τ _min, and the expected motion blur, $\epsilon,$ due to the actual time scale, τ _f. The combination of the flash light source, the camera, and the microscope gives $\epsilon < 1$ for the 20× magnification of the imaging microscope objective; hence, no motion blur is expected.

Table 1 The experimental parameters for stroboscopic flash recordings and for high-speed imaging

Full size table

The short illumination time of the fluorescence light source gives the possibility of choosing a very small temporal resolution δt, corresponding to the smallest timescale in the experiment. In Fig. 13b a step time of 10 ns was used and, to confirm its reproducibility, five images were recorded at each step. To control the delays between the drop, the illumination, and the camera, a pulse generator (Berkeley Nucleonics model 565) with a 250 ps precision was used. Additionally, a fully automated procedure was programmed in Matlab (The MathWorks) source code to control the hardware and to collect the images from the camera. In the experiments the initial phase of the drop was studied, recording only the first 60 μs of the drop formation. This resulted in a total of 30,000 images for a single experiment, taking about 6 h to record. Twenty-one images of the drop evolution are displayed as a time sequence in Fig. 13. The high-speed recordings were also performed with the Brandaris-128 ultra-high-speed camera (Chin et al. 2003) at a frame rate of 16 Mfps. To illuminate the drop formation, a high-intensity Xenon light source (Perkin Elmer MVS 7010 Xenon flash) was used. Due to the small field of view and the limited number of frames, multiple movies were required to visualize the entire drop formation. The total time of the drop formation is approximately 60 μs. With an interframe time of 62.5 ns, the Brandaris-128 camera acquires 128 frames in 8 μs in a single recording session. Therefore, 12 sequential movies must be captured to record the full extent of the drop formation process. The 12 movies were subsequently stitched together into a single continuous movie (which can only be done owing to the near-perfect reproducibility of the inkjet system). In addition, with the small field of view of the Brandaris camera at the 20× magnification, two series of movies were recorded at different axial distances from the nozzle. The results are displayed in Fig. 13a. If, however, the flow phenomenon is not repetitive, or uncontrollable, or unexpected, or when transients are present, in other words, when still photography and stroboscopic methods fail, then there is no other option as to resort to high-speed imaging, which will be the subject of the following Sect. 3.3.1.

3.3 High-speed imaging

Figure 14 gives an overview of commercial and specialized high-speed imaging systems. The left part of the diagram comprises commercial CCD and CMOS systems with an upper frame rate near 5,000 fps. These systems typically have a large number of pixels allowing detailed and high-resolution high-speed imaging. The center part of the diagram is occupied by high-end systems, roughly from a frame rate of 5,000 fps to 1 Mfps. As the throughput of the imaging system is limited, the number of pixels recorded drops linearly with increasing frame rate. The right part of the diagram contains ultra-high-speed camera systems at frame rates exceeding one million frames per second. These include highly specialized image sensors with in situ storage, digital rotating mirror systems, and image-intensified ultra-high-speed imaging devices with frame rates of up to 200 million frames per second.

3.3.1 CCD and CMOS systems

Charge-coupled device (CCD) chips (Nobel prize in Physics 2009 for Boyle and Smith for the invention of an imaging semiconductor circuit) are silicon semiconductor chips consisting of an array of photosensitive picture cells, called pixels. An array is built up of M columns × N rows of pixels. The incoming photons are converted into a charge (Nobel prize in Physics 1921 for Albert Einstein for the discovery of the photoelectric effect) by integrating the charge of each of the electrons contained in the potential well of the pixel. The deeper the potential well, that is, the more electrons can be contained in the pixel, the higher the dynamic range of the CCD chip. The dynamic range can be up to 17-bits deep, that is, more than 1.3 × 10⁵ separate gray scales can be identified. A row of pixels is read out by shifting the charge into the shift register of the CCD chip. The content of the serial shift register is then transferred to the camera circuit board where it is preamplified and digitized. A digital image is thus built up row by row in the frame buffer of the chip. Readout of a normal video-rate CCD chip is 20 ms, whereas high-speed imaging CCD chips have a special chip architecture and special A/D conversion architecture with a typical throughput of up to 5 Mpixels per millisecond, which corresponds to several Gigabits per second.

The sensitivity of a CCD chip is coupled to the noise floor within the individual pixels. Cooling reduces the thermal dark noise that collects over time, but for high-speed imaging with only very short readout times, that is not relevant. Signal noise and readout noise are intrinsic to the system and should be carefully assessed. CCD chips can be very sensitive, and unintensified CCD cameras are therefore used in surveillance technology. A light level of 0.0005 lux can be detected with a CCD, where 1 lux is already considered sensitive for consumer-type handycam CCDs. Pixels can be as small as 2–5 μm, making up a very compact chip. The pitch of the photosensitive area and the total pixel area along either the row or the column direction does not correspond to the full extent of the photosensitive part. This ratio is called the fill factor. Surface-coated microlens arrays can improve the effective fill factor to give a yield of almost unity. Blooming occurs when the potential well overspills and charge leaks to the neighboring pixels. As many of us use high-speed imaging in back-illumination, anti-blooming architecture is welcomed on-chip, and the microlenses also compensate for the loss of sensitivity as a result of anti-blooming technology, see Fig. 15.

While CCD chips have a better image quality because of their reduced noise component, chips fabricated using the recent Complementary Metal-Oxide Semiconductors (CMOS) technology are by far less expensive and faster and require less power to operate, which makes them popular for webcams and mobile phones. Each CMOS pixel has its own readout channel and can therefore be read out individually, and no shifting of the rows is necessary, which dramatically reduces the readout time of a subframe of the chip. As also the size of the data extracted from the region of interest is reduced, CMOS chips allow for a much faster readout, up to 1 million frames per second, before the upper limit of the throughput is reached. CMOS chips are, literally “a bit,” more noisy. Therefore, the pixels in CMOS high-speed imaging systems are often larger, typically of a size of 20 μm, to increase their sensitivity. For microscopic applications, this can be a problem as a higher magnification will be needed to obtain a similar field of view.

Color information in CCD and CMOS chip systems is typically obtained by placing a color filter array in front of the chip. The Bayer filter arrangement is the most popular, although different types of filters exist with a modification of colors and arrangement or using dichroic mirrors. Moreover, color information can also be obtained with multi-sensor cameras using dichroic and trichroic beam splitter prisms to separate out the colors per sensor. These systems typically have superior image quality and lower noise levels.

The Bayer filter consists of four color filters (red, green, green, blue—RGGB) per set of four pixels. An RGB image is recovered by demosaicing algorithms, which interpolate the set of red, green, and blue pixel values for each point. As each filter blocks approximately two-thirds of the incoming light, insertion of the color filter array goes at the expense of a factor 3 in sensitivity. Therefore, if color is not an essential ingredient for the flow visualization, it is recommended to use the monochrome models.

With decreasing interframe and even shorter exposure times, less photons reach the detector. The compromise between frame rate and sensitivity is a recurring problem in high-speed imaging. A similar compromise is found between sensitivity and magnification, which decreases the light collection efficiency quadratically. The solution can be found on either end: on the one hand increases the sensitivity of the detector and on the other hand increases the brightness of the image or the object through more intense illumination or better light collection efficiency.

The sensitivity of the detector is related to the noise level of the detector. Many of the sources of noise are intrinsic to the design of the imaging chip. Consequently, an increase in the sensitivity of silicon-based chip technology (CCD or CMOS) is difficult to accomplish. The most sensitive general CCDs are as sensitive as the most sensitive high-speed CCDs. One should therefore keep in mind that a high-speed event that is recorded in video rate at 25 fps, once recorded at 100 kfps, needs an increased illumination of at least three orders of magnitude based on its decreased interframe time, something which is non-trivial with even the brightest flood lights. Also keep in mind that if we would like to keep the field of view identical, we need to magnify by a factor 10 as a result of the reduced pixel count (see Fig. 14), leading to an additional loss of two orders of magnitude. Here, we should in principle revert to pulsed illumination instead, with all energy concentrated in the duration of the exposure. Xenon flash sources, Edgerton’s 1930s invention, provide ample energy (up to 0.5 J of photometric light output per pulse), are well controlled for syncing with the camera, and can be as short as several nanoseconds. The pulse repetition frequency (PRF) for high-intensity flashes is limited, however, typically to 10 Hz or less. Increased flash rates are possible at the expense of decreased output energy. Lasers are also used for illumination. The laser output is spectrally intense and primarily of interest for monochromatic applications such as high-speed fluorescence imaging. Copper-vapor lasers offer a high repetition rate of 40 kHz. High-brightness light-emitting diodes (LEDs) become increasingly popular for illumination. LEDs are readily available in a variety of colors (matching the spectral response of the imaging device) and relatively inexpensive. Although high-brightness LEDs already seem very bright to the naked eye, the real advantage for high-speed imaging applications is their use in a pulsed operation for a very short duration (order 100 ns) in a flash or burst mode. A dramatic increase in the intensity is accomplished by driving high currents through the diode; a pulse of up to 100 A can be driven through the LED without damaging it. The flash repetition rate depends on the LED housing’s ability to dissipate the heat that is formed; see Sect. 3.5 for testing the sensitivity of high-speed imaging systems.

One way to compensate for the loss of the number of photons is to simply enlarge the pixel size. This is common practice in CMOS-based chips, which suffer from a higher noise level than those equipped with a CCD chip. The typical size is 20 μm, but can be up to 66.5 μm (Etoh et al. 2002). It should be noted that the gain in sensitivity by a larger pixel size is counterbalanced by the fact that while imaging at a microscopic level, the required magnification decreases the illumination level by the same amount. In fact, in microfluidic high-speed imaging, one would be better-off with a smaller pixel size, as the requirements for magnification are then more relaxed and do not interfere with the limitations of optical diffraction.

Image intensifiers have been proposed to increase the sensitivity. First of all, we should note that because of the milliseconds lifetime of the image on the traditional P-43 phosphor screen (see Sect. 2.3), image intensifiers are not well suited for high-speed imaging applications. P-46 and P-47 phosphor screens on the other hand have a considerable shorter phosphorescence of order microseconds and could potentially be used for high-speed imaging applications (Hoess and Fleder 2003). The use of these phosphors is hindered, however, by the presence of a slowly decaying “tail” which leads to ghosting effects. More importantly, we should note that the signal level is indeed increased with increasing gain on the image intensifier but that this equally well holds for spurious noise that is amplified as well. The overall gain in sensitivity is therefore marginal, see Fig. 16. The noise is reflected in the image as a grainy overlay, typical for image-intensified images, and it therefore also reduces the resolution of the images. The main purpose of an image intensifier therefore remains to be to control the exposure time at nanoseconds timescale, as discussed previously in Sect. 2.3.

3.3.2 Triggering

Typically, triggering of a high-speed recording is essential, because we have a limited number of frames available for the recording. A proper positioning of the temporal region of interest is then required. If the number of recorded frames is not an issue and sufficient camera memory and storage space is available, triggering may still be useful, simply for convenience to instantly locate the high-speed event for further analysis. Triggering can be performed in several ways: manually using a trigger button, from a control device driving the system, and either mechanically, optically, or acoustically coupled to the electronics of the camera. Most digital high-speed cameras have a TTL trigger input. Optical triggering can be done with a laser photodiode combination, with the object interrupting the laser beam. Acoustic triggering can be accomplished using a microphone or a hydrophone. Note that the speed of sound both in air and in water is relatively low and that a trigger delay will be experienced depending on the distance between sound source and the detector. In the special case of ultra-high-speed imaging, especially those systems with a limited number of frames, such a delayed trigger, will always trigger too late.

The above description is termed a post-triggered acquisition where the frames are collected after the trigger is received. The trigger signal in this case is referred to as the start trigger. Sometimes we are also interested in the conditions preceding the event under study, and the trigger is only received after the event of interest. Or the event is too fast or too unexpected to trigger, for example, in crack formation (Kun et al. 2006). As an example, inkjet printing nozzle failures are extremely rare. On average, only once in a billion droplet ejections (only once during a 6-h continuous run), a disturbance could occur, for example, through air entrainment from a dust particle near the nozzle opening. In all these cases pretriggered acquisition is essential. In a pretriggered acquisition, the frames are already collected before the trigger signal is received. In such an application, the hardware initiates frame acquisition with a software function and stores the image data in a ring buffer in memory. The trigger signal in this case is referred to as the reference trigger. The principle is quite similar to that of a digital storage oscilloscope, images are continuously stored in the buffer, and once the total record length is filled, the oldest images are overwritten with the fresh images. Once the reference trigger is received, storage is halted, and the preset reference point for the trigger is set (with an arbitrary ratio of pre- and post-trigger frames), and images of the event before the actual triggering occurred are available.

The recording of Fig. 3, for instance, is triggered by the acoustic signal emitted by the microbubble formed after pinch-off of the downward microjet. The falling droplet and the formation of the cavity were recorded before the trigger signal was received. And while this highly controlled experiment could also have been recorded in a post-triggered acquisition, for example, by the falling droplet interrupting a laser--photodiode relay switch, all images of the implosion of the cavitation bubble of the snapping shrimp, Fig. 7, could only be taken with the pretriggering option in function. Triggering occurred on the loud impulsive snap of the animal, admittedly after gentle tickling with a paintbrush, which only helped trigger the snap, not the timing of it. In modern digital high-speed cameras, pretriggered acquisition is standard and these systems can be slaved to the experiment. Some high-speed camera systems, for example, rotating mirror cameras, are necessarily master to the experiment, and unless the high-speed event itself can be triggered, pretrigger images cannot be obtained. In some digital rotating mirror cameras, pretriggered acquisition can be implemented using smart electronic triggering.

The recorded frames need to be correlated in time. For this purpose a sync-out of the high-speed camera can be measured simultaneously with a trace related to the high-speed event, for example, hydrophone recording. In some systems this is already implemented in the software. As an example we show recordings of the interaction of a rising bubble with a hot-wire probe (Rensen et al. 2005), see Fig. 17. The purpose of the experiment was to obtain the energy spectra of a turbulent two-phase flow and to understand the two-way coupling of how rising bubbles modify the turbulence spectra and how the turbulent flow changes the bubble trajectories. This is a situation with a high gas fraction loading, which is really an experimental condition where optical techniques would fail miserably, hence the use of the hot-wire measurement technique. A typical trace is shown in Fig. 17, where we observe the drop in thermal conductivity of the hot-wire probe (for a duration of only 5 ms) as soon as the wire protrudes the gas core of the bubble. Each red dot in the recording indicates a single frame of the high-speed movie, which is strictly correlated in time. The purpose of this particular high-speed imaging experiment of the hot-wire/bubble interaction was to uniquely link the measured hot-wire traces to whether the bubble was penetrated by the hot wire (and recovered, as shown in Fig. 17a), whether the bubble bounced off the hot wire, or whether it splitted in multiple microbubbles as a result of the interaction with the heated wire (Bruun 1995). A perfect correlation of the timing is therefore essential.

3.3.3 Data

The data stream in high-speed imaging can be considerable. We should differentiate the data stream during acquisition and the data stream after acquisition. The first can be on-chip and should accommodate very fast direct transfer to memory, as it directly relates to the corresponding refresh rate for imaging. The second data stream relates to storage and usually goes at reduced speed through USB, FireWire, or Ethernet connection. Using flash SSD memory, the number of images that can be stored in memory is extensive, where 32 GB is no exception. The total data stream, however, does not depend on the selected frame rate, as the image size is inversely proportional to the frame rate, see Fig. 14. We do record a lot of images at higher frame rates, but the images are also smaller. However, the high number of frames does impose latency in data transfer when transferring image data for off-line analysis.

Another important aspect that should be considered here is where to store the acquired data. For example, a 2-s recording taken at 5,000 fps will take nearly 7 min to review at a playback speed of 25 fps. More importantly, with an uncompressed image size of 1 Mpixel, the total amount of data of that single event is 5 GB! A successful set of experiments in short time may result in a few hundred or so recordings corresponding to 500 GB of data, which theoretically may take up to 2.5 h to download over USB to an external hard drive with a transfer rate of 480 Mbps, but which typically proves to be very much longer in practice. Therefore, it is important to develop preanalysis software to quickly evaluate the recordings and distill the relevant observations, the true meaning of data reduction. Pretriggering, as discussed in Sect. 3.3.2, greatly helps in these efforts. In practice, however, the data reduction and image post-processing and analysis consume most of the work, and one may easily spend up to 2 months of work on analysis as a result of a single afternoon of experiments.

3.3.4 The problem of throughput

In Sect. 2.4 we have seen that there is a coupling between the temporal resolution and the spatial resolution. In high-speed imaging with commercial CCD or CMOS cameras, there is yet another important coupling between the two. They are coupled in the sense that higher frame rates can only be achieved by reducing the amount of readout pixels, as a consequence of the upper limit of the total throughput rate of the digitizer; only a fixed amount of data can be pushed through the A/D convertor of the imaging chip, typically of order of several Gbits/s, or one billion pixels/sec. Strictly speaking, the resolution remains the same if the magnification is unaltered, but in practice the tendency is to fill the field of view, which is then automatically associated with a loss in resolution once the frame rate is increased. A typical example is given in Fig. 18. Here two droplets are separated, and a microscopic thread is being stretched until it snaps. Surface modes, which can be described by a superposition of Legendre polynomials, are formed and decay following they own mode eigenfrequencies, setting up a display of droplet shapes. The fast dynamics is captured by the high-speed camera at 100 kfps, and all the sequential shapes look beautiful, but in all honesty, the image quality is not all that great. A single-shot photographic image well-timed with variable delays would do a much better job here.

Figure 14 shows an overview of how the total number of pixels in a frame recording depends on the frame rate for a set of popular high-speed imaging systems, taken from their datasheets. In the datasheet literature, the full frames are referred to as frames with the frame rate in frames per second (fps)], while the subdivided images are referred to as pictures [with the frame rate in pictures per second (pps)]. A standard digital high-speed CMOS system records 1,000 × 1,000 pixels (= 1 Mpixel) frames at a frame rate of 1,000 fps, but it quickly drops to about 100 × 100 pixels at a frame rate of 100 kfps. Frame rates, or to be precise, picture rates, of 1 Mpps (pictures per second) can be obtained, albeit at the cost of a much reduced resolution, typically a strip of 8 pixels wide. Although this may be sufficient for the imaging of one-dimensional motion, it is not practical in most situations.

3.4 Ultra-high-speed imaging systems beyond 1 Mfps

So what if we need to push the limits of current CCD and CMOS camera technology and perform high-speed imaging at exposure times shorter than 1 μs, and correspondingly at a higher sensitivity, and at frame rates higher than one million frames per second with a sufficient number of frames, typically one hundred or more. The simplest solution is to follow the classic example of Eadweard Muybridge and Berlyn Brixner by using a set of multiple cameras, as discussed in Sect. 1. The DLR German Aerospace Centre in Göttingen developed an 8-channel image-intensified ultra-high-speed framing camera for high-speed shock wave imaging, which was later commercialized through DRS Hadland as the Imacon 468 camera (Honour 1994). The Imacon 468 camera, see Fig. 19, utilizes a pyramidal beam splitter with an octagonal base that redirects the incoming image to eight individual image-intensifed CCDs. Cordin Company and Specialized Imaging offer cameras with similar beam-splitting configurations and 1 or 2 Mpixel chip size. LaVision’s UltraSpeedStar camera uses an image splitter into four channels with color iCCDs and then splits electronically the four RGGB channels to acquire 16 images at 1 Mfps. The image intensifiers can be gated as short as 5 ns, leading to a frame rate as high as 200 Mfps for the Imacon 200, and very high gain levels can be applied that allow for sufficient sensitivity at such short exposure times. Each intensified CCD can be individually triggered with high accuracy and flexibility. The Imacon 200 can be equipped with P-47 phosphor-backed image intensifiers, with rapid falloff of the phosphorescence, which allows a second exposure per channel, giving 2 × 8 frames in total with 1–4 μs in between successive exposures to prevent residual image of the previous exposure, termed ghosting of the image. As discussed in Sect. 3.3.1, overall the image intensifiers seriously degrade the image quality. And as a result of the beam splitting, a high gain needs to be applied to the intensifiers, decreasing the dynamic range as a result of increased noise levels. On the other hand, the system is unique in the 100–200 Mfps segment.

A rotating mirror camera does not suffer from the drawbacks of beam splitting or image intensifying and has a higher resolution, dynamic range, and frame count. A short exposure time is obtained by sweeping the image by rapid rotation of a turbine-driven mirror across a series of CCD sensors. The CCD sensors replace the traditional negative film used in the 1950s to record nuclear detonations. And while a nuclear detonation is bright enough to be captured with 3600 ISO number negative film, microfluidic applications with 100× magnification under a microscope certainly lack image brightness. Moreover, CCD sensors are much more flexible, can be accurately timed, and allow for multiple and repetitive exposures in short time. The Cordin 550 acquires 60 frames at 4 Mfps. The Brandaris camera, see Fig. 20, developed by University of Twente and Erasmus MC Rotterdam, is equipped with 128 CCD sensors, each of which can store six full-frame images in its onboard RAM. The helium-driven turbine can rotate at 20,000 rps, completing a sweep across the image arc in just 5 μs providing an interframe time of 40 nanoseconds. The Brandaris camera can therefore acquire a total of 768 frames at a frame rate of 25 Mfps in a set of 6 × 128 frames, and in segmented mode 12 × 64 frames or 24 × 32 frames (van der Meer et al. 2007; Gelderblom et al. 2012) or any permutation of the above, provided that a 20-ms readout time is required once all 128 channels are filled. A typical example of images taken with the Brandaris camera is included in Fig. 21. A third class of imaging systems utilizes in situ storage of image data on the CCD chip itself.

The Shimadzu HPV series camera (Etoh et al. 2002) can record 103 frames in a single experiment at a maximum frame rate of 1 Mfps with a fixed resolution of 312 × 260 pixels. Its operating principle is similar to a two-frame storage CCD chip used in PIV imaging (Adrian and Westerweel 2011); only here it uses 100 slanted storage pixels per one photosensitive pixel, see Fig. 22. This gives the sensor a rather poor fill factor of 13 % and limits its sensitivity, which is partly compensated by the large photosensitive pixel size of 66.5 μm. On the other hand, the camera operates as a traditional electronic camera, with flexible timing options, including pretriggering.

The future of ultra-high-speed imaging is probably in this direction. Recent developments in CMOS technology include backside-illuminated (BSI) image sensors. Back-thinned BSI sensors have an increased quantum efficiency, and through the absence of wiring on the backside, which can be exclusively placed on the front-side of the chip, they have in principle a fill factor of 100 %. This development has ultimately led to the design of the ISIS-V16 sensor with a tenfold increase in sensitivity which can now acquire 117 frames at a frame rate of 16 Mfps (Etoh et al. 2011). Another development is the stacked image sensor. Stacking is a straightforward extension of the BSI technology. Advanced circuit chips can be stacked on the front-side of the imaging chips. Commercial designs for consumer cameras are already marketed, and this technology can be applied to develop various highly functional image sensors for ultra-high-speed imaging. As for the in situ storage sensors, recent solid-state technology became available using special CMOS capacitors as the in situ memory. This system can operate at speeds exceeding 10 Mfps (Tichigi et al. 2012). Combinations of all these new developments are expected to reach 100 Mfps and beyond.

3.5 High-speed imaging: testing and calibration

A quantitative and objective determination of various physical parameters of the cameras (sensitivity, resolution, dynamic range, etc.) and light sources (brightness, spectral intensity, etc.) is welcomed. Testing conditions need to be constant and reproducible, and at the same time, it has to be flexible enough to fit different-sized sensor systems. That is why a black box design was developed that on one side was made to fit a light source, and on the other side, a sensor. The design is shown in side view in Fig. 23. The pathway of the light is as follows: Diverging light is emitted from the tip of a 7-mm optical fiber. The light from the optical fiber is attenuated by neutral density filters and a set of polarizers. The optical density of the filters ranges from 0.3 to 2.5. One of the polarizers is fixed, while the other can be rotated which allows for the fine-tuning of the light intensity within the range of a single neutral density filter. The combination of the filters and polarizers results in a dynamic light range covering three orders of magnitude. The transmitted light falls onto a diffusing glass, which will evenly illuminate a resolution target (1951 USAF glass slide, Edmund Optics), Fig. 24. The resolution target consists of eight groups of each six elements of line pairs with a resolution varying from 1 to 228 lp/mm. The green rectangle indicates the segment of the target that was used for the linearity, signal-to-noise ratio, sensitivity, and dynamic range measurements discussed below. The red squares show two groups of line pairs with different resolution. A translation stage allows for centering and a goniometric stage for rotational alignment in the object plane. An image of the resolution target is made by an objective lens (Carl Zeiss Makro-Planar T* f/2.8–100 mm) onto the sensor. The lens has the ability to make a one-to-one conversion of the object onto the sensor. The image distortions and luminance differences within the field of view caused by the lens are negligible. The lens and sensor are mounted onto a translation stage in order to focus the object onto the sensor.

Now we define the test procedure for six elementary high-speed sensor properties that can be quantified with the test setup and which should aid at the selection of the optimum sensor:

1.
Dark current

With a fully covered sensor, a single frame is recorded. The measured signal is the dark current that is caused by thermal energy in the CCD combined with the readout noise of the amplifier. The measurement will give two results: the background level and the noise. The noise is defined as the standard deviation of all pixel values, whereas the background level is the mean. A high background signal will decrease the usable number of gray levels and thereby reduce the dynamic range. When the noise is low, the background can easily be subtracted afterward without image degradation. Therefore, the standard deviation is more important than the mean value.

2.
Linearity

Several images are taken at different light intensities to measure the linearity of the sensor. The sensor is said to be linear when the output signal level follows a line with constant slope as a function of the incident light intensity. Linearity is desirable in quantitative measurements where the image intensity can be correlated to a concentration of a dye or density of molecular species.

3.
Resolution

An important factor for the image quality is the resolution. This was measured by taking images of the different groups of the resolution target. For each element the modulation transfer function (MTF) was determined. The MTF is a measure of the contrast of an image at a particular resolution. It is also referred to as the spatial frequency response.

$$MTF = \frac{({\rm signal}-{\rm background})}{({\rm signal}+{\rm background})} $$

(4)

4.
Signal-to-noise ratio

This ratio relates the signal level to the standard deviation of the background. The higher the signal-to-noise ratio, the better a signal or object can be distinguished from its background. The ratio is measured for different light intensities.

5.
Sensitivity

Sensitivity is the main property that needs to be evaluated for the different sensors. It is defined as the signal level with the background subtracted. Since light intensities in a high-speed imaging system are typically very low, the sensor should have a substantial signal level on top of the background level in order to distinguish between background and signal without considering the influence of noise. As with the linearity and signal-to-noise measurements, the signal is measured for different light intensities.

6.
Dynamic range

It is not just the background that suffers from noise. Signal noise also influences picture quality and generally increases with increasing signal level. The standard deviation of the signal level mainly determines the number of gray scales that an imaging system can distinguish. An evenly illuminated area of the sensor will give a distribution of pixel values which all correspond to the same gray level. The range of this particular gray level is given by the standard deviation of the pixel values. This determines the confidence interval of a pixel value, that is, the probability that a pixel value outside this interval does not belong to the same gray level. The total number of gray levels that fit within the complete range of the sensor is defined as the dynamic range.

With the full data set, different sensor types can now be compared. To summarize, we want the dark current to be low, the sensor linear, its resolution high, its signal-to-noise ratio also high, the sensitivity high, and the dynamic range large. This information together with several other considerations, such as size and price, in conjunction with operational complications, such as the need for cooling and possible electromagnetic interference, will then determine the optimum sensor for the applications, once it also fits the high-speed imaging considerations mentioned in the previous sections.

4 Combined high-speed imaging techniques

High-speed imaging and other experiments in fluid techniques are remarkable in their own entity, but combine them and you get something life-changing. Here we discuss various combinations of high-speed imaging with Schlieren and shadowgraphy, with particle tracking and particle imaging velocimetry, and we discuss high-speed fluorescence imaging.

4.1 High-speed shadowgraphy and Schlieren imaging

Both shadowgraphy and Schlieren imaging are well-known techniques to visualize density differences in a transparent medium (Settles 2001). Density differences cause a change in the light refraction index of the medium. Light traveling through the medium will experience a different refraction depending on the local density gradient of the medium. The Schlieren technique visualizes the spatial derivative of the refraction index, which can vary through a change in both density and temperature. Shadowgraphy is simpler to setup; however, it relies on the second derivative of the refractive index; hence, it is difficult to quantify and serves primarily as a powerful visualization tool. Both techniques are broadly used in compressible flows, in which pressure fluctuations and therefore density differences are common. Shadowgraphy and Schlieren are traditionally used in combination with high-speed flash photography in wind tunnel shock wave imaging (Settles 2001). One striking example is the photograph of the turbulent wake structure behind a supersonic riffle bullet by Mach and Mach (1889). The visualization of wake structures in liquid flows is more difficult. De Vries et al. (2002) was one of the first to report on high-speed imaging of wake visualizations behind rising spheres using a Schlieren setup. One important aspect is the non-intrusive nature of the Schlieren technique, as opposed to PIV (Brückner 1999) and dye injection (Schouveiler and Provansal 2002). The latter two are widely used and described in literature for the visualization of wake structures behind bodies [refs] for both solid spheres and bubbles; however, the introduction of particles and fluorophores interacts with the slip boundary layers and interfaces of the rising objects, changing the wake vorticity and bubble shape. On the other hand, density differences in the water created by the motion of spheres and bubbles are too small for Schlieren to be detected. Therefore, a vertical temperature gradient is imposed on the quiescent water in the water tank. A heating lamp is positioned on the top of the water tank, creating a constant vertical temperature gradient of 1.0 K/cm. The spheres and bubbles drag along water of different density, creating a local change in density, which can be visualized with the Schlieren technique. Of course, the introduction of high-speed imaging has also put a major challenge on the sensitivity issue of the setup.

Here we describe the construction of an orthogonal high-speed thermographic Schlieren imaging setup by transmitting the light originating from two LEDs (0.12 W) through a pinhole (diameter 1.5 mm) with a diffusor to create a near-perfect light point source. The two light bundles are collimated into two parallel light bundles with a diameter of 100 mm. The two light bundles are passed perpendicular to each other through a water tank. After passage through the water tank, the light bundles are focused onto a circular cutoff dot printed on a transparent plastic sheet (size range between 1 and 3 mm), and acting as the Schlieren knife edge. After the cutoff, the light bundles were directed onto a high-speed camera. In the post-processing a background image is subtracted from all recorded images to provide a smooth background in each image.

Figure 25 shows stereoscopic images of the wake structure behind falling spheres with densities approximately 4 % (Fig. 25a) and 5 % (Fig. 25b) higher than that of the surrounding liquid and, for comparison, that behind rising spheres with densities that are approximately 4 % (Fig. 25c) and 5 % (Fig. 25d) lower. In all these examples the sphere diameter is 6 mm. The wakes of the falling spheres appear to have a more irregular structure, and the path followed by these spheres shows much larger deviations from a straight vertical line. These examples show the characteristic wakes for spheres following a zigzag path, namely that the two counterrotating threads in the wake cross at the centerline of the zigzag. The presence of these threads of opposite-signed streamwise vorticity implies that the sphere experiences a lift force. As a consequence of the periodic crossing of the threads, this force is always directed toward the zigzag centerline, see for more details the work of van Wijngaarden (2005) and that of Veldhuis et al. (2005).

Shadowgraphy is a relatively straightforward technique for the visualization of shock waves in fluids. It is being used for the display of shock waves in aerodynamics and in wind tunnel supersonic flows. In medical therapy the accurate registration of the pressure waves, for example, in shock wave lithotripsy (SWL) (Bailey et al. 2003; Sapozhnikov et al. 2007), shock wave therapy (SWT) (Warden 2003), histotripsy (Zhen et al. 2004), or high-intensity focused ultrasound (HIFU) (Kennedy et al. 2003), is of prime importance for quality assurance of therapeutic devices. During megasonic cleaning in the semiconductor industry, shock waves may harm the fragile nanofabricated structures on the substrate. Figure 26a shows the wavefront of such a shock wave. It originates from the collapse near the substrate surface of a bubble cluster (red arrow) formed in cylindrical micropits (Bremond et al. 2006), which act as artificial nucleation sites for bubble cavitation. Here, an ultrashort flash photography method enabled shadowgraphy imaging of these waves. With the wave moving at the speed of sound (c _w = 1,500 m/s), the image was taken 220 ns after the bubbles collapsed. It indicates how fast these waves propagate at the small scale. If we were to visualize directly the propagation of such a shock wave, we could resort to stroboscopic laser or iLIF illumination (van der Bos et al. 2011) at fixed time delays (see Sect. 3.2) if the propagation of the pressure wave is reproducible. Fortunately, this typically is the case, except that the aftermath of the pressure wave passage, that is, cavitation cloud formation and thermal ablation, is more unpredictable and irreproducible. In this case we need to set aside the stop motion imaging and use high-speed imaging instead. If we were to image a 50-mm region of interest, then owing to Eq. 1 we need a non-trivial frame rate of 300 kfps. Earlier ultra-high-speed framing studies on shock wave inception of cavitation were performed by Ohl (2002) with the 8-frame Imacon 468. Figure 26b shows a high-speed visualization of a focused pressure wave generated with a piezoelectric device used for shock wave therapy (Dreyer et al. 2000). The movie was recorded at a frame rate of 1 Mfps, and focusing of the shock wave is observed, as well as formation and collapse of single-cavitation bubbles and the radial expansion of secondary shock waves from these microbubbles.

4.2 High-speed holography and interferometry

While Schlieren and shadowgraphy are popular visualization techniques, they suffer from a lack of accuracy due to their first-order and second-order dependence, respectively, of the refractive index following the Gladstone--Dale relation (Settles 2001; Merzkirch 1987). Moreover, Schlieren and shadowgraph methods do not offer enough resolution to measure quantitative details of shock waves, even in a simple geometry (Kleine 2010). Moreover, they cannot deal with the intrinsic difficulty of measuring with a limited depth of field which is dictated by the geometrical optics, see the review by Kitzhofer et al. (2011). The solution is to revert to measurements in wave optics using digital holographic imaging techniques, where the recorded phase image is an interference pattern from a superposition of a spherical scattering wave and a reference plane wave. The image contains information about the objects in the field of view with a nearly infinite depth of field (Kitzhofer et al. 2011; Arroyo and Hinsch 2008).

Digital holography is used in a variety of fluids applications, including shocktubes, reentry problems, and other compressible flows [see the excellent review on this topic by Kleine (2010)], and laser-induced processes such as ablation and optical breakdown in air or water. The latter, for example, suffers greatly from the formation of what appears to be chaotic multi-bubble clusters, shrouding part of the cavitation event and its growth. In order to capture the dynamics within the full context of space and time, holographic techniques have been employed making use of the three-dimensional character of image reconstruction of high-speed holograms, thereby obtaining views of the bubble distribution at different depths of field and at different viewing angles (Ohl et al. 1999). Similarly, digital holography can be used for microdroplet characterization and for microparticle flow velocimetry, which explains its extensive use in PIV applications, see Kitzhofer et al. (2011) and the references therein.

Interferometry is not been new to high-speed imaging, with the first interferometer built by Luwdig Mach in 1892 and independently by Zehnder in 1891. The Mach--Zehnder interferometer has regained popularity in today’s high-speed imaging community with new emerging fast camera technology (Kleine et al. 2012). Compressible high-speed flows are visualized with a series of interferometry techniques, including speckle interferometry and differential interferometry (Desse 2006). The introduction of fast, color, high-speed cameras has facilitated digital color interferometry and color holography alike; both are popular techniques in high-speed flows such as in transonic wind tunnels and shocktubes. At the much smaller scale, high-speed color interferometry has recently been used to measure the air layer thickness during drop impact on a solid surface at frame rates of up to 100 kfps (van der Veen et al. 2012).

4.3 High-speed particle tracking velocimetry

In the last few years 3D-particle tracking velocimetry (PTV) has become a powerful measurement technique in fluid mechanics. The rapid development of high-speed imaging has enabled a successful implementation of the technique in studies on turbulent motion of particles (Porta et al. 2001; Mordant et al. 2004; Toschi and Bodenschatz 2009), thermal convection flows (Gasteuil et al. 2007; Ni et al. 2012), and bubbly flows (Rensen et al. 2005; Martínez-Mercado et al. 2010; Martínez-Mercado et al. 2012). For the latter, the measured 3D spatial position of particles and bubbles and their associated time trajectories allow for a Lagrangian description which is the natural approach for the study of this type of transport mechanisms. Figure 27 sketches the positions of four high-speed cameras (Photron 1024-PCI) which were used to uniquely determine the positions of seeding particles and bubbles in the flow. The four cameras were viewing from one side and focused in the central region of a water channel filled with bubbles injected from 500-μm-diameter needles mounted on three capillary islands. Camera objectives with a 50 mm focal length allowed for a depth of field of 6 cm. The image sampling frequency was 1,000 Hz with a camera resolution of 1, 024 × 1, 024 pixels. The cameras were triggered externally in order to achieve a fully synchronized acquisition. We used the PTV software developed at ETH Swiss Federal Institute of Technology for camera calibration and tracking algorithms. For a detailed description of this technique, we refer to the work of Hoyer et al. (2005) and references therein. A 3D solid target was used for 3D spatial calibration. Different bubble concentrations were achieved by varying the air flow through the capillary islands. We performed experiments with dilute bubbly flows with typical void fractions α in the range 0.28–0.74 %. A monodispersed bubbly swarm with mean bubble diameter of 4–5 mm was studied. Typical Reynolds numbers Re are of the order 10³, the Weber number We in the range 2–3 (implying deformable bubbles), and Eotvos number Eo around 3–4.

Bubbles were detected within a volume of 16 × 16 × 6 cm³ with an accuracy of 400 μm. To illuminate the measuring volume, homogeneous backlight and a diffusive plate were used in order to get the bubbles contour imaged as a dark shadow. The image sequence was binarized after subtracting a sequence-averaged background; then these images were used along with the PTV software to get the 3D positions of the bubbles. We acquired 6,400 images per camera corresponding to 6.4 s of measurement (6.7 GB of image files). For the most dilute case (α = 0.28 %), around N _b ≈ 190 individual bubbles were detected in each image. This quantity dropped to nearly 100 for the most concentrated cases (α = 0.65 % and α = 0.74 %) as for the higher void fraction, various bubbles cannot be identified as single objects. For even higher gas fractions, hot-wire anemometry (CTA) was employed, see Sect. 3.3.2. If a bubble could be tracked for at least three consecutive time steps, we call it a linked bubble, and the bubble trajectories were coupled to the 3D PTV data for further detailed analysis.

4.4 High-speed particle imaging velocimetry

Combined with PIV, high-speed imaging is a powerful tool for quantitative analysis of the flow [see e.g., the recent work on flow behind blast waves by Murphy and Adrian (2010)]. As an example, we describe a micro-particle imaging velocimetry (μPIV) setup suitable for the analysis of the flow in microscopic channels in a single plane. The release of neutrally buoyant tracer particles within the flow is a procedure that aids enhanced visualization of the flow pattern. Such particles are expected to follow the path of the fluid with accuracy (Tilton 1999; Raffel et al. 1998). A conventional PIV setup makes use of a laser sheet to illuminate (fluorescent) tracer particles to visualize and analyze the flow. In the present μPIV setup, an objective with a small depth of focus and a continuous light source was used instead of a laser sheet. The main advantage of this setup was that the recording speed was not restricted by the amount of light emitted from the fluorescent particles. Therefore, recordings could be made both at high frame rate and for a prolonged time. A prolonged recording time allowed for ensemble averaging, which reduces the noise in the individual PIV recordings (Raffel et al. 1998). Here we show results from dentistry, notably the flow visualization of the irrigation in endodontic channels. Irrigation is a common procedure during a root canal treatment. An antimicrobial solution, typically sodium hypochlorite (NaOCl, bleach), is flushed through an open-ended, or a side-vented, needle into the confinement of the root canal, leading to an axisymmetric or non-axisymmetric jet, respectively. The aim of this study was to quantify the amount of shear exerted on the root canal wall in various confinements and to understand the irrigant refreshment mechanisms, in a range of Reynolds and Damköhler numbers, and to compare with CFD modeling (Boutsioukis et al. 2010).

The typical flow velocity u inside the root canal is 1 m/s, and its typical dimension l is 100 μm. Thus, following Eq. 1, the frame rate required to capture the dynamic flow behavior of needle irrigation techniques must be higher than 100 kfps, depending on the applied magnification. Here we use a high-speed camera capable of recording 100 frames at a maximum frame rate of 1 Mfps (HPV-1, Shimadzu). The camera was attached to a microscope (BX-FM, Olympus) for 1.25× to 20× magnification. Illumination was provided for bright-field imaging by a high-intensity continuous metal-halide cold-light source. The flow through the needles was driven by a syringe pump at flow rates in the range 1–10 mL/min. The corresponding Reynolds numbers span two orders of magnitude, 10²–10³. The flow was allowed to start up for three seconds before the start of a measurement. Monodisperse hollow glass spheres with a mean diameter of 10 μm and a mean density of 1.1 × 10³ kg/m³ were added to distilled water as tracer particles for μPIV. The tracer particles fulfilled the requirements for μPIV imaging: They are neutrally buoyant, and they have a diameter much smaller than the typical length scale. For this system the Stokes number St = (particle response time)/(typical flow time) < 1, even for the highest velocities.

Traditional PIV techniques rely on two consecutive photographs illuminated with a laser sheet and light-scattering particles to visualize a thin plane in the volume of interest. However, in μPIV such a laser sheet would have a thickness of the same order as the volume of interest. Therefore, here the total volume is illuminated, and the depth of focus of the microscope objective is used to image only a thin image plane. The depth of measurement dz of a μPIV system is then defined as twice the distance from the focal plane where the particles become sufficiently unfocused that they do not contribute to the velocity field (Meinhart et al. 2000). For an objective with NA = 0.25 imaging 10 μm particles scattering broadband light through air, the depth of measurement is approximately 100 μm, which indicates that the captured particle movement was limited to a plane with thickness of approximately one-fifth of the root canal diameter and 10 times the particle diameter. The flow direction and velocity magnitude are obtained from the recordings of particle motion by calculating the cross-correlation between (parts of) two consecutive frames. Subpixel accuracy can be obtained by fitting a Gaussian curve to the 2D cross-correlation. The accuracy can be increased further using image processing techniques and data validation strategies. From the obtained velocity components u and v, the vorticity ω = ∂v/∂x − ∂u/∂y can be obtained. This equation can be discretized and applied explicitly to the PIV velocity data.

Each recording consisted of 100 frames, which allowed for ensemble averaging (Raffel et al. 1998). A cross-correlation is made for each pair of frames, and these cross-correlations are added. This results in a higher peak in the final cross-correlations, which is equivalent to a higher signal-to-noise ratio. However, ensemble averaging reduces the temporal resolution and is therefore only suitable for steady flows. Non-steadiness of the flow was not observed within those 100 frames; therefore, it was assumed that ensemble averaging did not cause a loss of temporal detail. The streamlines representing the measured flow patterns for the two types of jets are shown in Fig. 28. The global features of the measured velocity patterns match those predicted by the CFD model (Verhaagen et al. 2012), showing two distinct velocity patterns: a straight, divergent jet in the case of the axisymmetric jet and a cascade of eddies declining in strength for the non-axisymmetric jet. The measured velocities were of the same order of magnitude as calculated by theory and the CFD model, although the highest velocities were not always observed in experiment.

The accuracy of the μPIV technique depends on many factors, both in the measurement equipment and in the analysis. The error in the analysis can be estimated by taking an image representative of the experiments that need to be analyzed and displacing this image over a certain number of pixels. The PIV analysis should result in velocities equal to the displacements applied to the image. The script used in Verhaagen et al. (2012) showed that the applied displacements could indeed be recovered, within an error of approximately 5 %, which can be attributed to noise. This error analysis also pointed out that the result of the PIV analysis is very sensitive to the amount of details in the image, as low-contrast areas in the image will result in a zero velocity. An extension to three-dimensional velocity measurements would be very useful considering the essentially three-dimensional flow inside a typical microfluidic channels. 3D-PIV in microfluidics is not easy to set up, but may be possible by estimating the out-of-focus component of consecutive 2D slices (Pommer et al. 2007; Poelma et al. 2011).

4.5 High-speed fluorescence imaging

Laser-induced fluorescence (LIF) is frequently used in fluids research, such as the inflow visualization of turbulent mixing (Dahm and Dimotakis 1987; Sreenivasan and Prasad 1989; Crimaldi 2008) and combustion processes (Eckbreth 1996; Kohse-Höinghaus and Jeffries 2002; Aldén et al. 2011), in (micro-)particle image velocimetry (μPIV) (Poelma et al. 2008; Adrian and Westerweel 2011), and in molecular tagging (Hiller et al. 1984; Gendrich et al. 1997; Miles et al. 2000; Ribarov et al. 2002). Molecular fluorescence imaging uses pulsed dye laser tuned for resonant spectroscopic excitation, while for the measurement of velocity fields, pulsed Nd:YAG lasers provide the non-specific fluorescence excitation of micron-sized or nanometer-sized tracer particles. The typical pulse repetition frequency (PRF) of these lasers is on the order of 10 Hz, and a separate laser cavity would be required for each individual frame to be able to perform LIF at high frame rates. To sketch the complexity of such a high-speed fluorescence experiment, we describe the joint effort of the Combustion Research Centers of Lund and Delft Institutes of Technology (Ding et al. 2000) to visualize time-resolved planar LIF in the Delft piloted turbulent jet diffusion flame (Nooren et al. 2000) at a timescale smaller than 100 μs, which corresponded to the smallest turbulent Kolmogorov length scales of the non-premixed flame. The laser system consisted of a Nd:YAG laser cluster with eight laser heads that could be individually triggered. Laser-induced fluorescence of OH radicals excited at a wavelength of 283 nm was imaged at a fluorescence wavelength of 309 nm. The fluorescence was imaged onto a custom-modified, 8-channel, ultra-high-speed camera (Imacon 468, DRS Hadland, UK) (Honour 1994), which consisted of eight UV-sensitive intensified CCDs and which could record the eight individual frames at a minimum interframe time of 10 ns, corresponding to 100 Mfps. The fluorescence images were used for flame front tracking, and the interframe time was set to 75–100 μs based on the turbulent timescales of the flame, which had a Reynolds number of 9,600. Figure 29 shows typical results of the time-resolved planar LIF measurement in the turbulent diffusion flame. It shows the turbulence–chemistry interactions, notably the local extinction of the flame and the timescales of its re-ignition. These flame dynamics measurements were also combined with PIV measurements, using two of the laser heads at 532 nm and using TiO₂ as tracer particles (d = 1 μm). These measurements showed the underlying mechanisms of extinction through the penetration of fuel vortices through the stretched flame front structure.

A series of eight frames are limited in view of more extended studies involving transients or harmonic responses, where 100 or so time-resolved recordings would be required. From a cost perspective, pulsed lasers are then considered unsuitable, especially when taking into account the required power supplies and cooling equipment. A CW laser does not reach the high peak intensity levels of a pulsed laser; however, it can utilize the full exposure time to excite the fluorescent molecules multiple times. Moreover, diode-pumped solid-state lasers come in a large variety of wavelengths, they have a small form factor, they do not require extensive cooling, and they are much more affordable than a pulsed laser system. Here we discuss the combination of a 5-Watt, 532-nm DPSS CW laser and the Brandaris-128 high-speed imaging facility (Chin et al. 2003), a system now capable of recording ultra-high-speed fluorescence at up to 25 million frames per second. To facilitate a fast rise time and short pulses, the laser was gated by an acousto-optic modulator (AOM). The light-transmitting efficiency of the AOM is 92 %, resulting in a maximum power of 4.8 W for a gate time of 5 μs, sufficiently long to illuminate all 128 channels of the camera with a 40-ns frame interval. At the same time, the AOM avoids overexposure and heating of the (biological) sample.

The modular design of the microscope allowed for coupling of the laser light into the optical path of the microscope via a dichroic mirror. A 60× water-immersion objective with a numerical aperture of 0.9, typically used for microbubble experiments, produces a spot with a full width at half-maximum (FWHM) of 1.0 μm and a maximum laser intensity of 72 MW/cm². The spot diameter can be increased by making the laser beam divergent before it enters the objective. When recording in bright field with the fluorescence filters in place, 42 % of the Xenon flash illumination is filtered out by the dichroic mirror and a 532-nm laser notch filter, maintaining sufficient light levels for bright-field imaging. With the fluorescence module in place, the Brandaris-128 configuration gives rise to three illumination options: bright field, fluorescence, and the combination of both illumination techniques. For each individual recording within one experiment, the illumination method can be predefined. In the case of six full-frame recordings for instance, the first three can be captured in bright field and the final three in fluorescence. Other options include alternating bright field and fluorescence or even 12 combined recordings of which each recording is split into two virtual segments: The first 64 frames are recorded in bright field, and the second 64 frames are recorded in fluorescence.

Coated microbubbles have been extensively studied and characterized in their role as ultrasound contrast agents, quantifying the perfusion of organs, for molecular imaging detecting molecular signatures of diseases with single bubbles in the body, and for drug delivery and therapy. A major drawback of bright-field imaging of microbubbles is the poor contrast of the bubble interface due to a complex interplay of refraction and scattering of the light transmitted in back-illumination. In addition, bright-field imaging lacks the ability to distinguish between liquid media with similar refractive index, which is crucial for the visualization of drug release. Thus, ultra-high-speed fluorescence imaging is a very promising experimental tool for the study of the above medical applications. Figure 30 shows a polymer-coated microcapsule (Philips Research) (Kooiman et al. 2009) imaged both in bright field and in fluorescence. The latter shows a distinct delineation of the wall boundaries, and in addition, a high contrast is achieved between the fluorescent liquid core and its surroundings. The microcapsules could be recorded with a frame rate of up to 10 million frames per second, giving a unique insight into the physical mechanisms of local intravascular drug delivery. Phospholipid-coated microbubbles loaded with liposomes are another important drug-delivery vehicle system (Lentacker et al. 2009), and the release of the drug-loaded liposomes is visualized in the lower section of Fig. 30. These images were recorded with a Photron SA 1.1 camera running at 200 kfps with CW illumination controlled by the acousto-optic modulator, τ _laser = 500 μs. The release is governed by the acoustical streaming pattern around the oscillating lipid microbubbles.

5 Summary

All of the above have given us insight into the design of a high-speed imaging experiment. Let us now summarize all aspects in the following parameter chart (Fig. 31). The chart defines selection criteria for the experiment, based on Eqs. (1) and (2), and is supplemented with data from a series of high-speed imaging systems. This chart has been designed with the examples provided in this review, with the camera systems used in the examples. It can be easily adapted at will, for example, to extend the parameter range or to include custom camera systems.

The frame rates are represented by the diagonal lines and range from 1 frame per second at the top left to 1 billion frames per second at the bottom right. Any combination of the typical velocity and typical length scale of the problem gives the desired frame rate option. The selection of the camera is governed by two limiting cases, that is, the minimum temporal resolution and the minimum spatial resolution. The minimum temporal resolution is dictated by the maximum frame rate, for example, 25 fps for video, 1 Mfps for the Shimadzu HPV-1, and 25 Mfps for the Brandaris-128 camera. Figure 14 has demonstrated that the number of pixels is significantly reduced with increasing frame rate. We set the lowest imaging frame size to 100 × 100 pixels, which for an entry-level CMOS high-speed camera corresponds to a frame rate of 35 kfps. For a high-end CMOS high-speed camera, the upper frame rate is then limited to 225 kfps. The minimum spatial resolution is given by the pixel size of the sensor and sets a limit to the minimum length scale that can be imaged in the experiment. We have set the minimum spatial sampling rate to 10× the Rayleigh criterion, in analogy to the temporal sampling rate in Eq. (1), and the maximum optical magnification was set to 100× . For the Shimadzu HPV-1, for example, with its pixel size of approximately 70 μm, the lower limit was therefore set to 14 μm. The CMOS cameras, with a typical pixel size of 20 μm, have no issues with a minimum length scale within the parameter range set in the chart of Fig. 31. The two limiting cases for each camera system are indicated with the colored solid lines. The shaded areas indicate the selection of a specific camera system, assigned by the corresponding color. Examples presented in this review are indicated in the chart. For example, Figs. 5, 6, and 28 are displayed along the vertical line with velocity 1 m/s, while the other examples of Figs. 5, 7, and 26b are indicated along the horizontal line with a typical length scale of 1–5 cm. It is seen that different camera systems are needed to image these events. This becomes even more evident for the two examples where scaling issues were present (Figs. 11 and 12). It can also be observed that several areas within the parameter space are exclusive to some specialized cameras. For example, micron-sized particles with a speed of 1 m/s can only be imaged by a Brandaris-128 type camera, while submillimeter particles in water traveling at the speed of sound are to be imaged with a framing camera of the type Imacon-200. The chart also shows that micron-sized events at a speed of 1,000 m/s cannot be comfortably captured by the current range of high-speed imaging systems, and until now we need to resort to single- or double-frame flash photography using iLIF, as was shown very recently for supersonic microjets by Tagawa et al. (2012). Emerging technologies in the near future will provide higher frame rates, and also better resolution and better sensitivity, which may, for example, lead to high-speed fluorescence imaging of flow phenomena at significantly increased frame rates. A better sensitivity in the end will be beneficial for the resolution anyway by a reduction in the pixel size. In the future we also hope for higher frame rates in more compact imaging systems.

The current possibilities of high-speed imaging systems are impressive, but new developments are imminent. New designs will in the very near future lead to increased frame rates, higher sensitivities, smaller pixel sizes, and an overall smaller form factor of the sensor. The improved technologies will allow us to work on unraveling new details in fine-scale acoustics, such as high-frequency medical imaging and surface acoustic wave (SAW) atomization. It will be possible to study shock wave propagation at much smaller length scales. New application fields will emerge in plasma physics, and eventually, we may be able to follow light propagation in real time, not only stroboscopically. With these kinds of phenomena at a typical timescale of a nanosecond, this requires high-speed imaging at a frame rate of 1 billion frames per second. These systems may indeed become available in the near future, although we may encounter severe physical limitations in illumination and in the present IC semiconductor technology.

The purpose of this review was to provide insight in defining criteria for high-speed imaging experiments in fluids, to give rule of thumbs for a series of cases. To complete, let us therefore now also summarize three common pitfalls in high-speed imaging. The first pitfall is that one may be inclined to think that the velocity dictates the frame rate. How trivial it may sound but it is the relevant timescale that dictates the frame rate, and the relevant timescale can be found from the ratio of the velocity and the length scale, see Sect. 2.2. The smaller the length scale, the higher the frame rate, so for microfluidic applications, this rule should definitely be particularly considered. Secondly, many researchers are tempted to use a high-speed camera no matter what, but in all honesty, high-speed cameras give lousy pictures because of their limited pixel count. If not ultimately necessary, it is advised to use still frames from high-speed photography, which give superior image resolution and significantly higher scientific value over the high-speed imaging frames. Agreed, we may need high-speed imaging to find all the details of the process under study, but in the end, single-flash or stroboscopic imaging will give more scientific satisfaction. And finally, please do not underestimate the illumination conditions for high-speed imaging. Many translate video recordings directly to high-speed recordings and assume that a factor 2 or may be 10 is required to compensate for light loss, whereas in practice it amounts to an increase of three or four orders of magnitude simply dictated by the frame rate, viz., interframe time, let alone with an additional order of magnitude for setting a relevant exposure time.

References

Adrian RJ, Westerweel J (2011) Particle image velocimetry. Cambridge University Press, Cambridge
Google Scholar
Aldén M, Bood J, Li Z, Richter M (2011) Visualization and understanding of combustion processes using spatially and temporally resolved laser diagnostic techniques. Proc Comb Inst 33:69–97
Article Google Scholar
Anna SL, Bontoux N, Stone HA (2003) Formation of dispersions using ‘flow focusing’ in microchannels. Appl Phys Lett 82:364
Article Google Scholar
Arroyo MP, Hinsch KD (2008) Recent developments of PIV towards 3D measurements. Topics in applied physics, vol 112. Springer, Berlin, pp 127–154
Bailey M, Khokhlova V, Sapozhnikov O, Kargl S, Crum L (2003) Physical mechanisms of the therapeutic effect of ultrasound (a review). Acoust Phys 49:369–388
Article Google Scholar
Bergmann RPHM, van der Meer D, Stijnman MA, Sandtke M, Prosperetti A, Lohse D (2006) Giant Bubble Pinch-Off. Phys Rev Lett 96:154505
Google Scholar
Bouakaz A, Versluis M, de Jong N (2005) High-speed optical observations of contrast agent destruction. Ultrasound Med Biol 31(3):391–399
Article Google Scholar
Boutsioukis C, Verhaagen B, Versluis M, Kastrinakis E, Wesselink PR, van der Sluis LWM (2010) Evaluation of irrigant flow in the root canal using different needle types by an unsteady computational fluid dynamics model. J Endod 36(5):875–879
Article Google Scholar
Bremond N, Arora M, Ohl C, Lohse D (2006) Controlled multibubble surface cavitation. Phys Rev Lett 96:224501
Article Google Scholar
Brixner B (1992) In: Dewey JM, Racca RG (eds) Proceedings of 20th international congress on high speed photography and photonics, vol 1801. SPIE, Bellingham, WA, pp 52–60
Brückner C (1999) Structure and dynamics of the wake of bubbles and its relevance for bubble interaction. Phys Fluids 11:1781–1796
Article MathSciNet Google Scholar
Bruun H (1995) Hot wire anemometry: principles and signal analysis. Oxford Univesity Press, Oxford
Google Scholar
Chen AU, Basaran OA (2001) A new method for significantly reducing drop radius without reducing nozzle radius in drop-on-demand drop production. Phys Fluids 14:L1–L4
Article Google Scholar
Chin CT, Lancée C, Borsboom J, Mastik F, Frijlink ME, de Jong N, Versluis M, Lohse D (2003) Brandaris 128: a digital 25 million frames per second camera with 128 highly sensitive frames. Rev Sci Instr 74(12):5026–5034
Article Google Scholar
Chomas JE, Dayton PA, May D, Allen J, Klibanov A, Ferrara K (2000) Optical observation of contrast agent destruction. Appl Phys Lett 77(7):1056–1058
Article Google Scholar
Cohen I, Brenner MP, Eggers J, Nagel SR (1999) Two fluid drop snap-off problem: experiments and theory. Phys Rev Lett 83:1147
Article Google Scholar
Collyer AA, Fisher P (1976) The Kaye effect revisited. Nature 261:682–683
Article Google Scholar
Cranz C, Glatzel B (1912) Photographic recording of ballistic abd other rapid phenomena with the aid of the quenched spark. Phys Gesell 14:525–535
Google Scholar
Crimaldi JP (2008) Planar laser induced fluorescence in aqueous flows. Exp Fluids 44:851–863
Article Google Scholar
Dahm WJA, Dimotakis PE (1987) Measurements of entrainment and mixing in turbulent jets. AAIA 25(9):1216–1223
Article Google Scholar
Davidson M, Abramowitz M (2002) In: Encyclopedia of imaging science and technology, pp 1106–1141
de Gans BJ, Duineveld PC, Schubert US (2004) Inkjet printing of polymers: state of the art and future developments. Adv Mater 16(3):203–211
Article Google Scholar
de Jong N, Frinking PJA, Bouakaz A, Goorden M, Schourmans T, Xu JP, Mastik F (2000) Optical imaging of contrast agent microbubbles in an ultrasound field with a 100-MHz camera. Ultrasound Med Biol 26:487–492
Article Google Scholar
de Jong J, de Bruin G, Reinten H, van den Berg M, Wijshoff H, Versluis M, Lohse D (2006) Air entrapment in piezo-driven inkjet printheads. J Acoust Soc Am 120:1257–1265
Google Scholar
Desse JM (2006) Recent contribution in color interferometry and applications to high-speed flows. Opt Laser Eng 44:304–320
Article Google Scholar
de Vries AWG, Biesheuvel A, van Wijngaarden L (2002) Notes on the path and wake of a gas bubble rising in pure water. Int J Multiphase Flow 28:1823–1835
Article MATH Google Scholar
Ding T, van der Meer TH, Versluis M, Golombok M, Hult J, Aldén M, Kaminski CF (2000) Proceedings of third international symposium on turbulence, heat and mass transfer, pp 857–864
Dong H, Carr WW, Morris JF (2006) Visualization of drop-on-demand inkjet: drop formation and deposition. Rev Sci Instr 77(8):085101
Article Google Scholar
Dreyer T, Krauss W, Bauer E, Riedlinger RE (2000) Proceedings of IEEE ultrasonics symposium, vol 2. pp 1239–1242
Eckbreth AC (1996) Laser diagnostics for combustion temperature and species, 2nd edn. Gordon and Breach, UK
Google Scholar
Edgerton HE, Killian JJR (1954) Flash! seeing the unseen by ultra high-speed photography. Boston C. T. Branford Co, Boston, p 41, 130
Google Scholar
Edgerton HE, Jussim E, Kayafras G, Hayafas G (1987) Stopping time: the photographs of Harold Edgerton. Harry N Abrams Inc, New York, p 45
Google Scholar
Etoh TG, Poggemenn D, Ruckelshausen A, Theuwissen A, Kreider G, Folkerts HO, Mutoh H, Kondo Y, Maruno H, Takubo K, Soya H, Takehara K, Okinaka T, Takano Y, Reisinger T, Lohmann C (2002) IEEE international solid-state circuits conference digest of technical papers, pp 46–47
Etoh TG, Nguyen DH, Dao SVT, Vo CL, Tanaka M, Takehara K, Okinaka T, Kuijk HV, Bosiers J, Lesser M, Ouellete D, Maruyama M, Hayashida T, Arai T (2011) IEEE international solid-state circuits conference digest of technical papers, pp 406–407
Fry SN, Sayaman R, Dickinson MH (2003) The aerodynamics of free-flight maneuvers in Drosophila. Science 300(5618):495–498
Google Scholar
Gasteuil Y, Shew WL, Gibert M, Chillá F, Castaing B, Pinton JF (2007) Lagrangian temperature, velocity, and local heat flux measurement in Rayleigh-Bénard convection. Phys Rev Lett 99:234302
Article Google Scholar
Gelderblom EC, Vos HJ, Mastik F, Faez T, Lucan Y, Kokhuis TJA, van der Steen AFW, Lohse D, de Jong N, Versluis M (2012) Brandaris 128 ultra-high-speed imaging facility: 10 years of operation, updates and enhanced features. Rev Sci Instrum 83:103706
Google Scholar
Gendrich CP, Koochesfahani MM, Nocera DG (1997) Molecular tagging velocimetry and other novel applications of a new phosphorescent supramolecule. Exp Fluids 23(5):361–372
Article Google Scholar
Glasheen J, McMahon T (1996) A hydrodynamic model of locomotion in the Basilisk Lizard. Nature 380:340
Article Google Scholar
Gordillo M, Sevilla A, Rodríguez-Rodríguez J, Martnez-Bazán C (2005) Axisymmetric bubble pinch-off at high Reynolds numbers. Phys Rev Lett 95(19):194501
Article Google Scholar
Hiller B, Booman RA, Hassa C, Hanson RK (1984) Rev Sci Instr 55(12):1964–1967
Article Google Scholar
Hoess P, Fleder K (2003) In: Takayama K, Saito T, Kleine H, Timofeev EV (eds) Proceedings on 24th international congress on high-speed photography and photonics, vol 4183. SPIE, Bellingham, WA, p 127
Honour J (1994) In: Proceedings on 21st international congress on high speed photography and photonics, vol. 2513 SPIE, Bellingham, WA, pp 28–34
Honour J (1997) Electronic camera for simultaneous framing and streak recording. In: Proceedings of SPIE 2869, 22nd international congress on high-speed photography and photonics, p 668
Honour J (2002) In: Ray SF (eds) High speed photography and photonics, vol. PM120, SPIE, Bellingham, WA, pp 134–149
Hoyer K, Holzner M, Luethi B, Guala M, Liberzon A, Kinzelbach W (2005) 3D scanning particle tracking velocimetry. Exp Fluids 39:923–934
Article Google Scholar
Hu D, Chan B, Bush J (2003) The hydrodynamics of water strider locomotion. Nature 424:663–666
Article Google Scholar
Hutchings IM, Martin GD, Hoath SD (2007) High speed imaging and analysis of jet and drop formation. J Imaging Sci Technol 51(5):438
Article Google Scholar
Igel EA, Kristiansen M (1997) Rotating-mirror streak and framing cameras. vol PM43, SPIE, Bellingham, WA
Google Scholar
Kaye A (1963) A bouncing liquid stream. Nature 197:100–102
Article Google Scholar
Kennedy JE, ter Haar GR, Cranston D (2003) High intensity focused ultrasound: surgery of the future? Br J Radiol 76(909):590–599
Article Google Scholar
Kitzhofer J, Nonn T, Brückner C (2011) Generation and visualization of volumetric PIV data fields. Exp Fluids 51:1471–1492
Article Google Scholar
Kleine H (2010) Filming the invisible time-resolved visualization of compressible flows. Eur Phys J Special Topics 182:3–34
Article Google Scholar
Kleine H, Olivier H, Tsuji K, Etoh K, Takehara K, Etoh TG (2012) Time-resolved Mach-Zehnder interferometry of shock waves. In: 28th international symposium on shock waves, vol 1. Springer, Berlin, pp 577–583
Kohse-Höinghaus K, Jeffries JB (2002) Applied combustion diagnostics. Taylor and Francis, New York
Google Scholar
Kooiman K, Böhmer MR, Emmer M, Vos HJ, Chlon C, Shi WT, Hall CS, de Winter SHPM, Schroën K, Versluis M, de Jong N, van Wamel A (2009) Oil-filled polymer microcapsules for ultrasound-mediated delivery of lipophilic drugs. J Contr Rel 133:109–118
Article Google Scholar
Kudo N, Okada K, Yamamoto K (2009) Sonoporation by single-shot pulsed ultrasound with microbubbles adjacent to cells. Biophys J 96(12):4866–4876
Article Google Scholar
Kun F, Wittel FK, Herrmann HJ, Kröplin BH, Måløy KJ (2006) Scaling behavior of fragment shapes. Phys Rev Lett 96:025504
Article Google Scholar
Lentacker I, De Smedt SC, Sanders NN (2009) Drug-loaded microbubble design for ultrasound-triggered delivery. Soft Matter 5(11):2161–2170
Article Google Scholar
Lohse D, Schmitz B, Versluis M (2001) Snapping shrimp make flashing bubbles. Nature 413:477
Article Google Scholar
Lohse D, Bergmann R, Mikkelsen R, Zeilstra C, van der Meer D, Versluis M, van der Weele J, van der Hoef M, Kuipers J (2004) Impact on soft sand: void collapse and jet formation. Phys Rev Lett 93(19):198003–1
Article Google Scholar
Longuet-Higgins S, Kerman BR, Lunde K (1991) The release of air bubbles from an underwater nozzle. J Fluid Mech 230:365
Article MATH Google Scholar
Mach E, Mach L (1889) Weitere ballistisch-photographische Versuche. Sitzungsber Kaiserl Akad Wissenschaft Mathematisch-Naturwisenschaft Classe 98:1310–1332
Google Scholar
Marey EJ (1868) Determination experimentale du mouvement del ailes des insectes pendant le vol. C R Acad Sci 67:1341–1345
Google Scholar
Marey E (1894) Le mouvement. G. Masson, Paris
Google Scholar
Marmottant PH, Hilgenfeldt S (2003) Controlled vesicle deformation and lysis by single oscillating bubbles. Nature 423:153–156
Google Scholar
Martínez-Mercado J, Chehata D, van Gils D, Sun C, Lohse D (2010) J Fluid Mech 650:287–306
Article Google Scholar
Martínez-Mercado J, Prakash VN, Tagawa Y, Sun C, Lohse D (2012) Lagrangian statistics of light particles in turbulence. Phys Fluids 24:055106
Article Google Scholar
Meinhart C, Wereley S, Gray M (2000) Volume illumination for two-dimensional particle image velocimetry. Meas Sci Technol 11:809–814
Article Google Scholar
Merzkirch W (1987) Flow visualization, 2nd edn. Academic Press, New York
MATH Google Scholar
Miles RB, Grinstead J, Kohl RH, Diskin G (2000) The RELIEF flow tagging technique and its application in engine testing facilities and for helium air mixing studies. Meas Sci Technol 11(9):1272–1281
Article Google Scholar
Miller C (1949) Half-million stationary images per second with refocused revolving beams. J Soc Motion Picture Eng 53:479
Google Scholar
Minnaert M (1933) On musical air-bubbles and the sounds of running water. Phil Mag 16:235–248
Google Scholar
Mordant N, Leveque E, Pinton J (2004) Experimental and numerical study of the Lagrangian dynamics of high Reynolds turbulence. New J Phys 6:34
Article Google Scholar
Murphy MJ, Adrian RJ (2010) PIV space-time resolution of flow behind blast waves. Exp Fluids 49(1):193–202
Google Scholar
Muybridge E, Mozley AV (1887) Human and animal locomotion. Dover, New York
Google Scholar
Ni R, Huang SD, Xia KQ (2012) Lagrangian acceleration measurements in convective thermal turbulence. J Fluid Mech 692:395–419
Article MATH Google Scholar
Nooren PA, Versluis M, van der Meer TH, Barlow RS, Frank JH (2000) Raman-Rayleigh-LIF measurements of temperature and species concentrations in the Delft piloted turbulent jet diffusion flame. Appl Phys B 71:95–111
Article Google Scholar
Ohl CD (2002) Cavitation inception following shock wave passage. Phys Fluids 14(10):3512–3521
Article Google Scholar
Ohl CD, Kurz T, Geisler R, Lindau O, Lauterborn W (1999) Bubble dynamics, shock waves and sonoluminescence. Phil Trans R Soc A 357:269–294
Article MathSciNet MATH Google Scholar
Ohl CD, Arora M, Ikink R, de Jong N, Versluis M, Delius M, Lohse D (2006) Sonoporation from jetting cavitation bubbles. Biophys J 91:4285–4295
Article Google Scholar
Oğuz HN, Prosperetti A (1993) Dynamics of bubble-growth and detachment from a needle. J Fluid Mech 257:111
Article Google Scholar
Parker V, Roberts C (2002) Rotating mirror and drum cameras. In: Ray SF (ed) High Speed Photography and Photonics, vol. PM120, Chap 10. SPIE, Bellingham, WA, pp 158–180
Patek SN, Korff WL, Caldwell RL (2004) Biomechanics: deadly strike mechanism of a mantis shrimp. Nature 428:819–820
Article Google Scholar
Pishchalnikov YA, Sapozhnikov OA, Bailey MR, Williams JJC, Cleveland RO, Colonius T, Crum LA, Evan AP, McAteer JA (2003) Cavitation bubble cluster activity in the breakage of kidney stones by lithotripter shock waves. J Endourol 17(7):435–446
Article Google Scholar
Plateau JAF (1873) Statique exprimentale et thorique des liquides soumis aux seules forces molculaires, Gauthier-Villard, Paris
Google Scholar
Poelma C, Vennemann P, Lindken R, Westerweel J (2008) Exp Fluids 45(4):703–713
Article Google Scholar
Poelma C, Mari JM, Foin N, Tang MX, Krams R, Caro CG, Weinberg PD, Westerweel J (2011) 3D flow reconstruction using ultrasound PIV. Exp Fluids 50:777–785
Article Google Scholar
Pommer MS, Kiehl AR, Soni G, Dakessian NS, Meinhart CD (2007) Proceedings of the 2nd IEEE international conference on nano/micro engineered and molecular systems
Porta AL, Voth GA, Crawford AM, Alexander J, Bodenschatz E (2001) Fluid particle accelerations in fully developed turbulence. Nature 409:1017–1019
Article Google Scholar
Prosperetti A, Crum LA, Pumphrey HC (1989) The underwater noise of rain. J Geophys Res 94:32–39
Google Scholar
Raffel M, Willert C, Kompenhans J (1998) Particle image velocimetry, 2nd edn. Springer, Berlin
Google Scholar
Rayleigh L (1879) On the capillary phenomena of jets. Proc R Soc Lond 29:71
Article Google Scholar
Rensen J, Luther S, Lohse D (2005) The effects of bubbles on developed turbulence. J Fluid Mech 538:153–187
Article MATH Google Scholar
Rensen J, Luther S, de Vries J, Lohse D (2005) Hot-film anemometry in bubbly flow I: bubble probe interaction. Int J Multiphase Flow 31:285–301
Article MATH Google Scholar
Ribarov LA, Wehrmeyer JA, Pitz RW, Yetter RA (2002) Hydroxyl tagging velocimetry (HTV) in experimental air flows. Appl Phys B 74(2):175–183
Article Google Scholar
Sapozhnikov OA, Maxwell AD, MacConaghy B, Bailey MR (2007) A mechanistic analysis of stone fracture in lithotripsy. J Acoust Soc Am 121(2):1190–1202
Article Google Scholar
Schouveiler L, Provansal M (2002) Self-sustained oscillations in the wake of a sphere. Phys Fluids 14:3846–3854
Article MathSciNet Google Scholar
Settles GS (2001) Schlieren and Shadowgraph techniques, Springer, Berlin
Book MATH Google Scholar
Shi XD, Brenner MP, Nagel SR (1994) A cascade of structure in a drop falling from a Faucet. Science 265(5169):219–222
Article MathSciNet MATH Google Scholar
Sreenivasan KR, Prasad RR (1989) New results on the fractal and multifractal structure of the large Schmidt number passive scalars in fully turbulent flows. Physica D 38(1–3):322–329
Article Google Scholar
Tagawa Y, Oudalov N, Visser CW, Peters I, van der Meer D, Sun C, Prosperetti A, Lohse D (2012) Highly focused supersonic microjets. Phys Rev X 2:031002
Article Google Scholar
Talbot WHF (1852) On the production of instantaneous photographic images. Phil Mag 4:73–77
Google Scholar
Thoroddsen ST (2002) The ejecta sheet generated by the impact of a drop. J Fluid Mech 451:373–381
Google Scholar
Thoroddsen ST, Takehara K (2000) The coalescence-cascade of a drop. Phys Fluids 12:1265–1267
Google Scholar
Thoroddsen ST, Etoh TG, Takehara K (2006) Crown-breakup by Marangoni instability. J Fluid Mech 557:63–72
Article MATH Google Scholar
Thoroddsen ST, Etoh TG, Takehara K (2007) Experiments on bubble pinch-off. Phys Fluids 19(4):042101
Article Google Scholar
Tichigi Y, Hanzawa K, Kato Y, Kuroda R, Mutoh H, Hirose R, Tominaga H, Takubo K, Kondo Y, Sugawa S (2012) IEEE international solid-state circuits conference digest of technical papers, pp 382–384
Tilton J (1999) Fluid and particle dynamics. Perrys chemical engineers handbook, vol 6.1 50
Toschi F, Bodenschatz E (2009) Lagrangian properties of particles in turbulence. Ann Rev Fluid Mech 41(1):375–404
Article MathSciNet Google Scholar
van der Bos A, Zijlstra A, Gelderblom E, Versluis M (2011) iLIF: illumination by laser-induced fluorescence for single flash imaging on a nanoseconds timescale. Exp Fluids 51:1283–1289
Article Google Scholar
van Hoeve W, Gekle S, Snoeijer JH, Versluis M, Brenner MP, Lohse D (2010) Breakup of diminutive Rayleigh jets. Phys Fluids 22(12):122003
Article Google Scholar
van der Meer SM, Dollet B, Chin CT, Bouakaz A, Voormolen M, de Jong N, Versluis M, Lohse D (2007) Microbubble spectroscopy of ultrasound contrast agents. J Acoust Soc Am 121(1):648–656
Article Google Scholar
van der Veen RCA, Tran T, Lohse D, Sun C (2012) Direct measurements of air layer profiles under impacting droplets using high-speed color interferometry. Phys Rev E 85:026315
Article Google Scholar
van Wijngaarden L (2005) Bubble velocities induced by trailing vortices behind neighbours. J Fluid Mech 541:203–229
Article MathSciNet MATH Google Scholar
Veldhuis C, Biesheuvel A, van Wijngaarden L, Lohse D (2005) Motion and wake structure of spherical particles. Nonlinearity 18(1):C1–C8
Article MATH Google Scholar
Verhaagen B, Boutsioukis C, Heijnen GL, van der Sluis LWM, Versluis M (2012) Role of the confinement of a root canal on jet impingement during endodontic irrigation. Exp Fluids 53:1841–1853
Google Scholar
Versluis M, Georgiev N, Martinsson L, Aldén M, Kröll S (1997) 2-D absolute OH concentration profiles in atmospheric flames using planar LIF in a bi-directional laser beam configuration. Appl Phys B 65:411–417
Article Google Scholar
Versluis M, Schmitz B, von der Heydt A, Lohse D (2000) How snapping shrimp snap: through cavitating bubbles. Science 289:2114–2117
Google Scholar
Versluis M, Blom C, van der Meer D, van der Weele K, Lohse D (2006) Leaping shampoo and the stable Kaye effect. J Stat Mech 280:P07007
Article Google Scholar
Versluis M, Palanchon P, Goertz D, Heitman I, van der Meer S, Dollet B, de Jong N, Lohse D (2010) Microbubble shape oscillations excited through an ultrasound-driven parametric instability. Phys Rev E 82:026321
Article Google Scholar
Wamel AV, Kooiman K, Harteveld M, Emmer M, ten Cate FJ, Versluis M, Jong ND (2006) Vibrating microbubbles poking individual cells: drug transfer into cells via sonoporation. J Contr Rel 112(2):149–155
Article Google Scholar
Warden SJ (2003) A new direction for ultrasound therapy in sports medicine. Sports Med 33:95–107
Article Google Scholar
Wijshoff H (2010) The dynamics of the piezo inkjet printhead operation. Phys Rep 491(4–5):77–177
Article Google Scholar
Worthington AM (1908) A study of splashes. Longmans Green and Co, London
Google Scholar
Worthington AM, Cole RS (1897) Impact with a liquid surface studied by the aid of instantaneous photography. Phil Trans R Soc Lond Ser A 189:137–148
Article MATH Google Scholar
Worthington AM, Cole RS (1900) Impact with a liquid surface studied by the aid of instantaneous photography; Paper II. Phil Trans R Soc Lond Ser A 194:175–200
Article Google Scholar
Xu L, Zhang WW, Nagel SR (2005) Drop splashing on a dry smooth surface. Phys Rev Lett 94(1–4):184505
Google Scholar
Zhen X, Ludomirsky A, Eun LY, Hall TL, Tran BC, Fowlkes JB, Cain CA (2004) Controlled ultrasound tissue erosion. IEEE Trans Ultrason Ferroelec Freq Contr 51(6):726–736
Article Google Scholar
Zijlstra AG, Ohl CD (2008) On fiber optic probe hydrophone measurements in a cavitating liquid. J Acoust Soc Am 123(1):29–32
Article Google Scholar

Download references

Acknowledgments

This hands-on review on high-speed imaging has not been possible without the numerous and valuable contributions of my dear colleagues and friends in the various fields of research. I would therefore like to acknowledge contributions and support from Detlef Lohse and Nico de Jong. I also enjoyed many scientific discussions with Claus-Dieter Ohl, Chao Sun, and Goji Etoh. I also gratefully acknowledge the valuable and extensive discussions with my former PhD students Erik Gelderblom, Arjan van der Bos, Aaldert Zijlstra, Bram Verhaagen, Christian Veldhuis, and Wim van Hoeve. Many undergraduate students have contributed in the course of the Physics of Fluids lab course and Experimental Methods in Fluid Mechanics course, and in particular I would like to thank Tim Segers and Hans Kroes for contributions to Fig. 1c, Sander van der Meer and Ramon van den Berg to Fig. 3, Arthur van Bilsen and Raymond Bergmann to Fig. 8, and Rik Groenen, Elbert van Putten, Erik Gelderblom, and Ramy El-Dardiry to Fig. 15. Furthermore, I would like to thank Devaraj van der Meer, Chien Ting “Cash” Chin, Frits Mastik, Charles Lancée, Martijn Frijlink, Elmer Koene, Rik Vos, Manish Arora, Rory Dijkink, Marlies Overvelde, Jos de Jong, Valeria Garbin, Siggi Thoroddsen, Flordeliza Villanueva, Xucai Chen, Peter Andresen$^{\dagger},$ Theo van der Meer, Marcus Aldén, Clemens Kaminski, Edwin van der Bunt, Jan Tukker, Mike Bailey, Larry Crum, and Tom Matula, and Hans Reinten and Mark van den Berg of Océ Technologies and Paul Duineveld of Philips for their contributions to this review. I would like to acknowledge various technical discussions with Frans Langeweg of Kodak MASD/Roper Scientific/Redlake, Vance Parker, Sid Nebeker, and Nathan Nebeker of Cordin Company, Tim Nicholls of Photron, Kinko Tsuji, and Yasushi Kondo of Shimadzu Corporation, and John Boaler and Heiner Voges of LaVision. The technical skills of Henni Scholten$^{\dagger},$ Gert-Wim Bruggert, Bas Benschop, Martin Bos, and of Jan Honkoop, Leo Bekkering, Wim van Alphen, and Cees Pakvis of Erasmus MC are gratefully acknowledged.

Author information

Authors and Affiliations

Physics of Fluids Group, MESA+ Institute of Nanotechnology, MIRA Institute of Biomedical Technology and Technical Medicine, University of Twente, P.O. Box 217, 7500 AE, Enschede, The Netherlands
Michel Versluis

Authors

Michel Versluis
View author publications
You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Michel Versluis.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary material 1 (ZIP 52,473 kb)

Rights and permissions

Reprints and permissions

About this article

Cite this article

Versluis, M. High-speed imaging in fluids. Exp Fluids 54, 1458 (2013). https://doi.org/10.1007/s00348-013-1458-x

Download citation

Received: 19 July 2012
Revised: 19 November 2012
Accepted: 09 January 2013
Published: 14 February 2013
DOI: https://doi.org/10.1007/s00348-013-1458-x

Keywords

Use our pre-submission checklist

Avoid common mistakes on your manuscript.

High-speed imaging in fluids

Abstract

Similar content being viewed by others

Brandaris Ultra High-Speed Imaging Facility

High-Speed Imaging of Shock Waves and Their Flow Fields

Measurement in opaque flows: a review of measurement techniques for dispersed multiphase flows

1 Introduction

2 High-speed imaging basics

2.1 Definitions

2.2 Frame rate

2.3 Exposure time

2.4 Magnification

2.5 Scale and scaling issues

3 High-speed imaging systems and techniques

3.1 High-speed photography

3.2 Stroboscopic imaging

3.3 High-speed imaging

3.3.1 CCD and CMOS systems

3.3.2 Triggering

3.3.3 Data

3.3.4 The problem of throughput

3.4 Ultra-high-speed imaging systems beyond 1 Mfps

3.5 High-speed imaging: testing and calibration

4 Combined high-speed imaging techniques

4.1 High-speed shadowgraphy and Schlieren imaging

4.2 High-speed holography and interferometry

4.3 High-speed particle tracking velocimetry

4.4 High-speed particle imaging velocimetry

4.5 High-speed fluorescence imaging

5 Summary

References

Acknowledgments

Author information

Authors and Affiliations

Corresponding author

Electronic supplementary material

Supplementary material 1 (ZIP 52,473 kb)

Rights and permissions

About this article

Cite this article

Share this article

Keywords

Search

Navigation