Introduction

Microplastics (MPs), plastic particles and fibres smaller than 5 mm, are ubiquitous pollutants found in water1, soil2 and air3. In all these environments, MPs have adverse effects, for example, they accumulate fungal pathogens in soil4, cause phytotoxicity on vascular plants5, enter marine top predators through trophic transfer6, and induce oxidative stress, inflammation, and metabolic disorders in humans7. It is widely acknowledged that a diverse combination of mitigation strategies is necessary to reduce MPs from entering the environment. Long-term strategies include shifting to a circular economy, developing new materials and sustainable products, and conducting educational work to change societal mindset8,9,10. Short-term measures include, for example, legislation to ban MPs in cosmetics, the use of substitute materials, or cleaning activities11,12. Filtration is another short-term means to reduce MPs close to their sources and entry points, but current filtration solutions are hindered by low retention efficiencies for MPs, as well as rapid clogging.

Washing machines are a major entry path for MPs, releasing between 10 and 120 g of MP fibres from textiles per person per year, and are one of the major MP sources13,14,15. Washing machines have coarse filters to protect the pump by retaining stones or coins, but no MP filter16. Hence, MPs are brought unfiltered into the sewage system. Wastewater treatment plants equipped with secondary and tertiary treatment stages retain 84–94% of MPs17. However, retained MPs accumulate in the sewage sludge, with 63–90% coming from washing machines17. The MPs are then reintroduced into the environment when the sewage sludge is disposed of on agricultural fields, as is common practice globally17,18. Therefore, despite the high retention rates within wastewater treatment plants, intercepting MPs before they reach the sewage system is a critical and underdeveloped intervention point.

We present a biomimetic, fish-inspired filter (FiF) with high filtration efficiency of standardised MP fibres and a self-cleaning process. Our filter module is based on the filter-feeding process in ram-feeding fishes because aspects of this filtration process, in principle, fits some of the requirements in washing machines such as particle size, flow regime, and filter size (Supplementary Note 1 and Supplementary Fig. 1). Ram-feeding fishes are pelagic fishes that use their forward motion to induce flow through the gill arch system19,20. The gill arch system is composed of gill arches with elongated gill rakers. The morphology of the gill arch system is species dependant with some having denticles on the gill arches and gill rakers19, while others produce mucus to accumulate particles21, have surface structures that induce local vortices to retain particles22 or use a combination thereof. In most filter-feeding fishes, the gill arches form a cone-shaped geometry that tapers down towards the oesophagus within the buccal cavity19,23,24,25. While swimming forward with an open mouth, a tangential flow transports the food particles towards the oesophagus, and the cleared water exits between the gill arches and under the operculum. This process was previously described as cross-flow filtration with variations found in some species called cross-step filtration20,22. Based on morphology and video analysis of their feeding behaviour19, it is likely that the filtration process in ram-feeding fish like anchovy, pilchard, and Atlantic mackerel, in fact, combines cross-flow and dead-end filtration. In these fishes, the gill rakers form a flat surface that induce particles rolling towards the oesophagus where they accumulate before being swallowed19. Particle rolling is enabled by the conically tapered geometry of the gill arch system. The geometry is described by an angle, the angle of attack α (Fig. 1A), and may range between 0°(dead-end filtration) and 90° (cross-flow filtration, Fig. 1B). Therefore, the filtration process was previously called semi-cross-flow filtration19 (Fig. 1B). Our FiF mimics the functional morphology in selected ram-filter-feeding fishes to take advantage of the semi-cross-flow filtration process. Furthermore, by varying parameters like filter size, mesh size or angle of attack in physical and numerical experiments, we demonstrate that filter performance depends on specific parameter combinations and can thus be tailored to a wide range of potential applications, including retention of MPs in washing machines. Our fish-inspired filter (FiF) is unique in its combination of (i) a cone-shaped filter element with a filter medium at a set angle of attack α to induce semi-cross-flow filtration, (ii) a filter housing with adapted inflow and separate permeate and concentrate outlet, and (iii) an adjustable periodic cleaning mechanism that delays clogging.

Fig. 1: Interaction of particles with filter media at different angles of attack α.

A Volume render cross-section through the frontal plane with view on the ventral side of the buccal cavity of Micro-CT scans the Indian mackerel Rastrelliger kanagurta. The blue arrow and blue line indicate the inflow and midline of the fish, and one side of α. The yellow, orange and red lines indicate the orientation of the four gill arches (GA1-4) and the other side of α. B Schematic drawing of (i) dead-end filtration with α = 90°, (ii) semi-cross-flow filtration with α ranging between 0° and 90°, and (iii) cross-flow filtration with α = 90°. Depending on inflow and the α of the filter medium, the tangential flow induces particles to stop or roll along the surface of the filter medium. C Probability of particles rolling after contact with the filter medium for each α (10°–60°), particle type (Brine shrimp eggs, Brine shrimp adults, MP fragments, MP fibres), and mesh size (53, 100, 300 µm). Polynomial regression, with y on x, was used to describe the influence of α on the probability of rolling. There are no results for the brine shrimp eggs and MP fragments because the mesh size was too big to retain the smaller particles. D Boxplots of all R2 of the Boltzmann fit to describe the motion of all four particle types. For more details of particle characteristics and results of preliminary experiments, see Table 2 and Supplementary Fig. 3.

Results

Effect of angle of attack on particle deposition during semi-cross-flow filtration

The gill arch system inside the buccal cavity is tapered towards the oesophagus, bilaterally symmetric, and consists of five gill arches that bear elongated gill rakers with denticles forming a mesh (Fig. 1A). In micro-CT scans of five ram-feeding fishes, the angle of attack α was measured between the hyoid bone and an extended line of the GR of the first four arches in virtual cross-sections of the fish’s head (Fig. 1A). Across all five species, α ranged between 4° for more anterior and 47° for more posterior gill arches (Supplementary Fig. 2). Based on the semi-cross-flow filtration process19, this would indicate that particles entering the fish’s mouth would experience higher tangential forces at lower α at the more anterior gill arches. Consequently, particles would roll along the gill rakers and denticles so that they accumulate close to the oesophagus due to higher α before being swallowed (Fig. 1A, B). In order to test this hypothesis, we observed particle rolling on filter meshes at α ranging from 0° to 60° under a range of other parameter variations in a series of flow tank experiments. We found that the share of rolling MP and non-MP particles —with no ‘permanent’ deposition and potential clogging of the filter mesh- increases with lower α, but the strength of that effect varies with particle type and mesh size (Fig. 1C). Round brine shrimp eggs, a part of the natural food of ram-feeding fishes, keep rolling up to an angle of 40° after which the probability to roll drops and reaches <30% at 60°. Brine shrimp adults show a similarly high probability to roll at low angles but a more linear decrease with higher α. The probability of rolling for MP fragments decreases in a cubic relation and is generally lower than for the natural particles. The probability of rolling is overall the smallest for the MP fibres compared to the other particle types. The probability of rolling drops from 20% at 10° to 0% at 40° with the 53 µm mesh size and from 37.5% at 10° to 0% at 50° with the 100 µm mesh size. With the 300 µm mesh size, the probability of rolling first increases from 50% at 10° to 70% at 20° and 30° and then decreases to 20% at 60° (Fig. 1C). Another difference between the brine shrimp adults in comparison to the brine shrimp eggs and MPs is the quality of the Boltzmann fit to the velocity profile of the particles when in contact with the filter mesh (Supplementary Fig. 3). For the brine shrimp adults, R² is 0.60 ± 0.34 whereas it is 0.95 ± 0.03, 0.97 ± 0.05, and 0.98 ± 0.06 for the brine shrimp eggs, MP fragments and MP fibres, respectively (Fig. 1D). This indicates a different behaviour of the motile brine shrimp adults in comparison to the other, non-motile particles. Hence, for MP retention, the angle of attack should preferably be lower than 20° to keep the MPs in suspension or rolling for longer and reduce clogging of the filter medium.

Design of fish-inspired filter element

In order to design a filter that separates particles based on semi-cross-flow filtration, we mimicked the gill arch system morphology with the filter element. The basic design of a filter element is radially symmetric and has several arched support structures that resemble the gill arches and can hold commercially available and standardised filter meshes (Fig. 2A, B). The filter mesh creates a flat filter medium as observed in the Atlantic herring, the Atlantic pilchard, and the Atlantic anchovy formed by the gill rakers and denticles (Fig. 2B). The filter element is described by an inflow opening, resembling the mouth, with diameter LM and an outflow opening, resampling the oesophagus, with diameter LC and a filter length of LF (Fig. 2A). Since LM > LC, the gill arches and the associated mesh decrease in diameter, which results in the angle of attack, α. By changing the filter element length or the opening diameter, α can be adjusted to lower or higher angles to increase particle rolling towards the outflow opening, while clean fluid exists laterally through the mesh. Overall, we built four different filter elements that covered three different length, i.e. small, medium and large, and two α, i.e. 11° and 22° (Table 1). Some of the species, i.e. Atlantic mackerel and Indian mackerel, also show shortened gill rakers and teeth that point inwards into the buccal cavity (Fig. 2C). These structures were mimicked by the hooked side of Velcro or 3D printed row of ellipsoids that can optionally be added inside any of the filter elements. For comparing the FiF element to a conventional dead-end filter design, one filter element had an α of 0° and no outflow opening. The length was chosen so that the filtration area AF was kept similar to the Large-11 filter elements to allow comparisons of these different designs (Table 1).

Fig. 2: Abstraction of morphological traits of ram-feeding fishes into CAD models and physical models.

A The cone-shaped gill arch system with a given angle of attack α is mimicked by the support structures of the filter elements (blue arrows). The filter element is described by an inflow opening diameter LM and an outflow diameter LC with a filter length of LF and a total length of LT. B The gill rakers (GR) and denticles (D) form the meshes (red box) and are mimicked with a purchased filter mesh that is glued onto the support structures. C Surface structures as observed in ram-feeding scombrids are mimicked by a row of ellipsoids that mimic the short GR, and hooked tape mimics the denticles and teeth, and is glued in the Large-11 filter element.

Integration into a functional filter housing

The filter element was encased in a filter housing consisting of an inlet, a flange, a transparent pipe for visual inspection, and an outlet (Fig. 3A, B). The outlet has two outlet pipes to separate the clean fluid that exits laterally through the filter element, i.e. the permeate (VP) with potentially lost MPs (MP), and the drainage of concentrate (VC) through the outflow opening that contains the accumulated particle mass (MC). Both outlet pipes are equipped with automated valves to mimic the periodic cleaning and swallowing in the fishes. During filtration mode, the concentrate valve is closed and the permeate valve is open which accumulates particles in the filter element and allows outflow of cleaned water through the mesh. For cleaning of the filter, the permeate valve closes while the concentrate valve opens so that the fluid with the particles is directed out of the concentrate outlet and the filter element is cleaned, here shown with MP fibres (Fig. 3C, D). In the fishes, periodic cleaning, i.e. swallowing, was observed in intervals between 0.17 s in the Atlantic herring, 0.27 s in the Atlantic herring, 0.53 s in the Atlantic mackerel and 3.7 s in the Indian mackerel19. For the FiF experiments, the cleaning intervals were set to clean after half of the test suspension passed and right before the total test suspension passed. Since volume flow depends on filter size and inlet type (Supplementary Fig. 5), the cleaning intervals were adjusted according to the particular filter setup and ranged between 0.6 s and 1.3 s.

Fig. 3: Design of the filter housing and cleaning mechanism.

A The modular filter housing consists of an inlet with a screw to de-air the system, a flange, a transparent pipe, and the outlet with a separation of concentrate and permeate. The FiF separates the feed suspension, which consists of a fluid volume (VF) and a particle mass (MF), into a concentrate volume (VC) with a particle mass fraction (MC) and the clean permeate (VP) with a second particle mass fraction (MP). The particle mass fraction that remains in the filter element is called the retentate (MR). The cleaning mechanism is regulated by the permeate valve and the concentrate valve at the respective outlets. B Set-up of the FiF with the long inlet and the Large-11 filter element. C Observations of MP fibres in the Large-11 filter element with the snail inlet at the start of the experiment, before cleaning, and after cleaning. D Observations of MP fibres in the Small-11 filter element with the snail inlet at the start of the experiment, before cleaning and after cleaning.

To investigate whether the inlet design and different cleaning setups will influence the flow patterns inside the filter element, we tested two different inlet geometries and two cleaning modes using computational fluid dynamics (CFD) (Fig. 4). The geometry consisted of a rotational symmetric Large-11 filter element. To quantify the effects on the flow, we measured the effective angle of attack α(eff). While the angle of attack α is defined by the filter element geometry and remains static, α(eff) is the resulting angle by the streamlines at contact with the filter mesh when the flow exits laterally through Gap 1–Gap 7 (Fig. 4B). A lower positive α(eff) should result in flatter, more parallel streamlines towards the filter mesh and, hence, further increase the probability of rolling for any particles that follow the streamlines.

Fig. 4: Flow through a 2D half of the FiF with a Large-11 filter element using computational fluid dynamics (CFD) in COMSOL®.

We compared the influence on flow behaviour of a long inlet (9° angled) with (A) no inner wall and D an inner wall during the three operating modes: (i) the concentrate outlet is closed and the permeate outlet is open (filtration mode), (ii) the concentrate outlet is open and the permeate outlet is open (single cleaning mode) and (iii) the concentrate outlet is open and the permeate outlet is closed (double cleaning mode). The support structured and gaps between the support structures of the filter elements are exemplarily indicated in Fig. Aiii). The light blue arrow indicates the direction of flow. B The angle of attack α (black, Aiii) is given by the filter element, whereas the effective angle of attack α(eff) changes depending on the orientation of the streamlines in relation to filter mesh (blue). Note that the colour scale indicating flow velocity is the same in (i) and (ii) but differs in (iii). The change of α(eff) at the filter mesh over the full length of the filter element for each operation mode is plotted for the setup with (C) no inner wall and E inner wall. The support structures of the filter elements are shown in grey with the separation into the seven gaps. The negative α(eff) values in (C) are caused by a recirculating vortex created across the filter element and reversed streamlines at gap1 (Supplementary Fig. 4).

The CFD results reveal that the presence of an internal wall in the inlet alters the local flow field and improves α(eff) at the filter mesh, especially at Gap1 and Gap2 (Fig. 4Ai, Di). Without the internal wall, the incoming flow is not deflected but hits the filter mesh in a straight line and creates a recirculation zone with backflow through Gap 1 and Gap 2 indicated by negative α(eff) (Fig. 4C, E and Supplementary Fig. 4). This dead zone reduces the effective filtration area, potentially leading to uneven particle accumulation and a less favourable flow distribution for semi-cross-flow filtration. In contrast, the internal wall guides the flow more uniformly along the filter surfaces, thereby leading to a low, but still positive α(eff). Besides the filtration mode (permeate outlet open, concentrate outlet closed, Fig. 4Ai, Di), we also tested two cleaning modes, i.e. single cleaning (permeate outlet open, concentrate outlet open, Fig. 4Aii, Dii) and double cleaning (permeate outlet closed, concentrate outlet open, Fig. 4Aiii, Diii). The three operating modes have a significant impact on the local flow orientation near the concentrate outlet. During both cleaning modes, backflow forms at Gap 6 and Gap 7, leading to a negative α(eff) (Fig. 4C, E). This effect is especially strong in the double cleaning mode and may be beneficial for the FiF’s performance. The reverse-throughflow could detach particles adhering to the filter mesh and flush them towards the concentrate outlet. As positive α(eff) increases from Gap 1 to Gap 7 during the filtration mode, most particles will likely deposit in the posterior region during semi-cross-flow filtration. Consequently, this backflushing effect may be particularly advantageous at these gaps to prevent clogging.

Proof of concept of the fish-inspired filter

Our FiF retains 99.6 ± 0.8% of MP fibres in the lab experiments (Fig. 5A, B). While this filtration efficiency ER is similar compared to the filter design based on dead-end filtration with 97.7 ± 3.1% for the same filtration area, our FiF delays clogging through a cone-shaped geometry in combination with a periodic cleaning mechanism, and collects up to 84.8 ± 3.5% of the retained 2 mm MP fibres, outside of the FiF in the concentrate (concentrate filtration efficiency EC). In the FiF with a conical-shaped filter element and an angle of attack α = 11°, the concentrate is collected through an outlet at the terminal end (Fig. 5A). The concentrate contained only 850 ml, i.e. 4.25% of the filtered fluid volume. At the same time, 14.3 ± 3.7% of the MP fibres remain as retentate in the filter element, which is approximately one seventh of the MP fibres in the filter element of the dead-end filter. In the dead-end filter, no MP fibres are collected outside the filter element as it has no concentrate outlet and all fluid has to pass the filter medium, which is typical for example in cartridge filters (Fig. 5C). This comparison shows that our FiF keeps the MP fibres rolling or in suspension, prevents the accumulation of MP fibres in the filter itself, and allows removal of the MP fibres through periodic cleaning. The majority of MP fibres are collected in the concentrate outside the FiF.

Fig. 5: Comparison of the fish-inspired filter (FiF) with a dead-end (DE) filter element.

A Atlantic mackerel Scomber scombrus with open mouth during filter-feeding and CAD design of the Large-11 FiF element. B Performance comparison of the DE and FiF filter element design based on MP fibre share in permeate, retentate and concentrate. C Common engineered filter cartridge of a MP washing machine filter and computer-aided design (CAD) of a filter element based on that design with the same filtration area as the FiF element. For filter element dimensions, see Table 1.

Parameter adaptation for the application in washing machines

Based on the conditions found in washing machines, a filter should fit the available space in the washing machine housing, not add additional drag for the pump, withstand changing volume flow and flow velocities, and have high filtration efficiencies (see washing machine requirements in Supplementary Note 1). FiF performance was evaluated on three parameters: filtration efficiency of MP fibres retained in the FiF (ER), i.e. the concentrate and retentate, MP fibres retained in the concentrate (EC), and yield in concentrate (ηC), which includes the fibre mass in the concentrate (MC) and the concentrate volume (VC). Yield in concentrate is especially relevant for applications in washing machines, because a small concentrate fluid volume is preferable to reduce post-treatment and user interaction.

First, we decreased the initially designed filter element length LF of 102.7 mm (large) and designed two smaller sizes of 77.7 mm (medium) and 52.7 mm (small) (Table 1). Even though a lower α would increase particle rolling, the length of the filter and therefore α is limited by the available space in the washing machine. Therefore, we kept α = 11° (denoted by ‘11’ in ‘Large-11’, ‘Medium-11’ and ‘Small-11’ filter types) as presented in the initial experiment to compare the FiF with the dead-end filter (Fig. 5). While E**R remains relatively similar in Large-11 with 93.2 ± 2.8%, Medium-11 with 94.0 ± 4.2% and Small-11 with 96.0 ± 5.5%, the MP fibre share in the concentrate E**C increased from 35.0 ± 12.6%, to 50.6 ± 9.4% in Medium-11 up to 65.3 ± 13.7% in Small-11 (Fig. 6A). The yield in concentrate ηC also slightly increased with 83.2 ± 6.5% in Large-11, 84.9 ± 10.3% in Medium-11 and 88.8 ± 15.3% in Small-11.

Fig. 6: Filtration performance of the FiF (each N = 5) based on fibre share in permeate (brown), retentate (green), and concentrate (blue).

Yield in concentration ηC is indicated by the mean with error bars (black). Variations include (A) filter size with Small-11 (S), Medium-11 (M) and Large-11 (L) filter element, snail inlet and mesh size 100 µm, B angle of attack α with the small filter element, snail inlet and mesh size 100 µm, C mesh size with the Large-11 (left) and Small-11 (right) filter element and snail inlet, D inlet type (swirl, long inlet at 9° with internal wall, snail) with Small-11 filter element and mesh size 100 µm, and E bio-inspired surface structures on a Large-11 filter element (100 µm mesh size) with half way hooked tape (L-50), full way hooked tape (L-100) and bumpy structures (L-B), and the snail inlet. Surface structures made from hooked tape and 3D-printing mimic teeth and bumpy gill rakers found in ram-feeding mackerels (Fig. 2). Please note that the experiments presented in E were done with a different test volume and, therefore, cannot be directly compared to the other results.

When increasing α to 22° in the Small filter element (‘Small-22’), we observed that E**R decreased from 96.0 ± 5.5% to 84.5 ± 6.7%, E**C decreased from 65.3 ± 13.7% to 37.9 ± 14.8%, and ηC decreased from 88.8 ± 15.3% to 62.0 ± 14.8% (Fig. 6B), suggesting that a larger angle of attack may adversely affect performance.

As there are not only MP fibres of different sizes in the washing machine effluent, but also sand, dust, pollen, hair and detergents (Supplementary Note 1), we tested three mesh sizes (53, 78 and 100 µm) for the Large-11 and two mesh sizes (53 and 100 µm) for the Small-11 (Fig. 6C). Although mesh size is expected to influence filtration performance parameters, the changes showed no trend related to an increase or decrease in mesh size (Fig. 6C).

To verify the influence of different inlet geometries on filtration performance, as was indicated by the CFD experiments, we tested the long inlet at 9° with an internal wall (long inlet), a swirl inlet with a helical internal wall and a snail inlet (Fig. 6D). The swirl inlet and the long inlet performed similarly with E**R being 96.0 ± 5.5% and 96.2 ± 3.6%, E**C being 65.3 ± 13.7% and 66.8 ± 7.8%, and ηC being 88.8 ± 15.3% and 89.9 ± 9.4%, respectively. The snail inlet showed a less high E**R with 85.5 ± 14.1% and ηC with 67.4 ± 23.0%, but E**C was slightly higher with 68.0 ± 38.3% (Fig. 6D).

We also investigated the relevance of surface structures on the inside of the filter element (Fig. 2C) because some fish species showed surface structures directed perpendicular to α (Fig. 2C). We fitted the Large-11 filter element with three different surface structure designs but the results suggest that these did not improve performance (Fig. 6E).

Discussion

The best-performing FiF combination, based on the lowest share of MP fibres in the permeate and the highest yield in concentrate, is the Large-11 filter element with a mesh size of 78 µm, in combination with the swirl inlet. It retains over 99% of the MP fibres with 0.8 ± 2.2% left in the permeate (E**R, Fig. 6C). The yield in concentrate ηC is 97.5 ± 5.6%. However, the best performing FiF configuration based on MP fibre share in the concentrate is the Small-11 filter element with a mesh size of 53 µm in combination with the swirl inlet with 79.8 ± 6.8% (Fig. 6C). This may indicate that the cleaning mechanism covers a larger share of the filtration area in the Small-11 filter element. Specifically, the shear forces in semi-cross-flow filtration will move the particles along the filter mesh at a low α until they deposit near the concentrate outlet as observed in the Large-11 and Small-11 filter elements (Fig. 3C, D), similar to the forces in cross-flow filtration26,27. Consequently, the reversed flow near the concentrate outlet during the double cleaning mode, as shown in the CFD simulations (Fig. 4), detaches more particles, flushes them out of the filter element, and increases the share of MP fibres in the concentrate. Both best performing FiFs used the swirl inlet, which might be due to the creation of a rotational, likely turbulent flow that further reduces the positive effective angle of attack α(eff) laterally towards 0° so flow becomes more parallel to the filter medium. In cross-flow filtration, turbulent flow is preferred over laminar flow because it increases the shear rate and prevents particle deposition on the filter mesh27. The poorest performing FiF combination is the Small-22 with ER being 15.5 ± 6.9% and ηC of 62.0 ± 14.8%. This indicates that a higher angle of attack substantially reduces the benefits of semi-cross flow filtration (Fig. 1).

Another advantage to cross-flow filtration is the low concentrate volume and, therefore, an increased yield in concentrate ηC. In established cross-flow ultra-, nano- and microfiltration processes, the concentrate volume ranges between 10% and 50% of the feed, depending on system design28,29,30. With the conical FiF and the periodic cleaning, we achieve a concentrate volume of 4.9 ± 1.6% of the feed across all FiF combinations (N = 70). A smaller concentrate volume lowers post-treatment or disposal costs, improves the cleaning efficiency, and reduces fouling through smaller stagnant volume and cake layer formation. Additionally, the FiF collects most of the MP fibres outside the filter housing, and only one seventh of the MP fibres remain in the filter element (Fig. 5B). This indicates that clogging may be delayed by up to a factor of seven. An increase in cleaning frequency could prolong the delay; however, this might come at the expense of a high fluid volume in the concentrate. A thorough cleaning in which the filtration process was set back to starting conditions was not observed as all FiF combinations still had some MP fibres remaining in the retentate (Fig. 6). During examination of the filter elements after the experiments, we noticed that some MP fibres were stuck in the pores in areas glued to the support structures. Remaining MP fibres may necessitate that the filter element be removed for cleaning, which is common practice in many filtration processes, such as dead-end filtration31. But since most MP fibres are collected outside and the dirt-holding capacity is not limited to the volume of the filter element, cleaning time and required energy are kept to a minimum. By collecting most MP fibres outside the FiF, wear on the filter medium from mechanical cleaning (e.g. scrapes or brushes) is reduced, helping to overcome a key limitation of conventional domestic filters31.

Besides the FiF presented here, other bio-inspired filtration processes also show potential for technical applications. The particle separator by Piedrahita et al.32 is inspired by cross-flow filtration in pump-feeding fish. In simulations, a maximum of about 70%32 and in experiments with prototypes, a maximum of 43% of particles were retained33. The concept of ‘ricochet separation’ was discovered in manta rays by Divi et al.34, which describes a mechanism in which particles bounce off the filter lobes, i.e. specialised gill rakers. A simplified filter lobe geometry was used to manufacture surface patterning on membranes to manipulate local flow fields and inhibit particle deposition35. In small-scale experiments under cross-flow, the membranes with a pore size between 3.5 and 10.5 µm removed 97.6% of MP particles of 700 nm in diameter from water36. In 2025, the filtration process was also virtually tested in CFD simulation for an application in fuel filtration, where it showed a significant reduction in pressure drop37. The ‘cross-step filtration’ is another process found in ram-feeding fishes such as the American paddlefish and was first described by Sanderson et al.23. The proof-of-concept was shown in abstracted models for the collection of harmful algae38 and fibre collection in washing machines16. The commercial VORTX filter by CLEANR®, a spiral adaptation of this principle, reportedly achieves 90% efficiency for MP fibres ≥50 µm, a 300% longer life and operates without a replaceable filter39. While promising, its performance data are currently based on manufacturer claims and focus primarily on domestic washing machine applications. These examples reflect the diversity of bio-inspired filtration designs serving a wide range of applications—from nanoparticle removal to domestic washing machine filtration. This diversity not only highlights the versatility of biological inspiration but also demonstrates a growing demand for bio-inspired solutions to complex separation challenges. In this context, the FiF offers a valuable addition and represents an alternative route, particularly suited to applications requiring coarse fibre separation, low concentrate volumes, and modular cleaning systems.

Currently, the MP fibre length is larger than all tested mesh sizes and the rod-shaped MP fibres can only pass when they encounter the mesh perpendicularly. Smaller MP fibre sizes should be tested to see if the semi-cross-flow process also keeps smaller MP fibres in suspension or if a smaller mesh size is required. The cut-off size of retained MP fibres is a crucial value to determine if the FiF can compete with other washing machine filters. So far, it was suggested that a filter should retain >80% of MP fibres larger than 100 µm to be competitive40,[41](https://www.nature.com/articles/s44454-025-00020-2#ref-CR41 “McIlwraith, H. K. et al. Capturing microfibers – marke