Measuring the Effectiveness of Three Dimensional Deterministic Lateral Displacement in Low Viscosity Liquids Brandon Gomes [email protected] Sophie Slutzky [email protected] Kaley Ubellacker [email protected] Jackson Weaver [email protected] Sharon Zhang [email protected] New Jersey Governor’s School of Engineering and Technology 2016 July 21, 2016 Abstract by analyzing particle motion in a different fluid, namely, water. Analyzing disparities between the data sets revealed the effects on particle motion through a fluid with a high Reynolds number, as well as how the separation mechanics of the passing fluid were altered according to 3D DLD. With more research, three dimensional deterministic lateral displacement (3D DLD) has the potential to greatly improve the efficacy and flexibility of microfluidic devices for separation within various fields of science, notably microbiology. As a particle passes through a set array of obstacles submerged in liquid, its type of motion varies according to its specific properties. Utilizing the differing movements, 3D DLD can be used to separate various particles according to size, deformability, and other properties. The project focuses on testing the competence of 3D DLD systems in a low viscosity liquid. Similar to the data collecting and data processing procedure used in a previous 3D DLD experiment utilizing corn oil, this was accomplished Analysis of the experimental data revealed important notes regarding particle sorting in lower viscosity fluids, many of which directly correlate with similar corn oil results. Studies in two planes of motion revealed distinct points for particle separation regardless of the type of motion, and further comparisons of forcing, critical, and migration angles demonstrated sophisticated relationships that are especially useful in particle separation. A prominent issue with the trials included the formation of rust, requiring the addition of vinegar 1 to counter and forestall the effects. Ultimately, the research supported the increased efficiency and accuracy of 3D DLD mechanisms but demonstrated that water is much less consistent in comparison to corn oil as a separating fluid. 1 1.1 lated to a microscopic set-up to function in the same way based on larger scale research. Another goal of this experiment is to compare the results of 3D DLD when used with water, a low viscosity liquid, with the results obtained using corn oil, a much more viscous liquid. Due to water’s low viscosity, the system has a higher Reynolds number, and the driving force, specifically gravity, has a much more significant effect on the movement of the particles. Introduction Goals The study focuses on the effectiveness of a 3D DLD system in separating particles based on size. 3D DLD is an expansion of two dimensional deterministic lateral displacement (2D DLD), a separation method used in microfluidics. 2D DLD is often used in binary separation, where it creates two different streams of distinct particles, often separating them by size, shape, texture, or material. However, because of the planar motion of particles within a 2D DLD setup, separating more than two different types of particles in a single fluid is extremely difficult and inaccurate. The primary purpose of the experiment is to overcome the limitations of 2D DLD by adding a third range of motion for the particles to move through. In order to accomplish the task, particles of three different sizes are sent through a 3D DLD system, and their relative pathways are recorded to determine the three distinct streams of particles. The 3D DLD system consists of an obstacle array with long cylindrical rods placed in a rectangular grid. The experiment is conducted in a macroscopic set-up, which allows for the manipulation of both the slope angle and the orientation of the array. This directly relates to a main ambition of microfluidics as a whole to create a feasible macroscopic model which can be trans- 1.2 Reasons for Development of 3D DLD A major motivation for the development of 3D DLD is that it can easily be applied to the purification of environmental and biological fluids [1]. One specific application is separating healthy and diseased blood cells, which is possible due to blood’s low viscosity and ability to move through microchannels without clogging [1]. Another advantage to 3D DLD is that it has the ability to separate many varieties of particles at a time and to replace the more complex and expensive separation methods that are currently in use. As a result of its simple design, 3D DLD has the potential to separate particles in a large variety of liquids, regardless of their composition and viscosity. 2 2.1 Background Microfluidics The principles of 3D DLD are founded on microfluidics, which focuses on the manipulation of fluids typically in the range of microliters to picoliters through channel networks [2]. Microfluidics seeks to control fluid flow on the microscale to provide 2 several advantages, such as increased range of capabilities, less material requirements, and more manageable compact units. Microfluidics additionally allows for detection with higher resolution and sensitivity, as well as improved cost efficiency. Exploiting smaller scale management and established characteristics of fluids, microfluidic mechanisms offer increased control and overall effectiveness. One of the characteristics of fluids vital to the principles of microfluidics and particularly relevant to this research is laminar flow. In laminar flow, fluids flow in parallel, and the only mixing that occurs results from the diffusion of molecules across the interface [2]. All in all, the networks of channels capitalizing on the varying properties of liquids and particles in microfluidics are a key component of separation during 3D DLD. 2.2 inertial microfluidic techniques, in addition to hydrodynamic filtration and pinched flow fractionation. Hydrodynamic filtration withdraws a small amount of liquid from a mainstream to align particles with side branch channels [3], and pinched flow fractionation introduces particles to a pinched segment where they are isolated with a spreading flow profile [4]. 3D DLD draws on the results of the various methods to improve upon prior techniques and optimize efficiency and accuracy in separation. 2.3 Two Dimensional Deterministic Lateral Displacement The most basic form of deterministic lateral displacement implements two ranges of motion along the x-axis and y-axis. The main focus of the separation technique is to create equivalent migration paths using laminar flow to process particles through an array of obstacles. A set of obstacles is arranged on one plane, and each successive row of obstacles is shifted horizontally by a steady factor: ∆λ , where λ is the center-tocenter distance between obstacles [5]. Ideally, in 2D DLD, particles experience a consistent routine of motion as they progress through the gaps between obstacles. The routine of motion, developed by each specific particle as it passes through the array, varies depending on individual properties, thereby contributing a key component to particle separation utilizing 2D DLD. The ultimate goal is to understand various particle interactions with the array and take advantage of them within the context of characteristic separation in order to improve the functionality of that separation. Compared to other sorting methods, 2D DLD offers various crucial advantages. Unlike other systems, it does not depend on Methods of Particle Separation Separation in microfluidics utilizes two different types of methods: active and passive. The sole difference between active methods and passive methods is that active methods involve the use of external fields to drive the separative displacement, while passive methods do not. Active methods include dielectrophoresis, magnetophoresis, and acoustophoresis, implementing electric charges, magnetic fields, and ultrasound waves, respectively, to separate particles. While active methods hone in on flow field fractionation, passive methods incorporate hydrodynamics and particle-solid interactions based on inertial motion as a means to separate particles within the fluidic system [1]. Deterministic lateral displacement serves as one passive method among other 3 diffusion or multipath averaging, eliminating the randomness and presenting a highly deterministic event. Additionally, DLD is much less susceptible to clogging, which is especially useful in biological settings and allows for continuous particle separation. However, the efficiency of the deterministic lateral displacement method can still be further maximized. One of the areas for improvement is precision. Although 2D DLD is far more exact compared to other separation techniques, there still remains a fair margin of error. Random diffusion across streamlines continues to moderately degrade resolution. Although high flow speeds would minimize the issue, separation in a 2D DLD system functions best at low flow speeds. Furthermore, the systems are not immune to jamming and cannot support non spherical particles without utilizing complicated post structures. There is much room for deterministic lateral displacement methods to be further researched, consolidated, and ultimately finalized. 2.4 Therefore, the combination of the two angles would allow for three separate critical angles to be found for three different particles. When a certain size particle’s critical angle is below the forcing angle, it will experience locked motion. If the other two particles have greater critical angles than the same forcing angle, the three dimensional geometry allows for the two other particles to be separated by their out-of-plane motion in the xz-plane, allowing for more specificity than 2D DLD [1]. 2.5 Collision Interactions The collision model assumes two different types of interactions with the posts: purely hydrodynamic or physical contact. Purely hydrodynamic interactions describe interactions in which the displacement going into the collision is equivalent to the displacement out. For the touching collision, the two displacements will be asymmetrical, with the position at which the particle exists the apparatus being set at one specific value [7]. For each given interaction, the particle trajectory is considered either reversible or irreversible. A particle’s trajectory is considered reversible in scenarios where the obstacle does not affect the particle’s intended motion. However, for each irreversible collision, there exists a critical offset which differs from the initial offset prior to interaction with an obstacle, giving rise to a net lateral displacement necessary for particle separation [7]. Overall Concept The deterministic lateral displacement is formed using cylindrical obstacles to affect the in-plane and out-of-plane motion of particles. The motion is altered by two angles: a rotation angle and the slope angle. Therefore, the gravitational force, which drives each particle through the system, becomes a three dimensional vector based on the two angles. The particles’ interactions with the rods are calculated in threedimensions due to the gravitational force, thus creating 3D DLD [6]. For 3D DLD, the forcing angle α, defined as the angle between the particle path vector and the z-axis in the yz-plane, is based on both the rotation and slope angle. 2.6 Types of Motion The particles exhibit two types of motion as they travel through the rods: zigzag and locked (see Figure 1). The two forms of motion allow for the separation of parti4 cles, as they create two different streams and points of exit from the cylindrical array. work of Dr. German Drazer and Siqi Du, as the system uses water as the liquid that the particle flows through, rather than corn oil. Corn oil has a much higher density and therefore a much lower Reynolds number. The Reynolds number is a measurement of turbulence within a system, and plays a significant role in the inertia of the particles. The Reynolds number also plays a role in the Navier-Stokes Equations, which describes the motion of a particle in a fluid. The Reynolds number measures the proportions of the viscous forces to inertial forces acting on a particle traversing the system. It helps to indicate above which values the particle will experience turbulent flow and below which it will experience laminar flow through the system. With water having a lower density and higher Reynolds number, a higher degree of chaos is expected, which may cause the particle to act differently within the system from the system with corn oil. The force angles may have different effects on the particles because there will be a different critical velocity value for the particle at which it switches types of movement through the system. Therefore, the critical angle of each particle will differ based on the fluid it flows through [6]. Figure 1: Adapted from [1]. Particle motion projected onto the yz-plane shows the migration angle β and the forcing angle α. The red particle moves in locked mode and is confined to a single column, whereas the blue particle moves in zigzag mode and travels across multiple columns. Both models of motion are made possible by the phenomenon called directional locking, where if the force angle is less than a specific value, known as the critical angle, the particles will follow the array of poles, creating a diagonal path. Directional locking is due to the touching force, where all particles will exit the collision at the exact same point, creating a specific pathway, which they will follow. When the angle is greater than or equal to the specific critical angle, the particles will bounce around and follow a zigzag pattern similar to a pinball machine [7]. 2.7 3 3.1 Separation Using Three Dimensional Deterministic Lateral Displacement Apparatus Preparation First, the apparatus that would be used in the 3D DLD system was assembled. Stainless steel poles, six feet in length, were cut into 15 centimeter rods, which would act as the cylindrical obstacles in the array. Theory For the experiment specifically, different results are expected from the previous 5 Two acrylic plates (see Figure 2a) designed in Solidworksr were constructed, each with a 20 x 20 grid of holes. The holes were spaced 4 mm apart, each with a diameter of 2 mm. A total of 200 cut rods were then inserted into both acrylic plates in the central 10 columns of the array, creating a 10 x 20 grid of rods. The array plates were then drilled into place, at a separation of 14 cm, into a square acrylic plate (see Figure 2b). Another acrylic plate was attached to the largest base plate by hinges, allowing for the array to be lifted at various slope angles. The square plate and base plate were loosely screwed together to allow for the rotation angle to be adjusted. A large rectangular acrylic plate was used as a base plate, and its surface was marked with angles in 10◦ increments, ranging from 0◦ to 90◦ . Another acrylic plate was attached to the base plate by hinges, allowing for the entire array to be lifted at various slope angles while still maintaining a flat bottom surface. The square plate with the array mounted on it and the base plate were loosely screwed together to allow for the rotation angle to be easily adjusted using manual force (see Figure 2b). A small hole was drilled into both of the unhinged corners of the base plate, and floss was threaded through and tied to a long metal bar. To adjust the slope angle, the floss was either wound, increasing the angle, or unwound, decreasing the angle. Then, a tank with a total volume of one cubic foot was filled to a height of eight inches with tap water, and the completed apparatus was entirely submerged in the liquid. The floss and rod were placed outside the tank, and the floss was made taut, acting as a type of pulley. In order to adjust the slope angle, the floss was either wound, increasing the angle, or unwound, decreasing the angle. (a) (b) Figure 2: (a) The two acrylic plates holding the stainless steel rods were designed using Solidworksr . (b) The apparatus from the view of the camera. The slope angle φ is measured along the center of the bottom acrylic plate. 6 Figure 3: The assembled rod array and the two other acrylic plates. The plate in the top left is marked with angles in increments of 10◦ . 3.2 (a) Rust Removal and Prevention After submerging the apparatus in the water, oxygen bubbles formed on the steel rods of the array, revealing the presence of an oxidation reaction (see Figure 4c). The oxidation as a result of the water’s reaction with the steel thus produced rust, a significant issue to the function of the apparatus as a whole. Due to the change in texture of the rods, particles became attached to the steel rods upon impact rather than passing through the array, dramatically varying the results of the trials. The presence of rust clearly affected the roughness of the steel rods, thereby altering the path of particles and leading to inaccuracy. In order to remove the rust from the steel rods, the entire apparatus was placed in a vinegar-water solution and cleaned with gauze. To prevent rust from appearing again, the water in the tank was replaced with a vinegar-water solution consisting of 21.48 L water and 100 mL vinegar, approximately a 215:1 water to vinegar ratio. The addition of vinegar lowered the slightly basic pH of the water into a neutral range, eliminating the potential for oxidation reac- (b) (c) Figure 4: (a) The apparatus submerged in water in the fish tank. A stand was used to make maintain balance of the tank. (b) Washing the apparatus in a vinegar-water bath to clean the rust off of the steel rods. (c) Oxidation of the steel rods was detectable by the bubbles forming around the rods. 7 tions without drastically impacting the viscosity. Additionally, the vinegar-water solution was monitored closely and changed daily in an attempt to maintain the cleanliness and quality of the liquid. 3.3 ment within the array and their point of exit was captured by the camera to allow for further review and analysis. The type of particle movement, zigzag or locked, was also observed in order to determine the value of the critical angle for each particle size at the given slope angle. At the smaller rotation angles, the vast majority of particles exhibited the properties of zigzag motion. The later transition from zigzag to locked motion signaled that the angle of displacement of the particles was approaching the value of the critical angle. Therefore, as the rotation angle was increased in increments of 10◦ , the particles passed through the intermediary state, where they switched from zigzag to locked motion. During the intermediary states, an additional trial was conducted at a 5◦ increase in order to find a more accurate value for the critical angle and locate a more exact point at which the motion of the particles underwent the transition. Once the particles of a certain size had sufficiently transitioned into locked motion, with the vast majority of the trial particles following the path along the rods, no more trials were conducted for that specific size and slope angle. Locked motion indicated the critical angle had already been achieved, meaning additional trials at larger rotational angles would have yielded the same result. Completion of Trials The testing process for 3D DLD involved manipulating three different variables: the particle size, the rotation angle, and the slope angle. A total of 44 trials were completed, consisting of three different sizes of particles, 22 rotation angles, and two slope angles. The particle diameters varied between 1.59◦ , 2.88◦ , and 3.16◦ , with Particle 1 being the smallest and Particle 3 being the largest. In each trial, 20 spherical Nylon particles of the predetermined size were inserted into the first column to the right of the array, and their motion through the apparatus was recorded. First, the mechanism constructed from dental floss attached to the base of the apparatus was rewound to the appropriate slope angle. The trials began by testing a constant slope angle of approximately 28◦ in conjunction with the manipulation of the other two variables. Afterwards, another series of trials mimicked the first set but instead used a slope angle of approximately 37◦ . The slope angles were deliberately chosen at an assumed safe range above the particles’ hypothesized critical angles. For each given slope angle, the rotation angle was adjusted beginning with 10◦ followed by steadily increasing 10◦ intervals until the last trial with a rotation angle of 80◦ . Before each session, a camera was set to record the trial. After visually specifying which particle size was being tested, the 20 particles were individually dropped into the array using tweezers. The particles’ move- 3.4 Review and Analysis Process Once data collection was complete, the analysis was split into two sections. The first part consisted of motion analysis in two dimensions, specifically within the in-plane (yz) motion. The forcing angle can be determined in terms of the components of gravity on the xand y-axes (see Figure) as follows [6]: 8 where C is the number of columns that the particle traversed and R is the number of rows that the particle traversed. The second part of analysis consisted of video analysis of particle motion in the xy-plane. Each trial was videotaped individually, with a total of twenty particles per trial. The camera was positioned parallel to the top of the apparatus so that the video could later be analyzed digitally in order to determine the path of the particles. Furthermore, the distance traveled by the particles in the x- and z-directions allowed for additional separation of the particles which were indistinguishable in the xz-plane analysis. Comparing the out-of-plane motion was accomplished by projecting the particle path vector in the xy-plane onto the xaxis and taking the magnitude of that vector. Videos were filtered after compilation from the camera, and a program written in MATLAB was used to generate frames from each video. Every fifth frame was saved as an image, and all saved images from the same video were saved into a folder with a unique video ID. The images were then further filtered so that the final image sequences consisted of only the frames showing the particles entering and exiting the grid. The data was then collected by manually analyzing the images in an open source image processing software, ImageJ, using manual tracking. The manual tracking process for each video sequence recorded the path vector of each particle, as well as the vector in the x-direction along the steel rods. In order to convert the projection length calculated by the ImageJ tracking from pixels to millimeters, another standard vector spanning the distance between four steel rods was recorded for each video. The length of this vector, which was known to be 18 mm, was then used to scale the projection length Figure 5: Adapted from [1]. The components of the gravitational force in the axis defined by the apparatus, as well as the slope angle θ and rotation angle φ . gz = g cos θ gx = g sin θ cos φ gy = g sin θ sin φ (1a) (1b) (1c) gy /gz = tan α = tan θ sin φ (2) The entrance and exit points of each particle in the array were recorded. Compilation of the row and column exit data enabled estimation of the critical angle for initial separation of particles into two streams, one containing a particle of one size and the other containing particles of the two remaining sizes. Three different critical angles were calculated for each slope angle, one corresponding to each specific particle size. The exit point was also used to find the migration angle of each particle, which was calculated by: β = arctan C R (3) 9 10◦ increments, the forcing angle decreased. Once the forcing angle approached a certain range bounding the critical angle, the particles’ tendencies to travel in locked mode increased abruptly. into millimeters. The particle’s displacement along the x-axis could then be calculated by the projection equation: |projb x| = x·b |x| 18 mm |s| (4) 4.2 where x represents the particle’s path vector, b represents a directional vector along the steel rods, and s represents the standard vector. 4 4.1 Separation Streams in 2D In order to determine the critical angle, a range was first pinpointed from the experimental data by calculating the probability of crossing (Pc ), defined as the ratio of the number of particles in zigzag motion to the total number of particles in a trial[1] [1]. The data showed clear transitions from a very large probability of crossing (Pc = 1) to a seemingly unapparent probability of crossing zero (Pc = 0) over two or three consecutive rotation angles for each particle size under a fixed slope angle. Results and Discussion Tracking Particle Motion Through Steel Rods The out-of-plane motion of each particle was recorded by the row or column which the particle exited from at the end of each run, and the motion in the y-direction was tracked by the video recording. Within the steel rod grid, two modes of particle motion were evident. Initially, the particle moved in zigzag motion, in which it followed the direction of the gravitational force pulling down on it, thus following a generally downwards path that spanned multiple columns and crossed a large portion of the array. The particles usually exited on the side of the array or in one of the leftmost columns at the bottom of the array. Particles may also travel in locked motion, following the direction of the lattice rather than that of gravity. Particles in locked motion moved through all nineteen rows of lattice but stayed strictly within the first column. The critical angle was the angle at which particles of a certain size switched from zigzag mode to locked mode and was the primary method of separation in two dimensions. As the rotation angle increased in Figure 6: The probability of crossing Pc for each particle size in different combinations of slope angles and rotation angles. Trials highlighted in green indicate when the particle is traveling in or close to locked mode, and trials highlighted in yellow indicate when the particle is traveling in zigzag mode. The trial highlighted in red indicates a potentially faulty result. Experimentally determined ranges for the critical angle α of a certain particle size can be found by taking the corresponding forcing angles of the trials where Pc most closely bound 0.5. For a slope angle of 28◦ and particles of diameter 1.59 mm, 2.38 mm, and 3.16 mm the respective ranges for α are: 10 respectively: 10.3◦ < α < 14.9◦ 18.9◦ < α < 21.3◦ ◦ 21.3 < α < 26.7 Particle Size 1.59 mm 2.38 mm 3.16 mm ◦ For a slope angle of 37◦ and particles of diameter 1.59 mm, 2.38 mm, and 3.16 mm the respective ranges for α are: ◦ 7.5 < α < 14.5 Critical Angle in Water 11.6◦ ± 2.3◦ 19.8◦ ± 1.6◦ 25.4◦ ± 2.1◦ The estimated critical angles for the same sizes of particles in corn oil are respectively [1]: ◦ 14.5◦ < α < 17.7◦ 20.6◦ < α < 30.0◦ Particle Size 1.59 mm 2.38 mm 3.16 mm Critical Angle in Corn Oil 6.7◦ ± 1.7◦ 10.0◦ ± 1.5◦ 12.6◦ ± 1.7◦ Figure 7: The probability of crossing Pc against the forcing angle α. Different particle sizes are marked by different colors and shapes. Plotting Pc against forcing angles and isolating the transition stage trials produces a linear regression. The regression enables the estimation of the critical angle, determined by the value of the forcing angle when the probability of crossing is 0.5 The graph also shows a distinct jump in Pc , indicating that the transition between modes occurs at a specific range of angles for each particle. The graph also shows that the critical angle tends to increase as particle size increases, which is consistent with the previous experiment using corn oil [1]. The estimated critical angle for particles of 1.59 mm, 2.38 mm, and 3.16 mm are Figure 8: The migration angle β as a function of the forcing angle α for all three particle sizes: 1.59 mm, 2.38 mm, and 3.16 mm. Particle motion may also be studied by plotting the migration angle β against the forcing angle α, which shows that the particles of each size begin to move in zigzag motion only after the respective critical angle, which is consistent with the particle motion exhibited in oil [1]. With a few exceptions, particles traveling at forcing angles less than the critical angle exhibit locked motion, with β = 0◦ . 11 4.3 Separating Particles in Three Dimensions Figure 9 shows that in the xy-plane, there existed critical angles in which one particle size was split into a separate stream but the remaining two particle sizes still experienced zigzag motion, highlighting the ability of the experiment to choose specific particles to separate into distinct streams. While the in-plane motion of the particles could not have been distinguished, Figure 10 shows that their out-of-plane movements were clearly distinct. For instance, at a forcing angle of 22◦ , particles of diameter 3.16 mm would be separated into a zigzag stream, but particles of diameter 1.59 mm and 2.38 mm would both remain in the same locked stream in the 2D DLD plane. However, Fig. depicts a clear difference in ∆x/∆z between particles of diameter 1.59 mm and 2.38 mm, thus splitting them into separate streams. Further separation of the particles in the third dimension is done in two planes. First, the projection of the particle motion vector in the xy-plane onto the x-axis is determined. The projection, or the particle’s movement along the x-axis (∆x), can be normalized along the particle displacement in the z-axis (∆z) to obtain ∆x/∆z. The ratio is effectively describing the particle’s movement in the xz-plane. Plotting ∆x/∆z as a function of the forcing angle α allows for further separation of particles locked within the same column, a motion that would be indistinguishable in 2D DLD. 4.4 Comparing Particle Motion in Water to Particle Motion in Corn Oil The critical angles for water are higher for each particle size, which was expected since the Reynolds number for water is significantly higher than that of oil. The difference in critical angle indicates that the gravitational force pulling on the particles met less inertial resistance in water than in oil, and the greater perturbations within the water allow the particles to remain in zigzag motion at greater forcing angles. Conversely, the greater inertial forces present in oil, which has a higher viscosity, minimize the perturbations caused by the force of gravity moving the particle and locking the particle within the steel rod lattice more easily. (a) (b) Figure 9: The xz-plane motion graphed as a function of the forcing angle α for slope angles of (a) 28◦ , and (b) 37◦ . 12 (a) (b) (c) Figure 10: The migration angle β as a function of the forcing angle α for particles of diameter (a) 1.59 mm (b) 2.38 mm (c) 3.16 mm. Noticeable plateaus in migration angle are observed for certain intervals of the forcing angle. 13 While the critical angles were greater for particles traveling through water than in oil, the general separation mechanism of 3D DLD functioned similarly in both fluids. In particular, an individual analysis of the migration angle β against the forcing angle α of each particle shows the directional locking behavior, defined as the occurrence of a fixed migration angle for a certain interval of forcing angles [1]. Similar to oil, the directional locking intervals decrease in number but increase in range as particle size increases. The definition of the plateaus, however, is much clearer in the data for oil than that of water. Additionally, it is evident that particles have migration angles of approximately 0◦ at forcing angles less than the critical angle, which suggests that the turbulent nature of water does not significantly alter the separation mechanism from that in oil. 4.5 changed every day that new trials were completed. As a result, each new day of trials may have had a slightly different viscosity liquid to travel through. Another potential source of error is the initial dropping of the particles into the array. Each particle was placed into the apparatus using tweezers, and as a result, they did not all enter with the same initial force or velocity. Small changes in how the tweezers were held or the location of the tweezers could have affected the overall motion of the particles as well. Human error is also present in the analysis of the particles using ImageJ. Due to the manual tracking process that was employed, some of the motion vectors may have been inaccurate. Some of the trial video recordings were not completely in focus, and a small speck, air bubble, or reflection in the tank could have been mistaken for the particle, leading to an incorrect reading. Sources of Error During the analysis of the collected data, discrepancies were found in select trials concerning the type of movement and the migration angle of particles. The error could be attributed to multiple possible sources of error. First, due to the original rusting of the steel rods in the array, a rice vinegar-water solution was composed to clean the array and prevent future rust. The vinegar present in the solution had a slightly higher viscosity than water, which could have affected the motion of the particles. In addition, the rust may have adjusted the texture of the rods. Such textural change would also have had an effect on the movement of the particles, as they could have stuck to the rods or changed their trajectory due to increased friction. The vinegar solution also posed a possible problem because the liquid in the tank needed to be 5 Conclusions As particles were placed into the array and passed through the series of obstacles, two types of motion were observed, zigzag and locked, which enable particle separation. Generally, particles demonstrated more inclination towards locked motion as the forcing angle approached the critical angle, often with greater rotation angles. The experimental data supported th, revealing that the probability of crossing decreased suddenly from one to zero over the span of the trials. After calculating critical angles for given particle sizes according to their forcing angles and plotting the probability of crossing against forcing angles isolated from the transition stage, a linear regression was observed. This further distin14 guished that the critical angle typically increases as particle size increases, strongly suggesting that water serves as a successful separation medium. Aside from the comparison of forcing and critical angles, the relationship between forcing and migration angles also revealed important notes concerning the data sets, namely, that particles tended to lock at forcing angles less than the critical angles. In general, particles moving according to forcing angles less than the critical angle exhibited locked motion. ing particles were blurry. Acknowledgments This paper and knowledge acquired through research would not have been possible without the extensive time and effort contributed by a great account of people and institutions. Specifically, the authors of this paper would like to express thorough gratitude to all their mentors. The special efforts of mentor and PhD candidate Siqi Du were invaluable to the experimental process and deep analysis required for success, and the authors would like to further acknowledge with great appreciation his considerable help, support, and enthusiasm in all project matters. Additionally, they extend their thanks to Residential Teaching Assistant Alissa Persad for her facilitation in review and overall focus. Furthermore, the authors express the utmost gratitude to Dr. German Drazer for allowing the use of his facilities, without which this opportunity would not be possible. They would also like to extend thanks to all the staff of the Governor’s School of Engineering and Technology for not only the materials and basis creating thie research project, but also the remarkable experience as a whole. Their hard work and detailed planning provided the foundation for all the studies described within the paper. Particularly, the authors would like to express deep appreciation to Dean Jean Patrick Antoine and Dean Ilene Rosen, without whom the research would not have been pursued. Finally, they respectfully acknowledge the generous donations of the sponsors of the Governor’s School of Engineering, enabling the pursuit of interests in science, mathematics, engineering, and Furthermore, utilizing the three dimensional coordinate system enabled greater separation capabilities in a given column, successfully expanding the range of motion and accuracy in comparison to 2D DLD. While there still remains areas for improvement in sorting particles according to DLD, such as in cases where two particles experience the same type of motion, 3D DLD permits analysis in two planes of separation with distinct displacements. Additionally, appealing to one of the primary goals of the research to identify discontinuities between oil and water used in 3D DLD, less inertial resistance was met in water due to lower viscosity provided for higher critical angles, also allowing zigzag motion in greater forcing angles. However, the difference in inertial resistance did not have a significant impact on the mechanism as a whole, as proven within the overall directional locking behavior of the particles. Even so, oil exhibited more consistent separation, especially when separation involved smaller particle sizes. Possible sources of error include material malfunctions, particularly viscosity and trial reset discrepancies regarding the water-vinegar solution and steel rod rusting, precision in original particle placement within the array, and human error concerning analysis, especially when videos track15 technology: Rutgers, the State University of New Jersey; Rutgers School of Engineering; South Jersey Industries, Inc.; Printrbot; and Lockheed Martin. [4] M. Yamada et al., “Pinched Flow Fractionation: Continuous Size Separation of Particles Utilizing a Laminar Flow Profile in a Pinched Microchannel,” Anal. Chem., vol. 76, pp. 5465-5471, Sept. 2004. References [5] L. R. Huang et al., “Continuous Particle Separation Through Deterministic Lateral Displacement,” Science, vol. 304, pp. 987-990, Dec. 2003. [1] S. Du and G. Drazer, “Deterministic separation of suspended particles in a reconfigurable array,” J. Micromech. Microeng., vol. 25, pp. 1-8, Oct. Month, 2015. [6] S. Du and G. Drazer, “Gravity driven deterministic lateral displacement for suspended particles in a 3D obstacle array,” unpublished. [2] G. M. Whitesides, “The origins and future of microfluidics,” NATURE, vol. 442, no. x, pp. 368-373, July 2006. [7] T. J. Bowman et al., “Inertia and scaling in deterministic lateral displacement,” Biomicrofluidics, vol. 7, pp. 114, Dec. 2013. [3] M. Yamada and M. Seki, “Hydrodynamic filtration for on-chip particle concentration and classification using microfluidics,” Lab on a Chip, vol. 5, pp. 1233-1239, Sept. 2005. 16
© Copyright 2026 Paperzz