Recent developments in 3D-printed membranes for water desalination

The recognition of membrane separations as a vital technology platform for enhancing the efficiency of separation processes has been steadily increasing. Concurrently, 3D printing has emerged as an innovative approach to fabricating reverse osmosis membranes for water desalination and treatment purposes. This method provides a high degree of control over membrane chemistry and structural properties. In particular, when compared to traditional manufacturing techniques, 3D printing holds the potential to expedite customization, a feat that is typically achieved through conventional manufacturing methods but often involves numerous processes and significant costs. This review aims to present the current advancements in membrane manufacturing technology specifically tailored for water desalination purposes, with a particular focus on the development of 3D-printed membranes. A comprehensive analysis of recent progress in 3D-printed membranes is provided. However, conducting experimental work to investigate various influential factors while ensuring consistent results poses a significant challenge. To address this, we explore how membrane manufacturing processes and performance can be effectively pre-designed and guided through the use of molecular dynamics simulations. Finally, this review outlines the challenges faced and presents future perspectives to shed light on research directions for optimizing membrane manufacturing processes and achieving optimal membrane performance.


Introduction
Membranes serve as a vital component in the process of water treatment and purification, where they play a crucial role in separating impurities and contaminants from water, ensuring the delivery of safe and clean drinking water.The remarkable growth of membrane technology in the separation industries is largely attributed to substantial advancements in materials and manufacturing techniques.These developments enable the scalable manufacturing of membrane materials for precision separation.Before delving into the recent advancements and future prospects of printing as a membrane manufacturing method, it is crucial to understand its historical context [1][2][3].The advent of asymmetric membranes marked a significant milestone in the widespread adoption of membranes across multiple industries.Asymmetric membranes are characterized by a thin, selectively permeable layer supported by a porous substrate [4][5][6][7][8].This innovative design enhances separation processes by capitalizing on the distinctive attributes of the thin, selective layer.Simultaneously, it benefits from the structural stability and support offered by the underlying porous material.The development of asymmetric membranes revolutionized the capabilities and applications of membrane technology, enabling its successful integration into diverse sectors.Furthermore, the nonsolvent-induced phase separation process, renowned for its simplicity, facilitated the cost-effective fabrication of membranes with extensive surface areas.This breakthrough in reverse osmosis (RO) membrane manufacturing not only improved performance but also established RO as a viable and efficient technology.Compared to conventional separation methods [9][10][11][12][13], RO offers a fundamentally simpler and more convenient design and operation [14].Additionally, by combining RO with other separation technologies, highly efficient and selective hybrid processes can be developed [15].The advancement in controlling polymer precipitation also paved the way for the creation of asymmetric membranes using various materials [6,8,16].
While asymmetric membranes represented a significant advancement in membrane preparation technology, they still fell short of meeting certain industrial requirements.Although RO showed potential for cost savings, its selectivity needed enhancement to enable efficient desalination of seawater in a single pass.This necessitated the development of new materials and manufacturing processes to enhance the desired property.The breakthrough came with the introduction of thin film composite (TFC) membranes, pioneered by Cadotte [17].Although not the first TFC membrane [18], it has emerged as a pivotal membrane platform with a distinctive feature: the separation of the selective layer from the support layer.The choice of materials for the selective layer is broad, often including more exotic options, as this layer is typically thin and can be precisely tailored by changing its chemistry.Consequently, TFCs found applications in various fields, including nanofiltration (NF) [19,20], thereby expanding the reach of RO technology beyond seawater desalination.TFC membranes have delivered significant improvements in ion selectivity and water permeance by orders of magnitude [21,22].However, numerous studies [23][24][25] investigated novel membrane materials for separation processes.Nevertheless, fewer studies have addressed the critical issue of manufacturing these membrane materials for practical applications.The potential improvements in performance are likely associated with advancements in manufacturing techniques, chemical structures, and other influential factors.
Additive manufacturing, commonly known as '3D printing' , has emerged as an increasingly progressive manufacturing method adopted across diverse industrial domains.One of its advantages lies in its capability to fabricate complex geometric structures, using a variety of materials in a single step.The origins of 3D printing can be attributed to Hideo Kodama, a trailblazer who revolutionized the field through the utilization of ultraviolet light for polymer solidification, establishing the fundamental principles for stereolithography (SLA) [26].The precise and controllable nature of this approach has attracted considerable interest, not only for printing traditional materials but also for cutting-edge nanomaterials.To date, during its development, 3D printing has made notable advancements in achieving higher resolution, enhanced scalability, quicker printing speeds, and reduced material consumption.To accelerate the manufacturing process using 3D printing, it is vital to investigate and analyze the parameters that influence both the manufacturing process and the performance of the resulting membranes.In experimental work, it poses challenges to simultaneously investigate manufacturing methods and all potential factors that can affect both the membrane manufacturing process and the resulting membrane performance.
Also, it is important to note that achieving the desired membrane performance through experimental trial and error can be a challenging and costly endeavor.However, molecular dynamics (MD) and density functional theory (DFT) simulations provide an alternative approach for obtaining a mechanistic and physical understanding of manufacturing processes and membrane performance in advance.By utilizing these simulations, we can remarkedly reduce both the time and cost associated with the experimental process, thereby expediting the design and development of novel membranes.Until now, MD/DFT simulations have been extensively employed to study the transport of water or solvents through various types of membranes.MD/DFT simulations have demonstrated substantial value in simulating water or solvent transport across diverse membrane types and assessing the feasibility of synthesizing structures, respectively.As a result, simulation techniques have gained widespread use in exploring and investigating membrane behavior.MD simulations provide crucial insights into the dynamic behavior of molecules within membranes and their transport properties [27][28][29].On the other hand, DFT simulations provide valuable information about the electronic structure, stability, and energetics of synthesized membrane structures [30][31][32].Together, MD and DFT simulations serve as powerful tools for comprehensively studying and understanding membrane properties and manufacturing processes.
This review article focuses on the recent advancements in 3D-printed RO membranes for water desalination and wastewater treatment.Section 2 provides a brief overview of the current progress in membrane manufacturing technology specifically tailored for water desalination purposes.Section 3 presents an in-depth analysis of the latest advancements in the preparation of 3D-printed membranes.Additionally, it showcases five distinct types of special 3D-printed RO membranes and elucidates their respective membrane performance.In section 4, the article delves into how membrane manufacturing processes and performance can be predicted and evaluated using MD/DFT simulations.The section further explores the underlying factors that contribute to the enhanced performance of these membranes.Lastly, section 5 addresses the challenges and future perspectives of membrane manufacturing, along with strategies  [34].CC BY 4.0.),interfacial polymerization (The image are reproduced from [35].CC BY 4.0.), and molecular layer by layer (The image are reprinted from [36], Copyright (2021), with permission from Elsevier.), and 3D printing (The image are from [37].Reprinted with permission from AAAS.).The current advancements in membrane manufacturing using 3D printing techniques are noteworthy, such as 2,2-bis(3-amino-4-hydroxyphenyl) hexafluoropropane (HFP-mAP) based membranes, polyamide membranes, and CNT/polyamide membranes (The image are adapted from [37][38][39][40][41].).Reprinted from [38], Copyright (2015), with permission from Elsevier.From [37].Reprinted with permission from AAAS.Reprinted with permission from [39].Copyright (2018) American Chemical Society.Reprinted from [40], Copyright (2019), with permission from Elsevier.Reprinted from [41], Copyright (2020), with permission from Elsevier.
to optimize membrane performance.To facilitate researchers' understanding, we have crafted a concise summary figure, depicted as figure 1, illustrating the topics presented in this article.

Current progress in membrane manufacturing technology
The membrane industry has witnessed rapid growth, resulting in the emergence of several conventional manufacturing processes that play a crucial role in producing a substantial number of available membranes.For the sake of conciseness, this discussion will specifically concentrate on polymeric membranes manufactured in flat sheet platforms at high volumes.Nonwoven membranes, which have already been reviewed in other studies [42], will not be included in this discussion.Our primary emphasis in this section will be on the most commonly used manufacturing techniques on both lab scale and industrial scale, including phase extrusion, phase inversion, interfacial polymerization (IP), molecular deposition layer by layer (mLBL), and 3D printing.

Extrusion
There are three primary methods commonly used for fabricating flat sheet films through extrusion: blown films, cast films, and extruded films.Blown films, also referred to as tubular films, are produced by extruding plastic through a tubular die and followed by stretching and blowing the film with air.As shown in figure 2(a), this process creates a thin-walled film, often utilized for poly(lactic acid) films [43].Cast films, on the other hand, are created using a similar extrusion technique.In this method, polymer pellets or powder are melted while traveling along the flights of an externally heated screw.Upon reaching the end of the screw, the molten polymer is forced through a flattened die, which can divide it into multiple layers.If different types of polymers, or polymers with various colorants and additives, are extruded into a single film, multiple extruders can be fed into one die.Cast film extrusion finds extensive application in producing polypropylene (PP) films for packaging purposes [44][45][46].Additionally, extrusion is commonly employed in manufacturing polytetrafluoroethylene and other fluorinated polymer membranes.It is important to mention that extrusion processes are not limited to the fabrication of flat sheet films.For instance, ceramic membranes  [47], Copyright (2019), with permission from Elsevier; the abbreviations GA, PL, and GN stand for graphene, paraffin liquid, and graphene nanosheets, respectively; (b) nonsolvent phase inversion casting process, reproduced from [34].CC BY 4.0; (c) general synthesis protocol for PA TFC membrane preparation using interfacial polymerization, reprinted from [28], Copyright (2017), with permission from Elsevier; (d) schematic of molecular deposition layer by layer to prepare PA TFC membranes, reprinted from [36], Copyright (2021), with permission from Elsevier; (e) electrospray printing process for PA selective layer.From [37].Reprinted with permission from AAAS. and some fiber membranes are also manufactured using extrusion techniques.This involves the extrusion of a polymer dope, followed by its precipitation in a nonsolvent bath.This combined method allows for the production of hollow fiber membranes with specific properties and characteristics.

Phase inversion
Phase inversion is indeed a prevalent and widely used method in membrane manufacturing, serving as the foundation for creating various types of asymmetric polymeric membranes.Typically, this is achieved through nonsolvent-induced phase separation [48][49][50].As shown in figure 2(b), the process involves casting a polymer dope solution onto a film using a dye, followed by immersion in a coagulation bath containing a nonsolvent.Within this bath, an exchange takes place between the solvent and nonsolvent, leading to the precipitation and formation of a thin polymer film [34,50].Phase inversion has been pivotal in membrane manufacturing.It was initially used in producing the first asymmetric RO membrane (cellulose acetate).
Subsequently, it has remained a key method for crafting various asymmetric membranes.The method's extensive adoption can be attributed to its broad compatibility with various polymers.The popularity of phase inversion in membrane manufacturing stems from its broad compatibility with numerous polymers, as well as its cost-effectiveness and efficiency.However, it is important to highlight several drawbacks.Firstly, the selective layer of these membranes tends to be relatively thick due to the integrated structure, with the dense selective layer gradually merging into the supporting structure.Secondly, the support structure itself represents a drawback as most of the material used contributes no separation function, making it an inefficient use of resources, especially when expensive materials are involved.

Interfacial polymerization (IP)
IP, a process that confines a chemical reaction at the liquid-liquid or liquid-air interface, offers significant advantages in the controlled fabrication of films, capsules, and fibers used as electrode materials and separation membranes.As depicted in figure 2(c), the term 'TFC' membrane originally referred mainly to PA TFC membranes specifically designed for RO and NF applications.However, this definition is too narrow and limited in scope.In fact, a TFC membrane can be defined as any membrane consisting of a chemically distinct thin film selective layer supported by a mechanically supportive layer.For instance, let us consider the synthesis of a PA membrane [51][52][53][54] using m-phenylenediamine (MPD) and trimesoyl chloride (TMC) monomers respectively dissolved in two immiscible solvents, including aqueous solution and organic solvent.In this method, a microporous polysulfone substrate layer is submerged in the MPD solution.This allows MPD to infiltrate the TMC region, while the TMC remains within the organic phase due to its hydrophobicity.Consequently, the IP process takes place across the MPD/TMC interface within the organic solution.As the polymerization reaction advances, the produced PA membrane acts as a barrier, inhibiting the intermixing of MPD and TMC monomers.This phenomenon leads to a self-limiting process, where further polymerization is restricted and the growth of the PA membrane is controlled.From a manufacturing standpoint, IP has emerged as a highly effective method for producing large quantities of membranes rapidly.This technique utilizes a single chemistry platform, allowing for efficient and streamlined production processes.The benefits of this technique have not been as successfully extended to other membrane materials.For instance, TFC hollow fibers have been produced successfully in the field of hollow fiber manufacturing.Nonetheless, IP remains the prevailing method for manufacturing TFC membranes and holds the potential to facilitate the production of other condensation polymers [55].Overall, TFC membranes have found extensive application in RO and NF.

Molecular deposition layer by layer (mLBL)
Commercial RO membranes at the forefront of technology employ a TFC design, which consists of a densely crosslinked polyamide (PA) selective layer on a microporous polymeric substrate.This selective layer facilitates the separation of water molecules from solutes.The structural properties of the PA layer, including its chemistry, homogeneity, and molecular topology, play a crucial role in determining the permeability and selectivity of the TFC membrane.Currently, the PA selective layer is typically produced through the IP of MPD and TMC.As shown in figure 2(d), this process forms a highly crosslinked PA network that is fully aromatic.However, due to the rapid reaction between the diamine and TMC, the system reaches a gel point quickly, which restricts the diffusion of both monomers.Consequently, IP results in selective layers that exhibit significant depth heterogeneity, limited control over film thickness, and ridge-and-valley structures [56][57][58].These characteristics complicate our understanding of membrane properties and impede our ability to modulate the permselectivity.
The mLBL deposition is an alternative process that offers advantages over traditional IP.It involves a sequential reaction between a diamine and a TMC, resulting in the formation of dense networks, chemically robust, and highly crosslinked.Unlike IP, mLBL deposition does not encounter the same kinetic and mass transfer limitations.This technique operates at monomer scales, enabling the creation of ultra-thin selective layers.It offers precise control over morphology, local chemical composition, and thickness that were previously unattainable.This method has been developed and utilized for the preparation of PA RO membranes, leading to the production of membranes with homogeneous morphologies [59][60][61][62][63].A bilayer polyelectrolyte assembly composed of poly(acrylic acid) (PAA) and poly(ethyleneimine) is strategically applied onto a poly(acrylonitrile) (PAN) supporting layer during the manufacturing process, making effective use of electrostatic interactions.The initial deposition process enhances the connectivity between PAA and MPD monomers by leveraging the carboxylic acid groups present in PAA.These carboxylic acid groups have the ability to establish hydrogen bonds with MPD's amine groups, thereby reinforcing the overall structural integrity.The PAN support layer, decorated with the polyelectrolyte bilayer, is immersed alternately in solutions of MPD and TMC, followed by appropriate solvent washing after each dipping step.This process leads to the formation of cross-linked and multilayered PA membranes.The molecular layer-by-layer process is repeated until the desired thickness of the PA membrane is achieved.To maintain compatibility with the surface chemistry of membranes created through the IP process, the polymerization reaction is intentionally halted during the deposition step involving TMC.This suspension of the polymerization reaction ensures that the TMC deposition is in harmony with the underlying surface chemistry of the membranes.

3D printing
In comparison to the traditional methods mentioned earlier in manufacturing membranes, 3D printing offers several advantages.It enables the fabrication of TFC membranes using any processable-solution polymer, allowing for greater versatility in material selection.Additionally, 3D printing has the potential to improve resolution to the micro/nanometer level, enhancing permeability without compromising selectivity [37,64].Several additive manufacturing techniques have been employed in membrane fabrication, as discussed in previous works on membrane 3D printing techniques [65][66][67][68].These techniques include powder bed fusion-selective laser sintering (SLS) [69], binder jetting (BJT) [68], and vat photopolymerization (VP).VP utilizes a vat containing a liquid photopolymer, which is cured layer by layer through exposure to ultraviolet light.This additive manufacturing process allows for the creation of the desired part with precise control and accuracy.In contrast to traditional processes like BJT and PBF, the VP process involves the curing of photopolymer liquids.This approach results in superior surface quality and lighter products compared to powder-based printing techniques.Moreover, the VP method is versatile and can accommodate a wide range of materials, offering flexibility in material selection.Additionally, it provides different levels of resolution and accuracy, allowing for tailored printing performance.The common VP techniques contain SLA [68,70], digital light processing [68], material jetting (MJ), and inkjet printing (IJP) [71,72], continuous liquid interface production [68,[73][74][75][76], two-photon polymerization [68,77,78], material extrusion (ME) and fused deposition modeling/fused filament fabrication [68], drop on demand, continuous inkjet [79], electrospray printing (ESP) [37,39,[80][81][82][83][84][85], polyjet/multijet fusion (MJF) [86][87][88], and other techniques [89,90].Clearly, various printing methods have the potential to impact membrane performance differently.For instance, Cai et al [91] conducted a systematic benchmark and comparison involving PA 12 parts fabricated through SLS and MJF.This comprehensive study encompassed the physicochemical analysis of raw powder materials and their respective printed specimens, as well as an evaluation of the mechanical performance and printing characteristics of the produced objects.The findings of this study indicated that the SLS-printed merlion exhibited greater profile deviations compared to its MJF-printed counterpart, particularly in regions featuring sharp contours.A comprehensive exploration of these printing processes, including their advantages and disadvantages, can be found in a previous review paper [2].
In the experiment, as shown in figure 2(e), a 3D printing technique called ESP was used to deposit PA membranes [37].The process involved a rotating drum, which was grounded at one end and connected to two needles at the other end.The two needles were used to extrude two different monomer solutions.One solution contained MPD dissolved in water, and the other hexane solution contained TMC.MPD/TMC ratios could be adjusted by governing monomers' molar numbers.When the MPD and TMC solutions were splashed from the two needles, they accumulated on the collector surface, which was the rotating drum.After that, MPD and TMC monomers reacted randomly with each other to form the PA membrane.The needles moved across the collector surface to make certain that the membrane was deposited evenly across the entire substrate.The 3D printing process continued until the target thickness was achieved.Compared to other methods, the resulting membrane formed by the technique had a more uniform structure and chemistry.

Advanced 3D printing for mixed matrix membranes
Three-dimensional printing stands as a disruptive and innovative technology renowned for its exceptional precision and control in the production of macroscopic objects boasting intricate shapes and geometries [92][93][94][95].Enhancing water treatment processes necessitates precise tuning of the membrane's microstructure during the printing process.A proposed mitigation strategy [96][97][98] outlined in the literature involves incorporating nanofillers, such as carbon nanotubes (CNTs) or metal-organic frameworks (MOFs), into various versatile and moldable bulk materials.This is particularly applicable to film materials characterized by excellent flexibility and significant permeability.In a recent study, Li et al [99] utilized advanced 3D printing techniques to fabricate mixed matrix films comprising MOF and thermoplastic PA 12 powder as the matrix material.This innovative approach involved the incorporation of five different types of MOF crystals as fillers in the PA12 matrix.The findings revealed that membranes featuring a grid pattern with the smallest pore size demonstrated exceptional performance as easily collectible adsorbents.These membranes exhibited both a satisfactory maximum adsorption capacity and an impressive adsorption rate.In the printing process described, voids may emerge as a result of rapid sintering or melting of powder particles through transient laser heating.When polymer particles and MOF fillers are combined for the printing process, these voids can serve as open channels, facilitating the integration of MOF fillers into the polymer matrix.In turn, it leads to an increased contact area between the fillers and the surrounding environment.The quantity and dimensions of these voids can be conveniently regulated by fine-tuning laser sintering parameters such as laser power, scanning speed, and hatching space.
Additionally, Elsaidi et al [100] engineered mixed matrix membranes comprised of MOF-based thin films.They achieved this by integrating HKUST-1 nanoparticles into an intrinsic microporosity (PIM-1) matrix, utilizing the innovative 3D ESP technique.The reduction in membrane thickness served to enhance gas permeance without compromising the integrity of the fabricated membranes, ensuring they remained free from defects while maintaining robustness.Subsequently, the research explores the systematic impact of casting concentration and the number of electrospray cycles on both membrane thickness and the performance of CO 2 separation.The outcomes of the study revealed that employing a low concentration of polymer of PIM-1 or a PIM-1/HKUST-1 solution (0.1 wt%) resulted in the production of thin-film composite membranes with a thickness measuring less than 500 nm.However, these membranes exhibited subpar CO 2 /N 2 selectivity due to the presence of microscopic defects.To mitigate these microscale defects, the study opted to raise the concentration of the casting solution to as much as 0.5 wt% when fabricating thin-film composite mixed matrix membranes within the 2-3 µm thickness range.As a result, these thin-film composite mixed matrix membranes exhibited a remarkable threefold increase in CO 2 permeance compared to the pristine PIM-1 membrane and showcased their potential for enhanced gas separation performance.
The authors acknowledge the existence of numerous fabrication methods for membranes, which encompass a wide range of techniques such as atomic layer deposition, spray coating, chemical vapor deposition, electrospinning, dip coating, and several others.Certain methods typically refer to techniques widely employed in industrial settings for large-scale production.These methods are often well-established, cost-effective, and readily scalable when compared to some of the alternative techniques mentioned earlier.Nonetheless, this review primarily focuses on laboratory-scale printing methods, regardless of their potential for large-scale manufacturing.By exploring various manufacturing methods, the authors aim to offer insights and information that may be particularly relevant to industries and businesses engaged in membrane production.Our intention is to furnish a comprehensive understanding of state-of-the-art membrane manufacturing techniques.

Manufacturing progress
In order to determine the most suitable technique for membrane fabrication, it is necessary to establish the critical metrics to be considered, as outlined in table 1.These metrics should be evaluated from the perspective of membrane applications.However, 3D printing techniques have a significant advantage in these key metrics.A detailed description of 3D printing techniques can be found in the review paper [2].Thus far, several 3D printing techniques have been considered as potential alternatives to address the limitations associated with conventional membrane manufacturing methods.These 3D printing techniques offer the potential to overcome certain shortcomings and provide innovative solutions in membrane production.Because of the significant progress in 3D printing techniques towards higher speed, lower cost, higher accuracy, better material processability, greater scalability, better structural integrity, and improved resolution [2,66,101,102], the compatibility of 3D printing techniques with common membrane materials has led to their exploration in the formation of specifically designed membrane components.This compatibility allows for the incorporation of 3D-printed elements into the overall membrane structure, enabling the customization and optimization of membrane properties for various applications.
To provide a clearer overview of the progress made in the field of 3D-printed membranes, we present a summary of the current published works that utilize printing approaches for membrane fabrication.Multiple research efforts have focused on utilizing IJP and ESP techniques for membrane production, primarily due to their capability to create thin films and mixed matrix membranes.Additionally, PBF-SLS and ME techniques have been explored as potential methods for fabricating porous membranes.These techniques offer promising possibilities for the development of innovative membrane structures and functionalities.While it is true that most PBF-SLS, ME, and VP methods may not typically achieve the resolutions required for membrane fabrication, they have been successfully employed for other purposes in membrane production.These techniques have been utilized to create base layers, surface patterns, and membrane filaments, which can serve as essential components in the overall membrane structure.By combining these techniques with other fabrication methods, it is possible to achieve the desired membrane properties and functionalities while leveraging the strengths of each approach.Here, we mainly focus on the PA membrane manufactured by 3D printing techniques.Case 1: Badalov et al [38] conducted a study where they showcased the controllable printing of a PA selective layer using an IJP technique with an aqueous solution of MPD. Figure 3(a) depicts the process of printing and subsequent IP.Initially, the asymmetric polysulfone support was wetted with deionized water.Following that, a 2.5% w/v aqueous solution of MPD was printed onto the support, forming a layer of MPD solution.The printing step was repeated for multiple layers, enabling the creation of selective layers with different crosslinking densities and thicknesses.Subsequently, the printed membrane underwent treatment with a TMC solution, following the same step employed in IP.This treatment further enhanced the crosslinking and stability of the membrane structure.Analysis using x-ray photoelectron spectroscopy revealed that membranes with a higher number of printing layers exhibited a higher degree of crosslinking.This suggests that the increased number of printing layers promoted stronger intermolecular interactions and enhanced the overall stability of the membrane structure.Furthermore, contact angle measurements confirmed that these membranes exhibited enhanced surface hydrophobicity.Additionally, the study revealed that an increase in the number of printing layers led to membranes with reduced water permeance and improved NaCl rejection.This suggests that control over the number of printing layers can be utilized to fine-tune the performance of the membranes for specific desalination applications.The permeance and salt rejection data indicate that membranes manufactured using IP generally demonstrate higher rates of permeance and rejection compared to the majority of inkjet-printed membranes.This disparity can be attributed to the much lower resolution of IJP techniques, which hinders the precise deposition of aqueous MPD droplets during the membrane fabrication process.
Case 2: Using the ESP technique, Chowdhury et al [37] utilized a drum-based printer to synthesize a PA selective layer through the separate spraying of TMC-Hexane solutions and MPD aqueous.Figure 3(b) illustrates the electrospray setup and the resulting PA selective layer.The support layer was firmly affixed to a rotating drum, which served the purpose of collecting the monomer solutions during the electrospray process.Polymerization occurred when the spraying monomer overlapped with the deposition of the other monomer on the support, allowing for the chemical reaction to take place and facilitate the formation of the polymer.
In Chowdhury's work, a lipophilic ionic liquid was introduced into the TMC/hexane solution with a concentration of 1 µl ml −1 .This addition enabled the generation of a cone-jet spray within the voltage range of 4-7 kV.Additionally, this study demonstrated that the 3D-printed PA membranes exhibited exceptional resolution and achieved precise control over the thickness of each printing layer, with an accuracy of 4 nm.Furthermore, Chowdhury successfully applied the electrospray technique to deposit the PA selective layer on various substrates.This demonstrated that the thickness of the PA layer remained consistent regardless of the material used for the substrate, highlighting the versatility and reliability of the electrospray method.In terms of roughness control, Chowdhury achieved an impressive surface roughness of approximately 4 nm, along with a uniform thickness of 20 nm.These printed membranes exhibited superior permselectivity compared to commercially available benchmarking PA membranes, indicating their enhanced performance and suitability for various applications.Moreover, these printed membranes displayed smoother surfaces in comparison to interfacial polymerized PA membranes, which often exhibit a ridge and valley morphology.The printed PA membranes' performance was determined to be adjustable and on par with commercially available PA membranes, further emphasizing their efficacy in water desalination applications.Case 3: Ma et al [39] employed the ESP technique to modify a conventional drum-based 3D printer, transforming it into a multi-nozzle system specifically designed for synthesizing a PA selective layer.They accomplished this by separately spraying a 0.2 wt% TMC organic solution and a 2.0 wt% MPD aqueous solution.The electrospray setup and fabrication process of the PA membrane are depicted in figure 3(c).The multinozzle system comprised six glass syringes positioned alternately with an intersyringe spacing of 3.2 cm.To ensure a consistent spray onto the receiving substrate, their translational movement was controlled by a stepping motor.The PES membrane, employed to collect the electrosprayed microdroplets, was affixed to a rotating drum.Excess water on the surface of the PES membrane was eliminated using dust-free paper, while the interior of the membrane retained its wet condition.During the electrospray process, the drum rotated at a constant speed of 100 rpm, while each syringe maintained an injection rate of 1.2 ml h −1 .The resulting membranes exhibited a remarkably smooth surface, characterized by 1.2 ± 0.2 nm average roughness.This is remarkedly different from the TFC membrane synthesized through conventional IP with identical monomer concentrations.The latter exhibited a surface roughness of 58 ± 2 nm, characterized by ridges and valleys [39].The utilization of electrospray-assisted IP led to the enhanced monomer dispersion and a more stable reaction across the interface.As a result, an ultrathin PA rejection layer with was successfully created, presenting as the minimal surface roughness.By accurately regulating the electrospray time, rejection layers with thicknesses varying from sub-10 nm to a few tens of nanometers were successfully achieved.This method enables more efficient utilization of monomers, minimizing waste and promoting environmentally friendly production practices.
Case 4: Ma et al [40] conducted a separate study where they explored the integration of CNT nanofillers into electrospray PA TFC membranes.Their findings revealed substantial enhancements in membrane permeance while maintaining excellent salt rejection capabilities.In a continuation of their previous research, the authors employed a similar electrospray setup, depicted in figure 3(d), for the fabrication of PA-CNTs NF membranes.However, in this study, they utilized lower monomer solution concentrations.The incorporation of CNTs into the membranes was analyzed using transmission electron microscopy, which indicated that the addition of CNTs had no impact on the PA selective layer' growth rate.The PA-CNTs membranes demonstrated a remarkable six-fold increase in permeance while keeping an equivalent level of salt rejection.It is noteworthy that these NF membranes, which incorporated nanofillers, exhibited higher water flux compared to commercial NF270 and NF90 membranes.However, the salt rejection of the PA-CNTs membranes was comparatively lower than that of the commercial membranes.
Case 5: Yang et al [41] conducted research on the application of electrospray for the printing of PA NF membranes.The research conducted by Yang et al utilized similar printing processes and parameters for the electrospray fabrication of PA NF membranes as those employed in the electrospray process for PA RO membranes.This electrospray procedure is illustrated in figure 3(e).In their study, they utilized a piperazine (PIP) aqueous solution and TMC mixed with a hexane-acetone solution as the materials for membrane synthesis.The researchers successfully demonstrated the capability to adjust both permeance and salt rejection through the manipulation of PIP concentration and electrospray time, similar to the PA RO membranes.This study provides valuable insights into the utilization of electrospray for fabricating PA NF membranes using TMC and PIP as the key components in the membrane synthesis process.The precise control over PIP concentration and electrospray time provided the researchers with the ability to fine-tune the performance of the NF membranes based on specific requirements.This optimization process enabled them to strike a balance between achieving the desired water permeance and ensuring the necessary salt rejection capabilities.

Design guide via MD simulations
Due to the tunability provided by 3D printing, there is an increasing demand for a thorough understanding of how chemical composition and manufacturing methods can be finely adjusted.This adjustment is crucial to attain the desired membrane properties, which are characterized by both high permeability and high selectivity.It is crucial to delve into the structural aspects of membranes and identify the underlying reasons behind their performance.This understanding will enable researchers to optimize membrane design and manufacturing parameters, ensuring that the desired properties are achieved and maintained consistently.
For instance, from a manufacturing standpoint, it is essential to investigate how different manufacturing methods impact the properties of PA membranes.Despite the existence of various experimental membranes synthesized through different manufacturing techniques, it is crucial to explore how these manufacturing methods specifically impact the properties of PA membranes.By studying the relationship between manufacturing techniques and membrane properties, researchers can gain insights into the structural changes, morphology, and performance characteristics of PA membranes.This knowledge can then be utilized to optimize the manufacturing process, leading to improved PA membrane properties such as permeability and selectivity.From a chemical composition standpoint, several factors influence the properties of a PA membrane.The degree of cross-linking (DC), MPD/TMC ratio, thickness, and monomer mixing types play significant roles in determining these properties.The DC affects the membrane's mechanical strength, stability, and resistance to degradation.A higher DC generally results in improved membrane durability.Secondly, the MPD/TMC ratio, which represents the relative amounts of MPD and TMC, affects the membrane's permeability and selectivity.Adjusting this ratio can fine-tune the separation performance of the membrane for specific applications.Additionally, the thickness of the PA membrane influences its mechanical properties, such as flexibility and resistance to fouling.Thicker membranes tend to be more robust but may exhibit lower permeability.Lastly, the choice of monomer mixing types, including the selection of additives or blending with other polymers, can impact the membrane's chemical resistance, hydrophilicity, and surface charge, thereby affecting its separation capabilities.Careful consideration of these factors is crucial in tailoring the properties of PA membranes to meet specific requirements in various applications.
Besides, the experimental capture of all internal structures and water and solutes' dynamics behavior has proven to be challenging, leading to a lack of comprehensive understanding regarding optimized membrane transport [103].MD simulations, on the other hand, offers a powerful approach to investigate how membranes' microstructures and synthesis conditions work on transport properties.Through MD simulations, it becomes possible to gain mechanistic insights into membrane transport that would be otherwise unattainable through experiments alone.Studies utilizing MD simulations have demonstrated their effectiveness in elucidating the intricate details of membrane behavior and providing valuable information on transport mechanisms [28,104].Thus, MD simulations serve as an alternative pathway for achieving a comprehensive mechanistic understanding of membrane transport.Importantly, MD simulations have emerged as a widely adopted tool for studying water transport across various membranes utilized in water treatment applications.This includes membranes such as MoS 2 or MoSe 2 nanoporous membranes [105,106], FT-30 RO membranes [107], PA RO membranes [108][109][110][111][112][113][114][115][116][117], nanoporous graphene membranes [118,119], MOF membranes [120], polymerized fullerite membranes [121], CNTs/PA RO membranes [122], and graphitic carbon nitride membranes [123].By employing MD simulations, researchers have been able to explore the intricate dynamics and transport phenomena associated with these membranes, providing valuable insights into their performance and guiding the design of advanced membrane materials for water treatment purposes.
Indeed, MD simulations can play an important role in guiding and designing the synthesis of experimental membranes before conducting actual measurements.With the help of MD simulations, researchers can delve into and forecast the membrane's molecular-level transport behavior.This approach yields valuable insights into synthesis parameters and conditions that can result in desired properties and optimized transport performance.Besides, this computational approach enables the identification of potential membrane materials, optimization of pore size, evaluation of surface modifications, and examination of other factors that may influence the membrane's performance.By harnessing the predictive capabilities of MD simulations, researchers can make informed decisions and streamline the experimental membrane synthesis process.This approach ultimately increases the likelihood of achieving the desired membrane properties and facilitates more efficient water treatment applications.

Amine monomer types
The production of highly cross-linked PA membranes often involves the use of two reactive monomers: TMC and amine monomers.It is essential to understand how amine monomers affect the properties of PA membranes and to optimize their composition in order to achieve optimal membrane performance.However, conducting experiments to test numerous amine monomers can be challenging.To overcome this challenge, MD/DFT simulations provide a valuable approach for calculating energy barriers and investigating the diffusion of amine monomers in organic solutions at the molecular level.These simulations yield valuable insights into the behavior of various amine monomers, facilitating the selection of the most suitable candidates for further experimental investigation.
By calculating the energy barriers and diffusion rates of amine monomers in organic solutions, valuable insights can be gained regarding their potential impact on the properties of highly cross-linked PA membranes [124][125][126].MD/DFT simulations offer a means to model and simulate the behavior of atoms and molecules based on their interactions and motion.By conducting simulations with various amine monomers, their diffusion within the organic solution and associated energy barriers can be observed [127,128].As shown in figure 4, these measurements and simulations provide information on the mobility and interactions of amine monomers, which are critical factors in membrane performance.
Understanding the energy barriers and diffusion rates of different amine monomers enables the identification of those more likely to interact favorably with the reactive monomers, such as TMC, during membrane manufacturing.This knowledge aids in selecting the most suitable amine monomers to optimize membrane performance.Moreover, MD/DFT simulations offer insights into other membrane properties, including structural stability, permeability, and selectivity.By studying these properties, the selection of amine monomers can be further refined to achieve desired membrane characteristics.It is important to note that while MD/DFT simulations provide valuable predictions and insights, experimental validation remains crucial.The simulations guide experimental design and offer a preliminary understanding, but actual testing and characterization of the membranes are necessary to confirm predictions and assess practical performance.In short, utilizing MD simulations to calculate energy barriers and amine monomer diffusion in organic solutions presents a promising approach for evaluating and optimizing the properties of highly cross-linked PA membranes.It enables more efficient exploration of different amine monomer candidates, ultimately leading to the development of membranes with improved performance.

Mole ratio of two reactive monomers
In highly cross-linked PA membranes, the cross-linking reaction takes place between two monomers, depending on the availability of reaction sites within the monomers.For instance, a ratio of 4:1 of MPD to TMC in experimental protocols is used to form the PA layer [37,[129][130][131].Therefore, it is important to examine the impact of having an excess of either MPD or TMC.In terms of structure, as shown in figure 5(a), our previous studies have employed MD simulations to investigate four MPD/TMC ratios: 1:1, 1:4, 3:2, and 4:1 [85].However, the polymerization reaction' stoichiometry actually requires an ideal ratio of 3:2.Therefore, it is important to examine the impact of having an excess of either MPD or TMC.In terms of structure, previous studies have employed MD simulations to investigate four different MPD/TMC ratios: 1:1, 1:4, 3:2, and 4:1 [132].These studies have provided insights into the impact of varying MPD/TMC ratios on several membrane properties.As displayed in figure 5(b), one crucial finding is that the distribution of pore sizes was found to be affected by the MPD/TMC ratios.Moreover, the permeability of water molecules, water flux, rejection, and membrane compaction were observed to be different for each ratio shown in figures 5(c) and (d).Of particular significance, observed from figure 5(e), the MPD/TMC ratios can also affect the yield strength of the membranes, which is essential in high-pressure RO processes.Importantly, Zhang et al [133] also identified variations in the PSD of dry PA membranes with different monomer ratios.These variations led to distinct water cluster distributions and water flux and influenced the wetting properties of hydrated PA films.Therefore, understanding the influence of MPD/TMC ratios on membrane properties and performance is crucial for optimizing the design and functionality of PA membranes used in various applications.By studying the impact of different MPD/TMC ratios on these various aspects, researchers gain insights into the structure-function relationships of highly cross-linked PA membranes.This knowledge allows for the optimization of membrane synthesis protocols and the design of membranes with tailored properties for specific applications, including water treatment, desalination, and gas separation.

Degree of cross-linking (DC)
For cross-linked PA membranes, controlling the DC during the polymerization process is crucial for optimizing membrane performance in the manufacturing stage.As shown in figure 6(a), previous studies [132] have also explored the impact of the DC on various aspects of the membrane, such as atomic composition, microstructure, density, PSD, water dynamics behavior, the diffused water molecules, salt rejection, water flux, and compaction.The results indicated that increasing the DC from 40.07% to 96.26% led to an increase in the composition of carbon (C), oxygen (O), and nitrogen (N) [132].However, the carboxyl groups (-COOH) composition decreased.Moreover, PA membranes with different DCs exhibited significant differences in PSD curves (figure 6(b)), with pore sizes ranging from 2.5 to 20 Å.As the DC increased, the PSD became more homogeneous throughout the membrane due to the cross-linking reaction occurring uniformly.Notably, the PSD curves shifted towards the left shift with increasing DC.When the DC increased from 40% to 96%, the peaks shifted significantly from 0.14 to 0.32 in the PSD curves, and the corresponding pore sizes decreased noticeably from 4.625 to 3.625 Å.This indicated a reduction in the network pores' size, although they still remained larger than the water molecule's diameter.Moreover, as the DC increased, as shown in figure 6(c), the number of water molecules and flux decreased under the same pressure.In addition, figure 6(d) indicated that when the DC reached 60% or higher, the membrane completely rejected salt ions within the timescale of our MD simulations.Conversely, under high pressure, some ions could travel through the PA membrane, and the rejection decreased as pressure increased.Likewise, the different DC can also influence the yield strength of the membranes, as exhibited in figure 6(e), which is essential in high-pressure RO processes.Additionally, Zhang et al [133] conducted research indicating that membrane swelling exhibited a linear dependency on the DC, and a greater propensity for swelling in the PA film correlated with higher water flux.Overall, controlling the DC in the manufacturing of cross-linked PA membranes is crucial for optimizing their properties, such as atomic composition, microstructure, PSD, and water and salt permeation characteristics.These findings can help guide the design and development of PA membranes for various applications, such as water purification and desalination processes.

Thickness
Controlling thickness is a key strategy to regulate the PA RO membranes' permeance and selectivity.Numerous reports based on simulations and experimental measurements [37, 59, 61, 63, 64, 107-113, 115, 116, 122, 129, 130, 133-150] have consistently indicated that PA membranes with varying thicknesses exhibit different membrane performance.Importantly, as shown in figure 7(a), our previous studies [151] have highlighted the significant influence of thickness on membrane density, microstructures, flux, diffusion coefficient, permeance, and permeability.Structurally, membranes with varying thicknesses exhibit distinct patterns in their PSD curves (figure 7(c)), indicating structural differences.Firstly, across membranes with different thicknesses, the pore diameter corresponding to the peak in their PSDs remains consistent at approximately 3.125 Å.Secondly, an increase in membrane thickness leads to a higher fractional content of cross-linked network pores, resulting in more homogenous PSDs.Thirdly, the swelling ratio (figure 7(b)) exhibits a nonlinear decrease with increasing thickness, while water diffusivity, water flux, and water permeance (figure 7(d)) for pure water and figure 7(e) for brine water) also follow a nonlinear decreasing trend.Importantly, the research mentioned above has identified a critical threshold thickness of 15 nm for the PA layer.Once the threshold is surpassed, the membrane microstructures remain unchanged, indicating that further increases in thickness do not substantially alter the membrane's properties.In summary, precise control over the thickness of PA RO membranes plays a vital role in regulating their performance.Membrane thickness affects various factors, including membrane density, swelling ratio, PSD, diffusivity, water flux, and water permeance.Understanding these interrelationships can facilitate the design of membranes with desired properties tailored for specific applications in water treatment and desalination processes.

Manufacturing methods
Various approaches have been employed for the manufacturing of PA RO membranes, as aforementioned.However, there has been a noticeable lack of systematic studies investigating the impact of various manufacturing methods on the properties of PA membranes, including water permeability, structural features, rejection, porosity, and transport mechanisms.Conducting experimental work on different manufacturing methods while maintaining consistent results has proven to be a challenging task.Recently, our research has utilized MD simulations [152] to investigate the impact of three manufacturing methods: 3D printing, mLBL, and IP on membrane performance and microstructures, which is exhibited in figure 8(a).The results have demonstrated that PA membranes produced through different manufacturing methods exhibit distinct atomic compositions, stacking modes of benzene rings, water flux, salt rejection, and compactness due to variations in their DC and distribution of cross-linking sites.Specifically, as shown in figure 8(b), PA membranes formed with IP display a more inhomogeneous PSD due to limited cross-linking within MPD/TMC interfacial layers.On the other hand, PA membranes generated using mLBL deposition or 3D printing techniques offer the advantage of a more homogeneous PSD.This is attributed to the dispersed distribution of cross-linking reaction sites throughout the membrane structure.Additionally, compared to mLBL and 3D printing, PA membranes produced through IP contain more pores with larger diameters.Overall, these findings highlight the importance of considering the manufacturing method when designing and optimizing PA RO membranes for water desalination [153].As for performance, as observed in figure 8(c) (pure water) and figure 8(d) (brine water), and figure 8(e), different manufacturing methods can affect water flux and salt rejection.By gaining a molecular-level understanding of how different manufacturing techniques influence membrane properties, we can make informed decisions to improve the performance and efficiency of water desalination processes.Further research and experimental validation will be necessary to confirm and expand upon the insights obtained from MD simulations, leading to advancements in membrane manufacturing and water desalination technologies.

Current challenges and future perspective
Membrane technologies are now recognized as highly effective solutions for tackling some of the most urgent global challenges and facilitating the development of innovative industrial processes crucial for sustainable industry growth [154].However, the challenge of accurately predicting the intricate relationship between membrane structure, process parameters, and desired properties has hindered its widespread application.This conundrum has impeded the effective design of new membranes with specific properties, thus limiting their potential impact.Such a challenge may be related to three key aspects.Firstly, the high dimensionality of membrane design features, encompassing both intrinsic information and extrinsic conditions, plays a crucial role.The intrinsic information includes factors such as the types of reactive monomers and chemical structure [155], surface area [156], and pore size [157].The extrinsic conditions involve parameters like reaction temperature, concentration [158], and pH [159].Secondly, the vast design space of potential materials poses a significant obstacle, making it difficult to thoroughly screen through all the options available.Finally, the last one is compounded by a lack of understanding regarding the complex membrane systems' chemistry and their overall physics.This limited understanding adds to the difficulties faced in overcoming the challenges associated with membrane design and development.
In addition to experimental membrane properties, theoretical methods such as DFT calculations [160] and MD simulations [161] provide the ability to predict membrane performance without the need for actual membrane synthesis.These computational simulation methods offer a comparatively low-cost approach to obtaining answers.However, it is important to note that the computational expense increases as the complexity of systems grows.This limitation significantly restricts simulations in terms of both time and small length scales.Therefore, while computational methods offer valuable insights, there are practical limitations when it comes to simulating highly complex systems with fine-grained details within limited time and spatial constraints.The diversity of membrane backbone materials, along with the range of available additives and the complex fabrication process, necessitates the rational design of 3D-printed membranes.Therefore, developing efficient methods to overcome the resource and time constraints associated with the iterative trial-and-error approach becomes crucial.
Machine learning (ML) has become an invaluable tool within the material science and chemistry communities.Its remarkable capability to analyze and glean insights from intricate, voluminous, and multidimensional datasets has led to significant advancements in various facets of chemical separations.ML's ability to process information efficiently has accelerated progress and unlocked new possibilities in the discovery of novel membrane materials and optimizing chemical synthesis processes.By harnessing ML techniques, researchers can more efficiently explore and identify promising membrane materials, reducing the need for extensive experimental iterations and streamlining the overall design process [162][163][164][165][166].
Figure 8. MD simulations were employed to investigate the impact of various manufacturing methods on membrane metrics, reprinted from [152], Copyright (2023), with permission from Elsevier.(a) Visualization of atomistic membrane models for PA was generated using three different manufacturing methods: IP, mLBL, and 3D printing.These models were created at different degrees of cross-linking, specifically at 78.4%, 84.6%, and 90.1% cross-linking degrees, respectively.(b) Microstructures for PA membranes produced with different fabrication methods described by PSDs; (c) and (d) relationship between water flux and pressure in pure water and brine water, respectively; (e) the rejection of PA membranes, formed using three different methods, is analyzed as a function of applied pressures.
Recently, ML has found application in various areas, including the design of gas separation membranes [167][168][169][170], the prediction of water permeance and salt rejection for RO and NF membranes used in water treatment [171][172][173][174][175], and assisting in the fabrication of UF membranes [176,177].However, the effectiveness of these models has often been hindered by unclear classification and incomplete inputs.Furthermore, previous studies have primarily focused on predicting water flux and removal efficiency.
The current research on ML in water treatment, specifically at the membrane level, has primarily focused on identifying optimal preparation conditions to enhance membrane structures and separation performance.Additionally, a few studies have delved into ML techniques for optimizing both membrane and process levels in a comprehensive, multiscale approach.Despite these recent developments, there remain several challenges that must be addressed to further propel the utilization of ML in the optimization of polymer membranes for water treatment.Factors such as temperature, pressure, and monomer compositions and types significantly influence membrane morphologies during the preparation stage, ultimately impacting their separation performance.ML studies have the potential to unravel and quantify the intricate and often elusive relationships between membrane morphologies and membrane fabrication conditions, which are not easily measurable through conventional means.Another significant challenge in water treatment is membrane fouling [28].When foulants adhere to an uneven membrane surface, it can have adverse effects such as concentration polarization and reduced water permeation.Considering the wide range of inorganic and organic foulants, future ML research should delve into the comprehensive exploration of foulant-membrane interactions, and their impact on separation performance.
Molecular simulation has become a widely recognized and established computational approach that enables the exploration of structure-properties relationship from a bottom-up perspective in membrane field.This approach proves valuable in facilitating the design and discovery of novel membranes for high efficiency water treatment [28].While fundamental studies employing molecular simulation provide valuable microscopic insights, they are often constrained to specific membrane types and come with high costs.In addition, the current landscape of ML modeling approaches specifically aimed at designing new membranes for water purification is relatively limited.In addition to process optimization and preparation, the careful selection of membrane materials holds significant importance in determining the overall separation performance.Consequently, there is a pressing demand for ML studies that leverage molecular representation to establish direct correlations between separation performance and membrane structures.Such studies would enable the rational and expedited design of innovative membranes capable of providing high-performance solutions for water treatment.Furthermore, recent studies, as depicted in figure 9, have emphasized the need for future advancements in ML for water treatment.Specifically, there is a growing recognition that the development and optimization of membrane synthesis and process design should embrace multiscale modeling for a more comprehensive approach.This is crucial because the separation performance is heavily influenced by a multitude of interconnected properties and processes.These encompass the chemical monomer structures, synthesis techniques, fabrication methods, intricate interactions among membrane-solute-solvent, as well as the operational parameters governing the 3D-printed membrane process.The incorporation of these factors represents just a fraction of the intricacies involved in this field.Currently, the comprehensive consideration of all potential factors within a single study presents technological challenges, requires a significant time investment, and demands substantial resources.To address these obstacles, several directions are proposed for future exploration of ML in the context of 3D-printed membrane separation.
(1) Separation genome.In ML studies, databases play a vital role as the most important resource.However, when it comes to the complete reporting of membrane performance and properties, technological challenges arise due to the extensive diversity of experimental fabrication methods, types of polymers used, and variations in separation conditions for membranes.Compounding this issue are the inconsistent fabrication methods in the report, additives, characterization methods, and solvents/nonsolvents conditions, resulting in scattered data throughout the literature.The limited availability and inconsistent data could potentially impede ML efforts in performance and separation predictions.Recently, the Open Membrane Database (OMD), available at www.openmembranedatabase.org, is a continuously expanding repository of more than 600 water purification and desalination membranes.These membranes are carefully collected from reputable sources, including peer-reviewed journals, patents, and commercial product data [180].The user-sourced, open-access database serves as a valuable resource for benchmarking novel RO membranes against the current state of the art.It enables researchers to conduct meta-analyses and establish structure-synthesis-performance relationships, which are vital for pushing the boundaries of membrane development.These functionalities are crucial in advancing the field and driving innovation in membrane technology.However, the OMD specifically focuses on gathering and providing data related to RO membranes.Therefore, it is crucial to develop comprehensive databases that encompass membranes' properties obtained from experiments and simulations, which we refer to as membrane separation genome.To address the challenges related to the scarce and inconsistent data, it is highly recommended to curate standardized and extensive experimental databases.These databases should systematically compile information on various membrane types, separation conditions, fabrication methods, and performance metrics.
Indeed, the inclusion of molecular membrane models and appropriate equilibration schemes within these databases is of utmost importance.By incorporating such information, these resources would significantly enhance computational reproducibility and facilitate investigations on membranes for separation purposes.Researchers would be able to leverage these computational databases to conduct comprehensive ML analyses and high-throughput screening, ultimately resulting in the discovery of promising 3D-printed membranes.The integration of molecular models and equilibration schemes within these databases would facilitate more accurate simulations and predictions, enabling researchers to explore a wide range of membrane design possibilities efficiently.By establishing such comprehensive repositories in both measurements and simulations, researchers can access a wealth of reliable and consistent data for their ML studies in membrane separation.This approach would not only enhance the efficiency and accuracy of future ML efforts but also foster collaboration and knowledge sharing among researchers in the field.
(2) Molecular representations.Once the membrane databases have been curated, the next step involves converting the inputs into digitized formats that are well-suited for ML purposes.This process involves the development and utilization of different representations at both the membrane and process levels.It is essential to initially characterize the nature of the membrane at the structural level.However, common membrane properties are relevant to its separation function and would serve as desirable descriptors, which may not be easy to measure or access through experimentation.Alternatively, molecular representation provides a broader and more comprehensive approach to capturing the chemical structures of membranes.To obtain numerical representations of these structures, their chemical compositions need to be digitally encoded using different molecular representation approaches, including molecular graphs [181] and SMILES [182].These approaches facilitate the transformation of chemical structures into digital formats that can be utilized in ML algorithms for further analysis and prediction in membrane separation studies.Further advancement in ML methods and molecular descriptors is necessary to quantify the distinct structural architectures of polymer membranes used in separation.Additionally, recently developed numerical representations of polymers, designed for diverse applications, hold potential relevance in describing polymer membranes [183][184][185].These innovative representations offer opportunities to capture the unique structural characteristics of polymer membranes used in liquid separation processes.By leveraging these novel numerical representations, researchers can enhance their understanding of polymer membrane architectures and further advance the field of membrane science and technology.In the context of membrane fabrication, the synthesis procedures are commonly documented in the form of text-based experimental sections.These sections outline the fabrication methods employed, characterization approaches utilized, and other additives related to the process.Such textual descriptions serve as a means to convey the step-by-step procedures and relevant details of the membrane fabrication process.
To utilize ML methods effectively, it is necessary to digitize the text-based information into a computer-readable format.Depending on specific performance objectives, an effective approach would be to compile a list of readily uncorrelated and available features that contain relevant information.As the ML model is constructed, it is possible to identify essential input parameters that significantly impact the target performance.In general, holistic representations should possess two key qualities: they should be interpretable, allowing for a clear understanding of the underlying physical and chemical aspects, while also enabling accurate ML predictions.The physical and chemical interpretability allows researchers to gain insights into the underlying mechanisms and behavior of the membrane systems.It enables the identification of key features and factors that contribute to the performance of the membranes.Additionally, these holistic representations should aid in uncovering fundamental insights into membrane fabrication and processing, providing valuable knowledge in those areas.This dual purpose of interpretability and predictive accuracy ensures that the developed ML models contribute to a deeper understanding of the underlying principles behind membrane fabrication and processing.
(3) ML algorithms.After curating membrane databases and performing inputs' featurization, the application of enhanced ML algorithms becomes crucial in establishing quantitative relations that effectively map these inputs to performance.Fortunately, there are currently several state-of-the-art ML algorithms available that are both versatile and accurate, making them suitable for ML applications in membrane separation.However, it is important to note that the current application of ML is primarily limited to aqueous separation and predominantly focused on the process level.To further advance the field, it is necessary to develop a wider range of network architectures and ML approaches that can effectively capture various membrane properties and improve the prediction of membrane performance.By expanding the scope of ML algorithms and exploring different network architectures, researchers can unlock new possibilities for optimizing membrane properties and enhancing separation processes across different applications.
Therefore, it is highly recommended that future researchers make a concerted effort to publicly share their database, code, and ML algorithm resources to the greatest extent possible.This promotes transparency and allows for reproducibility in scientific endeavors.Additionally, to enhance the volume of data available for ML, researchers may consider implementing transfer learning techniques that leverage pre-trained models.Exploring data augmentation methods by generating synthetic data from existing datasets can further augment the available training data.Lastly, after training the models, it is crucial to optimize them for achieving optimal operating conditions and maximizing separation performance.This can involve fine-tuning various parameters and configurations.Moreover, to validate the predictions made by the ML models, additional experiments should be conducted, thus ensuring the reliability and accuracy of the predictions.
(4) ML-improved high-throughput method.Until now, high-throughput screening techniques have remained largely untapped in membrane research.Only a limited number of studies have utilized high-throughput approaches at the membrane level, with one notable example being the exploration of different membranes.In order to fully harness the benefits of ML-enhanced high-throughput strategies, prioritizing the systematic and comprehensive fabrication and synthesis of membranes that encompass a wide range of diverse fabrication conditions and chemical diversity is crucial.This approach will enable the generation of extensive data sets that can be effectively utilized by ML algorithms to optimize membrane performance.Following that, the development of physically meaningful ML models becomes instrumental in exploring the diverse design space that encompasses fabrication techniques, membrane materials, and operating conditions, with the aim of achieving high-performance separation.These models will facilitate the identification of optimal combinations of parameters and guide the design of innovative membranes.By incorporating relevant physical principles into ML models, it becomes possible to extract valuable insights and make predictions that guide the optimization of membrane design and operational parameters.This approach facilitates the discovery of innovative and efficient separation processes.
In this context, Bayesian optimization emerges as a robust ML method capable of leveraging knowledge from previous measurements to guide the next optimized synthesis of membranes.By utilizing Bayesian optimization, valuable insights and trends can be extracted from past experimental data, enabling informed decisions on the most promising synthesis strategies for different materials.This iterative process leads to more efficient and effective material synthesis, ultimately accelerating advancements in the field.Furthermore, reinforcement learning, a methodology in which decisions are made on the basis of actions and rewards within a defined environment, can be adopted for materials synthesis planning.By framing the synthesis process as a sequential decision-making problem, reinforcement learning algorithms can enhance the selection of fabrication parameters and optimize the synthesis pathway.This optimization aims to achieve the desired membrane properties efficiently.This approach holds promise for accelerating the development of high-performance membranes through intelligent decision-making and exploration of the synthesis space.By utilizing reinforcement learning, decisions regarding fabrication and operating conditions can be made in a manner that optimizes membrane materials.This approach allows for an iterative process where the system learns from the outcomes of different actions, ultimately leading to improved decision-making and the development of highly efficient membrane materials.Overall, the ML-enhanced high-throughput approach has the potential to not only expedite membrane development by identifying the optimal synthesis planning but also significantly reduce the high costs associated with trial-and-error experiments.By leveraging ML algorithms and high-throughput screening techniques, researchers can efficiently explore the vast design space in fabrication techniques, novel membrane materials, and operating conditions.This systematic and data-driven approach minimizes the need for time-consuming and expensive trial-and-error experiments, leading to accelerated advancements in membrane development and improved cost-effectiveness in the overall research process.
These suggested directions outlined above present new prospects and challenges for the comprehensive application of ML in membrane separation.These advancements offer unique opportunities for the development of fundamental guidelines in the rational design of membranes at the molecular level.By leveraging ML techniques, researchers can effectively analyze and understand the intricate relationships between membrane structure and performance.Simultaneously, ML can be applied at the process level to aid in the design of membrane modules and optimize their overall efficiency [186].Through the integration of ML algorithms, data-driven insights can be extracted to enhance the performance, reliability, and cost-effectiveness of membrane separation systems.In essence, ML acts as a crucial bridge connecting the membrane and process levels, enabling a more seamless and practical development and commercialization of innovative high-performance membranes.ML techniques facilitate the efficient integration of membrane design principles at the molecular level with process optimization and module design at the system level [187].This holistic approach not only enhances the overall performance and reliability of membrane separation systems but also expedites the translation of research advancements into viable industrial applications.By leveraging ML, researchers and engineers can accelerate the discovery and development of novel membrane materials, leading to improved separation efficiencies and increased feasibility for commercial deployment.

Conclusions
This review article provides an overview of the past achievements and future prospects for 3D-printed PA membranes.The integration of 3D printing with membrane science, along with MD/DFT simulations, presents a compelling opportunity to revolutionize membrane manufacturing by enabling the design and production of membranes using a wide range of materials on a large scale.Presently, many studies in membrane science focus solely on material development, neglecting the crucial aspect of manufacturing innovations necessary to implement these materials at practical scales for their intended applications.Conversely, high-performance membranes achieved through experimental methods often rely on a trial-and-error approach.The high cost associated with experimental measurements makes it challenging to thoroughly test all potential membranes with superior performance.Therefore, the combination of 3D printing, ML, and MD/DFT simulations holds the potential to mature 3D printing as a widespread technique for membrane manufacturing, particularly due to its cost-effectiveness.Additionally, the amalgamation of 3D printing with ML and MD/DFT techniques can offer significant value by facilitating the use of emerging materials in the production of high-performance membranes.The inherent flexibility of 3D printing, combined with ML and MD/DFT techniques, offers a unique opportunity to integrate innovative materials into membrane manufacturing.This approach has the potential to broaden the range of commercial membrane offerings across various separation disciplines.By leveraging 3D printing's ability to create complex geometries and structures, along with the power of ML and MD/DFT techniques for material design and optimization, new and advanced materials can be seamlessly incorporated into membrane fabrication processes.This integration may enable the development of membranes with enhanced performance, improved selectivity, and increased durability, thereby addressing the specific needs of diverse separation applications.Consequently, the utilization of 3D printing, ML, and MD/DFT techniques in membrane manufacturing holds great promise for expanding the commercial availability of membranes and driving innovation in separation technologies.

Figure 3 .
Figure 3.Typical examples for 3D printed PA TFC membranes: (a) the MPD-impregnated support was subjected to IP after IJP was used to create different surface patterns with various fluorinated amine compounds.The lower pictures depict the computer-generated patterns that were employed for 3D printing, reprinted from [38], Copyright (2015), with permission from Elsevier; (b) ESP enables the fabrication of a free-standing and controllable selective layer of PA with adjustable thickness, from [37].reprinted with permission from AAAS; (c) ESP printed smooth PA selective layers from 4 nm to several tens of nanometers in thickness, reprinted with permission from [39].Copyright (2018) American Chemical Society; (d) ESP was utilized to print mixed matrix membranes consisting of PA-COOH functionalized carbon nanotubes (CNTs), reprinted from [40], Copyright (2019), with permission from Elsevier; (e) PA NF selective layer was printed using ESP on top of a span 80 interlayer.Reprinted from [41], Copyright (2020), with permission from Elsevier.

Figure 4 .
Figure 4. Examples from experiments and simulations highlight the impact of amine monomer types on membrane performance.(a) Visualization of interfacial polymerization between TMC and MPD and piperidine (PPR), and their corresponding water permeance and salt rejection, reproduced from [127].CC BY 4.0; (b) three membranes, namely PA, polyesteramide (PEA), and polyester (PE), were fabricated using different amine monomers, and their corresponding water permeance and salt rejection were assessed.Reprinted from [128], Copyright (2022), with permission from Elsevier.

Figure 5 .
Figure 5. MD simulations were employed to investigate the impact of the MPD/TMC ratio on membrane metrics, reprinted from [132], Copyright (2022), with permission from Elsevier.(a) Visualization for PA membranes with different MPD/TMC ratios, such as 1:1, 1:4, 3:2, and 4:1; (b) microstructure described by pore size distribution (PSD); (c) the water flux was analyzed as a function of pressure to examine its relationship and behavior; (d) salt rejection versus pressures; (e) relationship between yield strength and PA membranes with four MPD/TMC ratios.

Figure 6 .
Figure 6.MD simulations investigated the effect of the DC on membrane metrics, reprinted from [132], Copyright (2022), with permission from Elsevier.(a) Visualization for PA membranes with different DC, such as 40%, 50%, 60%, 70%, 80%, 90%, and 96%; (b) microstructures for PA membranes with different DC described by PSDs; (c) relationship between water flux and pressure; (d) salt rejection as a function of pressures; (e) relationship between yield strength and PA membranes having a different DC.

Figure 7 .
Figure 7. MD simulations investigated the effect of thickness on membrane metrics, reprinted from [151], Copyright (2023), with permission from Elsevier.(a) Visualizations were created to depict PA membranes with varying thicknesses, ranging from 4.0 nm to 32.5 nm; (b) swelling ratio as a function of membrane thicknesses; (c) microstructures for PA membranes with various thicknesses described by PSDs; (c) and (d) relationship between permeance/permeability and pressure in pure water and brine water, respectively.

Figure 9 .
Figure 9. ML is applied in water purification for both process and membrane levels.(a) Membrane synthesis and process for simultaneous rational design, reprinted from [178], Copyright (2020), with permission from Elsevier; (b) a schematic of hybrid mechanistic/data-driven model employed for multiscale optimization.Reprinted from [179], Copyright (2020), with permission from Elsevier.