ANALYSIS OF EJECTOR-STYLE MICROBUBBLE GENERATORS: MASS-TRANSFER PROPERTIES, MATHEMATICAL MODELING, AND DESIGN ALGORITHM By Ziwei Wang A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Chemical Engineering – Doctor of Philosophy 2022 ABSTRACT Gas-to-liquid mass transfer is a rate-limiting step for many commercial-scale operations in the chemical, biochemical, pharmaceutical, and wastewater-treatment industries. The use of microbubbles with a diameter on the100 µm scale has been shown to provide high volumetric mass-transfer rates due to its high gas contact area per volume. However, the use of microbubbles in commercial processes has been hampered by the lack of design algorithms with which to fabricate high-performance, microbubble-sparged gas-liquid contacting equipment. The goals of this study were to identify the type of microbubble generator best suited to provide high volumetric mass transfer rates in commercial-scale equipment, characterize the mass-transfer properties, develop models able to predict the mass-transfer rate as a function of the key independent variables, and use the models to develop a design algorithm suitable to use microbubble sparging in industrial processes. The study began with a literature review of microbubble generators that considered factors including the mechanism, safety, cost, and scalability, with the goal of identifying generators suited to cost-effectively provide extremely high mass transfer in commercial-scale equipment. Microbubble generators that used liquid turbulence were found to have the best combination of properties for such applications. In collaboration with the Michigan Biotechnology Institute, a 300-L bioreactor was customized for use with either a RiverForest microbubble ejector and a conventional ring sparger. E.coli batch growth experiments were conducted to compare the growth rates using the two aeration methods. The E.coli growth rate observed during microbubble aeration was about twice that observed with the traditional ring sparger. Mathematical models describing the performance properties of both a microbubble ejector and a Modified Jameson Cell were developed. The models included energy requirements, mass transfer rates, gas and liquid flow patterns, and clearance of spent bubbles. The models predicted that the ejector would be more energy-efficient for applications requiring higher mass-transfer rates and lower gas volume fractions, whereas the Modified Jameson Cell would be more energy-efficient for applications requiring lower mass-transfer rates and higher gas void fractions. A novel flow system was developed to measure the mass-transfer rate of microbubble produced by an ejector generator. A mathematical model was developed to reproduce experimental trends and estimate the effective microbubble diameter generated as a function of the gas and liquid flow rates. New axial mixing and two-phase friction factor correlations were developed for the model fidelity. The results were used to develop a correlation to predict the effective microbubble size as a function of system properties. The predictive power of this correlation has utility for industrial process design and scale-up applications. The friction factor and microbubble diameter correlations developed in this study were used to develop additional models to simulate the microbubble mass-transfer in large reactors that are sparged with arrays of microbubble ejectors. Collectively, the models developed in this study provide powerful new design tools that enable rational design, optimization, and scale-up of ejector microbubble sparger arrays for commercial-scale reactors that require high mass-transfer rates. ACKNOWLEDGEMENTS I would like to thank Dr. Worden for his guidance on this project. None of these would be possible without his effort and input. I would like to thank Dr. Saffron, Dr. Jaberi, and Dr. Lira for their guidance and supports. I also would like to thank Michigan Biotechnology Institute’s help on the E. coli experiment. AgBioResearch provided funding for this research. I am also lucky that Dr. Alexandra Zevalkink and her students are in the same office with me. Thank you, Alex, Sev, Bonnie, Ashiq, Towhid, Eleonora, Ashwini, Pengpeng, Gill, Mack, Mario, Corey, David, Megan, Dr. Aahrooo, Dr. Cody… You amazing people made my PhD journey so much more enjoyable. Also, I would like to thank Bill and Matt. It was my pleasure to overcome all these hardships with you two gentlemen. I also would like to thank Neda and Iman for their help and encouragement during my study here. Also, thank you Rachel for being here for me. But most importantly, I would like to thank my family for their support, love, encouragement, and confidence in me that motivated me to overcome the difficulties I faced during my study here. My parents, Yanping and Yingjuan, came from a humble background and they overcame enormous hardship to raise me. My grandparents, Zutong, Jingxian, Tongqin, Yimei, raised me till I was old enough for school when my parents were busy starting up their careers to support me. Their cumulative efforts made me what I am today. My paternal grandmother, Jingxian, passed away one month before my defense. I wish to let her know how much I love her and I want to thank her for all her love and care for me throughout all the years. iv TABLE OF CONTENTS LIST OF SYMBOLS ................................................................................................................ vi LIST OF ABBREVIATIONS ..................................................................................................xii CHAPTER ONE: LITERATURE REVIEW ............................................................................. 1 CHAPTER TWO: MBG CAPABILITY USING E. COLI GROWTH AS INDICATOR ....... 27 CHAPTER THREE: THEORETICAL COMPARISON OF MBG COST-EFFICIENCY BETWEEN EJECTOR (MBE) AND MODIFIED JAMESON CELL (MJC) ........................ 41 CHAPTER FOUR: BUBBLE SIZE PREDICTION FOR MICROBUBBLE EJECTOR ....... 58 CHAPTER FIVE: MASS-TRANSFER COEFFICIENT ESTIMATIONS FOR THE EXPERIMENT FLUID AND FRESH RO WATER .............................................................. 117 CHAPTER SIX: LARGE-SCALE BIOREACTOR SIMULATION USING MBES ........... 123 CONCLUSION ...................................................................................................................... 144 FUTURE WORKS................................................................................................................. 147 BIBLIOGRAPHY .................................................................................................................. 148 v LIST OF SYMBOLS A: absorbance a: interfacial area per unit reactor volume apipe: cross-sectional area of a pipe C: bulk concentration of the transferred gas dissolved in the liquid C*: equilibrium solubilities C0: constants correlated to the perforated plate geometry CC: constants correlated to the perforated plate geometry Cero: pipe erosion velocity constant Cf: Chen’s arbitrary constants cdye: dye concentration D: pipe diameter Dbub: bubble diameter D̃ b,m: dimensionless bubble diameter inside the MJC Dpore: pore diameter dmax: maximum stable bubble diameter dnoz: ejector nozzle diameter Ediss: dissipation energy Esurf: surface free energy per unit area, Eu: pressure loss coefficient vi Frg: Froude Number f: friction factor ffan: Fanning friction factor g: gravitational constant gc: conversion factor H̃ : dimensionless height inside the column Hpump: pump head Htank: vessel height h: axial position inside jet. hf: total fitting head-loss hjet: jet height value hpipe: friction energy head loss kfit: fitting loss coefficient kl: liquid-film mass transfer coefficient L: ejector length Ltube: tube length l: light path distance Nbub: total number of bubbles m: slope mg: gas molar flow rate vii nf: Chen’s arbitrary constants OUR: Oxygen Uptake Rate P: Power consumption Patm: atmospheric pressure Pfri: friction induced pressure drop Pgauge: gauge pressure Phyd: hydrostatic pressure Plap: Laplace pressure Ppump: pump power consumption Ptotal: total pressure Δp: pressure drop Δpm: porous plate pressure drop Δpmix: two-phase pressure drop Pe: Peclet number Q: volumetric flow rate Qread: rotameter flow rate reading Re: Reynolds number r: radial position inside the jet rbub: bubble radius tpore: pore depth viii tres: residence time of the stagnant fluid inside the void zone u: superficial velocity uero: pipe erosion velocity ujet: jet velocity at any point inside a jet umax: maximum liquid velocity at the jet centerline V: reactor volume Vall: total volume of the mass-transfer unit cell Vcone: jet cones volume Vejector: ejector volume Vmax: maximum theoretical reactor volume Void: volume between the jet cones Ṽ: dimensionless superficial velocity Vana: van der Waals constant Vanb: van der Waals constant W: power consumption We: Weber number Wec: Critical Weber Number Wejet: Jet Weber Number X: cell concentration Xo: percentage of gas being reacted ix yO2: O2 content inside the gas ε: volume fraction ρ: density β: square root of the quotient between pore area and plate area ϕ: energy dissipation rate per volume 𝛈: pump efficiency σ: surface tension μ: viscosity μnet: cell net growth rate εatt: molar attunement coefficient subscript: Big: Lab-made microbubble generator (10 mm) bottom: bottom of a large reactor g: gas i: gas species l: liquid mix: gas-liquid mixture r: radial direction of the flow River: Riverforest microbubble generator (7 mm) Small: Lab-made microbubble generator (7 mm) x top: top of a large reactor z: axial direction of the flow θ: angular direction of the flow xi LIST OF ABBREVIATIONS CPLJ: Confined plunging liquid jet DOP: dissolved oxygen profile DOS: dissolved oxygen sensor FOAK: First-Of-A-Kind HUSH: hollow cylindrical ultrasonic horn ISR: industrial-scale reactor MBE: microbubble ejector MBG: microbubble generator MJC: modified Jameson cell PFR: Plug Flow Reactor RO: Reverse-osmosis RTD: Residence time distribution SDS: sodium dodecyl sulfate VDC: Vessel Dispersion Coefficient VFD: variable frequency drive VVM: volume of air per volume of reactor per time xii CHAPTER ONE: LITERATURE REVIEW 1.1 Background Overview Many commercially important chemical and biochemical reactions that are carried out in a continuous liquid phase consume sparingly soluble gaseous reactants (e.g., O 2). The rate of gas- to-liquid mass transfer is frequently the rate-limiting step when these reactions are conducted in commercial-scale reactors. An effective process-intensification strategy for such reactions is to accelerate the mass-transfer step by dispersing the gas phase as microbubbles on the order of 100 μm in diameter. However, industrial implementation of this strategy has been hampered by the lack of a published design algorithm for commercial-scale microbubble-sparged reactors. One goal of this dissertation is to assess the suitability of alternative microbubble-generation methods for cost-effectively achieving high volumetric mass transfer rates (kla > 1000 h-1) in commercial-scale reactors. Emphasis is placed on methods that use liquid turbulence to generate microbubble dispersions, such as ejector and modified Jameson cell generators [1]. Published engineering correlations are used to predict and compare performance metrics of these systems, including volumetric mass-transfer rates, power-to-volume ratios, and suitability for scale-up. 1.1.1 Current Need for Rapid Gas-liquid Transfer Many multiphase chemical and biochemical reactions involve gaseous reactants that have low solubility in a continuous liquid phase. Examples include fermentation reactions that consume O 2 and/or CH4 carried out in aqueous solutions under mild temperatures and pressures. The 1 equilibrium solubilities (C*) of O2 and CH4 at 35°C and one atmosphere pressure are 33 mg/L (1.1 mM) and 17 mg/L (1.0 mM), respectively.[2] A recent DOE ARPE-E research program funded efforts to develop economically competitive fermentations that achieved an extremely high volumetric mass transfer rate (Q̇) for CH4 of Q̇CH4>50 gCH4/(Lreactorˑh) [3]. The volumetric mass transfer coefficient (kla) needed to achieve this QCH4 value can be estimated using the design equation for interphase gas mass transfer, Q̇ = kla(C*- CL), where kl is the liquid-film mass transfer coefficient, a is the interfacial area per unit reactor volume, CL is the bulk concentration of the transferred gas dissolved in the liquid, and (C*- CL) is the driving force for mass transfer. When gas mass transfer is rate-limiting, CL ≈ 0, so Q ≈ klaC*. Using this equation, the abovementioned ARPA-E technical target translates into a kla value of 2,900 h-1[3]. 1.1.2 Industrial Practices and Challenges on Gas-intense Operations The literature indicates that achieving kla values greater than 1000 h-1 cost-effectively is challenging. Commercial-scale aerobic fermentations are typically O2-mass-transfer limited, [4, 5] and stirred-tank reactors are generally used in such cases, because they achieve significantly higher kla values than columnar bubble columns or airlift reactors [4]. The intense turbulence generated by a stirred tank’s impeller breaks large bubbles into smaller ones, thereby decreasing the average bubble diameter (Dbub). Because a is given by Equation 1-1, where ε is the gas volume fraction, for a given ε value, decreasing Dbub by a given percentage increases a, and thus kla, by the same percentage. 2 6𝜀 𝑎= (1 − 1) 𝐷 However, using stirred tanks to achieve extremely high kla values scales up unfavorably. For geometrically similar tanks, power consumption (P) increases with impeller diameter to the fifth power, whereas reactor volume, V, only increases with impeller Dbub to the third power [6]. Moreover, kla has a relatively weak dependence on P/V (e.g., the square root of P/V) [5]. As a result, the power costs required to generate high kla values in large, stirred-tank reactors can be prohibitively high. Typical ranges of P/V values used in commercial scale reactors are reflected in literature reviews of gas-liquid mass transfer in stirred vessels. Van’t Riet’s review presented results over a P/V range of 0.4 to 10 kW/m3, which corresponded to kla values over a range of 72 h-1 to 2200 h- 1 [5]. Garcia-Ochoa’s review correlated data in seven studies over a P/V range of 0.2 to 1.1 kW/m3, which corresponded to a kla range of 22 to 150 h-1 [4]. Both of these kla ranges are well below the primary technical target value (2900 h-1) specified in the ARPA-E program, [3] suggesting that the conventional stirred tanks are not well suited to achieve the target kla value cost-effectively in commercial-scale reactors. 1.1.3 Microbubble Advantages on Gas-liquid Mass Transfer Microbubble dispersions offer several advantages over conventional bubbles for intensifying a wide range of industrial processes.[7, 8] Microbubbles’ extremely small Dbub (on the order of 100 μm) imparts several desirable properties, including extremely high large a and kla values 3 (Equation 1-1). At such small Dbub values, the intra-bubble pressure exceeds the pressure of the surrounding liquid by an amount that is inversely proportional to Dbub, as described by the Young- Laplace (plap) equation, where rbub is bubble radius and σ is surface tension. 2𝜎 ∆𝑝 = (1 − 2) 𝑟 This “Laplace pressure” accelerates a microbubble’s C*, (C*- CL), mass transfer rate, and rate of shrinkage as gas is transferred out of the microbubble. Moreover, microbubble shrinkage culminates in a violent bubble collapse that generates localized thermal gradients and free radicals[9, 10] that can be used to accelerate some chemical reactions, to accelerate biomass pretreatment[11], and to accelerate degradation of toxic compounds in wastewater treatment.[12] Microbubble sparging has demonstrated great potential in gas-intense operations where gas mass transfer is usually the rate limiting step [13, 14]. Microbubble spaging has been to used intensify a variety of processes, including biomass pretreatment[11], degradation of toxic compounds in wastewater,[12] biological water treatment [9, 15-17], fish farming [18, 19], nicotine biosynthesis [20], ocean vessel design[21], and various bioreactor processes [7, 22-25]. 1.2 Microbubble Generation Techniques and Technical Benchmarks Alternative methods to generate microbubbles for specific applications have been reviewed [26-29]. Parmar and Majumder described alternative microbubble-generation methods for waste treatment applications, but their study did not estimate power requirements or cost efficiency of the alternative methods for generating a target kla value [26]. Zimmerman et al. (2008) introduced 4 multiple innovative ways to produce microbubbles at a relatively low rate, but this article did not discuss the feasibility of achieving extremely high kla values in commercial-scale reactors or power requirements [27]. Terasaka et al. (2011) reviewed multiple microbubble-generation methods and conducted experiments to measure the O 2 transfer rate at a given power consumption level, but their reactor column was only 0.056 m 3, and the feasibility of scaling the system up was not considered [28]. Burns et al. reviewed some microbubble-generation methods based on electrolysis and rapid depressurization of liquid that was supersaturated with a gas. However, the methods Burns et al. discussed are utilized in separations instead of large-scale, gas-intense reactor processes [30]. The energy requirement from electrolysis-based microbubble-generation methods could also be too high for mass transfer processes. Rodriguez-Rodriguez et al. discussed various microbubble-generation methods for small-scale medicine productions as well [29]. Ansari et al. investigated microbubble-generation system that used individual tubes having an inner diameter on the order of 100 μm that were individually cut using a laser. This microfluidic fabrication approach would not be sufficiently scalable for commercial reactors [31]. The abovementioned studies have not satisfactorily addressed the important question of which microbubble-generation method(s) is best suited for commercial-scale chemical or biochemical reactors that require sparingly soluble gases to be transferred cost-effectively at extremely high kla values. The goal of this study is to address that question, which will contribute toward the ultimate objective of developing the first algorithm for design and scale-up of microbubble-sparged reactors that can meet ambitious to kla targets (>1000 h-1) cost-effectively at the commercial scale. 5 1.2.1 Bubble Size and Interfacial Area As described above, the average Dbub and resulting a values are primary performance metrics. However, because the power required to generate the surface area commonly represents a major component of the operating costs of reactors requiring high kla values, the ratio of power consumed to interfacial area generated is also extremely important. In addition to the average Dbub, the breadth of the Dbub distribution can also significantly affect system performance. Previously, two- phase flow microbubble-generation methods have been criticized for having a broad Dbub distribution[32], which would make it difficult to measure the Dbub via experiments. 1.2.2 Bubble Rising Velocity Bubble terminal rise velocity is an important factor for microbubble mass transfer because it governs gas-liquid contact time [33]. Bubble rise velocity is correlated to drag force and buoyancy force, which are further related to bubble size [34]. For rising bubbles whose Reynolds number (Re) is less than about 1.5 the rise velocity is reasonably represented by Stokes’ Law [35], which indicates that the rise velocity of bubbles varies inversely with the square of the bubble diameter [36]. For an aqueous solution at about 25ºC, this Re regime has an upper bound of about 130 µm. This correlation means sparging smaller bubbles would drastically improve the residence time for two phases to react and reduce material waste, thus improving the cost-efficiency of the two-phase operations [15]. 6 1.2.3 Bubble Coalescence Bubble coalescence can substantially reduce the volumetric mass transfer rate in a microbubble-sparged reactor through two mechanisms. First, it results in a larger average Dbub value, and consequently, as shown in Eq. 1-1, smaller a and kla values. Second, by increasing Dbub, coalescence reduces the intra-bubble pressure, and consequently, the C* value and mass-transfer driving force. The rate of coalescence depends on two factors: the rate of bubble collisions and the fraction of bubble collisions that result in a coalescence event rather than an elastic collision. When two bubbles collide, if the thin liquid film separating them drains to a critical value, the liquid film will rupture, and the two bubbles will merge [37, 38]. Bubble coalescence mechanisms and dynamics have been investigated using X-ray imaging [39], the lattice Boltzmann method [40], and the population balance method [41]. Prince & Blanch studied the impact of bubble coalescence phenomena on bubble-column performance and developed model for microbubble coalescence and break-up rate [42]. The rate of bubble coalescence observed in pure water can be reduced substantially by adding a sufficient amount of electrolytes, whose charges result in electrostatic repulsion forces between adjacent bubbles that prevent intra-bubble liquid films from rupturing, thereby inhibiting coalescence [43, 44]. In a previous study conducted by Zahradnik et al., it was found that adding small amount of electrolyte to xanthan-alcohol system could reduce bubble coalescence to almost non-exist [45]. Surfactant is another method to stabilize the bubble sizes and prevents bubble coalescence[46]. The characteristics of surfactant-stabilized bubbles were investigated before by numerous research groups previously as well [47-50]. 7 1.2.4 Microbubble Generator Power Consumption Implementing microbubble generator (MBG) could also reduce operation and capital cost as well. Previous study conducted on bubble sparging column indicated that the liquid inside the system is well mixed, thus no further mixing apparatus is required inside a microbubble sparging column [51]. This finding indicates microbubble generation could replace impeller rotation as the method of improving gas-liquid contact. The installation cost, along the operating cost of impellers could be removed from the overall operation and capital cost. 1.3 Microbubble Generation Mechanisms Many microbubble-generation methods have been developed over the years based on different fluid dynamics principles. In this study, we focus on the microbubble-generation methods based on fluid turbulence due to their simplicity and cost-efficiency when scaling up. This study also includes a collection of other microbubble generation based on other principles but were left out of cost efficiency discussion due to limit in the implementation of these microbubble-generation systems on scale-up systems. 1.3.1 Turbulent Energy Based Turbulence energy types of microbubble-generation methods rely on turbulence energy of the fluids to break up two-phase fluids down into microbubbles. These types of microbubble- generation systems usually have gas and liquid flow through confined spaces at high liquid 8 superficial velocity in order to maintain a level of turbulence and shred larger air pockets inside the liquid into microbubbles. The efficiency of this process is dependent on the equipment dimensions and fluid flow rate, which will affect the bubble size and energy cost of the systems. The effect of nozzle shape and size on bubble characteristic has been previously investigated [52], providing a good basis for optimization and design. 1.3.1.1 Confined Plunging Liquid Jet and Modified Jameson Cell Confined plunging liquid jet (CPLJ) is a microbubble-generation method that utilizes shear and turbulence to create microbubbles inside a flotation cell [53]. Inside a conventional plunging jet bubble generator, liquid and gas flow downwards into a cell and bubbles would form due to complex hydrodynamics [54]. Over the years, researchers have developed industrial production units based on CPLJ such as Jameson Cell [55] and cyclo-microbubble Flotation Column [56]. Majumder et al. has measured the bubble size and gas-liquid interfacial area inside a down flow bubble column [57]. This microbubble-generation method has had scale up application such as mineral flotations [58] and ozone water purification processes [59, 60] due to its improvement on gas mass transfer. However, the bubble generated using conventional plunging jet would sometimes still generate bubbles on millimeter scale, which is too large compared to other microbubble-generation methods [57, 59]. For all plunging jet columns, the penetrating depth of gas should also be considered when designing the column. Since there is only one gas source from above, the total amount of gas 9 transferred into the column would deplete after a certain depth. After this depth, it is pointless to add in additional height to the column since the rest will not receive any aeration at all. The calculated depth can be expressed in the following form, where ug is gas superficial velocity across the column and ρg is the gas density: 𝑢 𝜌 𝑑𝑒𝑝𝑡ℎ = (1 − 3) 𝑘 𝑎𝐶 ∗ A modified Jameson cell (MJC) system described by Li et al. includes a column through which liquid flows in a downward direction and bubbles of down-flowing liquid reactant liquid through which bubbles of the reactant gas rise that combines features of a Jameson Cell and conventional bubble sparger [1]. The MJC system consisted of a liquid column used a conventional porous bubble sparger at the bottom of the cell, and a perforated plate for liquid jet generation purpose at the top. Coarse bubbles on millimeter scale would generate by the porous sparger and then float to the top of the cell due to buoyance forces. These coarse bubbles would then generate a layer of foams on top of tank column liquid. The liquid jet from above would breakdown the foam layer into microbubbles and the current would carry the microbubble downwards for gas transfer. The schematic can be seen in Figure 1-1. This MJC column design has already been utilized in a pilot- scale bioreactor according the patent US20140212937. According to the patent, the MJC column managed to produce microbubbles on average diameter of 60 μm while Hernandez-Alvarado et al. managed to produce microbubble with average diameter of 600 μm. Sodium dodecyl sulfate (SDS) was added as a surfactant to stabilize the bubble formation in both the patent (100 ppm) [1] and 10 the bubble size estimation study (10 ppm) [61] for the MJC. This adds in additional material and separation cost for the production and could also reduce the overall gas mass transfer rate [8]. The patent contained more SDS and managed to produce smaller bubbles. Figure 1-1. Simplified schematics of a MJC Microbubble Generator 1.3.1.2 Spiral Ejector For swirl ejector microbubble generators, the fundamentals behind microbubble formation vary from type to type. An earlier design of swirl microbubble ejector was used for wastewater treatment [28]. The main body of the swirl ejector is a swirling chamber. The gas and liquid streams were injected into a swirl chamber perpendicular to each other, and the gas-liquid mixture is broken down into microbubbles due to the centrifugal force acting on the liquid flow [62]. The liquid usually enters the chamber from the side while the gas enters from the bottom. The cylindrical 11 liquid flow inside a tube has been thoroughly investigated before via DNS simulations and numerical simulations [63-65]. A more recent invention by Hato claimed that by isolating the gas and liquid injection flow, the microbubble ejector could generate bubbles on nano-scales. This is contributed to the conservation of kinetic energy [66]. Swirl microbubble ejector has wide utilizations in dissolved air flotation operation such as water treatment [67, 68]. The same type of microbubble generator has also had applications in Wafer cleaning devices [69]. Another usage for spiral microbubble ejector is for underground waste water treatment [15]. The aforementioned water treatment system managed to produce microbubbles of diameters from 0.5 to 100 μm. Another investigation from Levitsky et al. managed to produce microbubbles on scale of 20 μm and the research group managed to obtain the bubble size distribution correlations with operating conditions [70]. Figure 1-2 demonstrates a simplified version of spiral microbubble ejector, where liquid is introduced on the top of the chamber while gas is injected from the bottoms. Figure 1-2. Simplified schematics of a spiral ejector style microbubble generator 12 Terasaka et al. conducted an investigation on water sludge treatment and water aeration rate and they have compared the effect of microbubble sparging methods on O 2 transfer rate. The study concluded that spiral ejector style microbubble-generation method can improve O 2 transfer rate better than other methods, albeit at higher energy cost[28]. Another paper by Li et al. on this water treatment method has managed to find the bubble size distribution under different combination of aeration methods and fluid components [71]. 1.3.1.3 Venturi A Venturi type microbubble generator consists of a hollow tube with a converging-diverging neck. This type of microbubble-generation techniques utilizes the cavities of the Venturi neck to break up the bubbles to a desirable size [72]. The liquid flow, upon hitting the narrowed neck, would increase in pressure and create a pressure wall due to Bernoulli effect. The ul can reach up to sonic speed at the entrance of the converging-diverging neck, which is not found in ejector MBG [72]. The gas mixed in would bounce off the pressure wall and create microbubbles due to shockwave [9] . The bubble size prediction based on fluid flow and instrument dimensions was previously obtained through image analysis [73]. A more recent study by Gordiychuk et al. has correlated the bubble size distribution with dimensionless operation parameters such as ε and Re using image analysis technique as well [74]. The Venturi system used managed to produce microbubbles on scale of 100 to 400 μm. The utilization of Venturi microbubble generator has shown improvement in yeast fermentation [23]. 13 Figure 1-3 shows a simplified demonstration of Venturi type microbubble generator, which comprises of a two-phase fluid inlets and a converging-diverging nozzle [26]. Figure 1-3. Simplified schematics of a Venturi style microbubble generator Patent US20160325242A1 indicated that Venturi type microbubble generator has applications in fields of furniture and decorations, such as bathtubs and swimming pools [75].The bubbles produced using this method is on scale of micrometer scale, which makes it less attractive than other bubble generation methods due to its larger bubbles [72]. The bubble movement, velocity, deformation, and collapse through the converge-diverge nozzle have been investigated through image analysis previously [76]. It was found out that decreasing Ql would increase bubble collapse, thus reducing the Dbub. This finding suggests a trade-off between production rate (correlated to Ql) and Dbub, which makes it difficult to optimize. Another problem with Venturi tube is the cavitation of bubbles inside the nozzle [72]. It would be counter-productive to use Venturi tubes to generate bubbles since the goal of utilizing microbubble 14 is to increase gas dissolution inside the liquid. If cavitation removes dissolved gas inside liquid due to cavitation, using Venturi tube would defeat the whole purpose of microbubble aeration. 1.3.1.4 Spinning Disk Sebba et al demonstrated the setup of a spinning disk gate microbubble generator in US5314644 [77]. Spinning disk gate microbubble generator utilizes shear momentum of liquid to disperse pocket of gas through porous materials. In this particular design, a motor is installed at the bottom of a cylindrical vessel, and a round porous plate is attached to the rotor. Gas-liquid- surfactant mixture was propelled into the rotating cylinder complex and broken down into microbubbles [77]. Previous experiments conducted on rotating disk microbubble generator has found that this bubble generation technique could increase O2 transfer rate at lower agitation rate [78]. This indicates that implementing a rotating microbubble generator would be more energy conservative compared to mechanical agitating aeration processes. The spinning disk microbubble generator has shown potentials in increasing biomass fermentation [24, 25] and improving hairy root metabolites growth rate [20]. This microbubble generator design, however, is relying on surfactants to stabilize the bubble, which will not be economically viable due to material and separation costs. Additionally, surfactant content inside microbubble flow could reduce the overall O 2 transfer efficiency [15]. 15 1.3.2 Microbubble Generators Based on Other Principles Microbubble-generation methods based on fluid-dynamics mechanisms other than turbulence have been developed. However, these methods are not considered well suited for large-scale industrial application for reasons described below. 1.3.2.1 Oscillation A fluidic-oscillator mechanism has been used to produce microububble streams based on surface instability and Coanda effect [79, 80]. A fluidic oscillator employs a fluid amplifier whose geometry triggers a side-to-side mechanical oscillation under the influence of a flow field [81]. The main body of some fluidic oscillators have no moving parts, which proponents claim makes them easy to manufacture, durable and cost-efficient [82, 83]. As a gas stream enters the liquid phase, the oscillations influence bubble formation dynamics, interrupting the bubble development process and creating smaller bubbles than would have resulted in the absence of the oscillation[22]. Zimmerman et al. described use of a fluidic oscillator for aeration operations [84] and its potential application to bioreactors [81] and coal flotation [85]. Al-Mashhadani et al. measured mass- transfer properties of oscillation-induced microbubbles in CO2 mass transfer inside a bubble column [86]. The use of microbubbles produced by fluidic oscillation in the pretreatment of lignin biomass has been explored on the basis that O2 radicals generated by microbubbles could increase reactivity of biomass [11] and biogas [87]. The bubble sizes generated via fluid oscillation have been correlated with the oscillation 16 frequency, gas flow rate, and oscillator design [88]. Brittle et al. showed that an ‘optimal’ oscillation frequency for liquid oscillator that results in minimal bubble size microbubbles [79]. However, bubble size was found to fluctuates erratically with minor changes in gas flow rate [89]. This finding suggested that rigorous control of gas velocity through the fluidic oscillators would be required to obtain optimal bubble formation. Moreover, the Conada-effect oscillation occurs at the length scale of individual bubbles, suggesting that scale-up of the method would require the number of precision-fabricated oscillators to increase linearly with gas throughput. Each fluidic oscillator unit presumably requires precision fabrication, and a large number of small fluidic oscillators. Such a scale-up strategy would seem impractical for commercial-scale reactors that operate at very large kla values. 1.3.2.2 Electrolysis Electrolysis can be used to form bubbles of H2at the cathode and O2 at the anode of an electrolytic cell. The size and density of the bubbles can be controlled by adjusting the electrolyte concentration, power output, and electrode material and local hydrodynamic forces [90, 91]. Microbubbles 20 µm in diameter have been produced using electrode having sharp tips [92]. The delivery rate of the bubble is also a function of electrode shape and local hydrodynamic forces. However, this approach is impractical for mass-production of microbubbles for large-scale industrial applications [91] for several reasons. The approach would be useful only in rare 17 applications in which simultaneous production of a 2:1 molar ratio mixture of H 2 and O2 gases is needed. The production of significant quantities of this explosive gas mixture would create a safety hazard. Moreover, the manufacturing costs of custom electrodes having sharp tips and electrical costs required to drive the hydrolysis reactions and the cost of the electricity would likely to be prohibitively high. 1.3.2.3 Microfluidic Microbubbles can be generated in a microfluidic flow system when a continuous liquid stream impinges on a continuous gas stream (e.g., in a T junction), and hydrodynamic forces break up the gas stream into discrete bubbles. Garstecki et al [93] investigated the bubble-formation mechanism in a microfluidic T-junction microbubble generator and developed a scaling law to predict the bubble size as a function of diameter of the liquid channel’s diameter and the gas and liquid flow rates. This microbubble-generation method has potential to make relatively small quantities of microbubbles having a uniform bubble diameter. However, the costs to microfabricate and to install large numbers of such microbubble generators and to pump large quantities of liquid through them in laminar flow would be prohibitively high for large-scale reactor applications. 1.3.2.4 Ultrasonic By attaching an ultrasound inducer to a steel capillary, microbubbles can be injected into high surface tension fluids due to shockwave-generated bubble collapsing [94]. The size of the 18 microbubble is controlled by inducer frequency and capillary dimensions. Pulse-induced bubble size prediction and formation from a capillary was investigated using both computational fluid dynamics and image analysis [95]. From photographic evidence obtained by the same research group, it was observed that a single stream of microbubbles is produced when the needle is vibrated at a certain frequency. This evidence indicates the total gas flow rate is very low and cannot sustain industrial scale production [94]. Bubble coalescence under ultrasound field could also be a concern when designing reaction systems [96]. The coalescence phenomena of microbubbles under ultrasound field and bubble breaking model were calculated based on inertial drainage model [38, 97]. Kobayashi et al. also discussed using ultrasonic irradiation to separate microbubbles on industrial scales should the demand rises [98]. An upgrade version of ultrasound microbubble generator was designed by Makuta el al. in order to improve the total gas production rate using cylindrical horn [32]. By using a horn instead of needle to sparge gas and to create ultrasonic vibrations, the total gas production rate is increased. From graphic evidence, the total gas production rate was improved since the gas is coming out of orifice pores whose total area is much larger than a sparging needle [32]. However, the increase in gas flow rate is insufficient according to a later research based on this type of design [99]. This research based on hollow cylindrical ultrasonic horn (HUSH) managed to improve the gas transfer rate and reduce coalescence through adding multiple orifices inside a HUSH [99]. Figure 1-4 demonstrates a simplified schematics for a HUSH horn. 19 Figure 1-4. Simplified schematics for the experiment setup by Tatsuya et al. [99], depicting the fundamental elements of an ultrasonic microbubble generator. Achaoul et al. have recently developed a new method to combine the electrolysis and ultrasonic microbubble-generation method [100]. This type of ultrasonic microbubble generator combines electrolysis microbubble-generation method and the above ultrasonic horn method to enhance the control over Dbub and liquid volumetric flow rate (Q). By synchronizing the electrode and ultrasound frequency, the Dbub and Q can be finely controlled [100]. Under such synchronized conditions, the system can produce pure H2 gas for energy usage purposes, yet it is not economical to invest in electricity in order to produce H2 gas. However, this type of microbubble-generation method is very energy intensive, thus not viable for industrial purposes. In addition, in order to control the bubble generation with precision, the ultrasound inducer must be finely tuned, which requires a PID controller. The article proposed the electrolysis-ultrasound production type pointed out that this hybrid microbubble-generation method is more suitable for medicine and biology field [100]. 20 1.4 Microbubble Ejector In a previous study, Shirtum et al. reported that highly turbulent, gas-liquid flow in a confined area, such as an microbubble ejector (MBE), could generate steady flow of bubbles on micron scales [101]. The bubble formation is contributed to the turbulence energy. As fluid turbulent energy exceeds the surface tension energy, the gas-liquid mixture would break up into microbubbles [102]. One of the MBGs utilizing turbulent energy is the MBE. A schematic of MBE can be seen in Figure 1-5 High velocity liquid flows through narrow channel and mix with gas that was transported in by line pressure difference. The gas-liquid mixture is then broken down into microbubbles due to high turbulence and the liquid-bubble mixture is then ejected into the vessel. A typical microbubble ejector contains a suction nozzle, a liquid nozzle, a mixing tube, and a bubble diffuser [103]. The liquid nozzle is connected to the inlet pipe of whatever the system the ejector is aerating. The gas nozzle is connected to gas pipeline. The mixing tube is narrower than the inlet nozzles in order to increase the fluid speed and create turbulence for bubble creations. The exit nozzle would sparge out bubble-liquid mixture into the reactor systems. An MBE has no moving part, thus making it very simple and cheap to maintain [104]. The two-phase flow dynamics and O2 transfer inside an MBE has been thoroughly studied before, making it much easier to design suitable MBE for different operating conditions [103, 105]. 21 Figure 1-5. Simplified schematics of an MBE microbubble generator The size of bubbles generated by fluid turbulence is heavily correlated to fluid superficial velocity. According to previous study on fluid characteristics done by Chen et al., the combination of high liquid (ul) and low gas (ug) superficial velocity would produce microbubbles with the smallest diameter [102]. Kim et al. have studied the gas-hold up, flow regime, and gas suction rate for two-phase flow inside an MBE, and the result could be used for design [105]. The bubbly flow throw narrow channel was investigated through computational fluid dynamics and numerical simulation [106]. Prediction of Dbub from operating conditions, characteristics, and equipment dimensions is feasible with the available knowledge. The pressure drop for two-phase flow has been previously studied through experiments and computational fluid dynamics [107]. The bubble coalescence inside a jet ejector has been investigated before through computational fluid dynamics [41]. Since both bubble size and pressure drop are available for estimations, the cost-efficiency estimation for MBE system 22 scale-up would be feasible. Patent US6017022 claimed that MBE has been utilized in industrial production, such as monomer processing, air stripping, and water sludge treatment [101]. 1.4.1 Overall Mechanics Behind Microbubble Ejectors The Dbub obtained in an MBE can be estimated using the correlation developed by Chen et al., which expresses the maximum stable bubble size in highly developed turbulent flow as a function of liquid and gas flow rates and pipe diameter [102]. For a fully developed, highly turbulent flow inside a pipe, the flow can be assumed isotropic. Under this condition, the three velocity components are equal, allowing the turbulent kinetic energy of the fluid, Eturb, to be written as shown in Equation 1-4, where ul is the liquid velocity, f is the friction factor of the liquid phase, ρl is liquid density, r, z, θ are direction of the flow: 1 3 3 𝑓 𝐸 = 𝜌 𝑢 +𝑢 +𝑢 = 𝜌𝑢 = 𝜌𝑢 (1 − 4) 2 2 2 2 The work needed to create a is defined as surface free energy per unit area, Esurf [102]. Esurf of bubbles with Sauter-mean diameter of Dbub inside a pipe with cross-sectional area of apipe can be calculated using Equation 1-3, where σ is surface tension. When the dispersed bubbly flow forms, Eturb (energy provided by the turbulent fluid) and the Esurf (energy needed to create bubbles) would be equal to each other as shown in Equation 1-4: 6𝜎 𝐸 = 𝑎 𝑢 (1 − 5) 𝐷 3 𝜌𝑢 6𝜎 𝑓 𝑎 𝑢 = 𝑎 𝑢 (1 − 6) 2 2 𝐷 23 1.4.2 Bubble Size Correlation Using Single-phase model Rearranging Equation 1-4 can yield the equation for theoretical Dbub inside a cylindrical pipe in the following form, where f is friction factor: 8𝜎𝑢 𝐷 = (1 − 7) 𝑓𝜌 𝑢 For f, Chen et al., suggested to use the liquid pipe flow friction factor, where both Cf and nf are arbitrary constants and Re: 𝑓 = 𝐶 𝑅𝑒 (1 − 8) For a single-phase liquid, Cf = 0.046 and nf = 0.2. The above equation can be rewritten in form of dimensionless number in order to minimize variables, where We is Weber number: 8𝜀 𝑊𝑒 = (1 − 9) 𝐶 𝑅𝑒 Chen et al., use of a friction factor correlation for a single liquid phase may introduce error into the equation when it is applied to two-phase flow. Garcia et al., proposed a friction factor equation for two-phase dispersed bubbly flow [107]. The mixture density, velocity, and Re, (ρmix, umix, and Remix) for a two-phase gas-liquid mixture are defined as follows, where dnoz is the ejector nozzle diameter: 𝜌 = 𝜌 𝜀 + 𝜌 (1 − 𝜀) (1 − 10) 𝑢 =𝑢 +𝑢 (1 − 11) 𝜌 𝑢 𝑑 𝑅𝑒 = (1 − 12) 𝜇 24 The resulting two-phase friction factor expression, fmix, was correlated to be: . . . 16.46𝑅𝑒 − 0.067𝑅𝑒 𝑓 = 0.067𝑅𝑒 + (1 − 13) [1 + (𝑅𝑒 /50) . ] . Replacing Equation 1-6 with Equation 1-11 and plugging the friction factor expression into Equation 1-5 gives a modified Chen correlation, which will be referred as ‘two-phase (2P) Chen correlation’. A comparison can be shown in Figure 1-6: 0.0007 5% Gas, Chen 5% Gas, New 0.0006 10% Gas, Chen 10% Gas, New 15% Gas, Chen 15% Gas, New 0.0005 Bubble diameter(m) 20% Gas, Chen 20% Gas, New 0.0004 0.0003 0.0002 0.0001 0 6 7 8 9 10 Liquid velocity (m/s) Figure 1-6. Bubble size simulation based on single (Chen) phase and two-phase friction factor The 2P Chen correlation predicts smaller Dbub than the original 1P (one phase) Chen correlation. At 10% εg and between 20 to 30 ft/s ul range, 2P Chen correlation predicts a 15% smaller Dbub compared to the 1P Chen correlation. The study to validate the differences between these two models will be addressed later in this study. 25 1.4.3 Jet Cone Angle Upon leaving the MBE, the high-speed bubbly jet entrains the stagnant fluids from surrounding and thus increases the overall liquid flow rate inside the jet, making conventional estimation of the jet flow rate based on constant volumetric jet flow rate vastly inaccurate. Lima Neto et al (2008) investigated the stagnant fluid entrainment at high void fraction and developed for a given jet radius and axial liquid velocity [108]. However, axial liquid velocity is difficult to measure or estimate inside an industrial reactor vessel. A more recent model based on computational fluid dynamics has developed an equation to estimate the change in volumetric flow rate correlated to inlet liquid velocity, which makes it much easier to estimate the jet Ql when entrainment is present [109]. The following simulation model assumes a Gaussian Distribution of liquid velocity inside a liquid jet. 𝑢 =𝑢 𝑒 (1 − 14) where ujet is liquid jet velocity at any point inside a jet, umax is maximum liquid velocity at the jet centerline, r is the radial position inside the jet, and h is the axial position inside jet. Cushman-Roisin (2010) assumed jet angle from side to opposition is considered to be around 24 degrees. At any given point inside the jet, the ratio between the radii and the height of the jet would be approximately tan(12°)≈0.2 [109, 110]. 26 CHAPTER TWO: MBG CAPABILITY USING E. COLI GROWTH AS INDICATOR To validate the effectiveness of MBEs for microbial fermentations, a comparison experiment between the MBE and the conventional air ring sparger was conducted in the Michigan Biotechnology Institute’s (MBI) fermentation pilot plant. In this comparison, E. coli cells inside a 300 L fermenter that was customized by MBI personnel to allow sparging by either a conventional ring sparger and an MBE. The growth rate of the cells during a sparged fermentation was measured as an indicator of the relative effectiveness of the two aeration methods. 2.1 Modeling Cell Growth under Mass Transfer Limitation Batch growth of microbes exhibits multiple phases, including the lag phase, the growth phase, the stationary phase, and the death phase. During the growth phase with sufficient nutrients (carbon sources, O2, etc.) cells typically exhibit a constant doubling time, and the cell death rate is minimal. As a result, the cell number increases exponentially. During this phase, cell growth can be modeled using a mass balance on cell concentration that assumes the growth rate is proportional to the cell concentration (i.e., first order kinetics): 𝑑𝑋 =𝜇 𝑋 (2 − 1) 𝑑𝑡 where X is the cell concentration and μnet is the net specific growth rate. 27 2.2 Seed Train, Seed Source, and Broth The cell growth inside the 300L fermenter was measured over a 24-hour period under two aeration methods: a conventional ring air sparger and MBE. The E. coli cell line (XL1-Blue competent cells) were grown following culturing protocols recommended for this strain [111, 112]. The original frozen cell tubes were transported using insulated container from MSU to MBI, where the cells were thawed and used to inoculate a seed train consisting of increasingly larger glass batch-culture vessels, as recommended by MBI fermentation experts. Seed-train cultures were grown in replicate to reduce the risk of the experiment being terminated by contamination or poor growth in a single vessel. Experimental details are provided in Appendix 1-1. Figure 2-1 below is a simplified schematics of the seed train with the actual picture of the pilot-scale 300L reactor vessel that was used for this experiment. Figure 2-1. Schematics of the seed train and the pilot-scale fermenter used for the E. coli cell growth experiment. 28 Inside the 300L bioreactor, the M9 broth was selected as the growth medium. Even though M9 growth medium does not provide higher specific growth rate compared to other medium such as LB or 2TY, the slower specific growth rate is beneficial for accurate measurement of growth rate [111]. Due to hour limits from MBI, the research team could only take samples for content measurement every hour. Thus, a 24 hours growth period with a lower growth rate was selected to ensure the growth period is long enough to fully develop the exponential phase. 2.3 Fermenter and MBG Installation The 300L fermenter was equipped for two aeration modes. A Riverforest YJ-8 MBE (Figure 2-2 left) was installed at the bottom of the vessel for microbubble sparging. Also, an O-ring gas sparger (Figure 2-2 right) was installed inside the vessel for conventional bubble sparging. The batch cell growth kinetics resulting from the use of each aeration methods were measured in separate experiments. The Riverforest YJ-8 MBE was connected to the main water line and the tank with NPT connector. On Figure 2-2 right, the black nozzle on the side is the gas entry point. Gas pocket is mixed with the high velocity liquid flow inside the MBE nozzle neck and shredded into microbubbles before been sparged into the main liquid body. The O-ring gas sparger was a stainless-steel tube shaped into a circular ring at the bottom. Small holes were drilled into the bottom of the ring. Compressed air that was pumped into the top of the sparger emerged from the holes as bubble streams that rose up through the growth medium. 29 The two spargers differed in the mechanism that controlled the bubble size. In the ring sparger, the bubble size is influenced by ratio of surface tension of the gas-liquid interface, which resists shearing of the emerging bubble off of the sparger hole, and the dynamic pressure of the liquid flowing around the bubble, which encourages shearing of the emerging bubble from the hole. The ratio of the surface tension to the dynamic pressure can be quantified using a Weber number [112]. In the MBE, intense turbulence of a gas-liquid dispersion flowing at high velocity through the MBE’s throat causes disintegration of preexisting conventional bubbles into microbubbles [102]. Figure 2-2. The aeration instruments for the cell growth experiment (L: MBE, R: Ring sparger) A schematic of the two different aeration strategies is shown in Figure 2-3. For microbubble sparging, an external centrifugal pump was used to recycle the fermentation liquid from the 30 reactor, through the MBE where air was added, and then into the reactor. The centrifugal pump was turned off during ring-sparger aeration, and a small stirrer (diameter < 2 cm) was installed at the bottom of the fermenter to encourage convective mixing within the reactor. In both cases, spent air was filtered and released through the top of the fermenter. Figure 2-3. Schematic of the 300L fermenter showing the two modes of sparging 2.4 Measurements and Results 2.4.1 Sampling Strategy Four hours after the bioreactor was inoculated with E. coli culture, measurements (OD600, dissolved oxygen, glucose level, temperature) used to track the fermentation’s progress commenced. Optical density of the fermentation broth at 600 nm (OD600) was measured and converted into cell density using a calibration factor: 1.0 OD600 unit value equals to 8 × 10 8 cells 31 per ml [113]. For the first five hours, no samples were drawn since the growth is very likely to be still in the starting phase. Starting from the fifth hour, liquid content was extracted every hour for the content measurement. Due to MBI’s operating-hour constraints we were unable to take data points throughout the entire 24-hour operating period. However, sufficient data points were obtained to compare the specific cell growth rate under different sparging systems. 2.4.2 Growth Rate Comparisons The following Figure 2-4 demonstrates the measured OD600 values measured for three different fermentation trials inside the 300L pilot-scale bioreactor. The OD600 values were measured and plotted as a function of time since bioreactor operation started. The previous growth equation was used to fit the curve to calculate the specific cell growth rate inside the bioreactor during different aeration modes. Three aeration modes were investigated. The first was conventional air sparging using the ring sparger shown in the previous Figure 2-3. When using MBE to aerate the tank, a high gas flow rate (0.1 VVM or volume of air per volume of reactor per time) and a low gas flow rate (0.01 VVM) were investigated separately since the literature indicated that different εg has impact on Dbub generated by MBE [102]. 32 E.Coli Growth curve 2.5 MBG-High gas flow 2 MBG-Low gas flow Air Sparging OD600 (A) 1.5 1 0.5 0 0 5 10 15 20 25 Time (hour) Figure 2-4. Measured OD600 values for the three aeration conditions inside the fermenter The following Table 2-4 contains calculated specific growth rate for the three different aeration mode. The final OD after an entire day of fermentation is also recorded. According to the previous Figure 2-4, only MBE at low gas flow managed to enter stationary phase while the other two aeration mode failed to achieve so. The cell growth rate for low gas flow MBE is double of the specific growth rate of air sparging. The calculated specific growth rate for high gas flow MBE is lower than low gas flow MBE as shown in Table 2-1. At higher gas flow rate, the predicted bubble size is larger and the overall gas-liquid contact area could be smaller [102]. This could contribute to the low cell specific growth rate. 33 Table 2-1. Calculated specific cell growth rate and OD 600 after 24 hours for the aeration methods OD after 24 hours Growth rate (/hour) Air sparging (0.1VVM) 0.47 0.174 Microbubble (0.1VVM) 1.7 0.24 Microbubble (0.01VVM) 2.25 0.355 2.4.3 Glucose Level The measured glucose level decreases as the specific cell growth propagates as it can be shown in Figure 2-5. The glucose level inside the low gas MBE was the lowest at the end of the 24-hour period. This glucose level measurement also supports the previous calculated specific growth rates. Since the low gas MBE aeration method had the highest specific growth rate, the broth should have lowest glucose content left due to consumption from cell growth. In fact, the glucose content in the broth during the low gas MBE aeration was almost exhausted by the end of the cell growth period. The glucose level was plotted over the OD600 value instead of time to ensure a fair comparison, since the specific growth rates are different and cell concentrations are at different. The initial slopes from the starting phase look similar for all three aeration modes. Once the OD600 value surpasses 0.5, the glucose consumption rates start to differ. Since ring sparger could not deliver much gas compared to MBE, no conclusion of glucose consumption at high OD600 could be drawn from its data. The slope for the low gas MBG is much steeper than high gas MBG. This implies that at low gas condition, more glucose was consumed to produce the same number of cells. 34 Both the OD600 measurement and the glucose content measurement suggests that MBG at low gas flow rate managed to transfer large amount of O2 into the fermenter to sustain high cell growth. The fact that low gas MBG used more glucose to produce unit amount of cell does not change the result that low gas MBG grew the cells much faster to a much higher ceiling after 24 hours. Glucose level over OD600 9 8 Glucose conecentration (g/L) 7 6 5 Control-Ring sparger 4 MBG-High gas flow 3 MBG-Low gas flow 2 1 0 -1 0 0.5 1 1.5 2 2.5 OD600 Figure 2-5. Glucose level within the liquid at different OD600 value 2.4.4 Temperature and pH Some irregularities from the previous results could be potentially attributed to the temperature and pH irregularities during the fermentation process. These irregularities could change the cell metabolism and change the specific growth rates of the cells inside the fermenter. The below Figure 2-6 demonstrates the pH change inside the fermenter as the cell growth 35 propagates. Under low εg MBE and ring sparger aeration, the pH of the bulk fluid inside the fermenter decreased while cell concentration increased. When using low gas MBE, the pH dropped from 7.2 to around 6 at the end of the 24-hour period. Under high gas MBE condition, however, the pH drastically reduced from 7.2 to 6.6, and then returned to 6.8. The optimal pH for E. coli cell growth is between 7.5 and 6.5 [114], thus the low gas MBG aeration might produce acidic products and reduced the pH of the bulk fluid. However, this highlights the O2 transfer rate provided by the low gas MBG, since this particular aeration method was capable of developing cells even outside optimal pH condition. pH change with OD600 7.4 7.2 Control-Ring sparger MBG-High gas flow 7 MBG-Low gas flow 6.8 pH6.6 6.4 6.2 6 5.8 0 0.5 1 1.5 2 2.5 OD600 Figure 2-6. pH changes within the fluid for the cell growth experiment at different OD600 level 36 The experiment also faced temperature irregularities as shown in Figure 2-7. At the start of the experiment, the thermos jacket for the fermenter was initiated at lower temperature than the ideal growth temperature of E. coli cells. The ideal growth temperature for E. coli cells 37 ℃, even though the cell is survivable between a wider range [115]. For the MBG experiments, the temperature of the thermo-jacket on the bioreactor failed to maintain the temperature to around 37 ℃. This effect could also have affected the specific growth rate and introducing errors in growth rate calculations. 43 41 39 Temperature (C) 37 35 33 31 Air Sparging 29 MBG-High gas flow 27 MBG-Low gas flow 25 0 5 10 15 20 25 Time (h) Figure 2-7. Temperature changes within the fluid for the cell growth experiment 37 2.5 Conclusion The mass transfer capability of MBE was compared to conventional ring sparger inside a 300L pilot-scale bioreactor. E. coli cell growth was used as an indicator of O2 mass transfer rate, as higher O2 mass transfer rate usually translates into higher specific growth rate. The cell concentration, dissolved O2 content, glucose level, temperature, and pH of the media were measured to observe the cell growth. When using high gas flow MBE, the cell grew 40 percent faster and the final cell concentration is 3.6 times higher after 24 hours than when using air sparger. When using low gas flow MBE, the cell grew at over double the speed and the final cell concentration was 4.8 times higher compared to using air sparger. The glucose content inside the media also supported the cell growth trends. Higher cell concentration corresponds to lower glucose concentration during measurements. Some irregularities in pH and temperature might had impact the growth of the cells. Based on qualitative analysis, MBE provides better gas-to-liquid mass transfer than traditional ring sparger aeration. However, in order to implement MBE for large-scale production, further quantitative analysis is still needed. 38 2.6 Growth Media Content The seed train protocol for the E. coli growth is shown here: 1. The thawed seeds were transferred to a culture plate for growth. The culture plate was left for 24 hours for growth. 2. The content on the plate is then transferred into two 5 ml test tubes. Lysogeny broth was provided by MBI and additional glucose solution was added to provide cell growth content. The test tubes were left for 24 hours for growth. 3. The content for both test tubes were measured after 24 hours. The tube with higher cell concentration was selected and the content of the tube was transferred into two 500 ml shake flasks. Additional glucose solution along with M9 broth were added to the shake flasks for cell growth. The two shake flasks were left for 24 hours for growth. 4. The content value for both flasks were measured after 24 hours. The flask with higher cell concentration was selected and the content of the flask was transferred to the 300 L fermenter. Additional glucose solution along with M9 broth were added to the fermenter as well before the final cell growth starts. The temperature for all cell growth during the seed train was 37 ℃ The M9 minimal mineral medium composition can be seen in the following tables from 2-2 to 2-4. The highlighted ‘M9 Salt solution’ and ‘trace metal solutions’ compositions are recorded in the following tables as well [116]. 300 L of the aforementioned M9 minimal mineral medium was added to the fermenter, with an additional 4g/L glucose also added as the carbon source before the fermentation started. 39 Table 2-2. Composition of one liter of M9 mineral medium solution 1L M9 mineral medium Water 867 ml M9 Salt solution (Table 2-2) 100 ml 20% Glucose 20 ml 1M MgSO4 1 ml 1M CaCl2 0.3 ml biotin (1mg/ml) 1 ml thiamin (1mg/ml) 1 ml trace metal solution (Table 2-3) 10 ml Table 2-3. Composition of one liter of M9 salt solution 1L M9 salt solution Water 1000 ml Na2HPO4-2H2O 75.2 g/L KH2PO4 30 g/L NaCl 5 g/L NH4Cl 5 g/L Table 2-4. Composition of one liter of trace metal solution 1L trace metal solution Water 1000 ml EDTA 5 g FeCl3 498 mg ZnCl2 84 mg 0.1M CuCl2-2H2O 765 µl 0.2M CoCl2-6H2O 210 µl 0.1M H3BO3 1.6 ml 1M CoCl2-6H2O 8.1 µl 40 CHAPTER THREE: THEORETICAL COMPARISON OF MBG COST- EFFICIENCY BETWEEN EJECTOR (MBE) AND MODIFIED JAMESON CELL (MJC) The previous E. coli cell growth study using MBE and ring sparger offered a qualitative insight on the benefit of using MBE for gas-intense operations. The next step of this study is to quantitively analyze and compare MBE with other microbubble generators. The results presented in the literature review section suggest that microbubble-generation methods that use turbulence to form microbubble dispersions have the greatest potential of large- scale reactor applications. This section reviews the mechanisms by which turbulence results in bubble break-up to form microbubbles and describes operational properties of two types of turbulence-based MBGs (MBE and MJC) that seem most likely to cost-effectively produce microbubbles on a scale suitable for use in commercial-scale reactors. To estimate the relative cost- effectiveness of these two types, information from the literature was used to calculate estimates of the interfacial area for mass transfer generated per unit of electrical power required. 3.1 MBE and MJC Simulation Settings and Fundamentals Even though Chapter One briefly covered the mechanisms of the MBG for both MBE and MJC, there were still some more technical details that needed to be addressed to ensure a simulation with sufficient accuracy and fidelity. 41 The following Figure 3-1 demonstrates the simplified schematics of two reactor columns being aerated by MJC and MBE. For the MJC column, coarse bubbles rise to the top and form a foam layer. High speed liquid jet from a perforated plate would disperse the foam layer and create microbubbles. The microbubbles were then dragged downward through the column. The MBE operation follows the setup from section 2.3. Slow liquid was boosted to high ul and absorbs gas from the side neck of the MBE. High kinetic energy from the liquid phase shredded the gas-liquid mixture into microbubbles. The bubbly flow was then ejected into the reactor column. Figure 3-1. Simplified schematics of MJC (L) and MBE (R) columns 42 3.2 Bubble Size Prediction and Power Consumption Prediction for MJC and MBE The maximum bubble diameter in a CPLJ cell is correlated to the Ql because the bubble generation is depending on two-phase instability, which is governed by Ql [53]. Evans et al. correlated the maximum stable bubble diameter (dmax) inside a conventional CPLJ cell as a function of Critical Weber Number (Wec), σ, ρl, and energy dissipation rate per volume (ϕ) [53]: . 𝑊𝑒 𝜎 . . 𝑑 = 𝜌 𝛷 (3 − 1) 2 Evans et al. also found that the Dbub inside a plunging column is 61% of the dmax [53]. A recent study Hernandez-Alvarado et al. [61] correlated the dimensionless Dbub inside the MJC, D̃ b,m, as a function of jet Weber Number (Wejet), gas Froude Number (Frg), dimensionless height inside the column (𝐻 ), ε, and dimensionless superficial velocity (𝑉 ) according to a recent investigation [61]. . . . . . 𝐷 , = 71.1𝑉 𝜀 𝐻 𝑊𝑒 𝐹𝑟 (3 − 2) MBG utilization has associated capital and operating cost with different types of MBGs. The mass transfer rate calculated from Chapter One would be a determine factor of the size and number of mass transfer vessels to achieve certain kla. If the maximum achievable kla is low for an MBG, then multiple mass transfer vessels would be required to achieve the target kla. This would increase the overall capital cost of the MBG method. For the operating costs analysis, costs such as overheads, labor costs, and raw materials costs that are not directly associated with MBG operation were assumed to be the same. The primary cost of operating the MBG investigated in this paper was the electrical power consumed by the pump to overcome the frictional pressure drop involved in generating the microbubbles. That 43 power may be calculated using the following equation, where Ppump is pump power consumption, Δp is the pressure drop of the system, and Ql is the liquid volumetric flow rate: 𝑃 = ∆𝑝𝑄 (3 − 3) The main pressure drop for MBE occurs in the nozzle, where the gas is added to liquid flowing at a Re high enough that its turbulence is sufficient to reduce bubble size to the desired diameter. That MBE friction-induced pressure drop (Δpe) can be calculated with the Garcia correlation, as it can be seen in Equation 3-4, in terms of ejector length L, dnoz, ρl, ρmix, ul, umix, Fanning friction factor ffan, and fmix. [107] 𝐿 𝐿 ∆𝑝 = 2𝑓 𝜌 𝑢 + 2𝑓 𝜌 𝑢 (3 − 4) 𝑑 𝑑 For the MBE, the operation mode and protocol were designed around the E. coli cell growth experiment. The MBE would sparge gas-liquid mixture into the reactor vessel from the bottom of the vessel. This means the MBE pump would need to work against the hydrostatic pressure. The reservoir water was stationary and the MBE pump would also need to increase the ul to high enough value to sustain microbubble creation. Even though the MBE is short, the two-phase flow friction pressure drop was also taken into consideration. The total pressure drop across the MBE body would equal to the summation of all three components: ΔPMBE = ΔpHydrostatic + ΔpFriction + ΔpKinetic. The main pressure drop inside a MJC occurs as the liquid flows through the perforated plate. The plate pressure drop Δpm can be calculated using Equation 3-5 from the dimensionless pressure loss coefficient Eu, ρl, and ul [117]: 44 𝜌𝑢 ∆𝑝 = 𝐸𝑢 (3 − 5) 2 The value of Eu is a function of plate geometry, where both C0 and CC are constants correlated to the perforated plate geometry, and β is the square root of the quotient between plate and total pore cross-sectional area, tpore is the pores depth, and Dpore is the pore diameter [117]: 0.178 𝐶 = 0.5 + (3 − 6) 4(𝑡 /𝐷 ) + 0.355 ⁄ . 𝐶 = 0.596 + 0.0031𝑒 (3 − 7) 𝐶 (1 − 𝐶 𝛽 ) 𝐸𝑢 = (3 − 8) 𝐶 𝛽 For the independent variable range selection, a lower boundary 20 ft/s (6.1 m/s) superficial liquid velocity was calculated based on pipe erosion velocity inside carbon steel pipes. Pipe erosion velocity is defined as velocity where larger flow speed could cause mechanical erosion inside the pipe [118]. Pipe erosion velocity, uero, is dependent on ρmix and arbitrary pipe erosion velocity constant Cero [(lb/ft-s2)0.5]for two-phase bubbly flow inside a pipe [118]: 𝐶 𝑢 = (3 − 9) 𝜌 For a continuous, non-corrosive, two-phase dispersed bubbly flow inside a pipe, Cero is between 150 and 200 (lb/ft-s2)0.5 [118]. At 10% εg, atmospheric pressure, and room temperature, the calculated ρmix is around 60 lb/ft3 [118]. Based on this information, the calculated erosion velocity is around 20 to 30 ft/s for the microbubble ejector. This liquid velocity range was used as the upper and lower boundary for MBE ul. 45 The calculation scheme for the simulation can be seen here: Figure 3-2. Calculation scheme for MBE and MJC cost-efficiency analysis 46 3.3 Comparison Benchmark and Algorithm A quantitative comparison of performance properties and cost-efficiencies between MBE and MJC methods was conducted using the mathematical models described above in Figure 3-2. Ql and Qg were selected to be the initial independent variables. These variables were expressed in dimensionless forms as Re and εg. The ul across the apipe of the ejector nozzle under a Ql was used to calculate the Re. The εg was calculated by dividing Qg by the sum of the Ql and Qg. The apipe for flow through the MJC’s porous plate was taken to be the sum of the apipe of all pores in the plate. In simulations to compare the predicted performance properties of the MJC and MBE systems, the same apipe, Qg and Ql, liquid Re, and εg were used for both MBGs. The range of independent variables chosen for the simulations were chosen to match those used in published MJC studies [61]. The Ql through the MJC column body was calculated from published data [61]. Table 3-1 contains the constant values used in the simulations. Table 3-1. Physical constants assumed for the simulations Constant Value Unit ρl 998 kg/m3 μ 9.00×10-4 N-s/m2 σ 0.072 N/m ρg 1.18 kg/m3 Hernandez-Alvarado el al. used a column of square base with 0.1 m width and liquid height of 0.59 m[61]. The range of superficial column liquid velocity was 0.04 to 0.08 m/s, and that of superficial gas velocity was 0.004 to 0.02 m/s. The calculated Ql was 3.14×10-4 to 6.28×10-4 m3/s; 47 the calculated Qg was between 3.14×10-5 and 1.57×10-4 m3/s. For this simulation, ul for the MBE was calculated based on the Ql and apipe, and apipe along with dnoz were calculated by known Ql and ul. Liquid flow rate was converted into ejector nozzle Re, while Qg was converted into εg to keep the independent variables dimensionless. Hernandez-Alvarado et al. also reported the individual pore diameter (0.8 mm) and applicable Wejet value range, which was between 5.4×104 and 2.2×105 [61]. By the definition from Hernandez- Alvarado el al., Wejet = ujet2Dporeρl/σ. The calculated ujet value range based on the Wejet definition and pore diameter is between 6.24 and 12.6 m/s. Assuming that these ujet values correspond to the same upper and lower boundaries of the published Ql value from the same literature, the area ratio between total pore areas and the cross-sectional area of the column body was calculated based on conservation of mass flow in and out of the column. The β value was calculated from the pore/column area ratio. The pore depth/diameter ratio was assumed to be 1, which falls within the valid parameter range provided by Malavais., et al [117]. The calculated area ratio was that column cross-sectional area was 156 times of the total area of all the pores on the perforated plate. These parameter values were used to calculate Dbub values for both for the MBG systems, using Equation 1-7 for MBE and Equation 3-2 for the MJC system, as a function of the liquid phase Re and gas void fraction. The Dbub and εg values were then used with Equation 1-1 to calculate a for gas mass transfer. Pressure loss and power consumption for both the MBE and MJC systems were then calculated by solving Equation 3-4 for MBE and 3-5 for MJC. The estimated a was normalized by the P required under the same flow conditions to give a measure of cost- 48 efficiency for mass transfer. A kla value of 2900 h-1 that was suggested by ARPA-E as a design target for Electro-fuels fermentations [3] was used to for additional design calculations. Using Henry’s Law, the ARPA- E target kla is converted into an Oxygen Uptake Rate (OUR) to calculate the maximum theoretical reactor volume (Vmax) that can be sustained by the MBG: 𝑄 𝑄 𝑉 = = (3 − 10) 𝑂𝑈𝑅 𝑘 𝑎𝐶 ∗ Equation 3-10 is used to calculate the theoretical maximum volume of the liquid that can be supported under certain gas and liquid flow rate conditions for both MBE and MJC methods. The results can be seen in the next section 3.4 Comparison Results Figure 3-3 is the simulated Dbub comparison between MJC and MBE methods. The MBE produces smaller bubbles at higher Ql. At higher εg, MJC method produces smaller microbubbles. As the ul increases, the Dbub produced by MBE reduces at third power due to the increase in turbulence energy provided by liquid phase. MJC-generated Dbub does not change much with respects to Re, but marginally increases as the εg increases. The strong effect of liquid Re seen for the MBE system can be attributed to Equation 1-7, which predicts that Dbub varies inversely with b ul 3 . 49 Figure 3-3. Db correlation to liquid Re and ε for MJC and MBE method Figure 3-4 demonstrated total predicted a as a function of both εg and Re. The Dbub results from Figure 3-3 were then used to calculate a value for the EJ and MJC systems as a function of Re and εg as shown in Figure 3-4. For the MBE system, a steadily increases as the Ql increases but remains almost constant when altering Qg. For MJC method, the total a increases as the Qg increases. The increase in a is insignificant as ul increases in MJC columns. Consistent with Figure 3-2, the MBE system is predicted to generate higher a value than MJC system at higher Ql and lower gas fractions and vice versa. 50 Figure 3-4. Gas-liquid area (a) correlation to liquid Re and ε for MJC and MBE method The primary operating cost for the two MBG systems would be electrical power required to generate the high liquid velocities required for bubble disintegration. The power requirements to overcome the Δp of the two systems were calculated and the result can be seen in Figure 3-5. For MBE system, a full balance on pressure-drop through the MBG using Bernoulli's principle was also conducted. Unless the Re was high, the power consumption for the MBE system would be higher than MJC. However, since the porous plate has a higher friction coefficient than straight pipe, the friction-induced pressure drop for MJC would increase faster than MBE system. However, it should be emphasized that these initial power calculations are approximate and further studies are required to more accurately predict the Ppump for the two systems. 51 Figure 3-5. Power consumption (Ppump) correlation to gas and liquid flow rate for MJC and MBE method The a was then normalized by the Ppump required under the same flow conditions. Figure 3-6 demonstrates the relative cost-efficiency for mass transfer as a function of the two independent variables. The a per Ppump input as a function of Ql. This parameter is an indicator of how ‘energy- efficient’ an MBG is; the total power consumption being the cost of the system while the a being the benefit. The higher this factor is, the more efficient a microbubble sparging system would be. MBE method has lower power consumption per unit volume comparing to MJC method at higher Re and lower εg. The threshold where MBE cost-efficiency surpasses is marked with a red line in Figure 3-6. Due to the significant quantity of parameters involved in this simulation, an arbitrary, 2nd order polynomial correlation was used to denote this threshold: Re L = 2.84×105×εg2 + 1.62×105×εg + 3.32 ×104 52 Figure 3-6. Power-efficiency (a/Ppump) correlation to liquid Re and ε for MJC and MBE Other than theoretical calculation for cost-efficiency, the scale-up difficulty of a microbubble sparging system must also be taken into consideration. An individual MBE can be modeled as an individual mass transfer cell inside a large reactor. Multiple MBEs in triangular pitch formation and in multiple layers can in theory aerate large volume of liquid inside a reactor. To aerate industrial-scale reactor vessels, MBEs can be installed next to each other horizontally and stacked vertically inside a single, large vessel as shown in Figure 3-7. Each MBE- generated jet cone can be modeled as individual mass-transfer cell and the cells can collectively provide high kla industrial-scale reactor vessel. 53 Figure 3-7. Vertically and horizontally stacked large scale MBE formation for fermentation. According to Li, multiple smaller, individual MJCs with their own stir blades and perforated plates need to stack on each other to achieve a height-to-diameter ratio on par with conventional bubble columns [1]. This stacking arrangement introduces additional capital cost in column in form of additional column material and construction cost compared to the formation shown in Figure 3-7. Also, SDS was added as a surfactant to stabilize the bubble formation in both the patent (100 ppm) [1] and the bubble size estimation (10 ppm) [61] for the MJC. Previous study on downflow column similar to MJC pointed out that liquid jet would lose penetration kinetic energy when hitting liquid phase, thus the MJC method could also face challenges when implemented on a tall column [54, 119]. Liquid jet’s kinetic energy carries the microbubbles downwards, while the buoyancy force of microbubbles counter-balances the downwards force. As the kinetic energy depletes, the downwards force would also decrease, thus decreasing the downward flow speed of the microbubbles and increase the bubble residence time. As the reactive gas inside the microbubbles are expended while descending into the column, the microbubbles are left with inert gas inside. The descending microbubbles with richer inert gas 54 could interact with the rising coarse bubbles with richer reactive gas. The microbubbles could dilute the coarse bubbles and reduce the overall mass transfer drive force of reactive gas as the bubble residence time increases. This interaction indicates that MJC MBGs can only be used on relatively short columns due to mass transfer reductions. The bubble coalescence inside a MJC column must be taken into consideration as well. Surfactant like SDS was used to improve the bubble sizes generated inside the MJC, albeit at the cost of mass-transfer drive force [8]. Inside an MBE bubble column, the bubbles would disperse away from each other once leaving the ejector, thus reducing the chances of bubble collision and coalescence. In the patent US20140212937, the author directly indicated that the bioreactors operating with the MJC will be relatively smaller compared to conventional reactor columns [1]. This means the capital cost for implementing MJC microbubble column would be high since multiple smaller columns with their own sparging systems must be installed in order to achieve same level of O2 transfer rate of ejector types. Previous simulation results from Figure 3-6 demonstrated that the operating cost of MBE method would be lower than MJC method at higher Re. All factors considered, MBE method is overall the more economically attractive method to achieve higher kla for gas-liquid reactions compared to MJC method. 55 3.5 Conclusion Microbubble-generation methods based on intense liquid turbulence appear to offer the best combination of practicality, scalability, and cost for generating extremely high kla values in commercial-scale reactors. Mathematical models based on literature data and empirical correlations for two turbulence-based methods (MBE and MJC) allowed the relative performance characteristics of these two methods to be predicted and compared. Key dependent variables (interfacial area and power consumption required to generate that area) were calculated as function of dimensionless independent variables (Re and ε). The MBE method was predicted to generate higher interfacial area values than the MJC method at higher Re and lower ε values, whereas the MJC method would generate higher interfacial area values than the MBE method at lower Re and higher ε values. Moreover, the MBE method was predicted to achieve higher maximum kla value than the MJC method under the conditions studied. For equivalent Re and ε values, the MBE method was predicted to give higher cost efficiency values (assessed as the ratio of kla to power consumption) across the entire range of variables studied. This result suggests that turbulence generated by high Re pumping of a gas-liquid mixture through a tube is more efficient at generating microbubble dispersions than the turbulence generated by high Re pumping of a liquid through an orifice, through a gas layer, and into the surface of a continuous liquid phase. From an operational perspective, the MBE system offers additional advantage over the MJC 56 system for large-scale reactors that require extremely high kla values. The first disadvantage arises from the countercurrent flow of the gas and liquid feed streams in the MJC system. When the gas phase is not completely consumed in the reaction (e.g., if air is used as a source of O 2) or if the reaction generates a gaseous product (e.g., CO 2), once the reactive gas has been transported into the liquid phase, the resulting spent microbubbles containing an unreactive gas would result. Coalescence of these spent microbubbles with one another would resulting in larger, faster rising spent bubbles that would have a longer residence time than the microbubbles and thus could occupy significant reactor volume and reduce the reactor’s volumetric productivity. Coalescence of spent microbubbles with the larger bubbles fed into the bottom of the MJC reactor would dilute the reactant gas in those bubbles and thereby decrease the mass-transfer driving force and volumetric mass-transfer rate. The proposed use of surfactants to reduce coalescence in MJC systems has the potential to increase production costs and complicate the product-purification process. 57 CHAPTER FOUR: BUBBLE SIZE PREDICTION FOR MICROBUBBLE EJECTOR 4.1 Background: Experimental Validation of the Bubble Size Prediction Model and the Two-phase Friction Factor Model The accuracy of the previous simulation results and the future implementation of MBEs during production requires the capability to predict the effective Dbub of microbubble dispersions generated by an MBE as a function of Ql and Qg values. An objective of this study was to develop that capability by measuring Dbub over a useful range of Ql and Qg values and then develop a mathematical model able to reproduce trends in the experimental results. Optical methods based on laser diffraction [120] or image analysis of high-speed camera images [121, 122] are commonly used to measure Dbub in aqueous dispersions. However, given the extremely small Dbub and high velocity with which the microbubble dispersion exits the MBE, these approaches could not be used without specialized equipment that was not available for this study. For that reason, a novel, indirect method was developed to estimate the effective Dbub generated by MBE. The method entailed measuring the mass transfer rate resulting from microbubble dispersions generated by an MBE, and then calculating the effective Dbub that would be predicted to give the measured mass transfer rate. 58 4.2 Indirect Method to Estimate Dbub from Dissolved Oxygen Profile When a highly turbulent stream of microbubbles exiting an MBE is injected into low- turbulence flow environment, such as a bubble column, the ul drops precipitously due to liquid- liquid entrainment [109]. As a result, the turbulent energy intensity that disperses larger bubbles into microbubbles within the MBE is dissipated, and bubble coalescence results in a dramatic increase in Dbub, and corresponding decrease in a, according to Equation 1-7. For this reason, measurement of Dbub is difficult in low-turbulence environments unless microbubble coalescence can be prevented. Bredwell and Worden [8] circumvented this problem adding a surfactant to the liquid phase prior to forming the microbubble dispersion. The highly polar surfactant molecules coated the microbubbles and created a local surface charge that repelled nearby microbubbles and thus inhibited coalescence. As a result, the authors were able to inject surfactant-stabilized microbubble dispersions into a flow system consisting (i.e., a bubble column) and measure the mass-transfer properties of the microbubble dispersion with minimal microbubble coalescence. They also used a mathematical model to analyze the experimental results and show that, although the surfactant layer provide the beneficial effect of stabilizing the microbubble dispersion against coalescence, it also had the detrimental effect of increasing the interphase mass transfer resistance and thereby decreasing the kl value.[8] In this study a novel method was developed to inhibit coalescence of a microbubble dispersion generated in an MBE and measure the mass-transfer properties of the dispersion without using a 59 surfactant. The approach involved injecting the microbubble dispersion generated in an MBE into a flow system that consisted of a cylindrical tube having the same inner diameter as the MBE that had ports into which an oxygen probe could be mounted. As oxygen transfer occurred the dissolved oxygen concentration in the liquid phase would change, and my moving the oxygen probe to different distances down the tube, the steady-state dissolved oxygen profile (DOP) across the tube that could be measured. Because the tube had the same inner diameter as the MBE, the degree of liquid turbulence would be expected to remain approximately the same in the MBE as across the entire flow system. As a result, the maximum stable bubble size, and hence the effective Dbub value, would also be expected to remain approximately constant across the flow system, even in the absence of a surfactant. Moreover, at the very high Reynolds numbers generated by an MBE, axial mixing would be expected to be negligible, allowing a relatively simple plug-flow-reactor (PFR) mixing model to be used to mathematically model two-phase flow through the system and predict the expected steady-state oxygen profile for an assumed Dbub value. A similar PFR model was shown to be valid for the flow system used by Bredwell and Worden. [8] 4.3 Mathematical Model of the DOP Inside the PFR System The mathematical model developed to calculate the DOP across the PFR system is described below. 60 4.3.1 Setup and Overall Structure The following physical phenomena were considered in developing the PFR model: 1. Both the Ideal Gas Law model and the Van der Waals model were used to simulate the effect of pressure changes on the specific gas volume of gas contained in the microbubbles. 2. Henry’s Law was used to calculate the equilibrium concentration of the transferred gas at the gas-liquid interface from the bulk concentration in the gas phase. 3. Constant kl value was assumed since the liquid and gas fluids were the same for all the experiments. For this study, air and water were selected for mass transfer operation. 4. PFR mixing model was assumed for the equipment, since the fluid velocity would be very high and flow through a narrow tube at very high Re. 5. Individual microbubbles were assumed to be spherical and identical. Since the previous Dbub Equation 1-5 was developed for Sauter mean bubble diameters. 6. Constant ul inside the tube, and the relatively low εg does not affects ul. This assumption is dependent on assumption 4. Some additional physical phenomena were taken into consideration after observation on previous assumptions. These are the aspects that worth investigating to ensure model fidelity: 1. The Dbub might change along the way. Mass transfer between gas and liquid phase would change the bubble gas species content. Pressure change along the PFR would also change the Dbub depending on the equation of state. 2. The internal pressure of the bubbles changes as the bubbles travel down the pipe due to 61 friction induced pressure drop (ΔPfri) and Dbub change. 3. The dissolved gas content, CL, including both N2 and O2, will change along the way as mass transfer occurs. This is the same for C* of both species since the pressure change as well. 4. The O2 ontent inside bubbles will change as well as the O2 is being dissolved. After examining the previous assumptions and crucial physical phenomena, a five variable ODE group was developed (Equation 4-1, 4-4, 4-9): The CL and C* for both O2 and N2, plus internal pressure. The five initial conditions are calculated using the following protocol: 1. Initial pressure was calculated based on preset distance (Equation 4-1). 2&3. Initial O2 and N2 molar content inside an individual gas bubble, calculated using ideal gas law and later Van der Waals, initial condition 1, and calculated initial bubble radius from Equation 1-7. 4. Initial dissolved O2 was maintained to be 5 mg/L using O2 starvation method with pure N2gas. A tank was used to collect water and N2microbubbles were sparged in the tank to reduce the DO level in the tank till 5 mg/L. 5. Initial dissolved N2 was calculated using pure N2 and initial pressure with saturation. 62 4.3.2 Flow Model Development Two-phase pressure drop was a core component of other physical phenomena such as mass transfer drive force and bubble sizes. Equation 1-11 is the fmix equation, while Equation 1-6 is the one phase ffan equation. Both equations were needed to calculate the pressure drop across the MBE, but only Equation 1-11 was needed to calculate the PFR pressure drop. The usage of Equation 1- 11 and Equation 1-6 was addressed in detail in section 3.2 (Equation 3-4), as MBE nozzle is only half mixture flow because the gas entry point is in the middle, while PFR is full mixture flow within the tube. The pressure drop across MBE was calculated using Equation 3-4, while the pressure drop across the entire PFR was calculated using the following equation: 𝜕𝑃 𝑑𝑃 𝜌 𝑣 𝑓 = =2 (4 − 1) 𝜕𝑧 𝑑𝑧 𝑔𝑑 The next equation is the individual bubble species content inside a single microbubble. The mass balance of and individual gas species i inside bubble with radii of rbub is shown here: 𝜕𝑛 = 𝑘 (4𝜋𝑟 )(𝑃 ∗ − 𝑃 ) (4 − 2) 𝜕𝑡 Since the mixing model was assumed to be PFR, the time derivative can be converted into a length derivative: 𝜕𝑧 1 𝑢 =𝑢, = (4 − 3) 𝜕𝑡 𝑑𝑡 𝑑𝑧 Combining the previous equations yields the following expression: 𝜕𝑛 𝑢 = 𝑘 (4𝜋𝑟 )(𝑃 ∗ − 𝑃 ) (4 − 4) 𝜕𝑧 Applying ideal gas law and Henry’s Law to the Equation 4-4 generated the ODE for single gas species content inside individual bubbles when using ideal gas model. Pbub is the internal 63 bubble pressure, Vbub is the individual bubble volume, ntotal is the total species concentration in the bubble, R is gas constant, and T is temperature: 4 3𝑛 𝑅𝑇 𝑃 𝑉 =𝑛 𝑅𝑇 = 𝑃 𝜋𝑟 ,𝑟 = (4 − 5) 3 4𝜋𝑃 However, per the previous discussion, ideal gas equation of state does not suit gas at high pressure very well. Van der Waals equation of state was then utilized in this study, where Vana was 1.34×10-9 m3/mol and Vanb was 3.5×10-6 m3/mol: 𝑛 𝑅 𝑇 3𝑉 𝑉 = +𝑛 𝑉𝑎𝑛 , 𝑟 = (4 − 6) 𝑃 𝑅 𝑇 + 𝑉𝑎𝑛 𝑃 4 The following Henry’s Law equation with Henry’s constant H was used to calculate the C* for both O2 and N2: 𝑛 𝐶∗=𝑥𝐻𝑃 ,𝑥 = (4 − 7) ∑𝑛 Combining the previous equations from 4-2 to 4-6 generated the mass balance of gas phase species inside individual bubbles. The gas phase species mass balance was conducted for both N 2 and O2 inside the bubbles: 𝜕𝑛 𝑘 (4𝜋𝑟 ) = (𝑃 − 𝑃 ∗ ) (4 − 8) 𝜕𝑧 𝑢 The internal bubble pressure should be the summation of friction induced pressure drop (Pfri), and Laplace pressure (Plap): 𝜌 𝑣 𝑧 2𝛾 𝑃 =𝑃 +𝑃 = 2𝑓 + (4 − 9) 𝑔𝑑 𝑔𝑟 Mass balance for the liquid phase was also developed between the liquid phase and the bubbles. In this case, the mass transfer equation on the right-hand side need to multiply the bubble density 64 (Nbub). Also, the direction of mass transfer needs to be inversed since for this equation, the gas species are moving from gas phase to liquid phase. For this equation, no additional simplification is necessary: 𝜕𝐶 , 4𝜋𝑘 = 𝑟 𝑁 𝐶 , − 𝐶∗ (4 − 10) 𝜕𝑧 𝑢 Likewise, there are also two variations for both O2 and N2 to fulfill the ODEs associated with initial condition 4&5. 4.4 Experiment Setup, Design Equations, and the Equipment Selections for the DO Experiment 4.4.1 Tubing For the tubing selection, the tube roughness and ffan needed to be investigated. In this particular PFR system, the ffan needs to be constant at Re > 50000 to ensure consistent ffan in the PFR for model accuracy. This requires a relative pipe roughness around 0.01 [123]. According to Moody chart in Figure 4-1, at d = 0.008 m, the true roughness should be larger than 8×10 -5. 65 Figure 4-1. Moody diagram [123] The commercially available plastic tubing material’s absolute roughness is around 0.002 mm or 2×10-6. At d = 0.008 m, the relative roughness is around 2.5×10-4. ffan is around 0.02. Some common material surface roughness can be seen in the following Figure 4-2 [124]. Figure 4-2. Common material surface roughness in both mm and inches 66 The main technical requirement that needed to be satisfied when selecting the plastic tubes was the pressure and temperature ratings for this experiment. Since this experiment was conducted at room temperature, temperature rating was not a huge concern for common commercial materials. The gas delivery pressure was around 70 to 80 psi due to high Pfri at the start of the PFR tube with a 5 meter long and 7 mm width. To sustain such high pressure of operation under the flow rate and tube length, a particular type of polymer tube with braid reinforcement was selected to ensure the structural integrity of the flow system was not compromised by the tubing strength under pressure. A smaller tube with similar braid reinforcement was used to connect the gas line to the regulators on the gas cylinders. 4.4.2 Pumping The pumping selection had two key parameters needed to be taken into consideration: fluid flow rate and the pressure drop. Both parameters affect the pump head and by extension the pump power requirement. Both static and dynamic heads were calculated for the pump head. There was no static head in the PFR setup since both the pump and water tanks were on the same level. The only significant static head was the height difference between floor and water outlet. This was mitigated by elevating the tube and the pump to ensure the water entrance and exit are on the same elevation. The dynamic head loss consisted of two parts: Pfri and fitting loss. Since the housing unit for sensors was a T-shape fittings along the way for probe insertion, the fitting loss needed to be 67 considered. The k-value for each straight-through T-fitting is 0.1, while it is 1.2 when it’s through a side branch. The total fitting head-loss, hf, was a function of velocity head and fitting loss coefficient (k-value) kfit: 𝑘 𝑢 ℎ = (4 − 11) 2𝑔 The friction energy head loss, hpipe, due to the pipe flow was calculated using the following equations as a function of fmix, dnoz, umix, tube length Ltube, and conversion factor gc: 𝐿 𝑢 ℎ =𝑓 (4 − 12) 𝑑 𝑔 From AiChe handbook, the total power consumption (W) equation for a pump with head Hpump is a function of fluid volumetric flow rate Q, fluid density ρ, gravitational constant g, and pump efficiency 𝛈: 𝑄𝜌𝑔𝐻 𝑊= (4 − 13) 3.6 × 10 𝜂 At 6.1 m/s ul, 10% εg, 8 mm dnoz, and 6 m Ltube, the calculated Hpump was 41.44 m, the pump power was 127.7 watts, and the pump load was 0.000307 m 3/s. Converting units to common pump vendor units, the units were: 135 ft, 0.17 hp, 4.86 gpm. The calculated pump had relatively small load but very high friction head. Conventional centrifugal pump does not deliver such high head at this flow rate range. Peripheral Impeller pump fits the description better-high pump head with relatively low flow rates [125]. The selected pump was PK 100 peripheral impeller pump from Pedrollo. The pump motor frequency was 60 hz. The rated amperage was 6.2 A at 220 V. The pump motor was rated for 1.5 68 hp (1.1 kW). The rotation speed of the pump motor was 3450 rpm. To control the pump operation speed and flow rate, a variable frequency drive (VFD) was installed to control the liquid volumetric flow rate. 4.4.3 Dissolved Oxygen (DO) Probe and Sensor The DO sensor (DOS) needed for this study should to be fast reactive, small, and pressure resistant. A galvanic DOS was selected for this experiment. Galvanic DOS requires no warm-up time, and is more stable and accurate at lower DO level than polarographic probes [126]. Galvanic probe requires constant liquid flow or disturbance since O 2 is actively being consumed by the galvanic cell, but that was not a problem for this study, since high ul water flow was being flushed around the DOS constantly during the experiments. Smaller outer diameter ensures the DOS can be inserted in-line without introducing a very large side for the T-connector housing unit. A large probe side neck inside a T-connector housing unit could change the turbulence of the bubbly flow. DOS marketed for aquatic studies are required to withstand high hydrostatic pressure in natural lakes/oceans. Atlas Lab-Grade DO Meter was selected for this experiment. This particular probe can withstand various temperature (1 ~ 60 ℃), high pressure (34 atm, seven time more than max pressure for this experiment), and the maximum DO level is 100 mg/L (3 times more O 2 at saturation equilibrium). The DOS was connected to a dedicated processor provided by the same vendor to compensate for pressure change throughout the experiment. The processor board was 69 connected to an Arduino board via co-axial cable. The connection scheme and Arduino codes were both provided by the same vendor. The galvanic cell is composed of pure silver cathode and zinc anode. The electrolyte was procured from Atlas Scientific as recommended by the vendor. The probe was then inserted into a T-connector housing unit via a dedicated ring cap to seal off the water. 4.4.4 T-shaped Fitting and Barb Hose Connectors A T-shaped fitting was chosen to connect the tubes and to insert the DOS as the housing unit. A 304 stainless steel connector with three female 1/2-inch NPT connectors was chosen as the main fitting. Two brass 5/16 inches inner diameter barb hose connector to 1/2-inch male NPT connectors were selected to connect the PFR tubes on both ends of the T-connector. Since the interior diameter of these T-connectors are larger than the flow diameters, flow through these connectors could cause some local mixing that disrupts the flow pattern inside the tube. To avoid such disruptions, the total number of the connector for probe insertion was minimized. During the experiment, the probe would stay inside the same T-connector, while the tubes connected on both end of the connector would change to different length so the housing unit and the DOS move to a different distance from the pump via changing the PFR length at both ends of the housing unit. The converters were attached to the T-connector, and the Atlas DOS was inserted into the connector after being secured using a dedicated ring cap to stop leakage. Additional Teflon tapes 70 were applied on all connectors throughout the PFR to prevent leakage under high pressure. The following Figure 4-3 shows the final product of the DO measurement housing unit using the T- connector. Figure 4-3. The DOS connected to the main flow line with two barb hoses for the PFR 4.4.5 Variable Frequency Drive A variable frequency drive (VFD) was used to control the Ql of the pump instead of bypassing or throttling to avoid complicated fluid dynamics or fluid heating up. The VFD needs to accept the to-wall power outlet (120 V, 1 phase) and converts the current to the pump motor (220 V, 3 phases). The VFD selection was based on the parameters of the pump motor. An ATO SKU GK3000-SP1S1-2d2 VFD was selected as the VFD between the mainline and the pump motor. 12 AWG wire that is rated for 20 amps was chosen as power line and a 20 amps circuit breaker was installed on the main-line. The pump was further grounded on the installation rack. The O-ring terminals were also welded on the connectors to ensure movement of the equipment would not risk disconnecting the lines. 71 4.4.6 Gas Flow Meter The experiment was conducted under higher pressure compared to atmospheric pressure. The value read on the gas flow meter, which was calibrated under atmospheric pressure, does not represent the true gas flow rate. A correction equation is used here to correctly correlate the read gas flow meter values to the true gas flow rates with regards to the gauge pressure Pgauge and rotameter flow rate reading Qread [127]: 𝑃 𝑄 =𝑄 (4 − 14) 𝑃 4.4.7 Pressure Probe A pressure probe from the same vendor that provided the DOS was procured to measure the in-line pressure at different position inside the PFR. The measured pressure profile was used for two purposes: 1. The DOS needed pressure compensation during operation to obtain an accurate reading of the true DO value under higher pressure at a given position inside the tube. The braid that was being used to reinforce the plastic tube could have changed the surface roughness and the friction factor so it would be imprudent to use the earlier equation to estimate the pressure. 2. In order to accurately estimate the Dbub inside the tubes, a f was needed to calculate the turbulent energy provided. The 2P-Chen correlation used during the model development was never verified inside a narrow tube like this. An experimental validation was needed to observe whether or not Garcia correlation is suitable for this experiment. 72 The probe used for the pressure profile measurement used was Atlas Mega A-100 pressure probe, from the same vendor that provided the DOS. The connection was via Arduino and the code was provided by the vendor as well. The probe is inserted into the same T-connectors via an NPT adapter. 4.5 Mass Flow Experiment to Calibrate the VFD Due to the employment of a VFD, the Ql when using the pump at various VFD speed setting needed to be verified using a bucket and stopwatch method. The results are recorded in the following Table 4-1 and 4-2. Table 4-1. Ql measurement of a 7 mm ID, 5 m long tube attached to a 7 mm MBE Time (s) Volume (ml) VFD Speed (hz) Ql (ml/s) ul (m/s) 26.1 6980 45 267 5.97 27.73 7440 45 268 5.99 26.69 7100 45 266 5.94 24.46 7310 50 299 6.68 20.21 5990 50 296 6.62 23.79 7010 50 295 6.58 22.25 6615 50 297 6.64 20.85 6800 60 326 7.28 19.76 6480 60 328 7.32 21.36 6860 60 321 7.17 73 Table 4-2. Ql measurement of a 10 mm ID, 4 m long tube attached to a 10 mm MBE Time (s) Volume (ml) VFD Speed (hz) Ql (ml/s) ul (m/s) 18.26 11390 60 624 7.94 18.2 11260 60 619 7.88 17.46 10580 60 606 7.72 16.03 9230 55 576 7.33 15.99 9200 55 575 7.33 16.49 9650 55 585 7.45 16.42 8690 50 529 6.74 15.96 8420 50 528 6.72 16.62 8670 50 522 6.64 19.88 9440 45 475 6.05 19.68 9360 45 476 6.06 19.75 9420 45 477 6.07 Based on the measurement of the Ql through the entire PFR under different VFD speed, the VFD speed for the various experiment parameters were selected. The frequencies of the pumps are selected to keep the superficial velocities of the water through the ejector nozzle to be 6, 6.6, and 7.2 m/s. The VFD speeds corresponding to these superficial ul were recorded in the following Table 4-3. Table 4-3. ul measurement of a 7 mm ID, 5 m long tube attached to various MBEs MBE Diameter (mm) 6 m/s 6.6 m/s 7 m/s 7 45 50 60 10 45 49 54 74 4.6 Mixing Model Experiment 4.6.1 Theory Marshall Bredwell calculated the Peclet number (Pe) of his column and found Pe was high enough to ignore side mixing [8]. Here is the protocol to obtain Pe from measurable values according to Fogler [128]. Initially, spike the flow system (column, PFR, etc) with a concentrated salt solution to obtain the concentration curve using the concentration measurement device (i.e. conductivity probe). The time at which the signal peaks, the time interval between data collections, and the signal readings could be converted into mean residence time (tm) and variance (σ). Using a closed-closed Pe correlation, variance and mean residence time can be converted into Pe [128]. First, collect the concentration profile C(t), which was referred as C-curve, from the photometer. Sum up the area under the C-curve to obtain the residence time distribution (RTD) curve E(t), which is now referred as E-curve. The equation for E-curve under a pulse-signal experiment is shown in Equation 4-15: 𝐶(𝑡) 𝐸(𝑡) = (4 − 15) ∫ 𝐶(𝑡) 𝑑𝑡 Then, integrate the RTD curve to calculate the tm and σ: 𝑡 = 𝑡𝐸(𝑡)𝑑𝑡 (4 − 16𝑎) 𝜎 = (𝑡 − 𝑡 ) 𝐸(𝑡)𝑑𝑡 (4 − 16𝑏) For a closed-closed system, the Pe correlation is: 75 𝑡 2 2 = − (1 − 𝑒 ) (4 − 17) 𝜎 𝑃𝑒 𝑃𝑒 C-curve, E-curve, and the integrations were done using trapz function in Matlab. The final Pe correlation solver was calculated using fzero in Matlab. If Pe is larger than 10, it would be safe to assume that the flow has very little to no back mixing and can be categorized as PFR condition inside a column [8]. However, Bredwell experiments were conducted inside a wide column with a relatively slow Ql across the column compared to the PFR [8]. The previous pulse-signal experiment setup was not suitable for this experiment. The ul was so fast that the injection system could not inject the salt tracer fast enough to consider the signal to be pulse through observation. A step input was chosen and the same procedure between Equation 4-14 and Equation 4-17 was used to calculate the Pe. The Pe threshold for plug flow under a step-signal input is 78.125. The reciprocal of this number is referred as Vessel Dispersion Coefficient (VDC). 4.6.2 Setup A custom-made photometer along with an appropriate dye were needed to measure the dye spike signal. The reason why photometer instead of conductivity meter was selected was because of the slow response time on the available conductivity meter. This experiment was very sensitive to the response time due to high ul and optical signal responds faster compared to the available conductivity probe. A generic photometer contains a light source, a sensor, and a processor to convert the sensor 76 signal to data that can be processed with data processing software. To ensure the measurements were accurate with little to no external interference, the light source, measurement section, and the sensor should be encased in a dark housing with on environment light shining upon it. The processor is not needed inside since it does not participate in the actual photon signal collection. The dye selected for this experiment should meet the following criteria: 1. It should be miscible with the main liquid feed, which is water in this case to ensure no secondary phase separation deviating the measurement accuracy. 2. It should be chemically safe to dispose through main pipeline instead of a dedicated hazardous chemical waste cabinet. The total Ql is too high and the logistics would be challenging to dispose high volume of hazardous dye water via a dedicated container. 3. It should not be reactive with O2 or N2, since oxidation or nitration reaction could potentially change the experiment results. There will be bubbles being sparged into the pipe during mixing model experiment and the mass transfer aspects should not affect the measurement results. After further review on existing literature, patents, and DIY instructions, the following equipment were considered for the photometer construction. The processor kit selected was ‘ELEGOO Mega 2560’ from Elegoo. This vendor provided the starting kit for an Arduino processor and assorted accessories. Arduino processor could convert the measured signal into data and be processed by Matlab. The accessories include essential items such as resistors, DuPont wire, various controllers, etc. The light sensor selected was an 818 Series Photodetector from Newport. The calibration was 77 done by the manufacturer before delivery. The photodetector contains a photodiode that was attached to a standard BNC connector. The BNC connector was connected to a digital Newport Power Meter (model 815). The power meter received the signal and converted the signal into analog output between 0 to 2 volts. The signal was then sent to Arduino and was recorded on the computer for analysis. An Arduino board could accept and record voltage between 0 and 5 V in 10 bits. 1024 data points are evenly distributed to represent the reading. In this case, value 0 on Arduino represents 0 V, while 410 represents 2 Vs. The dye selected was Allura Red AC (Red 40). This type of dye was used for food coloring and was considered to be safe to dispose without using a dedicated hazardous collection vessel [129]. Figure 4-4. Allura Red AC (Red 40) chemical formula and absorption spectrum[129] 78 The LED light source was a 520 nm RubyLux 2nd Generation All Green LED. The diameter of the light is 4.9 cm and the power requirement is 120 V. This light was selected because it emits 520 nm light, which is close to the 504 nm optimal absorbance wavelength of Red 40. A casing for the measurement section was produced in order to cut off environment lights and ensure the signal received by the light sensor is proportional to the light passes through the measurement medium. A circular hole with the same diameter as the LED light was cut on a piece of wood to stabilize the LED light. A channel was cut open in the middle of a wooden block to allow a transparent tube to flow through. A cuvette with bottom cut out was installed on the outside of the tube to ensure the light pathlength is straight. An additional layer was added in the middle with a small hole cut open to ensure the light pass through was both consistent and the light intensity is low enough to be recorded by the sensor. The schematic of the casing can be seen in Figure 4-5. Figure 4-5. The schematics of the prototype photometer casing along with light source and sensor casing. 79 Once all the equipment were selected, the equipment for the mixing model experiment was constructed. Figure 4-6 shows the simplified schematics of the equipment used for mixing model experiment. Water was boosted inside the impeller pump for high ul and then sent through the MBE. Since the PFR was very long, the starting pressure at MBE would be high around both the gas and dye injection point. High pressure gas inlet was connected with a pressure-resistant gas tank loaded with the dye to push the dye to the injection point after MBE. The fluid was then pushed through the PFR and detected by the prototype photometer. The optical signal for clear water and dye water were recorded using Arduino. Even though the DOS was installed during this phase, it was not used for measurement. The DOS was left in place to ensure the mixing model measured does include the DO measurement housing chamber shown in Figure 4-3 and maintains model fidelity. Figure 4-6. Schematics for the mixing model experiment 80 4.6.3 Calibration The analogue readings of the 815 Newport Power Meter on PC end were directly proportional to the light intensity received by the sensor. According to Beer’s Law, the green light intensity passing through the Red 40 dye water should be correlated to the Red 40 concentration. By calibrating the sensor voltage/Arduino reading of the light sensor with the Red 40, a theoretical absorbance coefficient of Beer’s Law could be calculated. If the sensor voltage reading follows Beer’s Law closely with respects to the Red 40 dye concentration, then the sensor voltage could be used to measure the Red 40 dye concentration. According to Beer’s Law, the absorbance, A, is a function of light path distance l, dye concentration cdye, and molar attunement coefficient εatt. 𝐼 𝑉 𝐴 = log = log =𝜀 𝑐 𝑙 (4 − 18) 𝐼 𝑉 The light received by the sensor was directly proportional to the voltage received by the Arduino board. Pure water’s voltage reading was used as reference point, and the dye reading voltage was used to calculate the A. With known cdye and l (the inner diameter of the tube), a εatt could be calculated. Rewrite the above equation yields exponential relationship between the sensor voltage and the dye concentration: 𝑉 =𝑉 10 (4 − 19) Baseline light intensity and its corresponding sensor voltage Vclear were measured by passing through tap water into the tube. The voltage measured from the sensor was recorded as the emitted 81 light flux from the light source. Different cdye of Red 40 were then passed through the tubes and the corresponding voltage readings are recorded and plotted against their concentration. The sensor voltage reading at different cdye was recorded in Figure 4-7. 2 1.8 1.6 1.4 Signal voltage (V) 1.2 1 0.8 y = 1.8509e-19339x R² = 0.9979 0.6 0.4 0.2 0 0 0.00001 0.00002 0.00003 0.00004 0.00005 0.00006 0.00007 Red 40 Concentration (g/L) Figure 4-7. Sensor signal voltage at different Red 40 cell concentration The calculated A of green light was plotted against the cdye. The trendline is relatively straight with an R-square of 0.9971. The slope of the trend line, 8426.2 L/g, equals to the product of l (0.065m) and the εatt. The calculated εatt is 12600 ± 400 L/mol-cm. By plotting the signal voltage against the cdye, the exponential value at log 10 of the curve was the negative product between the light path distance and the εatt. The calculated A was plotted against cdye in Figure 4-8 and the plot follows Beer’s Law closely. 82 0.6 y = 8426.2x - 0.006 0.5 Calculated Absorbance R² = 0.9971 0.4 0.3 0.2 0.1 0 0 0.00002 0.00004 0.00006 0.00008 Red 40 Concentration (g/L) Figure 4-8. Photometer calibration of Red 40 concentration with Beer’s Law In this case, the calculated εatt was 12670 L/mol-cm with R-square value of 0.9974. Both methods have good agreement with each other and has good R-square value. This indicates that 12600 L/mol-cm εatt value could be used to calculate the cdye inside the tube for the RTD. 4.6.4 Measurement and Experiment Results The measured Pe at ul of 7.2, 6.6, and 6.0 m/s are shown in the following three figures. During the experiment, the measurements are carried out till the dye ran out. At ul = 6.0 and ul = 6.6 m/s, the concentrated Red 40 dye solution sustained 10 trials. At ul = 7.2 m/s, the dye lasted 7 experiments before ran out. A simulated line Pe = 78.125 under step signal was constructed using the RTD equations to observe whether or not the measured RTD of the step signal mixing experiments fall under 83 turbulent regime or not [128]. The mean and standard deviation of the measured Pe were shown in the below Table 4-4. A customized Matlab algorithm was developed to fit the curves with a calculated Pe to the RTD curve. The algorithm measures where the inflection point is on the PFR RTD using step signal. The tm was calculated using ul and Equation 4-15 in order to calculate the Pe. Table 4-4. The calculated mean Pe and standard deviation for the three ul in the PFR ul (m/s) 7.2 6.6 6 Mean 92.1 91.2 104.8 STD 11.7 12.6 12.4 At all three ul, the measured Pe was higher than 78.125 (VDC <0.00128), which was the threshold for low to no back mixing according to Fogler and Levenspiele [128]. It was safe to assume the flow system for the mass transfer experiment was under plug flow conditions. 1 0.9 0.8 F (Dimensionless tracer 1 0.7 2 0.6 3 4 0.5 5 7 concentration) 0.4 9 0.3 10 6 0.2 8 0.1 Average 0 0 0.5 1 1.5 2 Φ (Dimensionless Time) Figure 4-9a. C-curve for ul = 7.3 m/s 84 1 0.9 F (Dimensionless tracer 0.8 1 0.7 4 7 0.6 8 0.5 10 6 concentration) 0.4 2 0.3 3 0.2 5 9 0.1 Average 0 0 0.5 1 1.5 2 Φ (Dimensionless Time) Fig. 4-9b. C-curve for ul = 6.6 m/s 1 0.9 F (Dimensionless tracer 0.8 0.7 3 0.6 4 0.5 5 6 concentration) 0.4 7 0.3 1 0.2 2 0.1 Average 0 0 0.5 1 1.5 2 Φ (Dimensionless Time) Fig. 4-9c. C-curve for ul = 6.0 m/s As shown in the following Figure 4-10, the three measured VDC under different VFD speeds were all lower than the maximum VDC allowed for PFR model (0.0128). The curves were all 85 sharper than the VDCMax curve with regards to the slope around the inflection point. 1 0.9 0.8 0.7 0.6 VDC = 0.0128 F0.5 60 hz 0.4 0.3 55 hz 0.2 50 hz 0.1 0 0 0.5 1 1.5 2 Φ (Dimensionless Time) Figure 4-10. Average C-curve for all three pump settings and the PFR threshold (VDC = 0.0128) 4.7 Pressure Drop Experiment 4.7.1 Theory The mass transfer experiment requires accurate locational pressure measurement or estimation since the DOS needs to compensate for pressure during measurement. Also, the previous two- phase pressure drop model was established based on carbon-hydrate and solid flow instead of water-air [107]. This discrepancy between the literature and the PFR warrants a thorough investigation to validate the pressure drop model. fmix is crucial for this study because it was included in both Dbub correlation equation and Pfri equation. This covers both cost and benefit of MBGs. Three different Ql and Qg were selected. 86 Including pure liquid flow, the experiment has 12 different combinations of gas-liquid flow rate that are shown below: Table 4-5. Flow rate settings for the pressure drop experiment Frequency (Hz) 45 50 60 ul(m/s) 6 6.6 7.2 Qread(LPM) 0 0.35 0.7 1 0 0.35 0.7 1 0 0.35 0.7 1 Qg(LPM) 0 0.84 1.68 2.4 0 0.84 1.68 2.4 0 0.84 1.68 2.4 ug(m/s) 0 0.33 0.67 0.96 0.00 0.33 0.67 0.96 0.00 0.33 0.67 0.96 εg 0 0.06 0.11 0.16 0.00 0.05 0.10 0.14 0.00 0.05 0.09 0.13 The ug and ul were plugged in Equation 1-11 to calculate the Remix and Equation 1-12 to calculate the fmix using Equation 1-13. By measuring the pressure drop of the two-phase flow inside the pipe, the slope of the pressure drops as the function of distance travelled could be used to calculate the fmix. Equation 4-20 shows that the two-phase pressure drop (Δpmix) as a function of fmix, L, pipe diameter D, ρmix, and umix: 𝐿 ∆𝑝 = 2𝑓 𝜌 𝑢 (4 − 20) 𝑔𝐷 The fmix correlation Equation 1-13 is similar to the single-phase ffan correlation Equation 1-8. If the term after the plus sign in equation 1-11 can neglected, then fmix equation would be highly similar to the ffan equation. A theoretical simulation of estimated fmix values at the experiment range of Ql and Qg was conducted to assess whether or not the part after the plus sign could be ignored when calculating the fmix as shown in Equation 4-21. The calculated Remix range from Table 4-5 was used as independent variables to calculate the fmix using both Equation 4-21 and Equation 1- 13 to observe the results. 87 𝑎 𝑅𝑒 − 𝑎 𝑅𝑒 𝑓 = 𝑎 𝑅𝑒 + ≈ 𝑎 𝑅𝑒 (4 − 21) 𝑅𝑒 1+ 𝑡 The simulation result supported the practice of using the simplified version of Equation 4-21 instead of the complicated version of Equation 1-13 when calculating fmix for the flow rates recorded in Table 4-5. The difference at any given Remix is under 1 percent compared to using the simplification method. 6.35E-03 6.30E-03 5% void, simplified Theoretical friction factor value 5% void, full 6.25E-03 10% void, simplified 6.20E-03 10% void, full 15% void, full 6.15E-03 15% void, simplified 6.10E-03 6.05E-03 6.00E-03 5.95E-03 5.90E-03 5.00E+04 5.50E+04 6.00E+04 6.50E+04 7.00E+04 Gas-liquid mixture Reynolds Number Figure 4-11. Theoretical simulation between the full fmix form and the simplified version 88 4.7.2 Setup A pressure probe was modified to fit the 0.5-inch NPT port and pressure reading was recorded with a modified Arduino circuit. The pressure probe was inserted in the same 0.5-inch NPT port where the DOS inserted. An adapter that enlarges the pressure probe connector was used as connector between the probe and the port. From observation, the data readings of the pressure gauge, regardless of digital or mechanical, were all oscillating while maintaining the same Ql/Qg combination. Regardless of the Ql, probe orientation, flow rate, the pressure reading would fluctuate. The Arduino circuit could record the measured pressure data, and the average of the recorded data over 10 seconds was established as the recorded value. The first experiment was to measure whether or not the measured pressure reading would change significantly as the liquid level inside the liquid feed tank reduces. The liquid feed tank can fit 50 gallons and the liquid level at max capacity was around 1 meter. The hydrostatic pressure was at most only 9.81 kpa, and the measured pressure drop due to friction loss was an order of magnitude larger than this value even at the slowest ul. The pressure probe was left at 3.5 meters away from MBG. The experiment started with liquid at max level, and the liquid was being pumped into a second vessel. As the liquid level reduces, the pressure change was recorded using Arduino. 89 22.6 22.5 22.4 Pressure (psi) 22.3 22.2 22.1 22 21.9 21.8 0 200 400 600 800 1000 1200 Data point Figure 4-12. Pressure reduction as the water level drops inside tank 1. The next phase was to measure the pressure drop inside the tube in order to calculate the f. The probe was left in same port location for the experiment, while the Ql/Qg combinations were changed to continuously measure the pressure readings at one location. 4.7.3 Experimental Result The pressure curves at different point inside the tube were measured and the results are shown in the following figures. The ul and ug settings were from Table 4-5. 12 different pressure drop curves were obtained for each of the three different MBGs. 90 60 ul=6, void=0% ul=6, void=5% 50 ul=6, void=10% ul=6, void=15% ul=6.6, void=0% ul=6.6, void=5% ul=6.6, void=10% ul=6.6, void=15% Gauge Pressure (psi) 40 ul=7.2, void=0% ul=7.2, void=5% ul=7.2, void=10% ul=7.2, void=15% 30 20 10 0 0 1 2 3 4 5 Distance from MBE (m) Figure 4-13. Pressure drops within PFR using the 7mm Riverforest MBE 60 ul=6, void=0% ul=6, void=5% ul=6, void=10% ul=6, void=15% ul=6.6, void=0% ul=6.6, void=5% 50 ul=6.6, void=10% ul=6.6, void=15% ul=7.2, void=0% ul=7.2, void=5% ul=7.2, void=10% ul=7.2, void=15% Gauge Pressure (psi) 40 30 20 10 0 0 1 2 3 4 5 Distance from MBE (m) Figure 4-14. Pressure drops within PFR using the 7mm lab-made MBE 91 40 ul=6, void=0% ul=6, void=5% ul=6, void=10% ul=6, void=15% 35 ul=6.6, void=0% ul=6.6, void=5% ul=6.6, void=10% ul=6.6, void=15% 30 ul=7.2, void=0% ul=7.2, void=5% Gauge Pressure (psi) ul=7.2, void=10% ul=7.2, void=15% 25 20 15 10 5 0 0 0.5 1 1.5 2 2.5 3 3.5 4 Distance from MBE (m) Figure 4-15. Pressure drops within PFR using the 10 mm lab-made MBE. From the above results shown in Figure 4-13 through 4-15, the effect of Ql has a higher impact on the in-line pressure when the measurement point was closer to the MBE nozzle exit. When the measurement point was closer to the PFR exhaust, this effect was reduced. This was expected as the ul has a squared effect on the Pfri for fluid flows inside the PFR as shown in Equation 4-20. The slopes of the 12 curves obtained above were used to calculate the fmix inside the PFR. Rearrange the Equation 4-20 yields the equation to calculate the fmix using the slope (m) of the pressure drop curve: ∆𝑃 𝜌 𝑣 𝑓 = 𝑠𝑙𝑜𝑝𝑒 = 𝑚 = 2 (4 − 22) ∆𝑧 𝑑 𝜌 𝑣 𝑓 =2 (4 − 23) 𝑚𝑑 92 The calculated fmix are summarized in the Table 4-6 below: Table 4-6. Calculated fmix as a function of both ul and ug ul (m/s) 6 Qg (LPM) 0 0.84 1.68 2.4 Remix 4.86E+04 5.13E+04 5.40E+04 5.63E+04 fmix 4.63E-03 4.38E-03 4.16E-03 3.94E-03 ul (m/s) 6.6 Qg (LPM) 0 0.84 1.68 2.4 Remix 5.34E+04 5.61E+04 5.88E+04 6.12E+04 fmix 4.63E-03 4.37E-03 4.16E-03 3.99E-03 ul (m/s) 7.2 Qg (LPM) 0 0.84 1.68 2.4 Remix 5.83E+04 6.10E+04 6.37E+04 6.60E+04 fmix 4.42E-03 4.18E-03 3.97E-03 3.80E-03 After plotting the measured fmix values against the estimated fmix values calculated from the Equation 4-21, one discrepancy stood out: Under similar Remix, higher εg fluid flow produces smaller fmix. According to the simulation from Equation 4-21, as the εg of the mixture increases, the reduction of fmix should be almost linear as shown in Figure 4-16. However, according to the experimental result, the actual fmix decreases significantly more than predicted as gas was introduced into the flow system. 93 0.005 0.0045 0.004 fmix fmix_exp 0.0035 fmix_new fmix_Simp 0.003 0.0025 4.50E+04 5.00E+04 5.50E+04 6.00E+04 6.50E+04 7.00E+04 Remix Figure 4-16. The measured fmix as a function Remix compared to Equation 4-21 Numerous previous literatures obtained mixture friction factor as a function of Remix. The form 𝑓 = 𝑎𝑅𝑒 is quite common among these correlations [130]. To address the significant effect of εg on the friction factor, an additional εg was appendaged in order to predict the effect of εg on fmix. The updated fmix correlation can be seen below: 𝑓 = 𝑎𝑅𝑒 1−𝜀 (4 − 24) The fmix was then processed using Matlab to fit the parameters for the three different MBEs used in this experiment. A surface fitting tool was used to accustom both independent variables. The following three surface plots depict the fitting of the experimental data on the surface plot generated using the fitted parameters a, b, and c. 94 Figure 4-17. Surface plot fitting of fmix using Equation 4-24 for 7 mm Riverforest MBE Figure 4-18. Surface plot fitting of fmix using Equation 4-24 for 7 mm Lab-made MBE 95 Figure 4-19. Surface plot fitting of fmix using Equation 4-24 for 10 mm Lab-made MBE The fitting tool reported the following results and the 95% confidence interval for the parameters. Table 4-7. Parameter fitting results for the three MBEs using Equation 4-24. MBG Parameter Average +/- Lower Bound Upper Bound River a 0.120 0.11 0.013 0.226 b -0.299 0.08 -0.381 -0.217 c 0.942 0.14 0.803 1.080 Small a 0.127 0.17 -0.046 0.299 b -0.305 0.12 -0.430 -0.180 c 0.815 0.21 0.604 1.027 Large a 0.520 0.46 0.064 0.975 b -0.416 0.08 -0.494 -0.337 c 1.224 0.21 1.010 1.437 Thus, the full modified expression of the friction factor correlation used in this study would be the following: . . 𝑓 , = 0.12𝑅𝑒 1−𝜀 (4 − 25) 96 . . 𝑓 , = 0.13𝑅𝑒 1−𝜀 (4 − 26) . . 𝑓 , = 0.52𝑅𝑒 1−𝜀 (4 − 27) For the following experiment data fittings and simulations, Equation 4-25 through Equation 4-27 were used to calculate the internal pressure drop for the PFR when aerated using different MBEs. For the pressure drop calculation, this fmix was used for the entire length of five meters of the tube. However, when calculating the pressure drop of ejector MBGs, only half of the length was calculated using this new fmix while the other half of the length pressure drop was calculated using ffan. This was covered in section 3-2 with Equation 3-4. This is because the gas is injected from mid-point of the MBG, thus only second half of the nozzle is two-phase flow. 4.8 DOP Experiment With the new fmix equation conducted, the mass transfer experiment could commence. DOS was inserted into the same location where pressure probe was to obtain the bubbly flow DO value. The DOS selected for this experiment required manual pressure compensation when using Arduino to processing the data. The true performance of the manual pressure compensation needed testing in order to validate the vendor’s claim on pressure compensation functionality on the DOS. The bulk fluid inside the reserve tank was flushed through the system to check whether or not the readings of the DOS at various location were proportional to their corresponding pressure. 97 4.8.1 Method Reverse-osmosis (RO) water was used for this experiment to minimize the impurity in the liquid phase that could potentially change the mass transfer or bubble coalescence property. Initially, the DO level inside the bulk RO water was measured as the baseline DO level. During the bulk fluid DO level measurement, the water pump was turned on to create some liquid flow and mixing inside the bulk fluid. The DOS used for this experiment is a galvanic probe and DO content was consumed by its galvanic cell during measurement. Liquid flow inside the tank would ensure the bulk fluid around the DOS does not exhaust DO content. The following Figure 4-20 from the DOS product brochure demonstrated that after 10 seconds without liquid movement to replenish DO content in the stagnant water, the DO level reading would decrease due to consumption and return inaccurate readings [131]. Figure 4-20. Demonstration of DOS reading reduction due to oxygen consumption During the DOP measurement, the DOS was inserted in the housing unit to record the DO 98 level at a certain location in the system. Since the liquid was flushing through the measurement point at very high ul, there was no need to worry about DO consumption from the probe causing wrong readings. During operation, the commands to compensate for pressure inside the PFR was manually entered to the Arduino board to account for the drastic pressure change within the long and narrow PFR during operation. The RO water was flushed through the water at 7.2 m/s, the highest ul and in-line pressure this experiment would conduct under. This ensures all the possible pressure level was examined to avoid out-of-bound conditions disrupting the expectations. The RO water was flushed through the pipe for extended period of time to ensure the DO reading on the probe was stable. The pressure profile was measured in the earlier Pfri experiment. During the operation, the gas valves were all turned off so the measurement was for pure liquid flow only. The liquid exhaust was moved to another tank so none of the fluid under measurement was returning to the original tank. Even though the gas valves were turned off, high speed liquid (7.2 m/s) jet hitting the liquid surface would create gas to liquid entrainment and potentially introducing fresh O2 or N2 to the experiment and create errors. 4.8.2 DOS behavior The measured bulk RO water concentration was 7.37 mg/L as shown in the below Figure 4- 21. The pump was turned on and the DO level stabilized after being submerged in the bulk fluid at around 25 seconds. This same RO water bulk fluid was then flushed through the pipe and the 99 DO level was measured at various places to observe the effect of pressure on the DO level reading. 8.6 8.4 DO level (mg/L0 8.2 8 7.8 7.6 7.4 7.2 0 10 20 30 40 Time (s) Figure 4-21. DOS stabilization when measuring RO water inside the tank The following Figure 4-22 demonstrates the linear correlation between the distance travelled inside the tube and the probe pressure. The previous experiment on friction factor also indicated that the relationship between the distance the bubbly flow travelled and the in-line pressure at the location is linear. The DO level relationship with the distance travelled can then be converted into the relationship between DO level and pressure using Equation 4.22. 100 35 30 DO level (mg/L) 25 20 15 10 y = -6.4883x + 34.95 5 R² = 0.9938 0 0 1 2 3 4 5 Distance travelled (m) Figure 4-22. The DOS reading at various locations in the PFR using same RO water The following Figure 4-23 demonstrates the relationship between the pressure and the DO level within the PFR. The DO values of the same fluid would increase as the liquid is being compressed at higher pressure. The relationship between DO level and the liquid pressure is almost linear with an R-square larger than 0.99. When measuring the DO values near the MBE, the pressure would be very high due to Pfri. When analyzing the true mass transfer rate of the O2 between gas and liquid phase, this Pfri would change the DO reading on the DOS when the probe was being inserted at the start of the PFR. The DOS software integrated pressure-compensation to account for this pressure change at different locations inside the flow line. 101 35 y = 0.6933x - 1.9411 30 R² = 0.9936 DO level (mg/L) 25 20 15 10 5 0 0 20 40 60 Pressure (kpa) Figure 4-23. The DOS reading of the same fluid under different pressure 4.8.3 DOP Measurement Sequence Method The DOP measurement was conducted under the same fluid conditions presented in Table 4- 5 from section 4.7.1. The predicted pressure at a given location inside the flow system was predicted using the conclusion drawn from section 4.7. Equation 4-25 through 4-27 were used to calculate the fmix, and the pressure within the tube was calculated using Equation 4-20. The sequence of the measurements was randomly generated from statistical design in order to mitigate random errors. For this experiment, three different gas and liquid flow rates were selected. The flow rates from low to high are now described as low, medium, and high. Three replicates 102 were produced for the same flow rate combination to reduce the variability and to improve the confidence of the results. A total number of 27 experiment measurements (3 replicates, 3 ul, and 3 εg) at a particular point along the PFR was conducted before the DOS was moved to a different location. By applying the experiment design parameters to Minitab, a random sequence order of the 27 experiments was produced. The result can be seen in the following Table 4-8. The values of the two independent variables corresponding to the flow rates were also included. Table 4-8. Sequence of measurement for PFR DOP experiment and the DOP experiment flow condition Order ul (m/s) εg Order ul (m/s) εg 1 Low Medium 19 Medium Medium 2 Low High 20 High High 3 Low High 21 Medium High 4 Medium Medium 22 High High 5 High Medium 23 High Low 6 Low Medium 24 Low Low 7 Low Low 25 High Low 8 Low High 26 Medium High 9 Medium Low 27 High Medium 10 Medium High 11 High High εg 12 High Low Low 5% 13 Low Low Medium 10% 14 High Medium High 15% 15 Medium Low 16 Medium Medium ul (m/s) 17 Low Medium Low 6 18 Medium Low Medium 6.6 High 7.2 103 4.8.4 Water Charging to Ensure Standard Initial DO Values. Since the DOS was left in the same port position when measuring the DO values for 27 different DOPs, it was crucial to ensure the initial DO values of the bulk fluid were the same for all 9 port positions. Since air was used as the O2 source, it was also important to derive the bulk fluid of O2 to some extent so there could be mass-transfer drive-force between the bubbles and the water. One method to achieve both of these goals was to sparge N2 microbubble into the bulk fluid and stop the sparging once the measured DO value of the bulk fluid was reduced to a certain value below air saturation DO value. Due to the lack of O2 inside the pure N2 microbubbles, the DO content inside the initial reservoir could be removed via liquid-to-gas mass-transfer. After considering the DOS’s capability, 5 mg/L O2 content was selected as the baseline bulk fluid initial DO value. This process is now referred as ‘water charging’. The following Figure 4-24 demonstrates the schematics of the experiment during the water charging phase. A 50-gallon liquid tank filled with RO water was selected as the reservoir and was connected to the impeller pump. The impeller pump was connected to a VFD. N 2 gas tank and the pump outlet were connected to the MBE and the N2 microbubbles were sent through the PFR. The DOS was left in place to observe the decrease in DO values of the bulk fluid. The tube was then recycled back into the same reservoir. As more and more O 2 molecules were removed from the bulk fluid within the reservoir by the N2 microbubbles, the DO level would keep decreasing till 5 mg/L was achieved and the N2 microbubbles sparging was stopped. 104 Figure 4-24. The schematics of the PFR during water charging operation Once the water was charged, the DO measurement experiment could begin. The Ql and Qg were controlled via VFD and gas glow meter. The 27 experiments ran continuously and the DOS was moved to another location. The below Figure 4-25 demonstrates the setup for this experiment. The previous tank with charged water was used as the reservoir and the air bubbly flow was discharged into a second tank. The second tank became the new reservoir for water charging in the next experiment cycle. 105 Figure 4-25. The schematics of the PFR during DOP measurement 4.8.5 Experiment result 4.8.5.1 General DOP for the Three MBGs. The DOP for the flow system when using the Riverforest MBG can be seen in the following Figure 4-26. The measured DO values increase as the bubbly flows progress down the flow line. At certain point along the line, the DO value stopped increasing due to saturation and reduction in mass-transfer drive force. As the pressure decreases, the saturated DO content in liquid phase would prevent mass transfer from occurring. Under the same ul, adding in more gas would increase the DO value inside the liquid phase. Similarly, under the same ug, higher ul would generate higher DO value. In this case, higher ug or 106 ul would produce a higher kla. The effect of ug on the kla was also higher than the effect of ul on it. As shown in the Figure 4-26, the three highest DOPs and kla values were all under high ug. The kla for the highest ul with medium ug is still lower than the lowest ul with high ug. 0.5 0.45 0.4 0.35 Low Liq-Low Gas Dissolved oxygen (M) Low Liq-Mid Gas 0.3 Low Liq-High Gas 0.25 Mid Liq-Low Gas Mid Liq-Mid Gas 0.2 Mid Liq-High Gas 0.15 High Liq-Low Gas 0.1 High Liq-Mid Gas High Liq-High Gas 0.05 0 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 Distance travelled (m) Figure 4-26. DOP for the PFR when aerated by the Riverforest 7 mm MBE The following Figure 4-27 shows DOP for the PFR when using the 7mm lab-made MBE. The same trend from the 7 mm Riverforest MBE DOP could be observed here. Higher ug or ul would increase the overall kla of the system, while ug has a higher impact on the kla than ul. 107 0.5 0.45 0.4 Dissolved oxygen (M) 0.35 Low Liq-Low Gas 0.3 Low Liq-Mid Gas Low Liq-High Gas 0.25 Mid Liq-Low Gas 0.2 Mid Liq-Mid Gas 0.15 Mid Liq-High Gas 0.1 High Liq-Low Gas High Liq-Mid Gas 0.05 High Liq-High Gas 0 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 Distance travelled (m) Figure 4-27. DOP for the flow system when aerated by the lab-made 7 mm MBE The maximum measured DO value for both the lab-made and the Riverforest 7 mm MBEs are similar to each other as shown in the following Table 4-9. The highest difference for the DO value between the two different MBEs was only 7.81%. Table 4-9. Measured differences between the maximum observed DO value Dissolved oxygen (M) Riverforest Lab-made small Difference (%) Low Liq-Low Gas 0.298 0.296 0.74 Low Liq-Mid Gas 0.344 0.344 0.21 Low Liq-High Gas 0.410 0.415 1.30 Mid Liq-Low Gas 0.316 0.298 5.82 Mid Liq-Mid Gas 0.356 0.366 3.06 Mid Liq-High Gas 0.422 0.455 7.81 High Liq-Low Gas 0.326 0.305 6.46 High Liq-Mid Gas 0.372 0.359 3.69 High Liq-High Gas 0.440 0.469 6.49 108 For the larger 10 mm lab-made MBG, the general pattern of the DOP is the same as the previous smaller 7 mm MBGs’ results. The DO values reached maximum at between 3 and 3.5 meters long. However, since the 10 mm tube was 1 meter shorter than the 7 mm tubes, the internal pressure inside the tube was smaller due to smaller Pfri. The measured DO values for the 10 mm tube were also lower than the 7 mm tubes. 0.4 0.35 0.3 Low Liq-Low Gas Dissolved oxygen (M) Low Liq-Mid Gas 0.25 Low Liq-High Gas 0.2 Mid Liq-Low Gas Mid Liq-Mid Gas 0.15 Mid Liq-High Gas 0.1 High Liq-Low Gas High Liq-Mid Gas 0.05 High Liq-High Gas 0 0 1 2 3 4 5 Distance travelled (m) Figure 4-28. DOP for the flow system when aerated by the lab-made 10 mm MBE 4.8.5.2 kla Quantitative Analysis for the Flow Systems and Bubble Size Equation Assessment A critical part of this study was to analyze whether Chen’s correlation from Equation 1-8 and the lab-developed correlations Equation 4-24 were accurate at predicting the Dbub produced by the MBEs. To accomplish this, the previous model equations from section 4.3.2 were fit to the 109 measured DOP. The specific equation utilized was the liquid phase dissolved oxygen Equation 4- 10. Matlab lsqcurvefit and ODE45 tools were utilized to find the optimal kl values. The individual DOPs were fit to the predicted model and extracted from Matlab to observe whether the Equation 4-10 fits the measured DOP. The four parameters that were being fit were the mass transfer coefficient kl, the power factor on ul power factor on ul, and the initial factor for the bubble size correlations. These are the four arbitrary constants that were affecting the bubble size correlation as shown in the previous Equation 1-7. The three factors from Equation 1-7 are highlighted in red here. 8𝜎𝑢 𝐷 = (1 − 5) 𝑓𝜌 𝑢 The result of the least-square fit can be seen in the following Table 4-10 for both Chen’s Equation 1-8 and the new developed 2-phase bubble size Equation 4-24. 110 Table 4-10. The result of the fitted parameters for both flow models The square-2 residual norm for all the parameter fittings were also extracted from Matlab and shown in the below Table 4-11. 111 Table 4-11. The square-2 residual norm for the fitting results for both models and all MBEs A Student T-test was conducted to observe whether the Equation 4-24 yielded significantly different parameter fitting results when compared to the original Equation 1-8. The results can be seen in the following Table 4-12: Table 4-12. Student T-test result on the fitted parameters for both old and new bubble size prediction equations The previous T-test results demonstrated that the mean of fitted parameters for both models were within close range of each other when modeling for the DOP using Equation 4-24. However, when the pressure inside the tube was measured, there were significant differences between using 112 Equation 1-8 and using Equation 4-24. To maintain the consistency of the model and calculations, the fmix correlation 4-24 was recommended to model. The following figures demonstrates the DOP result of the model fitting. The dashed red lines are the simulated DOP using Equation 1-8. The solid blue lines are the simulated DOP using Equation 4-24. Figure 4-29. The DOP fitting result for the two models using Riverforest 7 mm MBE 113 Figure 4-30. The DOP fitting result for the two models using lab made 7 mm MBE Figure 4-31. The DOP fitting result for the two models using lab made 10 mm MBE When observing the results from Table 4-10, the fitted kl values were found to be influenced heavily by their corresponding εg more than their Remix as shown in Table 4-13. According to the 114 prediction from Equation 1-7, higher εg would lead to larger generated Dbub and smaller a. However, from observations from Table 4-10, higher εg also contributes to higher kl values. If lower εg yields higher a but lower kl, then the overall kla value could increase or decrease depending on specific conditions. This unexpected ‘trade-off’ is detrimental to kla prediction. Table 4-13. Parameters from Table 4-10 rearranged by void fraction independent variable One hypothesis to explain why smaller bubbles have lower kl is that some debris from the experiment such as oil, plastic, or metal contaminated the RO water and increased the gas-liquid membrane shell thickness [132]. To observe whether the bulk fluid used in the mass transfer experiment was contaminated, a kl measurement was conducted to compare the bulk fluid kl against pure RO water kl. 115 4.9 Conclusion A mass-transfer experiment was conducted inside a new PFR system to observe if Fanning or mixture friction factor usage provides better prediction for Dbub generated with MBE. The mixing model and friction factor within the new flow system were obtained to ensure model fidelity. By measuring the DOP of the liquid within the PFR system, the Dbub can be indirectly calculated. New axial-mixing and friction factor models were also developed to improve the model fidelity. Novel photometer as designed and manufactured to measure the mixing model and found that the new PFR system has little back and axial mixing. A pressure sensor was installed along the flow system to measure the pressure profile and the pressure drop along the flow system. The pressure profile was used to calculate the fmix of the two-phase flow inside the PFR and to evaluate the literature fmix accuracy. The literature fmix model was found to be inaccurate and a new fmix model was developed for this study. The finding on the DOP was that using a new fmix does not improve the model accuracy when predicting MBE Dbub. The lab made MBE nozzle has comparable performance when compared to the commercial MBE unit used for Chapter Two. From the parameter fitting result, smaller Dbub would generate higher a but lower kl in the PFR when changing the fluid flow rate settings. A further study was conducted to investigate whey improving Dbub would actually reduce kl and negate the benefit of smaller Dbub on kla. 116 CHAPTER FIVE: MASS-TRANSFER COEFFICIENT ESTIMATIONS FOR THE EXPERIMENT FLUID AND FRESH RO WATER 5.1 Background The findings from the previous mass transfer experiments indicated that the measured kl values for smaller bubbles are smaller compared to larger bubbles. A hypothesis was that some debris inside the bulk fluid was acting as barrier between gas and liquid phase [132]. To validate this hypothesis, an experiment to compare the kl values between pure RO water and experiment bulk water was conducted. The fundamental behind this experiment was to measure the kla of a fluid system with same a and to calculate the kla for the two RO water and bulk water. The fundamental equation to represent this study is the basic gas-liquid mass balance equation: 𝑑𝐶 = 𝑘 𝑎(𝐶 ∗ − 𝐶 ) (5 − 1) 𝑑𝑡 By integrating the previous equation, the theoretical DOP when aerating with constant C* gas can be shown in the following equation: 𝐶 ln 1 − = −𝑘 𝑎𝑡 (5 − 2) 𝐶∗ 117 5.2 Experiment Setup The design equation Equation 5-2. from the section 5.1 laid out multiple technical requirements for the equipment of this experiment. These technical details are being discussed in this section. The experiment liquid should be maintained at equilibrium concentration with air before the mass-transfer occurs. Using N2 microbubble or boiling to drive the initial CL further below air equilibrium CL could change some unknown properties of the bulk fluid. To avoid introducing further variables, air equilibrium concentration was selected as the initial CL and no further actions were conducted on the bulk fluid once they were collected from the liquid tank before the aeration started. Since the initial CL was kept at air equilibrium, pure O2, instead of air, was selected as the gas to provide mass-transfer drive force. To ensure the equilibrium concentration was truly constant, a steady O2 flow was applied to the liquid vessel’s head space. The outlet of the O 2 was kept at liquid level without touching the liquid. Since pure O 2 is heavier than air, the incoming O2 stream would remove air inside the liquid vessel to ensure a constant equilibrium O 2 concentration. The DOS was the same DOS used in the DOP experiment in Chapter Four. Since the probe is a galvanic probe, stirring was required to prevent DO level reduction due to DO consumption via DOS. Since both the DOS and the O2 tube line need insertion on the vessel, the lid of the vessel would require openings with sufficient area. After considering all the technical requirements, a reactor from MSU was utilized to conduct 118 this experiment. The schematics of the equipment and the equipment itself are shown in the following Figure 5-1. Figure 5-1. The schematics (L) and the equipment (R) of the aeration vessel used for the kla measurement experiment. RO water and the experiment bulk water were collected and transferred to the aeration vessel. The DOS was inserted into the liquid phase and the O2 was then inserted and secured along with the DOS. The impeller was then turned on and the data acquisition started. 5.3 Experimental Result The LHS of the previous Equation 5-2 was plotted against time and the linear region in the middle of the curve was regressed. The Figure 5-2 below demonstrates the first measurement of the kla inside the aeration vessel and the slope in the middle of the curve when using bulk fluid. The negative of the slope, which was the measured kla, was 1.23 s-1. The similar measurements were conducted five times for both RO water and bulk fluid. 119 The start and the end of the experimental data correspond to the beginning and the ending of the aerations. The data collected during these intervals were not corresponding to mass-transfer between pure O2 and water, since O2 was still building up in the head space. Thus, only the center part was used for linear regression calculation. Figure 5-2 shows the CL inside the liquid phase for the first measurement with bulk fluid. 0 0 500 1000 1500 2000 2500 3000 -0.05 -0.1 Experimental Precition -0.15 ln(1-C/C*) -0.2 -0.25 -0.3 -0.35 -0.4 -0.45 Time (s) Figure 5-2. The experimental data and the linear regression at the center of the curve for the first bulk water kla measurement The below Table 5-1 shows the measured kla values for both the RO water and the bulk fluid. A Student-T test was conducted on the two groups of data to evaluate whether these two kla values are significantly different from each other. 120 The average kla values for the RO water was 1.47±0.16 s-1. The average kla values for bulk fluid was 1.21±0.12 s-1. Both fluids had five measurements and the degrees of freedom was 8. The calculated Student-t experiment score was 2.8687, corresponding to a 98.6% confidence interval. Based on this calculation, the kla values for RO water and bulk fluids are different from each other. Since the measurements were conducted in the same vessel, it can be concluded that the kl values for the RO water and bulk fluids are different from each other. This is further supported by the literature that the impurities from rust metals inside the bulk fluid might be reducing the kl values and reducing the benefit provided by smaller bubbles. The T-connector housing unit had oil applied to the connector as corrosion prevention, and the pump was manufactured with cast iron for the pump body so these could have introduced impurities in the bulk water as the experiment progressed. Table 5-1 Measured values and T-score calculation for the RO water and bulk fluid kla values. 121 5.4 Conclusion A kla measurement experiment series was conducted on the experiment water used for Chapter Four and pure RO water to observe whether or not there are significant debris within the experiment water that increased the microbubble shell thickness. The method was to measure the DOP when aerating two kinds of water within the same vessel using pure oxygen. The finding was that measured kla from experiment water was lower than pure RO, thus confirming the experiment water for Chapter Four had debris and increased microbubble shell size and reduced kl at smaller Dbub. 122 CHAPTER SIX: LARGE-SCALE BIOREACTOR SIMULATION USING MBES 6.1 Background In the previous chapters, an ejector mechanism was identified as best suited for cost- effectively generating extremely high mass transfer rates in commercial-scale reactors. Then, the performance characteristics of MBE were measured and used to develop mathematical models estimate the effective microbubble size generated by MBE as a function of the key independent variables. In this chapter, those mathematical models are used to develop, for the first time, a design and scale-up algorithm for implementing MBE in commercial-scale reactors. The gas input from one single MBE nozzle is insufficient to fully aerate an industrial-scale reactor (ISR). However, a rational scale-up strategy could be developed based on the concept of a mass-transfer cell, which is defined as the volume that would be satisfactorily aerated by a single MBE. Then arrays of MBE could be arranged into rows of mass-transfer cells, and multiple rows could be stacked as needed, as shown in the following Figure 6-1. 123 Figure 6-1. Sample layout of an ISR aerated using multiple layers of MBEs Reactor system as sophisticated as such demands simulation models with substantial level of details that represents fluid dynamics inside the reactor. For this section, numerous aspects of the performance of such ISR with MBE nozzle formation were examined and compiled into intuitive results on performance that demonstrates what researchers and engineers would expect from an ISR aerated by MBE nozzle formation. 6.2 Jet Cone Height, Angle, and Stacking Strategy The previous mass transfer experiment was conducted on a long, narrow PFR equipment that has the same inner diameter with the MBE. During industrial operations, an MBE would sparge bubble into a vessel with larger diameter than its inner diameter. Liquid jet entrains the stagnant fluids from surrounding and thus increases the overall liquid flow rate inside the jet [109], making estimation of the jet flow model based on constant jet Ql vastly inaccurate. 124 6.2.1 Stagnant Fluid Residence Time Lima Neto et al. investigated the stagnant fluid entrainment at high void fraction and developed for a given jet radius and axial liquid velocity [108]. However, axial liquid velocity is difficult to measure or estimate inside an industrial reactor vessel. A more recent model based on computational fluid dynamics has developed an equation to estimate the change in volumetric flow rate correlated to inlet liquid velocity, which makes it much easier to estimate the jet volumetric flow rate [109]. The following simulation model assumes a Gaussian Distribution of liquid velocity inside a liquid jet. 𝑢 =𝑢 𝑒 (6 − 1) where ujet is liquid jet velocity at any point inside a jet, umax is maximum liquid velocity at the jet centerline, r is the radial position inside the jet, and h is the axial position inside jet. Cushman-Roisin (2010) measured jet angle from side to opposition to be around 24° [109]. At any given point inside the jet, the ratio between the radii and the height of the jet would be approximately tan(12°)≈0.2 [109, 110]. The relationship between the umax and the ul at the MBE nozzle exit can be developed via momentum balance [109]: 5𝑢 𝑑 𝑢 = (6 − 2) ℎ Combining Equation 6-1 and 6-2 would generate the expression for the liquid entrainment volumetric flow rate (Qent) equation of a jet cone at a combination of ul, h, and dnoz: 125 ℎ 𝑢 𝜋ℎ 𝑢 ℎ𝜋𝑑 𝑄 = 2𝜋𝑢 𝑒 𝑟𝑑𝑟 = 2𝜋𝑢 × = = (6 − 3) 100 50 10 When multiple liquid jets are propelled parallelly into the bioreactor, the regions in between the jet cones would have relatively low ul compared to the jet zone. Through liquid entrainment, these stagnant fluids can be recycled into the jets and creates a certain ‘residence time’ for itself. This residence time indicates how long the organic cells inside these void zones will be out of nutrients such as O2. This residence time should be carefully designed so that the cells do not starve of nutrients. To calculate this residence time, both the Qent and the void volume (Void) must be calculated. For this simulation, it was assumed that the jet nozzles were packed in such way that the top of the jet cones touch each other on the edge at where 90% of the reaction gas is depleted. The jet cones were packed in a triangular pitch as shown in the below Figure 6-2. Figure 6-2. Top-down view of a triangular pitch jet cones with tight-pack arrangement 126 The entrainment rate around a jet cone creates the Qent of the void region. The void region volume can be calculated via subtraction of jet cone volume from the cylinder volume of a unit cell. The volume of the cone can be determined by the following equation, where Vcone is the volume of the unit jet cones, mg is the gas molar flow rate, OUR is the oxygen uptake rate and Xo is the target percentage of gas being reacted (for this simulation, the target set was 90%): 𝑚 𝑋 𝑉 = (6 − 4) 𝑂𝑈𝑅 For the simulation, all the jet cones were assumed to have similar shape and size, thus the h, jet cone radius R, nozzle distance L, and subsequently the Void between the jet cones could be calculated as soon as the jet cone volume was calculated since the jet cone angle is known to be 12 degrees from center line [109]. the cylindrical and cone volume equations are shown as below for a unit cell, where Vall is the total volume: 1 𝑉 = 𝜋𝑅 ℎ (6 − 5) 3 √3 𝑉 = 𝐿 ℎ = 2√3𝑅 ℎ (6 − 6) 2 1 𝜋 𝑉 = 2√3𝑅 ℎ − 𝜋𝑅 ℎ = 2√3 − 𝑅 ℎ (6 − 7) 3 3 Now that both the Void and the Qent have been calculated, the residence time of the stagnant fluid inside the void zone, tres, can be calculated as: 𝜋 2√3 − 𝑅 ℎ 𝑅 𝑡 = 3 = ≈ 7.7 (6 − 8) 0.1𝑢 𝜋𝑑ℎ 𝑢 𝑑 127 6.2.2 Pressure Drop Estimation In order to correctly estimate the cost-efficiency of an ISR aerated by MBE (MBE-ISR), the operating cost due to pump power consumption of liquid delivery should be calculated. When calculating the Ppump to operate the MBE-ISR, the same principle for calculating the MBE power cost inside the PFR also applies. Instead of Pfri, the hydrostatic pressure (Phyd) would be the main pressure component that was associated with the geometry of the reactor vessel. Inside the PFR, the tube was long and subsequently the Pfri was high. Inside an industrial-scale reactor vessel, the liquid depth would be high and the hydrostatic pressure would be high. The Ppump for individual MBE would be: 𝑃 = ∆𝑃𝑄 = ∆𝑃 + ∆𝑃 𝑄 (6 − 9) Pfri loss was calculated using Equation 3-4. Phyd was calculated using the calculated depth of the MBE nozzle location. The detail of Phyd calculation would be covered in the later section. 6.2.3 Bubble Size Estimation and Interfacial Area Dbub is an important indicator of the performance of a microbubble generation system since it affects both the bubble rising velocity and the a [34, 36]. The earlier Dbub prediction model results from section 4.8 indicated that using Equation 1-8 produces decent predictions on Dbub and kla of microbubbles exiting MBE nozzles. For the purpose of this simulation, Equation 1-8 instead of Equation 4-24 was used to predict Dbub. 128 6.2.4 Oxygen Transfer Rate Gas-dissolution is usually the rate-limiting step during bioreactor operations since the rate of gas dissolving is usually lower than the rate of cell metabolism [13, 14, 133]. In this case, OUR could be approximated as mass transfer limited, where dissolved gas solubility CL is almost zero: 𝑂𝑈𝑅 = 𝑘 𝑎(𝐶 ∗ − 𝐶 ) ≈ 𝑘 𝑎𝐶 ∗ (6 − 10) The mass transfer coefficient kl was assumed to be constant in this simulation. At mass transfer limitation, the dissolved gas concentration inside liquid phase should be approximately zero since the microorganisms are consuming nutrients faster than the gas inflow replenishes it. The equilibrium gas concentration C* is dependent on the inner pressure of the gas bubbles, which is further dependent on the position of the bubble inside the column due to the influence of depth on hydrostatic pressure. According to Henry’s law, C* for O2 is proportional to both the total pressure Ptotal and the O2 content (yO2) inside the gas. 𝐶∗ = 𝑦 𝐻 𝑃 (6 − 11) The liquid entrainment would increase the total Ql as the jet progresses inside the cone. Meanwhile gas consumption and pressure change would alter the C* as the jet progresses since reactive gas is being consumed. These two factors would change the a as the jet rises as well. The total pressure at the exit of the nozzle and the termination point of the jet cone would be different due to the difference in Phyd and depletion of reactive gas. This would change C* as the jet progresses. Since both a and C* are not constant across the jet cone, Equation 6-10 was no longer an accurate representation of the actual OUR inside the jet cone. This issue was resolved 129 by utilizing an analog to the log mean temperature difference to represent the mass transfer drive force difference along the jet cone: 𝑎 𝐶∗ −𝑎 𝐶∗ 𝑎𝐶 ∗ = (6 − 12) 𝑎 𝐶∗ ln 𝑎 𝐶∗ 𝑂𝑈𝑅 = 𝑘 𝑎𝐶 ∗ (6 − 13) The following Figure 6-3 provides an illustration of the log mean mass transfer drive force for a triangular pitch formation. Figure 6-3. Sample triangular pitch ejector cones and the log mean mass transfer boundary The yO2 at the jet bottom (exit of the nozzle) was the O2 content of the inlet gas, which was 0.2 for air for the purpose of this simulation. At the top of the jet cone, yO2 would decrease due to dissolution of gas inside the liquid phase. At 90% exhaustion of reactive gas, the yO2 value at the 130 top of the jet cone would be: 𝑦 − 0.9𝑦 0.1𝑦 𝑦 , = = (6 − 14) 1 − 0.9𝑦 1 − 0.9𝑦 The εg at the top of the jet cone would decrease since Ql would increase due to entrainment and Qg would decrease due to gas consumption. Applying Equation 6-3 and the denominator of Equation 6-14 into void fraction equation would yield the εg at the top of the jet cone: 𝑄 1 − 0.9𝑦 𝜀 = (6 − 15) 𝑄 + 0.1𝑢 𝜋𝑑ℎ The final expression for the top and bottom aC* in Equation 6-12 would look like the following, where: 6𝑢 𝑎 𝐶∗ = 𝑦 𝐻 𝑃 (6 − 16) 𝑑 𝑢 6 𝑄 1 − 0.9𝑦 0.1𝑦 𝑎 𝐶∗ = 𝐻 𝑃 (6 − 17) 𝑑 𝑄 + 0.1𝑢 𝜋𝑑ℎ 1 − 0.9𝑦 6.2.5 Dissipation Energy Ejector style microbubble generator relies on liquid phase turbulence energy to create microbubbles [102, 134]. The same turbulence could cause cell damage and reduce the overall production inside the bioreactors [135, 136]. This means any optimization processes should take the level of cell damage into consideration since increasing turbulence without taking cell damage into consideration would be counter-productive. Different research groups have studied the source and effect of shear damage on various kinds of cells inside different types of devices. Millward & Bellhouse (1994) investigated turbulence- 131 induced damage on Mammalian cell under laminar flow regime inside a membrane bioreactor [137]. Li et al. (2011) studied the effect of temperature on the cell viability reduction due to shear damage [138]. It was discovered that increasing medium temperature would make the cells more susceptible to shear damage. The effect of stirring-induced shear on cell viability has also been investigated before on insect cells [136]. Multiple papers were also published on how to evaluate shear stress inside various reactor vessels. Sánchez Pérez el al. (2006) developed a theoretical correlation of stirrer shear rate inside a STR for both laminar and turbulent flow regimes [135]. Liu et al. (2013) investigated the shear stress distribution, turbulent kinetic energy, and energy dissipation rate inside a two-phase dispersed bubbly flow inside confined nozzle spaces and developed, and discovered that the turbulent energy is the main cause to the cell death inside a bioreactor [139]. However, Liu et al. (2013) did not develop an explicit equation that’s suitable to estimate the level of turbulence the microorganisms are undergoing inside ejector nozzles. Reichmann et al. utilized a way to estimate the dissipation energy exerted on the fluids when investigating two-phase reactions inside bioreactors [140]. The same equation was used in other studies on bioreactors that involves in high turbulence flow conditions [141, 142]. The dissipation energy Ediss is a function of Δp, Ql, ρl, and ejector volume Vejector. ∆𝑃𝑄 𝐸 = (6 − 18) 𝜌 𝑉 132 6.2.6 Simulation Setting For the following simulation, the OUR, Ppump, and Ediss were evaluate for a bioreactor via simulation using Matlab as calculation and plotting software. Combinations of ul and dnoz were selected as independent variables. The range of ul was between 20 ~ 30 ft/s and the range of dnoz selected was between 1 ~ 10 cm. In order to maintain the jet nozzle formation shown in Figure 6- 2, the jet cone size, and subsequently jet height and distance between jet nozzles, were kept as a constant inside the bioreactor. At the bottom of the bioreactor, the εg was set to be 0.1. The Vcone and h for the mass transfer unit cells was calculated using Equation 6-4 with known mg and OUR. mg was a known variable due to constant εg at the bottom. Matlab fzero function was used to calculate OUR and jet height via the following method: 1. Assume the jet cone OUR is the OUR calculated using Equation 6-10 based on the flow conditions at the exit of the ejector nozzles. Use this to calculated an initial guess jet height required by fzero function. 2. Assume a new jet height value hjet. The total pressure used in Equation 6-11 for the bottom and top of the jet cone are with this assumed jet height in a vessel with height Htank would be: 𝑃 = 𝜌𝑔𝐻 (6 − 19) 𝑃 = 𝜌𝑔 𝐻 −ℎ (6 − 20) 3. Plug in Equation 6-19 and 6-20 into Equation 6-16 and Equation 6-17 will yield the components for Equation 6-12 and 6-13. In this way the OUR could be calculated based on 133 Equation 6-13 4. With both OUR and Qg calculated, the Vcone, and subsequently hjet could be calculated. The calculated jet cone height needed to match the assumed hjet from step 2. In Excel this can be accomplished via Goalseek function. In Matlab this was accomplished with fzero function. 5. Consolidate step 2 ~ 4 to formulate the equations into a one-variable equation where hjet was the only unknown. Use the initial guess calculated from step 1 for fzero function to calculate the actual hjet. The calculated hjet, based on the chosen pair of ul and dnoz, was the true hjet of all the mass transfer units inside the bioreactor for the chosen independent variables. 6. As the nozzle formation progresses upwards inside a large-scale bioreactor, the ug would change as well due to Phyd change as the liquid gets shallower. For the upper-level jet nozzles, solve for ug with a different pressure to compensate for this change. 7. With ug and hjet calculated, an accurate estimation of OUR, W, and Ediss could be obtained. Preliminary results obtained via this method indicated that Dbub does not change much as at different pressure, which supports the previous assumption. 8. The number of jet nozzles per level was calculated by rounding down the ratio between total bioreactor cross-sectional area and the cross-sectional area of maximum jet cone. This ensures the entire column as covered with mass transfer units in order to maximize the production rate of the bioreactor. 9. OUR and power consumption values were summed for the entire bioreactor vessel. The simulation software was capable of calculating cumulative dissipation energy, albeit instantaneous 134 maximum dissipation energy inside the bioreactor was more representative to the maximum strain cells undergoes. The reason why maximum was chosen was because the summation of the Ediss does not reflect on the level of strains microorganisms undergo individually. The following values for chosen for the constants that were added for this section Table 6-1. The new variable sheet for large-scale simulation Variable Value unit Hen,298: Henry's constant at 298 Kelvin 1.28E-05[143] mol/m3-Pa kl: mass transfer coefficient 0.0001 m/s l/d: nozzle length to diameter ratio 5 Patm: atmospheric pressure 101325 Pa 3 R: gas constant 8.314 m -Pa/K-mol T: Temperature 300 K Vreactor: reactor volume 500 m3 H/D: height/diameter ratio of the bioreactor 2 XO2: oxygen reaction extent 0.9 yO2:oxygen content of inlet gas 0.2 σ: liquid surface tension 0.072[144] N/m 6.3 Simulation Results The following Figure 6-4 shows the Dbub corresponding to different ul and dnoz. For the chosen range of independent variables, the bubbles produced were all on micrometer scale. Dbub sizes increase as the ul decreases, mostly due to the reduction in available turbulence energy. dnoz does not change the bubble size much. 135 Figure 6-4. Theoretical Db generated inside the MBE-ISR Figure 6-5 demonstrates that the calculated hjet as a function of both ul and dnoz. Increasing ul marginally decreases the calculated hjet. Increasing dnoz would raise the calculated hjet significantly since the εg at the bottom of the bioreactor was a constant, ug and mg would increase significantly, extending the size of the cone the two-phase flow could sustain. 136 Figure 6-5. Theoretical hjet as a function of both ul and dnoz inside the MBE-ISR The Ppump at various level of ul and dnoz is shown in Figure 6-6 The total W of the entire bioreactor suddenly increases to a higher level as dnoz decreases. This sudden increase is due to the addition of a new layer of nozzles inside the bioreactor. The W of the system decreases marginally as the dnoz decreases until the sudden bump due to increase in number of nozzles. As the ul increases, the total W increases as well due to higher Ql. 137 Figure 6-6. Total power consumption as a function of both ul and dnoz inside the simulated MBE- ISR Similar trends in Figure 6-7 can be observed for the jet nozzle number as a function of ul and dnoz. Due to the increase in dnoz, the total Qg, along with the Vcone, would increase. This would reduce the number of nozzles needed to aerate the entire vessel, as shown in Figure 6-7 Raising ul also marginally increases the total number of nozzles as well. This is surprising mainly because as ul surges, ug and Ql should increase as well at constant bottom εg. This trend should reduce the total number of jet nozzles as the Vcone should increase, but the simulation result in Figure 6-7 demonstrates otherwise. A conclusion that could be drawn is that increasing ul at constant dnoz inside a tight pack MBE-ISR would reduce the Vcone due to entrainment-induced mass drive force change. 138 Figure 6-7. Total number of MBE nozzles needed to aerate the entire MBE-ISR as a function of both ul and dnoz Figure 6-8 is the calculated tres as a function of ul and dnoz. This parameter indicates how long the microorganisms would be deprived of nutrients until they are recycled into the jet cones. Compare this result with the total number of nozzles in Figure 6-7, it was found out that adding a new layer of nozzles as the Vcone decreases would increase the tres. Increasing ul would marginally decrease the tres. Two lines were marked out on Figure 6-8 to indicate where the sudden drops in tres occur. These two lines describe the ul and dnoz combination that produces the exact mass-transfer unit cell that fills up the entirety of the BIG reactor with no wasted head-space. This is the optimal ul and dnoz combination the BIG reactor should be operating at, since this combination would minimize the wasted space. The first step would be selecting the layer of nozzles wanted for the reactor. The first red line where ul = 215.15dnoz-11.988 corresponds to two layers. The other ul = 315.29dnoz-11.736 139 corresponds to three layers. The second step would be selecting a dnoz. This design decision is usually influenced by the available MBE nozzles on the market. The last step would be calculating the ul based on these two equations. This way, at 10% εg, the selected ul and dnoz combination would create mass transfer unit cell jet cones that fill up the entirety of the MBE-ISR with no wasted head space. Figure 6-8. Stagnant fluid residence time inside the MBE-ISR as a function of both ul and dnoz Figure 6-9 is the maximum Ediss inside the bioreactor at any given ul and dnoz. For the present work, only the maximum level of Ediss was calculated. Smaller dnoz has higher Ediss due to higher pressure drop at inside the bioreactor. Increasing the ul only marginally increases Ediss. 140 Figure 6-9. Maximum Ediss as a function of both ul and dnoz inside the MBE-ISR Figure 6-10 is the calculated total OUR as a function of ul and dnoz. The total OUR increases at lower dnoz. OUR also increases significantly as ul increases. This tendency in OUR favors scale up Ql during continuous operation since higher Ql provides higher OUR for a MBE-ISR. Both OUR and Ppump shares the same trend but on a different scale so it was difficult to predict whether or not a combination of independent variables would be cost efficient. To visualize this effect, a new parameter, OUR/ Ppump, was introduced as an indicator to the cost efficiency of the system, where OUR is the benefit, and W is considered to be the cost. 141 Figure 6-10. Total OUR inside the bioreactor as a function of both ul and dnoz inside the ISR Figure 6-11 demonstrates the cost-efficiency for a BIG bioreactor by plotting OUR/ Ppump as a function of both ul and dnoz. As a new layer of nozzles was added to the system due to shrinking Vcone, the cost efficiency value undergoes a marginally small drop. This trend indicates that adding a newer layer of nozzles reduces the cost-efficiency somehow, mainly due to a newer layer just aerating a small volume at the top of the column was not very beneficial. 142 Figure 6-11. Total OUR/Ppump inside the bioreactor as a function of both ul and dnoz Power per reactor volume is also an important factor that needs to be concerned with when designing a large-scale bioreactor. Figure 6-12, shows the estimated values of power per volume under different conditions. This value shares the same trend with power consumption value. Figure 6-12. Power density inside the bioreactor as a function of both ul and dnoz 143 CONCLUSION This study started to explore whether or not microbubble ejector is a cost-effective method to improve the gas to liquid mass-transfer rate. E. coli cell growth experiment inside a 300L pilot scale reactor vessel confirmed MBE’s aeration rate advantage over conventional ring sparger and stirrer combination strategy. MBE managed to grow E. coli cells at double the growth rate compared to the conventional stirrer. Afterwards, a critical literature study was conducted to study the mechanism of various MBGs and explored the cost-efficiency of a modified Jameson Cell system from Lonza Tech and the microbubble ejector used for this study. The theoretical study showed that MJC system has better cost-efficiency at higher gas void fraction, while the ejector has higher cost-efficiency at higher Re. The intersection threshold of which one particular MBG is better than the other one was also developed and can be helpful to engineers during decision making phase when designing gas- intense operations. Following up the theoretical simulation of microbubble ejector performance was the experiments to validate the actual performance capability of a microbubble ejector. The most aspect of the MBE performance that requires validation was the produced microbubbles’ Dbub upon leaving the ejector nozzle. A new two-phase friction model was developed and this following study was to investigate whether the two-phase friction model represents the true bubble size better than the old one-phase Fanning friction model. Due to the technical challenge to measure the bubbles 144 directly using optical signal, an alternative indirective bubble size measurement method was developed. By sparging air bubbles into oxygen-deprived water, the DOP can be used to estimate the mass-transfer characteristics of the microbubbles. By working backwards from the mass- transfer properties of the bubbly flow, the Dbub of the microbubbles produced by MBE can be indirectly measured. This indirectly measured microbubble diameter could be used to verify the aforementioned bubble size correlation equation. To ensure the DOP actually represents the mass transfer properties of fresh microbubbles leaving the MBE, the flow system was modified. The key to ensure the full DOP behaves as if the bubbles just left MBE, the flow system was set to be a long tube so the superficial liquid velocity and the kinetic energy associated with it could be constant. As long as the liquid phase kinetic energy is constant, the bubble sizes should be maintained to be constant. Since the equipment was different from existing models, the actual mixing model and friction factor needed to be actually measured instead of deducted using existing literature. A custom photometer was designed and manufactured to measure the Residence Time Distribution of the flow system to calculate the mixing model of the custom system. A photometer was used because normal conductivity probe and salt tracer used for RTD measurements are too slow to react to the high-speed fluid in the custom flow system. The conclusion was that PFR flow model fits the new flow system’s RTD. Pressure drop profile was used to calculate the measured two-phase friction factor; it was discovered that the Garcia correlation did not fit this experiment’s data well. A new pressure drop 145 correlation was developed to account for the effect of void fraction on two-phase friction factor. The new correlation was used in the following mass-transfer experiment to account for the change in mass-transfer drive force in the mass balance between bubbles and the liquid phase. A collection of mass-transfer equations was developed to estimate the change in the DO value in the liquid phase. A galvanic DOS was used to measure the DOP and the results were fitted to the previous DO model. The parameters were fit and the fitting results indicated that using a new two-phase friction model did not significantly improve the fitting parameters when compared to using the old Fanning friction model. However, since the newer friction model fit the pressure profile better, for the sake of consistency and fidelity, the new friction model is still recommended when prediction bubble size for MBE. Finally, the knowledge and conclusion obtained during the mass-transfer experiment was used to simulate the performance of an industrial-scale reactor aerated by MBE nozzle formations. The final simulation results can be used to help engineers and scientists to design reactor vessels with high gas-transfer demand and improve the production rate of two-phae reactions. An unexpected conclusion reached was that the lab-made 7 mm MBE performance is on par with the market-ready Riverforest commercial 7 mm model. 146 FUTURE WORKS During the scale-up process for ejector-sparged industrial-scale reactor, both the mass transfer property and the geometry of the mass transfer unit cell needed to be characterized. The mass transfer property of microbubble ejector was characterized using the steady state dissolved oxygen experiment. However, the geometry of the mass transfer unit cell was not investigated thoroughly to observe the effect of gas fraction on the cell geometry. The existing literature that was used to derive Eqn 6-3 was based on liquid-liquid entrainment. The 12° angle used for Eqn 6-1 through 6-3 was measured when turbulent liquid jet was injected into liquid body. No gas fraction was present in the study that identified the 12° angle within the jet cone side profile. Given the importance of this angle on the geometry of the mass transfer unit cell, it is crucial to accurately measure the true angle from the jet cone centerline to the cone boundary. Similar strategy used by Cushman could be applied to approach this investigation. A camera can be used to identify the boundary between the liquid jet and bulk fluid body with some image processing techniques such as increasing image contrast. 147 BIBLIOGRAPHY 1. Li, X., SYSTEM AND METHOD FOR IMPROVED GAS DISSOLUTION. 2014, LanzaTech New Zealand Limited: United States. 2. International Critical Tables. 1928: National Research Council. 3. Reducing Emissions Using Methanotrophic Organisms for Transportation Energy (REMOTE). CFDA Number 81.135 2013; Available from: https://arpa-e- foa.energy.gov/FileContent.aspx?FileID=4f84a273-85d7-447c-9ffc-811282a97eb0. 4. Garcia-Ochoa, F. and E. Gomez, Bioreactor scale-up and oxygen transfer rate in microbial processes: An overview. Biotechnology Advances, 2009. 27(2): p. 153-176. 5. Vantriet, K., Review of measuring methods and results in non-viscous gas-liquid mass- transfer in stirred vessels. Industrial & Engineering Chemistry Process Design and Development, 1979. 18(3): p. 357-364. 6. Bredwell, M.D. and R.M. Worden, Mass-transfer properties of microbubbles. 1. Experimental studies. Biotechnology Progress, 1998. 14(1): p. 31-38. 7. Bredwell, M.D., P. Srivastava, and R.M. Worden, Reactor design issues for synthesis-gas fermentations. Biotechnology Progress, 1999. 15(5): p. 834-844. 8. Bredwell, M.D., and R.M. Worden, Mass Transport Characteristics of Microbubbles: Experimental Studies. Biotechnology Progress, 1998. 14. 9. Agarwal, A., W.J. Ng, and Y. Liu, Principle and applications of microbubble and nanobubble technology for water treatment. Chemosphere, 2011. 84(9): p. 1175-1180. 10. Takahashi, M., K. Chiba, and P. Li, Free-radical generation from collapsing microbubbles in the absence of a dynamic stimulus. Journal of Physical Chemistry B, 2007. 111(6): p. 1343-1347. 11. Mulakhudair, A.R., J. Hanotu, and W. Zimmerman, Exploiting microbubble-microbe 148 synergy for biomass processing: Application in lignocellulosic biomass pretreatment. Biomass & Bioenergy, 2016. 93: p. 187-193. 12. Ren, Y.Z., et al., Oxidation of Phenol by Microbubble-Assisted Microelectrolysis. Chemical Engineering & Technology, 2011. 34(5): p. 699-706. 13. Donati, G. and R. Paludetto, Scale up of chemical reactors. Catalysis Today, 1997. 34(3- 4): p. 483-533. 14. Yasin, M., et al., Microbial synthesis gas utilization and ways to resolve kinetic and mass- transfer limitations. Bioresource Technology, 2015. 177: p. 361-374. 15. Li, H.Z., et al., Characteristics of Micro-Nano Bubbles and Potential Application in Groundwater Bioremediation. Water Environment Research, 2014. 86(9): p. 844-851. 16. Temesgen, T., et al., Micro and nanobubble technologies as a new horizon for water- treatment techniques: A review. Advances in Colloid and Interface Science, 2017. 246: p. 40-51. 17. Gurung, A., O. Dahl, and K. Jansson, The fundamental phenomena of nanobubbles and their behavior in wastewater treatment technologies. Geosystem Engineering, 2016. 19(3): p. 133-142. 18. Srithongouthai, S., et al., Control of dissolved oxygen levels of water in net pens for fish farming by a microscopic bubble generating system. Fisheries Science, 2006. 72(3): p. 485- 493. 19. Endo, A., et al., DO-increasing effects of a microscopic bubble generating system in a fish farm. Marine Pollution Bulletin, 2008. 57(1-5): p. 78-85. 20. Zhao, B., F.A. Agblevor, and J.G. Jelesko, Enhanced production of hairy root metabolites using microbubble generator. Plant Cell Tissue and Organ Culture, 2014. 117(2): p. 157- 165. 21. Hashim, A., et al., Review of Micro-bubble Ship Resistance Reduction Methods and the 149 Mechanisms that Affect the Skin Friction on Drag Reduction from 1999 to 2015. Jurnal Teknologi, 2015. 74(5): p. 105-114. 22. Al-Mashhadani, M.K.H., S.J. Wilkinson, and W.B. Zimmerman, Airlift bioreactor for biological applications with microbubble mediated transport processes. Chemical Engineering Science, 2015. 137: p. 243-253. 23. Kaster, J.A., Michelsen, D.L., Velander, W.H., Increased Oxygen Transfer in a Yeast Fermentation Using a Microbubble Dispersion. Applied Biochemistry and Biotechnology, 1990. 24/25: p. 469--484. 24. Bangalore, D.V. and D.D. Bellmer, Microbubbles for enhancement of oxygen transfer in xanthan gum fermentation. Chemical Engineering Communications, 2006. 193(10): p. 1232-1252. 25. Zhang, W., Z.H. Li, and F.A. Agblevor, Microbubble fermentation of recombinant Pichia pastoris for human serum albumin production. Process Biochemistry, 2005. 40(6): p. 2073- 2078. 26. Parmar, R. and S.K. Majumder, Microbubble generation and microbubble-aided transport process intensification-A state-of-the-art report. Chemical Engineering and Processing, 2013. 64: p. 79-97. 27. Zimmerman, W.B., et al., Microbubble Generation. 2008. - 2. 28. Terasaka, K., et al., Development of microbubble aerator for waste water treatment using aerobic activated sludge. Chemical Engineering Science, 2011. 66(14): p. 3172-3179. 29. Rodriguez-Rodriguez, J., et al., Generation of Microbubbles with Applications to Industry and Medicine, in Annual Review of Fluid Mechanics, Vol 47, S.H. Davis and P. Moin, Editors. 2015. p. 405-429. 30. Burns, S.E., S. Yiacoumi, and C. Tsouris, Microbubble generation for environmental and industrial separations. Separation and Purification Technology, 1997. 11(3): p. 221-232. 150 31. Ansari, M., H.H. Bokhari, and D.E. Turney, Energy efficiency and performance of bubble generating systems. Chemical Engineering and Processing - Process Intensification, 2018. 125: p. 44-55. 32. Makuta, T., R. Suzuki, and T. Nakao, Generation of microbubbles from hollow cylindrical ultrasonic horn. Ultrasonics, 2013. 53(1): p. 196-202. 33. Kulkarni, A.A. and J.B. Joshi, Bubble formation and bubble rise velocity in gas-liquid systems: A review. Industrial & Engineering Chemistry Research, 2005. 44(16): p. 5873- 5931. 34. Parmar, R. and S.K. Majumder, Terminal rise velocity, size distribution and stability of microbubble suspension. Asia-Pacific Journal of Chemical Engineering, 2015. 10(3): p. 450-465. 35. Nedderman, E.I. STOKES' LAW FOR SOLID SPHERES AND SPHERICAL BUBBLES. 2021 [cited 2021 Jan 22]. 36. Talaia, M.A.R., Terminal Velocity of a Bubble Rise in a Liquid Column, in Proceedings of World Academy of Science, Engineering and Technology, Vol 22, C. Ardil, Editor. 2007. p. 264-268. 37. Azad, M.A.K. and S. Razia, A NUMERICAL MODEL FOR BUBBLE SIZE DISTRIBUTION IN TURBULENT GAS-LIQUID DISPERSION. 2010. - 24. 38. Postema, M., et al., Ultrasound-induced coalescence of free gas microbubbles, in 2004 IEEE Ultrasonics Symposium, Vols 1-3, M.P. Yuhas, Editor. 2004. p. 1-4. 39. Kim, Y., et al., Coalescence preference in densely packed microbubbles. Scientific Reports, 2015. 5. 40. Chen, R., et al., Spatial and Temporal Scaling of Unequal Microbubble Coalescence. Aiche Journal, 2017. 63(4): p. 1441-1450. 41. Patel, D., et al., 2016. - 2016: p. - 19. 151 42. Prince, M.J., and H.W. Blanch, Bubble Coalescence and Break-Up in Air-Sparged Bubble Columns. AIChE Journal, 1990. 36(10): p. 1485-1499. 43. Zahradnik, J., M. Fialova, F. Kastanek, K.D. Green, and N.H. Thomas, The Effect of Electrolytes on Bubble Coalescence and Gas Holdup in Bubble Column Reactors. TransIChemE, 1995. 73(Part A): p. 341-346. 44. Kluytmans, J.H.J., et al., Gas holdup in a slurry bubble column: Influence of electrolyte and carbon particles. Industrial & Engineering Chemistry Research, 2001. 40(23): p. 5326-5333. 45. Zahradnik, J., M. Fialova, and V. Linek, The effect of surface-active additives on bubble coalescence in aqueous media. Chemical Engineering Science, 1999. 54(21): p. 4757-4766. 46. Anna, S.L., Droplets and Bubbles in Microfluidic Devices, in Annual Review of Fluid Mechanics, Vol 48, S.H. Davis and P. Moin, Editors. 2016. p. 285-309. 47. Amiri, M.C., and E.T. Woodburn, A Method for the Characterization of Collodial Gas Aphron Dispersions. Trans IChemE, 1990. 68(A): p. 154-160. 48. Jauregi, P., G.R. Mitchell, and J. Varley, Colloidal gas aphrons (CGA): Dispersion and structural features. Aiche Journal, 2000. 46(1): p. 24-36. 49. Save, S.V., V.G. Pangarkar, Characterization of Collodial Gas Aphrons. Chemical Engineering Communications, 1994. 127: p. 35-54. 50. Parmar, R. and S.K. Majumder, Hydrodynamics of Microbubble Suspension Flow in Pipes. Industrial & Engineering Chemistry Research, 2014. 53(9): p. 3689-3701. 51. Kawahara, A., et al., Prediction of micro-bubble dissolution characteristics in water and seawater. Experimental Thermal and Fluid Science, 2009. 33(5): p. 883-894. 52. Kim, S.J., et al., Microbubble-inducing characteristics depending on various nozzle and pressure in dissolved air flotation process. Ksce Journal of Civil Engineering, 2015. 19(3): p. 558-563. 152 53. Evans, G.M., G.J. Jameson, and B.W. Atkinson, PREDICTION OF THE BUBBLE-SIZE GENERATED BY A PLUNGING LIQUID JET BUBBLE COLUMN. Chemical Engineering Science, 1992. 47(13-14): p. 3265-3272. 54. Bin, A.K., GAS ENTRAINMENT BY PLUNGING LIQUID JETS. Chemical Engineering Science, 1993. 48(21): p. 3585-3630. 55. Clayton, R., G.J. Jameson, and E.V. Manlapig, THE DEVELOPMENT AND APPLICATION OF THE JAMESON CELL. Minerals Engineering, 1991. 4(7-11): p. 925- 933. 56. Li, B., et al., Cyclo-microbubble column flotation of fine coal. Separation Science and Technology, 2003. 38(5): p. 1125-1140. 57. Majumder, S.K., G. Kundu, and D. Mukherjee, Bubble size distribution and gas-liquid interfacial area in a modified downflow bubble column. Chemical Engineering Journal, 2006. 122(1-2): p. 1-10. 58. Evans, G.M., A.K. Bin, and P.M. Machniewskj, Performance of confined plunging liquid jet bubble column as a gas-liquid reactor. Chemical Engineering Science, 2001. 56(3): p. 1151-1157. 59. Jakubowski, C.A., et al., Ozone mass transfer in a confined plunging liquid jet contactor. Ozone-Science & Engineering, 2003. 25(1): p. 1-12. 60. Jakubowski, C.A., et al., Ozone mass transfer in the mixing zone of a Confined Plunging Liquid Jet Contactor. Ozone-Science & Engineering, 2006. 28(3): p. 131-140. 61. Hernandez-Alvarado, F., et al., Void fraction, bubble size and interfacial area measurements in co-current downflow bubble column reactor with microbubble dispersion. Chemical Engineering Science, 2017. 168: p. 403-413. 62. Ohnari, H., et al., 1999. - 46: p. - 244. 63. Facciolo, L., et al., A study of swirling turbulent pipe and jet flows. Physics of Fluids, 2007. 153 19(3). 64. Escue, A. and J. Cui, Comparison of turbulence models in simulating swirling pipe flows. Applied Mathematical Modelling, 2010. 34(10): p. 2840-2849. 65. Sloan, D.G., P.J. Smith, and L.D. Smoot, MODELING OF SWIRL IN TURBULENT-FLOW SYSTEMS. Progress in Energy and Combustion Science, 1986. 12(3): p. 163-250. 66. Hato, Y., MICRO-BUBBLE GENERATOR AND MICRO-BUBBLE GENERATION DEVICE. 2012: United States. 67. Li, P. and H. Tsuge, Water treatment by induced air flotation using microbubbles. Journal of Chemical Engineering of Japan, 2006. 39(8): p. 896-903. 68. Lawrence R. Wang, L.K., Mu Hao S. Wang,, GAS DISSOLVING SYSTEM AND METHOD. 1991, International Environmental Systems: United States. 69. Yamaguchl, T.H., MICRO BUBBLE GENERATING DEVICE AND SILICON WAFER CLEANING APPARATUS. 2013, Siltronic AG: United States. 70. Levitsky, I., D. Tavor, and V. Gitis, Generation of Two-Phase Air-Water Flow with Fine Microbubbles. Chemical Engineering & Technology, 2016. 39(8): p. 1537-1544. 71. Li, H.Z., et al., Subsurface Transport Behavior of Micro-Nano Bubbles and Potential Applications for Groundwater Remediation. International Journal of Environmental Research and Public Health, 2014. 11(1): p. 473-486. 72. Fujiwara, A., et al., Bubble breakup phenomena in a venturi tube. Fedsm 2007: Proceedings of the 5th Joint Amse/Jsme Fluids Engineering Summer Conference Vol 1, Pts a and B. 2007. 553-560. 73. Yin, J.L., et al., Experimental study on the bubble generation characteristics for an venturi type bubble generator. International Journal of Heat and Mass Transfer, 2015. 91: p. 218- 224. 154 74. Gordiychuk, A., et al., Size distribution and Sauter mean diameter of micro bubbles for a Venturi type bubble generator. Experimental Thermal and Fluid Science, 2016. 70: p. 51- 60. 75. Stevens, R., MICRO BUBBLE DEVICE, SYSTEMS AND METHODS RELATED THERETO. 2016: United States. 76. Zhao, L., et al., A visualized study of the motion of individual bubbles in a venturi-type bubble generator. Progress in Nuclear Energy, 2017. 97: p. 74-89. 77. Donald L. Michenlsen, F.S., Thomas M. Murpy, Microbubble Generator. 1994, Virginia Polytechnic Instistute and State University: United States. 78. Hensirisak, P., et al., Scale-up of microbubble dispersion generator for aerobic fermentation. Applied Biochemistry and Biotechnology, 2002. 101(3): p. 211-227. 79. Brittle, S., et al., Minimising microbubble size through oscillation frequency control. Chemical Engineering Research & Design, 2015. 104: p. 357-366. 80. Tesař, V. and M. Jílek, INTEGRAL FLUIDIC GENERATOR OF MICROBUBBLES, in Colloquium FLUID DYNAMICS. 2013: Prague. 81. Zimmerman, W.B., et al., On the design and simulation of an airlift loop bioreactor with microbubble generation by fluidic oscillation. Food and Bioproducts Processing, 2009. 87(C3): p. 215-227. 82. Rehman, F., et al., Fluidic oscillator-mediated microbubble generation to provide cost effective mass transfer and mixing efficiency to the wastewater treatment plants. Environmental Research, 2015. 137: p. 32-39. 83. Tesar, V., C.H. Hung, and W.B. Zimmerman, No-moving-part hybrid-synthetic jet actuator. Sensors and Actuators a-Physical, 2006. 125(2): p. 159-169. 84. Zimmerman, W.B. and V. Tesař, 2007. 155 85. Wang, J.Y., et al., Improving the performance of coal flotation using oscillatory air supply. Fuel Processing Technology, 2017. 165: p. 131-137. 86. Al-Mashhadani, M.K.H., H.C.H. Bandulasena, and W.B. Zimmerman, CO2 Mass Transfer Induced through an Airlift Loop by a Microbubble Cloud Generated by Fluidic Oscillation. Industrial & Engineering Chemistry Research, 2012. 51(4): p. 1864-1877. 87. Al-Mashhadani, M.K.H., S.J. Wilkinson, and W.B. Zimmerman, Carbon dioxide rich microbubble acceleration of biogas production in anaerobic digestion. Chemical Engineering Science, 2016. 156: p. 24-35. 88. Tesar, V., Microbubble generator excited by fluidic oscillator's third harmonic frequency. Chemical Engineering Research & Design, 2014. 92(9): p. 1603-1615. 89. Hanotu, J.O., H. Bandulasena, and W.B. Zimmerman, Aerator design for microbubble generation. Chemical Engineering Research & Design, 2017. 123: p. 367-376. 90. Chung, S.K., Y. Zhao, and S.K. Cho, On-chip creation and elimination of microbubbles for a micro-object manipulator. Journal of Micromechanics and Microengineering, 2008. 18(9). 91. Hammadi, Z., R. Morin, and J. Olives, Field nano-localization of gas bubble production from water electrolysis. Applied Physics Letters, 2013. 103(22). 92. Hammadi, Z., et al., Immobilization of a bubble in water by nanoelectrolysis. Applied Physics Letters, 2016. 109(6). 93. Garstecki, P., et al., Formation of droplets and bubbles in a microfluidic T-junction - scaling and mechanism of break-up. Lab on a Chip, 2006. 6(3): p. 437-446. 94. Makuta, T., et al., Generation of micro gas bubbles of uniform diameter in an ultrasonic field. Journal of Fluid Mechanics, 2006. 548: p. 113-131. 95. Najafi, A.S., Z.H. Xu, and J. Mashyah, Single micro-bubble generation by pressure pulse technique. Chemical Engineering Science, 2008. 63(7): p. 1779-1787. 156 96. Makuta, T., Y. Aizawa, and R. Suzuki, Sonochemical reaction with microbubbles generated by hollow ultrasonic horn. Ultrasonics Sonochemistry, 2013. 20(4): p. 997-1001. 97. Postema, M., et al., Ultrasound-induced microbubble coalescence. Ultrasound in Medicine and Biology, 2004. 30(10): p. 1337-1344. 98. Kobayashi, D., et al., Agglomeration and rapid ascent of microbubbles by ultrasonic irradiation. Ultrasonics Sonochemistry, 2011. 18(5): p. 1193-1196. 99. Tatsuya NUMAKURA, K.K., Toshinori MAKUTA, Development and Optimization of a Microbubble Generator with a Hollow Cylindrical Ultrasonic Horn. Journal of JSEM, 2014. 14(Special): p. 4. 100. Achaoui, Y., et al., Tunable microbubble generator using electrolysis and ultrasound. Aip Advances, 2017. 7(1). 101. Shirtum, R.P., SHEAR MIXING APPARATUS AND USE THEREOF. 2000, The DOW Chemical Company: United States. 102. Chen, X.T., X.D. Cai, and J.P. Brill, A general model for transition to dispersed bubble flow. Chemical Engineering Science, 1997. 52(23): p. 4373-4380. 103. Park, S. and H. Yang, Flow and Oxygen-Transfer Characteristics in an Aeration System Using an Annular Nozzle Ejector. Industrial & Engineering Chemistry Research, 2013. 52(4): p. 1756-1763. 104. Aly, N.H., A. Karameldin, and M.M. Shamloul, Modelling and simulation of steam jet ejectors. Desalination, 1999. 123(1): p. 1-8. 105. Kim, M.I., et al., Numerical and experimental investigations of gas-liquid dispersion in an ejector. Chemical Engineering Science, 2007. 62(24): p. 7133-7139. 106. Molin, D., C. Marchioli, and A. Soldati, Turbulence modulation and microbubble dynamics in vertical channel flow. International Journal of Multiphase Flow, 2012. 42: p. 80-95. 157 107. Garcia, F., et al., Friction factor improved correlations for laminar and turbulent gas-liquid flow in horizontal pipelines. International Journal of Multiphase Flow, 2007. 33(12): p. 1320-1336. 108. Neto, I.E.L., D.Z. Zhu, and N. Rajaratnam, Bubbly jets in stagnant water. International Journal of Multiphase Flow, 2008. 34(12): p. 1130-1141. 109. Cushman-Roisin, B., Turbulent Jets, in Environmental Fluid Mechanics 2010, John Wiley & Sons. 110. Kotsovinos, N.E., A note on the spreading rate and virtual origin of a plane turbulent jet. Journal of Fluid Mechanics, 1976. 77(2): p. 305-311. 111. Paliy, O. and T.S. Gunasekera, Growth of E. coli BL21 in minimal media with different gluconeogenic carbon sources and salt contents. Appl Microbiol Biotechnol, 2007. 73(5): p. 1169-72. 112. Besagni, G., F. Inzoli, and T. Ziegenhein, Two-Phase Bubble Columns: A Comprehensive Review. ChemEngineering, 2018. 2(2): p. 13. 113. Bicalho, R.C., et al., Susceptibility of Escherichia coli isolated from uteri of postpartum dairy cows to antibiotic and environmental bacteriophages. Part I: Isolation and lytic activity estimation of bacteriophages. Journal of Dairy Science, 2010. 93(1): p. 93-104. 114. Philip, P., et al., Parallel substrate supply and pH stabilization for optimal screening of E. coli with the membrane-based fed-batch shake flask. Microbial Cell Factories, 2018. 17(1): p. 69. 115. Doyle, M.P. and J.L. Schoeni, Survival and growth characteristics of Escherichia coli associated with hemorrhagic colitis. Appl Environ Microbiol, 1984. 48(4): p. 855-6. 116. Geerlof, A. M9 mineral medium. M9 mineral medium 2010 [cited 2017. 117. Malavasi, S., et al., On the pressure losses through perforated plates. 2012. - 28: p. - 66. 158 118. Madani Sani, F., et al., Review of the API RP 14E erosional velocity equation: Origin, applications, misuses, limitations and alternatives. Wear, 2019. 426-427: p. 620-636. 119. Kundu, G., D. Mukhherjee, and A.K. Mitra, Gas Entrainment And Depth Of Penetration In A Co-Current Gas-Liquid Downflow Bubble Column. Journal of Chemical Engineering of Japan, 1994. 27(5): p. 5. 120. Couto, H.J.B., et al., Micro-bubble size distribution measurements by laser diffraction technique. Minerals Engineering, 2009. 22(4): p. 330-335. 121. Lad, V.N. and Z.V.P. Murthy, Breakup of free liquid jets influenced by external mechanical vibrations. Fluid Dynamics Research, 2016. 49(1): p. 015503. 122. Vejražka, J., M. Zedníková, and P. Stanovský, Experiments on breakup of bubbles in a turbulent flow. AIChE Journal, 2018. 64(2): p. 740-757. 123. Moody, L.F., Friction Factors for Pipe Flow. Transactions of the American Society of Mechanical Engineers, 1944. 66. 124. Software, P.F.E. Pipe Materials and Common Pipe Roughness Values. 2022 [cited 2021. 125. DUMP, F. PUMP TYPES EXPLAINED. 2021 [cited 2022. 126. Sensorex. Selecting Among Dissolved Oxygen Measurement Methods (Optical, Galvanic, Polarographic). 2018 [cited 2021. 127. Instrument, B. Gas Correction Factors. [cited 2021 2021]. 128. Fogler, H.S., Elements of Chemical Reaction Engineering. 2nd ed. 1992, Englewood Cliffs, NJ: Prentice-Hall. 129. Inc, T.F.S., Food Dyes and Beer’s Law, T.F.S. Inc, Editor. 2020, Thermo Fisher Scientific Inc. 159 130. Márquez-Torres, L., et al., Comparison of 63 different void fraction correlations for different flow patterns, pipe inclinations, and liquid viscosities. SN Applied Sciences, 2020. 2(10): p. 1695. 131. Scientific, A., Atlas LG DO Probe, A. Scientific, Editor. 2020. 132. Doran, P.M., Preface to the Second Edition, in Bioprocess Engineering Principles (Second Edition), P.M. Doran, Editor. 2013, Academic Press: London. p. vii-viii. 133. Vantriet, K., REVIEW OF MEASURING METHODS AND RESULTS IN NONVISCOUS GAS-LIQUID MASS-TRANSFER IN STIRRED VESSELS. Industrial & Engineering Chemistry Process Design and Development, 1979. 18(3): p. 357-364. 134. Lewis, D.A. and J.F. Davidson, BUBBLE SIZES PRODUCED BY SHEAR AND TURBULENCE IN A BUBBLE COLUMN. Chemical Engineering Science, 1983. 38(1): p. 161-167. 135. Perez, J.A.S., et al., Shear rate in stirred tank and bubble column bioreactors. Chemical Engineering Journal, 2006. 124(1-3): p. 1-5. 136. Tramper, J., et al., SHEAR SENSITIVITY OF INSECT CELLS IN SUSPENSION. Enzyme and Microbial Technology, 1986. 8(1): p. 33-36. 137. Millward, H.R., et al., MAMMALIAN-CELL DAMAGE IN A NOVEL MEMBRANE BIOREACTOR. Biotechnology and Bioengineering, 1994. 43(9): p. 899-906. 138. Li, M.G., X.Y. Tian, and X.B. Chen, Temperature Effect on the Shear-Induced Cell Damage in Biofabrication. Artificial Organs, 2011. 35(7): p. 741-746. 139. Liu, Y., et al., Effects of bubble–liquid two-phase turbulent hydrodynamics on cell damage in sparged bioreactor. 2013. - 30(- 1): p. - 58. 140. Reichmann, F., et al., Gas-liquid dispersion in micronozzles and microreactor design for high interfacial area. Chemical Engineering Science, 2017. 169: p. 151-163. 160 141. Kockmann, N. and D.M. Roberge, Harsh Reaction Conditions in Continuous-Flow Microreactors for Pharmaceutical Production. Chemical Engineering & Technology, 2009. 32(11): p. 1682-1694. 142. Reichmann, F., F. Varel, and N. Kockmann, Energy Optimization of Gas-Liquid Dispersion in Micronozzles Assisted by Design of Experiment. Processes, 2017. 5(4). 143. Knowino. Henry's law. 2010 [cited 2022 2022]. 144. Pallas, N.R. and Y. Harrison, Colloids and Surfaces. 1990. p. 169–194. 161