Experimental system for investigating membrane gas–liquid mass transfer during a parabolic flight

Abstract

This research tests the hypothesis that oxygen mass transfer through a hollow fiber membrane contactor remains unaffected by changes in gravity. To validate this, oxygen mass transfer coefficients were measured under terrestrial, Martian, Lunar, and microgravity conditions simulated during parabolic flights. Baseline data from 15 terrestrial experiments were statistically compared via Student’s T-Test to results from 10 Martian, 10 Lunar, and 5 microgravity parabolas. The respective p-values of 0.41, 0.48, and 0.85, all well above the 0.05 threshold, indicate no significant difference in oxygen mass transfer across gravity levels. These findings support the robustness of membrane gas transfer technologies across varied gravitational environments, advancing their Technology Readiness Level. This outcome has important implications for life support and in-situ resource utilization (ISRU) systems in space habitats and promises to enhance membrane process efficiency in diverse terrestrial and extraterrestrial applications.

Introduction

The efficiency of many chemical processes, crucial for the sustainability of life in space, hinges on the rate at which reagent gases dissolve into liquid media. This is notably critical in bioreactor cell growth, where dissolved gases—either in deficit or excess—can significantly limit biological processes. Bioreactors, as highlighted in recent studies, are pivotal for long-term space missions, providing a mass-efficient method for producing essential resources such as food and oxygen (Alvarado et al. 2021; Fahrion et al. 2021). These systems necessitate the effective dissolution of feed gases into growth media and the proficient extraction of metabolic waste gases, a process complicated by the variable gravitational conditions encountered beyond Earth. One such process pioneered by Mango Materials makes use of methanotrophic bacteria to process methane gas into the biopolymer poly-3-hydroxybutyrate (PHB) in the presence of oxygen (Pieja et al. 2017). This polymer has wide ranging applications in additive manufacturing, textile, and as a food additive.

Traditionally, the dissolution of gases in bioreactors on Earth is achieved through bubbling, a method where feed gases are introduced at the bottom of a vessel typically through a diffuser and allowed to ascend, dissolving into the liquid. This method, however, faces efficiency losses due to the varying sizes of the bubbles and the gradient in hydrostatic pressure as they rise. A further complication arises when these bubbles reach the surface; unable to dissolve, they are often vented out, representing a loss of potential resources. Moreover, the phenomenon of bubble coalescence, more prevalent under microgravity conditions, exacerbates these inefficiencies by reducing the effective surface area available for gas exchange (Colin et al. 2008).

Gas exchange membranes, also known as membrane contactors, offer a promising alternative. They circumvent the limitations of bubbling by providing a consistent surface area for bidirectional mass transfer, allowing a more controlled and uniform diffusive transfer of feed gases into bioreactors as well as the extraction of waste gases (Hou et al. 2019; Schell 1984). The independent control of liquid and gas flow rates along the membrane surface is essential to minimize boundary layer thickness leading to higher gas mass transfer rates and efficiency. A comparative pilot study found that membrane aeration could transfer up to 6 kg O₂/kWh or 4 times more than bubble aeration (Côté et al. 2015). The membrane also allows slight pressurization of the liquid which increases gas solubility and hence gas dissolution rate as well as gas availability for microbes. Higher gas mass transfer rates, efficiency and gas solubility will lead to higher growth densities and ultimately more compact systems, which are essential for space applications. This technology not only competes effectively in terrestrial applications (Orgill et al. 2013), but also opens the door to recycling feed gases, significantly reducing operational costs, and removing potentially growth-inhibiting waste gases.

The authors have previously developed a membrane-aerated bioreactor (MABR) to enable the production of PHB polymers from in situ resources in space for a NASA SBIR Phase I (Pieja, et al. 2017) and II (Pieja et al. 2019) project. In the first phase of this process methanotrophs convert oxygen and methane into cell biomass. After growth stabilization ammonium nutrient dosing is switched off. This triggers the microbes to produce the PHB biopolymer which can easily be harvested by cell lysis through various means. It was successfully demonstrated that PHB can be produced by introducing gases into the reactor with membrane contactors instead of bubble diffusers.

This MABR system for the bioconversion of methane into PHB has a high infusion potential with existing and future life support, manufacturing, and waste systems. Methane is available from the Sabatier Reaction as part of the CO₂ processing assembly and is currently vented as a waste gas from the International Space Station (Greenwood et al. 2017). The methane could also come directly from the Martian atmosphere (Hintze et al. 2018) or from a Sabatier reactor fed with Martian CO₂ (Mumma et al. 2009). Upscaling these carbon sources to versatile PHB polymers would be a prime application of In-Situ Resource Utilization (ISRU). Membrane contactors are an excellent tool for in-situ resource extraction and recycling (Hou et al. 2019; Rongwong and Goh 2020; Chen et al. 2023; Bazhenov et al. 2018); they have been applied to recover ammonia from anaerobic digester centrate and concentrated nitrate brines (Schwiebert et al. 2024; Huo et al. 2020) and to recycle cyanide in metal extraction processes (Hammer et al. 2023).

Despite their potential, MABRs have not been tested in microgravity. While the ultimate goal is for long-duration testing on the International Space Station, NASA generally recommends testing technology first during shorter duration microgravity flight opportunities like the parabolic flight that was utilized in this study. Microgravity is known to alter microbial processes (Marotta et al. 2025; Li et al. 2022; Huang et al. 2018). While no effect of gravity on the gas mass transfer coefficient is expected as gas mass transfer in membrane contactors is purely driven by a difference in gas concentrations between the gas and liquid side, it is still important to validate this hypothesis before integrating this technology in other processes like bioreactors. Hence, the goal of this study is to validate that gas mass transfer rates are not affected by varying gravitational conditions. The main constraint of measuring mass transfer during parabolic flights is the limited time of consistent microgravity conditions of 15–28 s. Currently, no commercial methane sensors have response times shorter than this data collection window. Moreover, methanotrophs require both methane and oxygen to grow, a mixture that constitutes an important hazard to manage on a parabolic flight. Therefore, the effect of gravity strength on oxygen gas mass transfer only and into an abiotic medium was studied at the same temperature of 37 °C as the original culture. The short time window also excluded the use of dense polymeric membranes like polydimethylsiloxane membranes for instance.

Addressing this critical gap, our study introduces a novel system capable of measuring the mass transfer of oxygen gas through a microporous hollow fiber membrane under simulated Martian, Lunar, and microgravity conditions during parabolic flights. The system was designed to determine oxygen mass transfer rates in under 10 s with the meticulous control of physical parameters to mimic bioreactor conditions as well as eliminate as much as possible variability in mass transfer rates due to non-gravity effects. After a short section of membrane mass transfer theory, the characteristics of the different subsystems will be discussed in detail. In the results, the quality of the obtained gravity setpoints will be presented as well as the final data product of oxygen transfer rates for different gravity levels.

Materials and methods

Membrane theory

The microporous hydrophobic membranes examined in this study serve as selective barriers that are impermeable to liquid water but allow gases to cross its boundary. The primary advantage of these membranes over traditional gas transfer methods is their ability to provide a larger interfacial surface area within a more compact design. Unlike sparging, the interfacial area provided by the membrane is constant and does not depend on the flow rates of the liquid or gas. Conversely, tangential liquid and gas mass flow rates can be increased independently to increase gas mass transfer rates through the membrane, termed gas flux when flow rate is divided by the membrane area, which minimizes the boundary layer thickness. The mechanism driving the flux of gas between the liquid and gas sides of the membrane is the difference in partial pressures between the two phases. It is important to note that while partial pressures conceptually represent the solubility and concentration of dissolved gases, discussing solubility and concentrations is more practical for experimental purposes.

For a gas–liquid membrane contactor, the concentration of the gas in the liquid as a function of time is described by the following equation (Côté et al. 1989; Rezakazemi and Younas 2022):

Simply put, the dissolved concentration at time, t, given by depends on the saturation as well as the initial dissolved concentration at time zero, . The flux of oxygen is inhibited by the overall resistance to mass transfer , which is a function of the porosity, thickness and tortuosity of the membrane as well as the boundary layer resistances on the liquid and gas side. Lastly, a greater liquid volume to membrane area ratio elongates the mass transfer process. The calculation of the saturation concentration C^* is explained in Sect. 2.5.1. on oxygen sensor calibration.

Given that the concentrations of gases can be empirically measured, it may be practical to rearrange the equation as follows:

Notice that the membrane area and liquid volumes are constants which have been combined into a single term, the specific interfacial area, a. The inverse of the membrane’s resistance to flow, the overall mass transfer coefficient is the term that this work is focused on investigating.

Membrane specifications

Several membrane contactors were trialed before selecting the optimal module for the parabolic flight experiments. The selection criteria prioritized a high flux in a compact volume, as the microgravity periods were limited to approximately 15 s, which disadvantaged flat-sheet modules compared to hollow-fiber designs. The volume of water inside the module was also a key factor, since a larger internal volume slowed gas exchange and increased the difficulty of maintaining thermal stability. Additionally, because the operating water temperature was higher than ambient air, there was concern about condensation on the gas side. This was mitigated by selecting a transparent shell that allowed visual monitoring while keeping the form factor small enough to fit submerged within the custom water bath.

After evaluation, the 3 M™ Liqui-Cell™ MM-1.7 × 5.5 Series hollow-fiber module was selected (see Table 1 and Fig. 1). The fibers are potted at both ends in a sealed shell, with open-ended lumen flow for the test liquid and gas flow outside the fibers, enabling rapid gas exchange without dead-end limitations. The membrane material is polypropylene, with an outer diameter of 330 µm, an inner diameter of 220 µm, and a nominal pore size of 0.1 µm. The packing density and specific surface area are reported in Table 1.

Table 1.

Membrane contactor specifications, provided by (Estay et al. 2013)

Manufacturer	3 M™ Liqui-Cell™
Type	MM-1.7 × 5.5 Series
Membrane material	Polypropylene
Pore size	0.1 μm
Number of fibers	7400
Fiber type	X40
Surface contact area	0.58 m2
Fiber outer diameter	330 μm
Fiber inner diameter	220 μm
Shell inner diameter	0.043 m
Membrane contactor length	0.12 m
Packing density	43.6%
Specific surface area	3328 m2 m−3

Fig. 1.

Schematic of the membrane contactor used

Pneumatic subsystem

The pneumatic system played a crucial role in transitioning the experiment between oxygenation observation and the deoxygenation reset phase, effectively preparing the setup for subsequent observation cycles (Fig. 2). During the deoxygenation phase aimed at reducing the dissolved oxygen concentration to zero, nitrogen gas travels from its cylinder (C_N2) through an open cut-off valve (V₁), past a pressure gauge (P₁) that monitors cylinder pressure. It then proceeded through a pressure regulating valve (P_RN2) set to 30 PSI—this pressure level was confirmed by another gauge (P₂)—and continued through a second cut-off valve (V₂) for the pressure-regulated line.

Fig. 2.

Schematic of the pneumatic subsystem where the gas-side control occurred. Legend: C_O2 C_N2, pressurized gas cylinders; V_1-5 manually operated valves; P_1-4 pressure gauges; PR_O2 PR_N2 pressure regulator (set to 30 PSI); FC_O2 FC_N2 digital flow controllers; PC pressure controller, FM_E exhaust digital flow meter; FM₁ exhaust flow meter; T₂ temperature probe, S₃ oxygen sensor, M membrane element

The nitrogen then reached a digital flow controller (FC_N2) (Alicat, Tucson, Az), which, in coordination with its oxygen counterpart (FC_O2) (Alicat, Tuscon, Az), precisely managed the gas mixture entering the membrane. Similarly, oxygen from its cylinder (C_O2) passed through its sequence of components: valve (V₃), pressure gauge (P₃), pressure regulating valve (P_RO2), and gauge (P4), before reaching the line’s cut-off valve (V₄) and finally, FC_O2. It’s important to note that FC_N2 and FC_O2 were designed to alternate between nitrogen and oxygen supply, ensuring that only one gas type was introduced at any given time. Both controllers were set to a flow rate of 1 LPM, their effective maximum capacity.

From here, the selected gas—either nitrogen or oxygen—moved towards a three-way valve (V₅), typically directed towards the membrane but with an option to bypass if needed for troubleshooting. After passing through the membrane, the gas encountered an oxygen sensor (S₃) and an optical temperature sensor (T₂), then entered a digital pressure controller (PC) (Alicat, Tuscon, Az) that maintained the line pressure at 20 PSI. This pressure was found optimal to minimize the error on the gas mass transfer coefficient (see Sect. “Optimization”). Finally, the exhaust gas exited through a digital flow meter (FM_E) (Alicat, Tuscon, Az), passed an analogue flow meter (FM₁), and was then released to the aircraft cabin. The utilized flow rates were too low to substantially alter the cabin gas composition.

We also ensured the membrane was relatively dry prior to the experiment campaign. Condensation that formed on the shell side of the membrane module when the system was not in use was removed by pumping air that had been dried using a desiccant cartridge. During the experiment the pure gases coming out of the cylinders ensured the membrane stayed dry.

Hydraulic subsystem

The core of the experiment was the mass transfer loop, also referred to as the main loop, where the critical gas exchange process took place. This loop was initially filled with 560 ml of deionized water, with periodic replenishments from the hydraulic accumulator to compensate for any water loss through evaporation to the gas side of the membrane module. The diagram in Fig. 3 outlines the structure of this mass transfer loop and its key components, detailed as follows.

Fig. 3.

a Schematic of the hydraulic subsystem where the critical experiment components such as the membrane and oxygen sensors were housed. Legend: M membrane element; S_1,2 dissolved oxygen sensor; T_1,3 temperature probes; P_6,8 pressure gauges; P_7,9 pressure sensors; BP bypass valve; FM_L digital flow meter; FM₂ rotameter; P_MAIN main pump; HE heat exchanger; HA hydraulic accumulator; FS₁ flow-switch; V_C1,C2 check valves; V_HA valve. b The membrane module is visible within the reactor on the foreground, the suite of sensors is visible on the right and input gas and liquid lines are visible at the top

Starting from the bottom right of the schematic, the main pump (P_MAIN) (Procon) propelled the water through a quick connector, denoted as an arrow fitting into a triangle. This connector served as an interface point for external water sources, such as a bucket, facilitating the loop’s filling process. The liquid flow rate of the pump was set to 5 LPM as it was found to minimize the error on the gas mass transfer coefficient (see Sect. “Optimization”).

Proceeding from the quick connector was the stainless steel 316 shell-and-tube heat exchanger (HE) where the mass transfer loop fluid passed through the inner tube. The HE facilitated thermal exchange by circulating water from the thermal loop around the tube, effectively regulating the temperature within the mass transfer loop.

Next in line was the flow-activated switch (FS₁), which remained off under normal conditions and activated upon detecting a flow rate of 1 LPM or higher, according to the direction indicated in the schematic. This switch, in tandem with a counterpart in the thermal loop, ensured the activation of the Peltier device only when water circulated through both loops, preventing the risk of freezing or boiling due to stagnant water.

Following FS₁, we find a thermistor (T₃), which acted as the input for thermal control, and an analogue flow meter (FM₂). FM₂ served as the calibration reference for a subsequent digital flow meter. The loop also included an analogue pressure gauge (P₆) and a digital pressure sensor (P₇) both of which measured the pressure entering the membrane (M) (Liqui-Cel MM 1.7X5.5 Series Membrane Contractor G542, 3 M).

Positioned between these pressure gauges and the membrane was a pressure-triggered bypass valve, set at 30 PSI, designed to protect the membrane from excessive pressure by diverting the water through a check valve (V_C) before it reconnected to P_MAIN via another quick connector.

On the output side of the membrane, we installed two oxygen sensors (S₁ and S₂) for redundancy and a thermal probe (T₁), which was instrumental in normalizing the sensor data. A check valve (V_C2) led to the hydraulic accumulator (HA), set to a pressure of 20 PSI and integrated into the system through a solenoid valve (V_HA). This valve opened once every second during the system’s deoxygenation phase, allowing the HA to compensate for water losses due to evaporation or leaks and to eliminate air pockets within the system.

Following the oxygen and thermal sensors, another set of pressure gauge (P₈) and pressure sensor (P₉) recorded the membrane’s outgoing pressure. The system’s configuration was completed by a digital volumetric turbine flow sensor (FM_L, JLC International, New Britain, PA), strategically placed near the P_MAIN inlet.

Thermal subsystem

The thermal loop, which held approximately 4 L of water, played a crucial role in regulating the temperature within the mass transfer loop. This design choice leveraged the thermal mass of a larger water volume to achieve greater temperature stability, minimizing fluctuations and preventing direct contact between the mass transfer loop water and the copper tubing of the Peltier heat pump, which could have led to corrosion when exposed to deionized water and subsequent membrane fouling. The diagram in Fig. 4 represents the thermal loop and its key components, detailed as follows.

Fig. 4.

a Schematic of the thermal subsystem with the water bath and thermoelectric heater. Legend: WB water-bath; HE heat exchanger; FS₂ flow switch; Th thermoelectric actuator; FM₃ liquid rotameter; P₁₀ pressure gauge; T₄ temperature probe; P_TH thermal pump. b Front of the full system as it is standing up. The flight configuration has it laying on its back side. Refer to Table 2 for component details

Starting from the bottom right, the thermal loop initiated with the pump (P_TH) (Procon) propelling water past a temperature sensor (T₄) and a pressure gauge (P₁₀). The flow continued through an analogue volumetric flow meter (FM₃) and entered the Peltier device (Th). The inclusion of a flow-activated switch (FS₂), synchronized with a similar switch in the mass transfer loop, ensured the Th was activated only when water circulated through both loops and thus prevented parts of the thermal subsystem from freezing or boiling in the absence of flow. Subsequently, the heated water advanced into the heat exchanger module (HE) and then into the water bath (WB). The WB housed both the membrane and coiled gas tubing, maintaining uniform temperature across these critical components to avoid condensation that could affect gas mass transfer in the membrane module. A quick-connect access point between the HE and WB facilitated easy filling of the loop from an external water source, such as a bucket.

The flow rate within this loop was maintained at a constant 3 LPM, independently of computer control, to ensure consistent thermal regulation throughout the experiment. The thermoelectric heater’s operation was guided by a thermistor located within the mass transfer loop T₃, allowing for responsive temperature control. On average, it took the system about 30 min to reach the operational temperature of 37 °C from a starting point of 20 °C, and the error in temperature was ± 1 °C. Table 1, 2 and 3 summarize component specifications and range of controlled and monitored variables.

Table 2.

Legend of all components in the schematics: Pneumatic (P), Hydraulic (H), Thermal (T)

ID	x	Component	P	H	T	Specifications
M		Membrane		x		See Table 1
WB		Water bath			x	3.3 L capacity (empty)
HE		Heat exchanger			x
Th		Thermoelectric Heater			x	37 °C setpoint, ± 1 °C, 0.1 Hz control rate
HA		Hydraulic Accumulator		x		18 PSI
BP		By-pass valve		x		30 PSI trigger
Cx	O, N	Gas cylinder	x			2015 PSI, 8 L capacity
Vx	1–5, HA	Valve	x	x
VCx	1,2	Check Valve		x
Px	1–10	Pressure gauge/sensor	x	x	x	1 Hz sample rate
Px	MAIN, TH	Pump		x	x	(MAIN) 5 LPM, (TH) 1 GPM
PRx	N2,O2	Pressure regulator	x			30 PSI
Tx	1–4	Temperature sensor	x	x	x	1 Hz sample rate
Sx	1–3	Oxygen Sensor	x		x	2 Hz sample rate
FCx	N, O	Flow controller	x			0 or 1 LPM
PC		Pressure Controller	x			20 PSI
FMx	1–3,E,L	Flow meter	x	x	x	1 Hz sample rate
FSx	1,2	Flow switch		x	x	>  = 1 LPM ON trigger

Table 3.

Digitally controlled or monitored variables and their measured range

Variable	Unit	Range	Actuator	Sensor
Liquid temperature	C	20–40	Peltier heat pump	Thermal sensor T3
Liquid pressure	PSI	0–30		Sealed pressure sensors
Liquid volumetric flow	LPM	0–9	Procon pump	Turbine flow meter FML
Dissolved oxygen	%	0–100		Optical probes
Gas temperature	C	20–40		Optical probe
Gas pressure	PSIa	15–30	Alicat pressure regulator	Alicat pressure
Gas volumetric flow	LPM	0–1	Alicat flow regulator	Alicat flow sensor
Oxygen concentration	%	0–100		Optical probe

Oxygen sensor

Calibration

The saturation concentration for a given gas species in deionized water can be determined as a function of temperature from Henry’s constant from molarity :

where A and B are constants available on NIST (Oxygen 2023; Nitrogen 2023). To convert Henry’s law constant for molarity into a concentration-based constant one must account for the temperature-dependent density of water (U.S. Department of the Interior Bureau of Reclamation 1995).

Finally, knowing the gas pressure and molar mass of the gas yields the saturation concentration . Note that this is only true for a pure gas as is the case in our experimental setup at steady state.

The constants for oxygen are A = 0.0013, B = 1500; and for nitrogen they are A = 0.00065, B = 1300.

In instances where a complete oxygenation segment could not be measured, it was imperative to record a full cycle prior to experimentation to calibrate the data accurately. For the conditions outlined in this paper, oxygen saturation concentration was calculated to be 45.263 mg/L and nitrogen saturation concentration was calculated to be 20.334 mg/L. The first value was used to calibrate the upper limit of the oxygen sensors.

Prior to the experiment campaign, we performed some testing and sensor calibration. After ensuring that the system could maintain its target temperature, liquid and gas flow rates, we performed a two point dissolved oxygen saturation calibration. This calibration was performed at nominal temperature and pressure conditions.

Response Time

Accurate monitoring of transient responses in our system necessitates the use of gas sensors with rapid response times. It was crucial to differentiate between the sensor’s advertised response time, which often refers to the rate at which the electronic signal is sampled, and the sensor’s inherent physical response limitations. Many dissolved gas sensors, like galvanic sensors for instance, actually consist of a gas transfer membrane. This membrane has a certain resistance to gas transport and hence it takes often over a minute to obtain an accurate reading which is impractical considering the data acquisition window of maximum 30 s. In this study, robust optical oxygen sensors (OXROB10-HS, Pyroscience GmbH, Aachen, Germany) were utilized that do not have this limitation and have an advertised response time in water of 3 s.

The relevant response time, denoted by the time constant , should be determined by the sensor’s ability to reach a steady state when it is immediately transferred between a desaturated and a saturated solution. As Kirk et al. suggests (Kirk and Szita 2013), the sensor’s time constant should be significantly smaller than the time constant of the transient response being measured. More generally, it may be sufficient to follow the signal processing convention surrounding the Nyquist rate and require the sensor’s time coefficient to be smaller or equal to half of the system’s own time coefficient.

The experiment apparatus was modified slightly so that a cross junction could be quickly disconnected from the main loop. This cross junction housed an oxygen sensor at the bottom intersection, a cap at the top intersection, and quick connect male and female on either side. The system side had the remaining oxygen probe in its usual position and a cap to block out where the first probe used to be. The experiment was set up this way so that the liquid flow, gas pressure and liquid temperature all matched the parabolic experiment.

After the liquid reached its nominal temperature, the experiment began with nitrogen being introduced into the membrane until the sensors reached 1% DO, this marked the start of a dataset. At this point, the main pump was momentarily turned off so that the cross junction could be removed from the system, and the system was closed. The sensor therefore remained immersed within the junction. The pump was then restarted, the nitrogen supply was halted and replaced by oxygen for 150 s since the oxygen sensor on the system side stabilized at 100% DO within approximately 100 s. The main pump was then momentarily turned off so that the junction could be reintroduced. This had the effect to displace the oxygen-poor water within the junction almost immediately since the pump operated at 5LPM. The resulting sensor oxygenation signal was fitted to a logarithmic growth curve so that could be found. This process was performed 9 times. The data for the membrane oxygenation response was collected at the same time, during the time when the junction was disconnected.

In our experiments, the sensor’s time constant was found to be to be = 0.725 s, and a nominal oxygenation phase of the cycle had a time constant of = 19.458 s, which is 26.8 times larger than the sensor response time, so it was deemed not necessary to correct oxygen measurements for the sensor response time in these experiments.

Noise reduction

Because liquid temperature fluctuated within ± 1 °C during nominal operation, post-processing normalization of sensor readings was advisable for cycles ranging from near-zero dissolved oxygen to full saturation. Moreover, since sensor noise is greater at higher values of concentration, we segmented the data into the ascending parts of the signal and trimmed these segments to derive a curve that closely followed a logarithmic or sigmoidal growth pattern. This approximation follows the form:

A robust fit to this function served to reduce noise in the signal, and the extrapolated function could be extended indefinitely to ascertain the steady-state value. This steady-state value could then be compared with a calculated saturation concentration to scale the signal.

Experimental procedure

We opted to utilize a parabolic flight to conduct our experiment within different gravitational environments. The alternative suborbital flight option only allows testing in microgravity and a parabolic flight has the added advantage of being able to interact with the payload. While advanced automation is still advisable due to potential operator nausea, on a parabolic flight, the process does not have to be 100% automated. The flight was performed by a modified commercial airplane with limited seats and minimal insulation to maximize agility. By quickly modulating the angle of attack, parabolic flight profiles are obtained wherein the acceleration vector of the craft and all the contents within could be controlled such that it would simulate a lower gravity for several tens of seconds. The flight we had commissioned was scheduled to perform 6 sets of 5 parabolas for a total of 30, which comprised of 2 sets for each gravity environment: Martian, Lunar and microgravity. Between each set was a 5-min break, to allow the airplane to turn around completely to stay within the limited airspace that was reserved for the flight campaign. Time was also planned before and after all the parabolas where the airplane experienced an acceleration vector approximating terrestrial gravity at a similar altitude to where the parabolas were conducted, which was when baseline tests were recorded.

Since the experiment consisted of recording the rate of oxygenation within different parabolas, each test was divided into cycles, where one cycle involved both an oxygenation and deoxygenation phase. Each cycle began with a low level of dissolved oxygen (DO), which was always less than 5% DO or less than 2 mg per L. This was obtained by maintaining a constant flow of pure nitrogen in the gas line to purge oxygen from the liquid phase by diffusion until the lower threshold of 1% was reached, at which point gas flow was halted to conserve nitrogen gas. Two identical buttons were pushed by two operators in case one would get compromised by nausea after a call out from the flight operator that a target gravity level was reached. This signaled the start of a cycle, which commanded the control system to flow pure oxygen into the membrane’s gas side for 7 s, which was the oxygenation phase when the critical data was collected. After that period, the system automatically switched off the oxygen and enabled a flow of pure nitrogen gas into the gas phase to purge oxygen from the liquid phase, thus resetting the conditions for the beginning of the next cycle.

Results

Optimization

The liquid flow and gas pressure were optimized to minimize the variance between experiments as the ultimate goal of the experiment is to discern differences in mass transfer between different gravity levels. The mass transfer coefficient exhibited the lowest standard deviation when operating at 5 LPM at 20 PSI (Fig. 5). During these optimization trials, the oxygen flow was initially regulated at a steady 1.5 LPM; however, the resulting mass transfer rate proved insufficient for the short oxygenation window available during the parabolic flight. To address this limitation, the air valve control system was adjusted to allow the valve to open fully, enabling a higher flow rate exceeding 1.75 LPM. This increase in flow enhanced the mass transfer rate to 1.14 × 10⁻4 m s⁻1.

Fig. 5.

Comparison of the oxygen mass transfer coefficient throughout a range of liquid flow and gas pressure conditions. Sets of 5 experiments were collected for each condition, at the nominal water temperature of 37 °C. The mean is plotted as a dot, and ± 1 SD is represented as a line

Quality of gravity level

A typical parabolic cycle followed the same structure. The period before the simulated gravity was called the pull, at this point the setup experiences hyper-gravity of up to 2 g. At the apogee, the crew signaled “release” which was immediately followed by the period of simulated gravity which is called the fall. Hence, there was just a call and no accelerometer data is provided by the flight operators to the experimental teams. Therefore, it was decided to install a triaxial accelerometer (Model 3713F1110G, PCB Piezotronics, Inc., Depew, NY) close to the center of gravity of the system to at least verify quality of the gravity level. It was not used as a control signal as the reliability could not be tested in advance. The system was installed close to the center of gravity of the airplane right between the wings.

The duration of the fall period was variable and ranged from about 25 s for Martian gravity to 18 s for microgravity (Fig. 6b). The lower duration for lower gravity levels was to be expected based on the flight profile to achieve them. The target accelerations were Martian, Lunar, and microgravity; expressed as a ratio of Earth’s gravity these were 0.38, 0.16 and under 0.1 respectively. By taking the magnitude of our system’s triaxial accelerometer, we were able to validate that the average gravity level of the parabolas adhered to within 0.1 g of the target gravity during the critical experiment window (Fig. 6a). The standard deviation of measured gravity levels indicates a similar quality of simulated environments (Table 4).

Fig. 6.

a Range of accelerations measured throughout the parabolic flight. b Range of time within the critical part of each parabola. In each case, the mean is plotted as a dot, and ± 1 SD is represented as a line

Table 4.

Acceleration statistics for each flight segment

Gravity mode	Target	Mean	Std. deviation
Hyper-gravity		1.464	0.301
Terrestrial	1.000	1.004	0.058
Martian	0.378	0.385	0.054
Lunar	0.165	0.184	0.062
Micro	< 0.1	0.058	0.046

Data selection

We planned for 10 parabolas within each simulated gravity modality, but a dust containment breach in another experiment terminated the flight early, resulting in the cancellation of one of the planned microgravity parabolas.

As a further complication, the automated temperature control failed during the second set of microgravity parabolas (Fig. 7). It was determined from video footage of the reactor that a bubble on the thermal loop side, which was in a stable position in the reactor in Martian and Lunar gravity, got entrained into the thermal line during microgravity. The bubble likely got stuck and deactivated the flow switch which was installed to prevent overheating and boiling in absence of flow and hence in turn deactivated the Peltier device. Since temperature has a significant impact on diffusion rates, we decided to omit the last 4 microgravity cycles from the final analysis, bringing the total to 5 cycles for that gravity level. For the same reason, the post-parabola terrestrial baseline was also omitted from the final analysis. Since a temperature variation is present even in valid cycles, we have opted to perform a temperature correction by determining the percentage change in oxygen saturation from its nominal value when temperature is at 37 °C and using this change to normalize the dissolved oxygen values.

Fig. 7.

Recorded liquid temperature throughout the parabolic flight. The temperature stayed within 1 °C of 37 °C for most of the flight but temperature control failed around hour 3

The final selection of oxygenation datasets can be seen in Fig. 8, on primary inspection it appears that the reduced gravity did not produce results outside of the range observed under terrestrial gravity conditions. The oxygenation curves were aligned in the time axis by finding the x intercept of a linear fit to the straightest part of the curve, the resulting data is displayed in Fig. 9. These trimmed oxygenation curves were then used to calculate the mass transfer coefficient as described in Sect. “Membrane Theory”.

Fig. 8.

Oxygenation curve during each set of parabola. Each datapoint is marked by an x. The grey area on the graph delineates the path taken by oxygenation curves in the terrestrial case, this was overlayed to demonstrate that the similarity in oxygenation curves in of conditions. a Oxygenation curves under terrestrial gravity levels; b Martian gravity; c Lunar gravity, d microgravity

Fig. 9.

Trimmed and aligned oxygenation curves a 15 cycles in terrestrial gravity, b 10 cycles in Martian gravity, c 10 cycles in Lunar gravity, d 5 cycles in microgravity

Statistical analysis

The dissolved oxygen (DO) dataset was normalized and trimmed to retain only the most linear portions of the oxygenation curve between  and  (Fig. 9). These segments were used as input for the diffusion model to calculate the mass transfer coefficient  for each gravitational condition. Despite uneven sample sizes across gravity modalities, statistical differences between the distributions of  values were assessed using Student’s T-test. The results indicate that no pairwise comparisons yielded a p-value below the 0.05 threshold, meaning there is no statistically significant difference in oxygen mass transfer coefficients across any gravity condition (Fig. 10). Conversely, no pairs scored above 0.95, which would indicate strong statistical equivalence. Notably, microgravity results were most similar to terrestrial conditions, and Martian conditions were closest to lunar in their mass transfer behavior, but these trends were not statistically significant.

Fig. 10.

a Oxygen mass transfer coefficient within each gravity condition. The mean is plotted as a dot, and ± 1 SD is represented as a line. b Comparison of each gravity condition distribution with each other using Student’s T test. Values under 0.05 would be considered statistically significant for dissimilarity

Upon close inspection of the mass transfer coefficients (Fig. 10), the mean values obtained for each segment are similar, ranging from  to  m s⁻¹, indicating that the resistance to mass transfer was relatively uniform across gravity modes. To contextualize these results, the liquid-side mass transfer coefficient can be estimated using the Sherwood relation:

where  is the Sherwood number,  m is the fiber hydraulic diameter, and  m² s⁻¹ is the diffusion coefficient of oxygen in water at 37 °C. Using a Sherwood number of 5.75 from the leveled-off regime reported by Alhmiedy (Aljasem Alhmiedy et al. 2025), this gives  m s⁻¹, which is within 25% of the experimentally measured values in our system ( m s⁻¹). Reported values in previous studies using similar hollow fiber modules, such as those by Van der Vaart (2000), are also in the same range (0.7–1.4 × 10⁻⁴ ms⁻¹), demonstrating that our measured coefficients are consistent with the literature.

It is worth noting that the lunar gravity dataset exhibited a larger variance and standard deviation in  values compared to the Martian dataset, while the microgravity dataset showed no clear variance trend, potentially due to its smaller sample size resulting from the noted temperature anomaly. The terrestrial dataset demonstrated greater consistency and less scatter in oxygen transfer curves relative to other gravity conditions (Fig. 9). Rather than attributing these differences to gravitational effects, we propose that increased vibration and acceleration fluctuations during parabolic flight maneuvers under Martian, lunar, and microgravity simulations introduced additional noise and variability into the measurements. This hypothesis is supported by accelerometer data and Fourier analysis indicating that terrestrial conditions were significantly less noisy (data not shown).

Taken together, these findings statistically support our initial hypothesis that oxygen mass transfer through the membrane contactor is not measurably affected by variations in gravity over the tested regimes. The lack of significant difference in  across gravity levels strengthens confidence in the robustness of membrane-based gas–liquid mass transfer technologies for applications spanning terrestrial to extraterrestrial environments. The slight increase in variability observed in simulated reduced gravity is more plausibly attributed to experimental noise inherent in parabolic flight rather than true gravitational effects. This result aligns with theoretical expectations and elevates the Technology Readiness Level for deploying membrane systems in space life support and resource utilization systems, as well as other industrial processes where gravity varies.

Relevance of results

This study clearly confirms the use of membrane contactors will not introduce a gravitational effect into systems. The application that is targeted in this study is for gas delivery into bioreactors. Gas is a convenient feedstock for synthesis of more advanced chemicals and membrane contactors are the only way to efficiently deliver it to liquid media in the absence of gravity. Microbial metabolism is known to be affected by gravity, hence, any observed effect of gravity on synthesis in an MABR will be solely due to the biologic processes and not due to the gas delivery system. While this study focused on microporous membranes due to the short time window, the results are also expected to be applicable to dense membranes like polydimethylsiloxane (PDMS) which typically exhibit much lower gas fluxes.

Membrane contactors are not just beneficial for gas delivery, but due to the Chatelier principle, they can greatly enhance reaction rates by improving gas extraction as well. Membrane contactors have been shown to maximize biomethane synthesis when integrated in digesters (Lee et al. 2021). And this does of course not only apply to biotic processes. A major application that will be essential for space exploration and that is severely hampered by reduced gravity conditions is electrolysis for instance for the production of hydrogen and oxygen. Bubbles just collect at the electrodes in absence of gravity and impede further electrolysis. Recently, several studies (Kabir et al. 2025; Goldman et al. 2025) have proposed to integrate hydrophobic membranes at the electrodes to facilitate gas removal and have shown greatly increased efficiency. Those studies were performed in the presence of gravity, but it is now certain that if tested in different gravity conditions, the enhanced effect is solely from the extraction of the gas not the method it is extracted with.

Conclusion

We have demonstrated the design and optimization of a system capable of rapidly testing oxygen mass transfer rates, under 10 s, through membrane interfaces, using an innovative two water loop system to maintain thermal stability while reducing test water volume. This system could easily be adapted to test and compare different membrane systems, or the transfer rate of any pure gas, provided an inline sensor exists. Such a system could also be used as a high-throughput characterization tool to compare the performance of several new membrane modules and materials in a short amount of time.

We have shown that the oxygen mass transfer rate through a membrane contactor did not change significantly when subjected to lower gravity conditions. We believe this indicates that membranes can be used for gas exchange processes designed for space, and that the performance of such a system under terrestrial conditions should be representative of the performance in lower gravity. While the porous membrane contactor used in this experiment was selected for optimal parabolic flight performance, this finding should generalize to all types of gas membranes including dense membranes, like polydimethylsiloxane membranes for instance.

This is a major milestone towards integrating carbon upscaling MABRs in the ISRU architecture in space. As such we encourage further research into MABRs intended for lower gravity and propose that the next step should be a small demonstration bioreactor designed for long-lasting experiments on the International Space Station.

Author contribution

A.B. developed the control system and wrote the main manuscript and prepared all figures. J.L. supported system design and pre-flight as well as during flight certification procedures. J.B. and J.V. supported hydraulic and electronic system development. A.S., A.J. and N.F. assisted with flight operations. J.L.,A.P., J.M. and J.V. secured funding. All authors reviewed the manuscript.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Conflict of interest

The Authors declare no Competing Financial or Non-Financial Interests.