An open-source platform for 3D-printed redox flow battery test cells

The development of new, large-scale stationary energy storage technologies, such as redox ﬂ ow batteries, is vital to fully utilise renewable energy resources. However, test cells capable of assisting in this development can be prohibitively expensive and unreliable. Here, an open-source, low-cost, customisable 3D-printed test cell is presented as an alternative. These newly developed cells are designed to be printable using a ﬀ ordable desktop 3D-printers and readily available polymers. A simulation-led design optimisation yielded an improved internal manifold geometry that demonstrated improved real-world performance. The polymers used have been tested for chemical compatibility and through the use of advanced X-ray micro-CT, optimised parameters for 3D-printing have been identi ﬁ ed. This framework provides a straightforward process enabling researchers to produce robust cells at an extremely low cost, helping to democratise research and widen accessibility to ﬂ ow electrochemistry.


Introduction
In the midst of a worsening climate crisis, it is widely accepted that the pathway to a sustainable future involves a transition to greater utilisation of renewable energy resources. 1 These resources, such as wind, hydro, and solarwhilst being nonpolluting and abundantly availableare intermittent and difficult to predict. Coupled with a lack of adequate technology for large-scale energy storage, this makes the use of renewables in a stable and reliable power grid problematic. 2 Effective gridscale energy storage is one of the world's foremost challenges, requiring collaboration between institutions and researchers and the democratisation of new technologies. Recently, significant research has been carried out in an attempt to develop adequate large-scale electrochemical energy storage systems, with the redox ow battery (RFB) among the most promising for longer-term energy storage. [3][4][5] A typical redox ow battery consists of redox couples, dissolved in electrolyte solutions, and stored in separate reservoir tanks. During charge and discharge these electrolytes are pumped into half-cells, separated by an ion-exchange membrane where electrons are either consumed or generated within a porous electrode. 6 Fig. 1 shows a schematic of a basic redox ow battery cell and system. As a result of this unique working principle, power and energy are decoupled, i.e., power is determined by the cell, and energy by the tank size and electrolyte energy density, allowing these devices to be customised to meet a wide range of requirements. 7 Furthermore, in conventional RFB systems, such as Fe-Cr and all-vanadium, the electrodes are not chemically converted during charge and discharge, and these systems generally have a longer lifespan than other energy storage technologies such as Li-ion batteries. However, RFBs have several drawbacks that have limited their wider use, primarily their low energy densities, 7 high cost of some components required, (e.g., membrane materials), 2 and the lack of a "winning" or "best" RFB chemistry. 6 These issues have been addressed by researchers in several ways, most commonly through the development of new electrolyte and membrane technologies. 4,[8][9][10] Flow battery test cells used in the development of these new electrolytes tend to be expensive and provide limited scope for re-design, presenting a potential barrier-to-entry into the eld of ow battery research. Commercially available test cells with active areas of $25 cm 2 typically range in price from $$2300 to $4000 [11][12][13][14] and, in some cases, their sealing can be problematic. Here, we report the development of custom-made test cells for laboratory research utilising fused deposition modelling (FDM) 3Dprinting as a compelling alternative to these commercially available cells.
As the accessibility of 3D-printing has increased, 15 more researchers are recognising its potential to be used in electrochemical ow systems. 16 Recently, 3D-printing has been used to develop electrochemical ow systems which can be used for water splitting, electrosynthesis cells, electroanalysis reactors, fuel cells and electrolysers. 17,18 Some of these electrosynthesis and electroanalysis reactors were FDM 3D-printed from polypropylene (PP), incorporating small channels and reaction chambers that are analogous to the features found in RFB test cells. 19,20 Another general-purpose reactor was developed for electrosynthesis, electrodeposition, and electrochemical water splitting, with manifolds and turbulence promoters manufactured via FDM 3D-printing. 21 As part of its design process, the chemical compatibility of polyethylene-terephthalate and acrylonitrile-butadiene-styrene (ABS) were evaluated in several common chemical compounds and electrolytes.
Elsewhere, the chemical compatibility of a wider range of FDM 3D-printed materials with common solvents 22 and chemical solutions 23 were also investigated. The chemical resistance of the components was shown to vary between printed and unprinted laments due to microstructural defects which arise during printing.
3D-printing has also been utilised in the eld of electrochemical energy storage, with a range of 3D-printing technologies being utilised to fabricate electrode materials 24 for various energy storage devices such as Li-ion batteries and supercapacitors. 25 So far, few reports have been published on the potential of 3D-printing technologies applied to RFBs. 26 Test cells for use in a Zn-Ce system have been partially 3D-printed via FDM, incorporating ABS ow plates held together by a stainless-steel clamping structure. 27 A similar test cell has also been designed for a vanadium 28 and a non-aqueous cobalt 29 system, with ow and end plates manufactured via stereolithography 3D-printing. This 3D-printing technology has also been used to manufacture turbulence promoters to enhance reactant distribution 30 and static mixers in a slurry electrode to improve charge transfer. 31 3D-printing has also been employed in a heat-managing miniaturized RFB cell, integrating uidic networks manufactured via multi-jet fusion which were then coated with nickel. 32 These pioneering works have begun to explore the potential of 3D-printing in electrochemical ow reactors. 33 However, there is still scope for wider utilisation of the technology, particularly as applied to RFBs.
The freedom of design offered by FDM 3D-printing allows for the investigation of many different complex manifold and cell topologies that are difficult to achieve via traditional manufacturing methods. This advantage has been widely utilised in the full cell fabrication of other rechargeable batteries. 24 Additionally, parts can be produced at a low-cost and high speed 34 compared to the methods used to manufacture the majority of test cells. Some of these cells have been collated and can be seen in Table S1 in the ESI. † Furthermore, the majority of the existing devices mentioned can be difficult to reproduce without expertise in manufacturing or 3D-design. This study aims to provide researchers with a straightforward framework that can be used to easily 3D-print custom-made, low-cost test cells. These cells and methods are intended to assist in the development of improved ow battery electrolytes, materials, and cell designs.
This 3D-printing approach has been combined with a coupled electrochemical-computational-uid-dynamics (CFD) model and X-ray micro-CT to generate and print an optimised cell design. To the knowledge of the authors, this is the rst time that rapid prototyping and simulation have been employed to converge on a more optimised RFB cell design and demonstrate this improvement with real-world results. This framework provides a platform for altering cell design for parametric studies enabling fast-track development and optimisation of performance for a given electrolyte system. The resultant cell design of this study has been subjected to a set of electrochemical tests to fully assess its performance against commercially available test cells. Remarkably, the total cell cost was estimated to be 5% or less than the selling price of commercially available test cells. The polymers used (ABS and PP) are also tested in common ow battery supporting electrolytes to ensure chemical compatibility. Finally, by providing all source les, this work enables researchers with no expertise in 3Ddesign to download and print cells without the use of computer-aided design (CAD) soware.

Results and discussion
Polymer selection and compatibility testing Two potentially suitable FDM polymers, ABS and PP, were identied 16,17,20,27,35,36 for use in ow batteries. However, it has also been suggested that ABS may be less chemically resistant, and is only suitable for short-term testing with low acid concentrations. 27 To address this, and to identify suitable material-electrolyte combinations, a chemical compatibility study was carried out. Inspired by the previous works, 21-23 a number of commonly used aqueous and non-aqueous ow battery solvents and supporting electrolytes used in ow battery systems were selected and tested with these polymers. A total of six aqueous solutions (3 M sulfuric acid (H 2 SO 4 ), 1 M potassium hydroxide (KOH), 2 M sodium chloride (NaCl), deionized water (diH 2 O), 3 M hydrochloric acid (HCl) and 3 M methanesulfonic acid (MSA)) and four non-aqueous solvents (i.e., acetonitrile (ACN), propylene carbonate (PC), dichloromethane (DCM) and an ionic liquid, trihexyl tetradecyl phosphonium bistriimide ([P 66614 ][NTf 2 ])) were investigated. A concentration of 3 M sulfuric acid (H 2 SO 4 ) was chosen as it is the concentration used in the electrochemical modelling in the "test cell design" section and the real-world tests in the "charge-dischargetesting" section. Chemical incompatibility between a chosen polymer and electrolyte could result in leakage or absorption of electrolyte and failure of experiments. As such, three different printed samples of ABS and PP, each from a different manufacturer (i.e.,Verbatim, FormFutura and Ultimaker) were tested. Classication of the levels of degradation was based on a previous study 21 which stated that degradation in mass of <1% aer 1 week was acceptable, 1-5% was moderate, and >5% was severe. The samples were le in solution and the mass differences between the beginning and end of the test were recorded. A duration of one week was chosen to give a good representation of the degradation that may occur during typical usage identi-ed from literature (see Table S1 †). The results are shown in Fig. 2.
All ABS and PP laments showed no measurable volume change or swelling in aqueous solutions, and aer cleaning and drying, they exhibited less than 1% mass change. As such, they all appeared suitable candidates for use in the 3D-printing of ow battery cells using these electrolyte systems. However, as expected from previous studies, 21,22 the 3D-printed ABS rapidly soened and experienced severe mass gain, with dissolution and swelling, when submerged in the organic solvents DCM and ACN. ABS samples submerged in PC also exhibited the same behaviour. Thus, it can be concluded that ABS should not be used with any of these three non-aqueous compounds. Moreover, Formfutura ABS was found to discolour in MSA and H 2 SO 4 , alongside a small mass loss of Ultimaker ABS in MSA; however, the mass change in all instances was below 1% and thus deemed acceptable. The discolouration and mass loss are predicted to be due to the different plasticisers and antioxidants used by different manufacturers. These ndings indicate the potential for ABS ow frames to be used for long-term lab-scale studies with electrolytes containing higher acid concentrations.
PP showed greater compatibility with these solutions, exhibiting acceptable to moderate mass changes. However, some samples, such as Verbatim PP in DCM and Formfutura PP in PC, showed a larger mass change than the other two PP samples. This trend is repeated across a number of different PP samples, particularly those printed using Formfutura PP. In all aqueous and non-aqueous solutions (with the exception of NaCl) these samples increased in mass more than the other PP laments tested. This nding, along with a number of other uctuations between the same types of polymers from different manufacturers, would indicate that the different lament additives can result in a noticeable difference in chemical compatibility and warrants further study.

Test cell design process
An active area of 50 Â 50 mm was chosen for the design detailed in this work. These dimensions were chosen based on the dimensions of test cells described in literature (see Table S1 †).
Another key design choice was the overall structure of the test cell. Although other RFB cell structures exist, 37,38 the most common are the conventional "ow-through" and "zero-gap" cell structures (Fig. 3). In ow-through cells, the electrolyte ows through the electrode, which typically consists of a carbon felt material. In zero-gap cells, however, the electrolyte ows over the surface of thinner carbon paper electrodes through ow channels which are typically machined into graphite ow plates. 39 Whilst it is widely accepted that designs that include ow elds can achieve higher power densities and reduce overpotentials, 40 most commercially available cells 41 and larger industrial-size cell stacks 42 still use ow-through electrode congurations. Furthermore, zero-gap cells tend to be more expensive than ow-through cells, due to the complex machining and extra material cost of the graphite ow plates. 43 As one of the primary objectives of this work was to provide researchers with low-cost, easy-to-manufacture ow cells, a ow-through design was developed, however zero-gap designs will be explored in future work.
To improve reactant distribution and reduce concentration overpotential in the 3D-printed cell, the manifold design was optimised using a coupled electrochemical-CFD model developed in literature. [44][45][46] This model examined the ow of electrolyte through the porous carbon felt electrode in one half-cell (the negative half-cell) in 3D at a given state-of-charge (SOC), allowing cell designs that have the potential to improve reactant distribution to be investigated. This approach has been successfully employed by other research groups, with novel cell topologies resulting in greatly enhanced reactant distribution. 30 The parameters used were selected from literature 44,46 and an inlet ow rate and applied current density of 25 mL min À1 and 50 mA cm À2 were chosen, respectively. For the simulations, cell geometry has been simplied in the z-direction to a uniform thickness of 2 mm. The effect of the carbon felt electrode, with micron-sized bres, is modelled with homogenised macroparameters where permeability is described by the Kozeny-Carman equation, using the carbon bre diameter. Key model parameters are detailed in Table S2. † The cell was designed to be as compact as possible given the 25 cm 2 active area. A detailed cell design drawing can be seen in Fig. S3. † A more compact footprint results in a shorter print time, less material used and lower costs. Within these constraints, a simple manifold was designed. This design ensured the avoidance of large overhangs, which can prove difficult to 3D-print, and consisted of four equally spaced, 2 mm Â 2 mm ow channels between the electrode compartment and the in-and outlets. A velocity streamline plot of the electrolyte ow through this manifold design is shown in Fig. 4a. Due to the wide range of velocities seen, the logarithm of the velocity magnitude has been contoured in Fig. 4c.
Across most of the electrode area, this original manifold design achieved an even velocity prole, which can be seen from both the streamlines in Fig. 4a and from the contour plot of the log of velocity in Fig. 4c. However, regions of both higher and lower velocity are found to exist near the inlet and outlet channels which can be seen in Fig. 4c. The velocity prole close to the inlet into the electrode area appears to be critical, as it can have a dramatic effect on the distribution of the concentration of vanadium(III), V 3+ , across the entire working section, as can be seen in Fig. 4g. As V 3+ enters the cell during charging, it is reduced to vanadium(II), V 2+ , which results in a drop in V 3+ ions moving from inlet to outlet. Due to the regions of low velocity close to the inlet in between the channels, V 3+ is locally depleted, and the distribution doesn't recover uniformity. A more even distribution results in a cell with reduced concentration overpotential, allowing for higher power output. 47 Based on these results, the manifold design was rened, within the same footprint. The optimised manifold (Fig. 4b) consisted of a "diffuser" type design from the inlet and outlet nozzle to the electrode area. This diffuser was divided using 20  evenly spaced 2 mm Â 2 mm pillars in each manifold. These pillars were incorporated to reduce the size of overhangs in the manifold, whilst limiting design complexity to enhance printability. This design results in an increase in cross-sectional area, which decelerates the ow from the inlet, achieving a more even velocity prole at entry to the electrode area, as seen both in the straight and parallel streamlines of Fig. 4b and in the plot of the log of the velocity magnitude in Fig. 4d, in which the disturbances at the inlet from the original manifold design have been completely eliminated. As a result, a more uniform distribution of reactant concentration can be seen in Fig. 4h with the nonuniformity from Fig. 4g having been greatly reduced. This also resulted in an increase in the minimum reactant concentration seen within the domain to 726 mol m À3 , from 720 mol m À3 .
The optimised design also exhibited a reduction in pressure drop across the cell from 2.2 kPa (Fig. 4e) to 1.82 kPa (Fig. 4f). This pressure difference of 0.38 kPa equates to a reduction in pressure drop of $17%. As cell area and ow rates are scaled up to larger, commercial-scale cells and stacks, this pressure drop difference would become increasingly signicant, equating to large differentials in parasitic pumping losses when comparing systems. These improvements highlight the potential for this approach of iterative renement using rapid prototyping and CFD-electrochemical modelling to improve ow cell design, which will be re-visited in future work.

3D-printing parameters
FDM 3D-printing has many desirable traits when it comes to the manufacture of ow battery cells. FDM systems are low cost, 34 safe, and easy to use in a desktop or laboratory environment, 48 allow for many thermoplastic polymers to be printed, and have the ability to print complex internal geometries. However, an area of concern was the potential porosity of the parts produced. Microporosity of parts produced by FDM can occur as a result of the pores and voids which form as the thermoplastic polymers cool during the printing process, 49 potentially allowing electrolyte to permeate into the 3D-printed components. This could be further exacerbated by the back pressures exerted on the walls of the part by the uid ow, 50 and the potentially "aggressive" media used as electrolyte. 22 This effect can be mitigated by changing various printing parameters. Previously, increasing the material ow rate by applying an "extrusion multiplier" or "k-value" has been demonstrated to have perhaps the most signicant effect on the sealing of 3D-printed parts. 50 Increasing this property results in more polymer being forced through the nozzle per unit time, which can ll defects and reduce the permeability of the part. 22 In addition, other recommendations have been made, including allowing for diagonal lling in walls and distributing the seams for each layer randomly. 50 All of these recommendations have been implemented in the design of the half-cell shown during and aer printing in Fig. 5.
All printed components (both ABS and PP) for ow cells in this work were printed using a 0.4 mm nozzle, 0.2 mm layer height, 100% inll, and 105% ow rate (k ¼ 1.05). The print time for each ow frame is approximately 7.5 hours for ABS and 9.5 hours for PP. Detailed print settings can be found in the Methods section. These settings are easily achievable on any desktop 3D-printer and provide reasonably fast printing times, whilst minimising electrolyte permeation into the components.

X-ray micro-computed tomography
To probe the impact of changing the k-value on the porosity of the 3D-printed parts, six samples (PP and ABS for k ¼ 0.90, 1.00, 1.05) were printed, exposed to vanadium electrolyte, and scanned using X-ray micro-computed tomography (micro-CT). Acquisition and processing parameters are in the Methods section. 3D renderings of the solid plastic and the internal void space are shown in Fig. 6a and b, respectively. For both materials, increasing the k-value had a signicant effect on the internal structure of the 3D-printed parts, leading to a decreasing volume of void space, as shown in Fig. 6c. At each k-value, ABS contained more voids than PP and the effect of increasing the k-value from 0.90 to 1.05 led to a 75% decrease in internal porosity from 6.41% to 1.59% and a 96% decrease in internal porosity from 1.9% to 0.07% in ABS and PP, respectively. Although increasing the k-value above 1.00 is associated with a reduction in dimensional accuracy, X-ray micro-CT clearly demonstrates 3D-printed parts with lower porosity were accessible by a small increment in this parameter. Moreover, negligible vanadium (not reliably segmentable from high grayscale noise) was detected in PP samples, whereas ABS showed 0.42%, 0.11% and 0.03% vanadium deposits for kvalues of 0.90, 1.00, and 1.05, respectively (see the Fig. S2 †). It should be noted that vanadium deposits below the resolution limit may be present but undetected, and vanadium deposits in the ABS k ¼ 1.05 sample were minimal and may not represent the true level of ingress in the component.

Polypropylene electrolyte permeation
Having been previously used in other electrochemistry applications 19,20 and comparable RFB test cells using vanadium electrolyte, 51 as well as favourable performance in the chemical compatibility tests and low void space during the X-ray CT scans, it was expected that PP would be the ideal material for 3Dprinting ow cells. However, when these cells were tested with an electrode compression of 26% and electrolyte ow rate of 25 mL min À1 , a high level of permeation was observed within 1 hour. This effect is shown in Fig. 6d(i), where the previously clear polypropylene half-cell was discoloured aer absorbing electrolyte. Upon further investigation, the location of electrolyte permeation was identied by examining a cross-section of the half-cell, shown in Fig. 6d(ii). It appears that the permeation occurs at the bottom surface of the half-cell, which is in contact with the heated build-plate (65-85 C) during the printing process. This permeation was observed with half-cells printed using PP laments from all three of the manufacturers in this study. The cause of this permeation is unknown at this time, and, to the authors' knowledge, this behaviour has not been reported elsewhere in literature. An in-depth investigation is underway to identify the cause of this permeation along with strategies to prevent it, allowing PP to be used in a wider range of ow electrochemistry applications in the future. ABS cells were used in all subsequent tests in this work as these cells were not susceptible to electrolyte permeation. A cross-section of an ABS cell aer 12 hours under the same conditions used in PP testing is shown in Fig. 6d(iii) with no observable electrolyte permeation.

Cell assembly
The 3D-printed ow frames were designed to be easily assembled, with grooves for O-rings, gaskets and current collectors. A schematic view of all the components in the cell can be seen in Fig. 7a. Hand-cut expanded polytetrauoroethylene (ePTFE) with a thickness of 1 mm or ethylene propylene diene monomer (EPDM) sponge gaskets with a thickness of 1.5 mm were used between the ow frame and both the membrane and current collectors. The current collectors consisted of 4 mm thick brass disks, with a 20 mm protrusion to connect the cell to the battery test equipment. To make the cell as compact as possible, the diameter of the current collectors was 75 mm, and the O-ring inner diameter was 71 mm. To protect the current collectors from corrosion, a 0.6 mm thick graphite composite disk (PV15, SGL Carbon) was used. The components were clamped together using a G-clamp which was hand-tightened before adjustment to 10 Nm using a torque wrench. The clamping ensured adequate sealing of the cell and reduced contact resistance of the components, whilst the electrode depth was dictated by the thickness of the 3D-printed ow frames and gaskets ($8 mm total). More  information about these materials and their manufacture is in the Methods section. The circular shape of the current collectors ensured an even clamping force was applied around the perimeter of the electrode compartment. The sealing structure allows for low tolerances in both 3D-printed and non-3D-printed parts, reducing cost and complexity by avoiding the need for expensive manufacturing processes. Coupled with the optimised 3D-printing parameters, conrmed by X-ray micro-CT, these design choices resulted in a robust, leakproof cell that can be reliably used for testing. Furthermore, the assembled cell is compact and lightweight, weighing only 900 g when fully assembled. The cells were re-used multiple times during testing with no signs of mechanical damage, withstanding the clamping force applied. One area of concern regarding durability was the 3D-printed nozzles. The 3Dprinted nozzles designed for 4.8 mm bore tubing used in this work were relatively robust, however, certain users may want to ensure a longer lifespan, or require compatibility with different diameters of tubing. To facilitate this, versions of all the cell designs discussed in this work that allow for removable nozzles have been included. File details are in the ESI. †

Charge-discharge testing
To characterise the efficacy of the 3D-printed cells, appropriate operating conditions were identied from literature. 41,52-55 A commercially available cell with the same 50 Â 50 mm active area and similar electrode cavity thickness (C-Tech Innovation) was tested under the same conditions as the 3D-printed cells. Tests were carried out using a typical vanadium electrolyte, 1.5 M V 2+ and V 5+ dissolved in 3 M H 2 SO 4 , and a commonly used, commercially available Naon® 212 membrane. Two different electrode compressions, 26% and 42%, were examined using ow rates of 25 and 50 mL min À1 . For the chargedischarge tests, the cell was cycled at a constant current density of 50 mA cm À2 between 1.10 V and 1.75 V. These cut-off voltages equate to roughly 10% and 90% SOC of the cell with 26% electrode compression. These tests were all carried out using a bespoke ow battery test rig (shown in Fig. 7b and c) at $20 C. Component information and testing parameters are found in the Methods section.
A typical charge-discharge test using the 3D-printed ow cell is presented in Fig. 8a. Representative cycles from various tests can be seen in Fig. 8b-e. Fig. 8b and c show an increased cycle time and capacity in cells that have an optimised ow eld design. This effect can be explained by enhanced mass transport and therefore a more effective electrolyte distribution through the electrode (Fig. 8b and c), resulting in reduced concentration overpotential. This can also be observed in the modelling results from Fig. 4g and h, where larger concentrations of spent electrolyte are seen in the unoptimised design.
A similar pattern is shown when comparing Fig. 8b and c (42% electrode compression) to Fig. 8d and e (26% electrode compression). This performance improvement as electrode compression is increased, has been extensively investigated elsewhere in literature. 41,53,56 The commercially available cell had higher overpotentials than the 3D-printed cell. At 42% compression, charging began at a higher voltage in the commercial cell, $1.45 V vs. $1.40 V, and discharging began at a lower voltage, $1.44 V vs. $1.50 V. Furthermore, the cut-off voltages were reached more quickly in the commercial cell, $20 min sooner during charging and $23 min sooner during discharging, with a smaller capacity, 485 mA h less upon charge and 438 mA h less upon discharge. There are various potential explanations for higher overpotentials in the commercially available cell. Once such explanation is higher ohmic resistance of the cell. To investigate this, the cells were assembled without a membrane or electrolyte, and the resistances measured (see Methods section for procedure). The average resistance of the 3D-printed cells was 0.015 U, compared to 0.035 U for the commercial cell. This discrepancy in electrical resistance may be due to the materials used to manufacture the components (endplates and current collectors), or higher contact resistance between components in the commercial cell due to the different clamping methods. Another contributing factor to the lower overpotentials could derive from better reactant distribution in the 3D-printed cell designs. Both 3D-printed cell designs consequently achieve similar energy efficiencies (see the ESI † for equations) to the commercially available cell, with energy efficiencies (neglecting the power consumption of the peristaltic pumps) of 81.4% for the optimised design and 79.0% for the unoptimised design, compared to 78.0% for the commercially available cell. These energy efficiencies are plotted in Fig. 9a, along with several other similar cells from literature at 40-50 mA cm À2 . Each cell number corresponds to a row in Table S1, † along with more detail about the cell construction and performance. The 3Dprinted cells achieved satisfactory energy efficiencies especially considering the lack of optimisation of the membrane and electrodes. This is illustrated in the favourable comparison to cell 7 in (Fig. 9a), 60 which also used a Naon 212 membrane and a current density of 50 mA cm À2 . In future work, the performance of the 3D-printed cells will be improved by optimising the selection of membrane and electrode materials. Furthermore, no leaks from the cells occurred during the charge-discharge tests, depicted here or during leak tests which were carried out with electrolyte at a ow rate of 50 mL min À1 and electrode compression of 42% for up to 31 days (744 hours), demonstrating the efficacy of the 3D-printed cells for longer duration lab testing. These tests are compared with the highest reported electrolyte ow rate and longest reported test durations found for a number of similar cells from literature in Fig. 9b.

Cell cost analysis
Detailed cost analyses have been carried out to approximate the cost for researchers to manufacture the 3D-printed cells (details in the ESI †). Key costs are shown in Table 1. The total cell cost is estimated to be $107. Neglecting overheads, this results in a cost-saving of >95% compared to commercially available labscale test cells. [11][12][13][14] This signicant cost reduction makes the 3Dprinted cell developed and tested here a compelling alternative to researchers working on ow batteries.

Chemical compatibility testing
A similar procedure to that described by Marques-Montes et al. 21 was followed to determine the chemical compatibility of the materials used for 3D-printing. 2.85 mm diameter ABS and PP laments from Ultimaker, Verbatim and ABS were used for the tests and the manufacturers recommended print settings for speed and temperature were used when printing the samples, along with 100% inll, a k-value ¼ 1 and a 10 mm brim. The specimens were put into sample bottles with a height of 40 mm and a diameter of 20 mm which were lled with 20 mL of solution. 4 mm glass rods, with a length of 40 mm were used to submerge the specimens in solution. Each specimen was weighed to an accuracy of AE0.0001 g and inserted into a sample bottle. The samples bottles were stored in an INC/50 incubator (Genlab) at 25 C for 1 week. The sample bottles were shaken once every 24 hours during the test. The specimens were then rinsed in deionised water and dried in a vacuum oven at 40 C for 12 h. Any remaining moisture was removed under vacuum for 12 h. The mass of the part was then recorded, and the mass change calculated. More details about the samples used and the preparation of the solutions can be found in the ESI † section.

Electrolyte preparation
The electrolyte solution was prepared using a similar method to that described in literature. 57 For the supporting electrolyte, a 3 M H 2 SO 4 solution was prepared using concentrated sulfuric acid (99.99 wt%, Alfa Aesar) and deionized water (diH 2 O). Vanadyl sulfate oxide hydrate (99.9% metals basis VOSO 4 $4.25H 2 O, Alfa Aesar) was added to obtain a 1.5 M solution of VOSO 4 which was dissolved using a hotplate and magnetic stirrer at 50 C until the solution turned a clear blue   ) 107. 43 $ colour. The vanadium(IV), VO 2+ solution was then split into three-parts; two of which were put into the positive tank (100 mL) and one-part into the negative tank (50 mL). These solutions were then charged at 50 mA cm À2 (galvanostatic mode) until the voltage reached 1.75 V. The cell was then charged in potentiostatic mode until the electrolyte was fully charged and the current dropped below 4 mA cm À2 . This resulted in one-part vanadium(II), V 2+ in the negative half-cell and two parts vanadium(IV), VO 2 + in the positive half-cell. Half of the VO 2 + was removed and equal amounts of V 2+ and VO 2 + , 50 mL in each tank, were used for the subsequent experiments. Both tanks were kept under nitrogen at all times to prevent oxidation of the electrolyte.

3D-printing procedure
The cell was designed using Solidworks 2020 CAD soware (Dassault Systèmes). Ultimaker Cura slicing soware was used to generate G-code which can be read by 3D-printers. All cells printed in this work were manufactured using an Ultimaker S5 (Ultimaker) FDM 3D-printer using single extrusion. To ensure strong build plate adhesion, a 20 mm brim was used along with 3D-printing adhesives (Magigoo Original and PP). All prints used a nozzle diameter of 0.4 mm, a layer height of 0.2 mm and were printed with 100% inll and a ow rate of 105% (k ¼ 1.05). For all ABS laments, the nozzle temperature was 250 C and the print speed was 60 mm s À1 . A build plate temperature of 85 C was used for Ultimaker ABS and 100 C for the Verbatim and Formfutura ABS. For the Verbatim PP material, the nozzle temperature was 250 C, the print speed was 30 mm s À1 , the fan speed was 30%, and the build plate temperature was 85 C. Similar settings were used for the Formfutura "Centaur" PP; however, a lower build plate temperature of 65 C was used. The default print speed and temperature settings from the Ultimaker Cura were used for the Ultimaker PP. Retraction settings for the PP prints were also rened to 5 mm and an acceleration of 5 mm s À2 . To prevent vertical seams in the 3D-printed components, the "Z-seam" alignment in Ultimaker Cura was set to "random" and "randomize inll" was also selected.

Non-3D-printed components
For the current collectors, brass disks with a thickness of 4 mm, diameter of 75 mm and a protrusion of 20 mm Â 5 mm were used along with 0.6 mm thick PV15 graphite composite (SGL Carbon). The half-cells were separated by a commercially available Naon® 212 membrane (Sigma-Aldrich) which was held in place using gaskets hand-cut from either 1 mm thick expanded PTFE Fluolion 200 (James Walker) or 1.5 mm thick EPDM sponge (PAR group). Sealing was ensured by use of 75 mm-diameter nitrile O-rings with a 1.78 mm cross-section (Polymax). All components were held together using a simple clamp. For the tests carried out here, an Irwin G-clamp (RS Components), was used and tightened to 10 Nm using a torque wrench. An insulating material was used between the clamp and the current collectors to prevent a short circuit.

Membrane preparation
The Naon® 212 membrane was prepared based on standard procedures from literature. 58 The membrane was rst cut to size (ca. 60 mm Â 60 mm), allowing for a $5 mm overlap around all sides of the electrode compartment. The membrane was rst washed in 100 mL diH 2

Ohmic resistance investigation
The commercial test cell (C-tech innovation) and a 3D-printed test cell was assembled in the same conguration as described in the "cell testing" section using three pieces of graphite felt (GFD 4.65 EA, SGL Carbon) in each half-cell, however no membrane or electrolyte was used. The resistances were measured by a 4-point probe method using an Agilent 34401A Digital Multimeter (Agilent technologies). Measurements were taken three times per assembly. Each cell was then taken apart and reassembled three times to account for any variations in assembly. The average resistance of each cell was then calculated.
X-ray micro-CT acquisition, reconstruction, processing, and analysis Six samples in total, with optimised geometries for X-ray micro-CT scanning, were designed and 3D-printed using ABS and PP, each with ow rates of 90%, 100%, and 105% (k ¼ 0.90, 1.00, 1.05). The geometry chosen was a hollow tube with an outer diameter of 6 mm and wall thickness of 1.5 mm, such that each sample could be brought close to the X-ray source of the Nikon XT H 225 instrument used for all scans, giving rise to a voxel dimension of ca. 4.50 mm. The hole provided a pathway to ow vanadium electrolyte through at a ow rate of 10 mL min À1 for approximately 10 min, prior to all scanning, to potentially highlight areas of electrolyte permeation through residual vanadium in the resultant tomograms. All scans were conducted at an X-ray voltage of 60 kV and beam power of 7 W. 1401 projections, each with an exposure time of 1 s, were collected, giving an overall scan time of <25 min per sample. Given the small lateral dimensions of the samples, four scans were taken along the z-axis, a vertical distance of ca. 7.4 mm between them, constituting an overlap of 15% and total scan time for all six samples of <12 h. Reconstruction of the radiographic projections into 3D tomograms was performed in CTPro3D (Nikon) by nding the optimised centre shi but without beam hardening correction. Image processing and analysis was carried out in Avizo 9.4.0 (Thermo Fisher Scientic). Raw images were Gaussian-ltered (3D, kernel size of 1.5) and stitched together manually before segmentation by a standard watershed approach. 59 The stitched volumes were cropped at the top and bottom to remove edge effects. Phase fraction calculations were performed by simple voxel counting and all fractions are relative to the entire volume encapsulated by the plastic matrix (excluding the pore space in the hole and outside the cylinder).

Conclusions
A process for designing and 3D-printing a low-cost, open-source redox ow battery test cell has been detailed. The cell has been evaluated under demanding test conditions identied from literature (electrolyte ow rates of 50 mL min À1 and electrode compression ratios >40%) and has been shown to be leakproof during long-duration tests (up to 1 month). A coupled electrochemical-CFD model was used to investigate electrolyte ow through different manifold designs to converge on a more effective ow cell design which achieved improved reactant distribution. The 3D-printing platform was then used to validate the simulation results which were reected in real-world performance, the rst time such results have been reported in literature. This highlights the exibility of this method for ow cell production, allowing different designs to be easily evaluated experimentally. Furthermore, the results of these tests compared favourably to a commercially available redox ow battery test cell. This framework presents a compelling avenue for ow cell development and will be utilised extensively in future work to investigate larger-scale cells and new cell topologies.
Many of the techniques investigated in this work are also applicable to the wider eld of ow electrochemistry and can help facilitate the wider use of FDM 3D-printing. Methods of rening print parameters to reduce porosity have been validated by using X-ray micro-CT to show the effect of different kvalues on part quality. ABS has been identied as a well-suited polymer for 3D-printing ow cells and has also been shown to be chemically compatible with a number of commonly-used aqueous ow battery supporting electrolytes. Although PP showed promise during the evaluation of materials for 3Dprinting of ow cells, these parts experienced electrolyte permeation during leak testing. A more thorough examination of the causes and mitigation of this effect is the focus of current investigations.
FDM 3D-printing has been shown to present an extremely cost-effective method of ow battery test cell production. Work is on-going to scale up the 3D-printed test cells detailed here towards larger industrial-scale cells and stacks to investigate their economic viability. This pathway is illustrated in Fig. 10.

Conflicts of interest
There are no conicts to declare.