From materials to devices using fused deposition modeling: A state-of-art review

Fused deposition modeling (FDM) uses computer-aided design to direct a 3D printer to build successful layers of product from polymeric materials to generate 3D devices. Many reviews have been reported recently on the cutting-edge FDM technology from different perspectives. However, few studies have delved into the advances in FDM technology from materials to 3D devices. Therefore, in this work, with a bottom-up approach from materials (including commodities and nanomaterials) to printing process (including effort for fast printing, effort for resolution improvement, and simulations) and from printing process to 3D devices (including biomedical implants, topological structures, and multifunctional devices), it aims at reviewing the FDM technology developed over the past decades.


Introduction
Additive manufacturing, which is distinguished from subtracting manufacturing, uses computer-aided design to direct a 3D printer to build successful layers of product from powder, molten plastic, or other materials to generate 3D devices. As summarized in some excellent reports, there are numerous advantages of applying the state-of-the-art additive manufacturing in science and engineering research. After reviewing the current and emerging approaches in hydrogel design and bioink reinforcement techniques, Gaharwar and coworkers [1] confidently concluded that additive manufacturing is one of the successful methods for developing next-generation cellularized 3D scaffolds to mimic anatomical size, tissue architecture, and tissue-specific functions. Plocher and Panesar [2] pointed out that, in conjunction with mathematically based structural optimization approaches, the additive manufacturing provides a portfolio of solutions for next-generation lightweight structures. Culmone and coworkers [3] summarized various 3D printing methods in the field of medical devices. After examining the reasons behind choosing additive manufacturing technology to produce instruments for diagnostics and surgery, they concluded that additive manufacturing opens a door to a new approach in the production of medical devices, allowing the complexity of designs to be pushed to the extreme. Some other works have been carried out with interest in summarizing the applications of additive manufacturing in construction [4], aerospace [5], and polymer-fiber composites [6], just to name a few. Kim and coworkers [7] reviewed advanced quality control technology in the additive manufacturing process. To conclude, all sorts of products can be produced with a 3D printing, from medical implants to auto and aircraft parts, batteries, energy harvest devices, and commercial accessories for mobile phones. The printing process requires as little as 10% of the raw material expended in traditional manufacturing (i.e., subtracting manufacturing) and uses less energy than conventional factory production. What's more, a 3D printer can print multiple copies just like a photocopy machine and run-off three-dimensional products with precise dimensions. Therefore, the development of 3D printing techniques would enable each individual potentially be their own manufacturer as well as their own internet site and power company. Because of those extraordinary benefits, the 3D printing technology has been hailed as the new infrastructure for a Third Industrial Revolution.
Among those developed additive manufacturing methods and practical applications, fused deposition modeling (FDM) has been well-adopted in multiple applications as a 3D printing technology in prototype or endproduct [8][9][10]. During FDM printing process, polymeric material is sent to the nozzle through a wire feeding device, after which a heater converts the material from solid to molten state, and then the molten material is extruded by a nozzle onto a base plate. The nozzle moves horizontally on X-Y plane. After the completion of layer deposition, the base plate moves in a vertical direction. It descends a thickness of one layer in a predetermined increment. Then it continues the fused deposition until the entire 3D shape is completed. It is worth noting that, when printing, each layer is printed with a contour first and then with an internal filling, for the sake of saving printing time. The nozzle moving speed, temperatures, and layer height setting of the first layer at the start of printing are different from those of the other layers. The difference in the settings is highly recommended. This is because it can enhance the adhesion between the sample and base plate, further to prevent warpage.
FDM has become a more mature 3D printing method. Many reviews have been reported recently on the cuttingedge FDM technology from different perspectives. However, few studies have delved into the advances in FDM technology from materials to 3D devices. In this review work, it aims at reviewing the cutting-edge FDM technology developed over the past decades, with a bottomup approach from materials (including commodities and nanomaterials) to printing process (including effort for fast printing, effort for resolution improvement, and simulations) and from printing process to 3D devices (including biomedical implants, topological structures, and multifunctional devices), explicitly demonstrated in Figure 1.

Materials
Thanks to the improvement advances in FDM printing technologies, especially the Arburg Plastic Freeforming [11], the choice of polymeric materials has increased considerably to a wide range for FDM printer. These materials have to meet at least two criteria: (1) relatively low melting temperature (T m ) and (2) appropriately high glass transition temperature (T g ). This is because low T m makes the 3D printing more affordable; while with a high T g the polymer could solidify fast enough after the extrusion so that the shape can be held as printed [12][13][14]. It is worthy to mention that T g is a property of the amorphous materials or the amorphous portion of the semicrystalline [15]. When the ambient temperature is below T g , the molecular chains of amorphous materials or the amorphous phases are frozen in place, behaving like solid glass. Figure 1 shows plenty of polymeric materials suitable for FDM technology from commodity to engineering and from engineering to high performance polymers, such as PCL for topological self-interlocked structures [16][17][18], UHMWPE for artificial implant [19], nylon for lightweight structure based on polymeric alloys [20], PVA for medicine device [21], PU for energy absorber [22], elastomer Tangoblack for self-powered triboelectric touch sensor [23], PC for food and/or drug packaging and medical devices [24], PEEK for biomedical implants (such as bulk implants, orthopedic implants, prosthesis systems, just to name a few) [25][26][27][28][29][30], so on and so forth. It is worth noting that the listed materials in Figure 1 are all pure polymers, which provide great potentials for FDM endproducts. However, some of them are facing challenges when used in FDM process. To overcome those challenges, different types of fillers have been proposed to the rapidly developed market, such as particles, fibers, and nanomaterials [31].

Particle-reinforced polymeric materials for FDM
When applied to FDM printing process, challenges faced by some of the polymeric materials include: (1) the printing wire is easy to break and thus the extrusion is not smooth; (2) the molded FDM parts are easy to warp and thus they have insufficient strength and low surface quality. Therefore, particles have been added to pure polymers with the aim of improving the rheological behavior and thus enhancing the mechanical properties, or improving the surface quality of FDM products. For instance, Wu and coworkers [32] applied the organomodified montmorillonite (OMMT) to poly(L-lactide) (PLLA) to improve the rheological behavior of PLLA. It pointed out that OMMT were uniformly dispersed in the PLLA matrix in intercalation mode, and also the dispersed OMMT acted as a heterogeneous nucleating agent to promote the crystallization of PLLA. Tao and coworkers [33] added wood flour to PLA matrix. The results showed that the wood flour has changed the microstructure of material fracture surface, and the initial deformation resistance of the composite was enhanced as a result. Daver and coworkers [34] made a bio-interpretable cork-PLA by combining cork with PLA and adding tributyl citrate as a plasticizer, improving the elastic modulus and strength compared to pure PLA. Metal particles have also been used as fillers to FDM polymeric fillaments [35]. Nikzad and coworkers [36] added different volumes of metallic fillers such as iron or copper powder to ABS to make iron/ABS or copper/ABS composite. It was found that with the increase of metallic content in the composites, the tensile strength, flexural strength, and impact strength of the composites are lower than those of pure ABS materials, but the thermal properties have been significantly improved, leading to better surface quality.

Fiber-reinforced polymeric materials for FDM
Other issues faced by polymeric materials include the lowest interlaminar bonding strength and the highest mechanical anisotropy [37]. Thus, by modifying the polymer matrix with fillers, more isotropic and stronger FDM parts have been achieved. Quan and coworkers [38] mixed short carbon fiber (SCF) with ABS, which reduced the porosity of the molded sample and further improved the mechanical properties as well. When the SCF content was 30%, the tensile deformation of the molded sample strength and tensile modulus increased by 52% and 378%, respectively. Galatas and coworkers [39] printed ABS film first and then combined the film with a carbon fiber fabric to create a sandwich structure for strength and modulus improvement. Compared to pure ABS samples, the ultimate strength and Young's modulus of the sandwich structure were improved tremendously. For instance, the ultimate strength improved about nine times and the Young's modulus increased about 16 times. Gao and coworkers [40] explored the relationship between fillers and properties of FDM parts by adding different volume fillers (carbon fiber or talc powder) to PLA. They found that the filler has a significant effect on the formation of interlaminar bonding of FDM parts.

Nanomaterials for FDM
Nanomaterials are defined here as polymer matrix-based nanocomposites for FDM 3D devices. The nanoparticle inclusions are taken as the important potential fillers for enhancing physical and mechanical properties of polymer matrix. These inclusions include exfoliated clay, carbon nanotubes, carbon nanofibers, exfoliated graphite, nanocrystalline metals, and a host of additional nanoscale inorganic filler [41][42][43][44][45][46][47]. Once incorporated into polymeric matrix, as shown schematically in Figure 2, the types of morphology are usually categorized into three states: immiscible, intercalated, and exfoliated (miscible).
For the sake of properties enhancement, complete exfoliation of nanoparticle inclusions is highly desired after the incorporation process. Studies have been taken to develop nanomaterials for FDM with prospect to overwhelm the property or performance restrictions of traditional/conventional materials. For instance, Liu and coworkers [50] developed polymer matrix composites with carbon nanotubes and graphene nanoplatelets for FDM. The tensile test of FDM-printed dogbone (according to ASTM D638) showed that the mechanical property in the filament-oriented direction has been enhanced due to the incorporation of nanoparticles. Similar results have been reported in the references, showing mechanical property improvement in filament-oriented direction [51][52][53][54][55]. As explained in the Figure 2, complete exfoliation of nanoparticle inclusions would improve the mechanical property of filament. It would not be surprised to acknowledge the improvement of mechanical property in the filamentoriented direction. However, in the perpendicular direction, it is hard to improve the mechanical property by using nanoparticle inclusions. This is because the interfacial bonding between two adjacent printed layers plays a significant role in determining the properties in the perpendicular direction (i.e., normal to the filament orientation direction). During the formation process of interfacial bond, it undergoes stages from surface contacting, neck growth to molecular diffusion [56]. It is seldom to see the work on using nanoparticle inclusions to improve the surface contacting, neck growth, or molecular diffusion during FDM process.
Many other works have been reported to improve the processing or performance of 3D devices by using nanoparticle inclusions. Cobos and coworkers [57] added multiwalled carbon nanotubes (MWCNTs) and halloysite nanotubes to PLA matrix with the aim of improving its thermal properties. But because of the cluster of nanoparticles, defects exist within filaments if printed at a high scanning speed. In their work, maleinized linseed oil was added to the nanocomposites as plasticizer to mitigate the defects. It showed that the melt flow index increased approximately 46% from the capillary rheometry results. Bronze nanoinclusion was introduced to PLA matrix to improve the tribological properties of 3Dprinted devices, enabling them owning unique performance such as self-lubrication, vibration-damping, and corrosion resistance [35,58]. Also, graphene (GnP) filler was widely used in FDM to improve the dielectric and/or electromagnetic interference shielding properties of the printed devices [51,[59][60][61][62][63][64][65]. Since temperature gradient within a nozzle during printing process plays a great effect on the printing speed and product properties, great attention was paid to improve the thermal properties of polymeric filament, using cellulose nanofibrils [52], graphene nanoplatelets [66], graphene oxide [67], and MWCNT [68]. In addition, nanofillers have been used to improve the antibacterial and biocompatible performance of FDM-printed implants [69,70].

FDM printing
If the performance of a material determines the upper limit of the printed devices, appropriate setting of process parameters is an effective strategy to achieve the upper limit of the performance. There are a number of great work published on reviewing process parameter influences and optimization for FDM process [8,[71][72][73]. The parameters of FDM could be divided into four categorizes, including temperature, hierarchical, speed, and raster. However, how to optimize these parameters is not our interest in this work.
FDM prints functional devices using extrusion nozzles with about 0.1-1 mm width scales. FDM-like technique uses nozzles with centimeter width [74]. The Big Area Additive Manufacturing system, also with centimeter-width scale nozzles, can print 3D structures on the order of several meters [75]. Despite the record of high extrusion rate, solving the rate-resolution tradeoff problem is essential to accelerate the FDM adoption and industrialization.

Effort for fast printing
Since the first commercial FDM system sold by Stratasys, they are all gear-driven extrusion-based 3D printers. Filament is fed by gears into a chamber, where the filament is melted completely, as shown in Figure 3. The applied shearing force between the filament and gears drives melted polymer flowing through a nozzle. By multiplying the cross section area of the nozzle and extruding speed, it gives the extrusion rate. Nowadays, the size of a FDM filament becomes standard with a diameter of 1.75 mm. According to the law of conservation of mass, the extrusion rate of a FDM printer is mainly determined by the gear feeding speed. Efforts have been taken to increase the gear feeding speed, as shown in Figure 4, for the purpose of improving the extrusion rate and further increasing FDM printing rate [76,77]. For instance, a motor shaft (large circle) pinches filament against a ball race (small circle), driving it down through the chamber and pushing it out through a nozzle, as shown in Figure 4(b). Figure 4(c) shows another solution for increasing pinch force, thereby improving the filament feeding speed. Given the circumstances of intrinsic high resistance of thermal penetration of polymeric filament, however, higher filament feeding speed would result in a higher risk for extruder clicking, gears slipping, and nozzle clog [78].
Go and coworkers [79] have pointed out that an FDM printer can't print at a higher rate than the one dictated by its rate-limiting module; otherwise failure of a single module would occur, resulting in failure of the process altogether. This is because, for a given extrusion rate, the extruder must feed the filament at a commanded velocity, while providing sufficient force to overcome the flow resistance in the chamber. The chamber, in turn, must be capable of heating the filament to the target temperature at a specified heating rate. The motion system must then move the printhead (the extruder and chamber) at a rate commensurate to the exit velocity of material from the nozzle. Thus, they proposed a solution for fast printing, as shown in Figure 5. In addition to a conventional heating block within the printhead, a laser heating element was mounted on purpose, which ensures the melting of polymeric filament prior to entering the chamber (conventional heater block). They also proposed to use threaded filament rather than using regular filament products to increase extrusion force. The shear engagement area between filament and textured driving wheel has been remarkably increased [76]. As a result, the designed printhead was able to extrude materials at a The proposed laser-assisted printhead can print 3D devices with nozzle diameters of 1.0 and 0.5 mm. It has been approved that the printhead owns fast printing capability. However, can it still print fast if the nozzle diameter goes smaller than 0.5 mm? Besides, the FDM printing resolution in the XY axis is significantly greater (≈300 ηm) than what can be achieved with other techniques, such as SLA (≈10 ηm) and inkjet (≈50 ηm) [81]. An essential question for FDM printing is whether it can be made to extrude ever-smaller filament to achieve the micro-sized features that are desired for so many applications.

Effort for resolution improvement
In 3D printing, resolution is the quality, or level of details, at which the product is created. Higher resolution value means more printing details. In general, by using FDM, the movement of printhead on both X-Y and Z planes determines how fine the resolution would be. At its most basic level, the X-Y resolution is controlled by the size of nozzle and the movement of printhead. But for Z resolution, heat is the dominant factor in determining the outcome. This is because, due to heat transfer, a small layer height will distort previous layers. In this review work, it mainly focuses on the efforts for X-Y resolution improvement. In Figure 6, it shows typical FDM nozzles with different sizes as well as examples telling the resolution difference between 0.25 and 0.4 mm nozzles. Obviously, smaller nozzle gives higher printing resolution (i.e., more detailed prints). But, with a smaller nozzle size (0.2 mm diameter, for example), drawbacks of the printing process are blocking and buckling and even slippage of the wire on the pinch wheel [82]. This is because the filament is extruded to a very tight diametric tolerance with a gear-driven-based nozzle. A variation in filament diameter may cause blocking and/or slippage of the wire on the pinch wheel. Figure 4: Gear-driven extrusion-based nozzles for FDM printers: (a) drive gear having same diameter size as bearing roller; (b) drive gear having larger diameter size than bearing roller; (c) multiple drive gears and bearing rollers but drive gears having same diameter size as bearing rollers. Adopted from ref. [76] with modification. Figure 5: The feeding of threaded filament and the laser-assisted heating chamber. Adopted from ref. [80] with modification.
One of the great issues resulting from small nozzles is the build up of backpressure against the filament extruding. Regular solutions to this problem are either to print at a low speed or to avoid using small nozzles. At higher printing speed, the backpressure builds up by the geometry of nozzles. Thus, high speed printing is usually not suggested when using a nozzle with small size. Efforts have been made to overcome these challenges. Silveira and coworkers [84] proposed an interchangeable nozzle based on mini screw extrusion for the sake of fast printing when using a 0.4 mm nozzle. The difference between a commercial FDM nozzle and screw-based nozzle is shown in Figure 7. In Figure 7(a), the filament is driven by pin wheels and extruded through the nozzle at a relatively low speed, ensuring to be completely melted within the chamber. As shown in Figure 7(b), a mini screw extrusion-based FDM nozzle unit consists of a nozzle, screw, and barrel as well as peripherals. The printing material is driven forward by a mini screw, but extruded through the nozzle at a relatively high speed. By extruding PCL, the extruding rate has been improved to 65 cm 3 /h. After an optimization on configuration of screw section and screw geometry, the extruding rate has been greatly improved to 167 cm 3 /h for PEEK [85]. The configuration and geometry of screw-based FDM nozzle unit are illustrated in Figure 7(c). However, the disadvantage in screw-based nozzle unit is that it requires several types of heaters surrounding the barrel in order to ensure a complete melting. Consequently, the weight of a mini screw extrusion-based FDM nozzle is much higher than a filament-based nozzle. Because of the heavy weight, it leads to a large inertial force during a printing process, resulting in a difficulty in the control of printing accuracy [86][87][88].
It is worth noting that die swell exists during both filament-based and screw-based FDM. It occurs when non-Newtonian fluid flows in the mold in the extrusion process. Due to the shear stress formed along the die wall, the polymer chains align with the flow direction. When they reach the nozzle outlet, constraints disappear, and thus, because the conformational entropy tends to increase, these chains are driven to return to their original and misalignment state, resulting in the increase of volume (i.e., extrudate diameter) [89]. As a result, the size of the extrudate is larger than that of the die. The shape of the cross section also changes when non-Newtonian fluid is forced to extrude the exit die. To date, as reported, the smallest nozzle diameter was 0.15 mm. Although it has been claimed that FDM printer can print an object of 50 microns or even higher, it is of great importance to note that the claimed resolution is limited on a Z-axis. Due to die swell, the resolution or printing width on X-Y axis never reaches to 0.15 mm or higher. How to avoid die swell for FDM becomes a challenging question.
Electrohydrodynamic-based FDM (E-FDM) was proposed for the easy manufacturing of high-resolution patterns with sub 10 µm, by generating micro-scale liquid extrudate via the application of a high voltage between the nozzle and substrate, as shown in Figure 8. With this strategy, die swell has been well-avoided when printing high-resolution FDM products. The working principle is similar to that of electrospinning [90][91][92]. The difference is that the fibers formed in the electrospinning process are quite randomized and thus the fiber orientation could not be controlled precisely. However, in the E-FDM process, the fiber orientation can be controlled precisely and thus finest 3D structures could be printed [93].
The printed pattern stability and resolution are influenced by E-FDM parameters including temperature, feed-rate, pressure, voltage, plotting speed, and nozzle size [92,94]. Temperature is a unique factor that impacts the viscosity. In one of the studies, a temperature range from 235°C to 260°C was investigated for PLA material. As being pointed out, when temperature reached over 250°C, stable cone-jet becomes more difficult to maintain and resolution fluctuates more significantly [91]. Definitely, there is a critical point in the temperature range; in other words, the pattern and resolution can be improved when temperature increases as long as it is below the critical point. There is also a critical point for voltage and plotting speed [90]. With fixed voltage and plotting speed, higher pressure leads to stable jet and skewed cone-jet. Better printing resolution could be obtained with smaller nozzle size and lower feed-rate [95]. Currently, only PCL and PLA have been reported for biomedical devices by E-FDM. For instance, Wang and coworkers [94,95] fabricated controlled uniform 3D porous PCL scaffolds using E-FDM towards the realization of printing functional human organs. He and coworkers [92] constructed highresolution nanoscale patterns by simply controlling the movement of E-FDM collector according to a userspecific program. Wu and coworkers [96] produced controllable caplets for personalized medicine using E-FDM for the sake of safety and efficacy of the drug delivery system.

Simulation
Simulation is especially important for high-value components fabricated using the state-of-the-art 3D printing process. It helps to understand and visualize the complex thermomechanical phenomena taking place during 3D printing. For instance, through a numerical simulation on mixture of PLA-nHA, Li [97] successfully printed out FDM biomimetic artificial bone to meet the accuracy requirements of clinical bionic artificial bone replacement. Thus, it is highly important to take simulation strategy to understand the FDM processing by establishing simulation models from deposition flow to thermal behavior to structural performance, as shown in Figure 3.

On feeding process
On the feeding process, filament is fed into a chamber, in which it undergoes a reversible transition from solid to liquid. A lot of works have been carried out to understand the rheology of polymeric materials by using different statistical and heuristic methods [71]. With the aim of processing parameter optimization, these simulation works mainly focused on identifying key processing parameters by studying the thermal and thermomechanical behavior of the printed part, postdeposition of the feeding process. According to analytical or numerical thermal models, the relationship between printingprocess parameters and thermomechanical behavior could be established for better controlling product quality [98][99][100].
The feeding process involves multi-physics (including thermal and dynamic) phenomena on different spatial and temporal scales. It is worth noting that the extrusion is a local operation while the printing of an entire product is a global process. Physically sound models based on numerical simulations have been conducted to understand the feeding process and to gain parameter optimizations [101,102]. For instance, computational fluid dynamics (CFD) simulations of the deposition flow were proposed to predict the flows of melted filament inside the molten chamber, including the calculation of flow field, pressure drop, and melting distance to the entry of the chamber [103]. With CFD model, the impact of feeding process on dimensional accuracy and mechanical properties of the final product was thoroughly investigated. It found that the shape of printed strand can be affected by feeding rate, nozzle dimensions, and extruding force. Furthermore, the shape has a direct impact on the surface roughness of the printed part, cooling efficiency after the printing, and the interfacial bonding strength between adjoining strands and overlaid layers.

On interfacial bonding
After deposition process, the strand undergoes a transition from liquid to solid. Fast cooling of the printed strand causes solidification prior to the complete molecular interdiffusion with other strands, leaving voids or defects between adjoining strands and overlaid layers that do not have full mechanical properties [18]. As been pointed out, thermal behaviors of the filaments after deposition significantly determine the quality of the FDM 3D product [104]. This is because, during the formation of interfacial bonds, temperature profiles are involved in all interface wetting, molecular diffusion, and randomization [105]. As temperature profile being involved, the polymer chains continue to diffuse with an increase in the mechanical properties of the interface. Therefore, it is crucial to establish analytical models to predict thermal history of a strand after deposition.
There are a great number of excellent work reviewing the simulation tools for predicting thermal history of strands under FDM extrusion [106][107][108]. Here, we categorize them into three groups: (1) the one-dimensional transient heat analysis (1D THA), which only accounts for either the strand width or the layer height due to the onedimensional limitation [109]. (2) The two-dimensional transient heat analysis (2D THA), with which the strand has a rectangular cross section. Accordingly, the effect of conduction to the platform can be neglected and any contact resistance between strands could be negligible [110][111][112]. (3) The three-dimensional transient heat analysis (3D THA), which describes the real building up process of any cuboid by a FDM printer in the case of raster angle of 0 (90)°and filling ratio of 100%. Simulations were taken to investigate the influence of nozzle temperature, base plate temperature, environment temperature, layer thickness, printing speed, geometry dimension and the resolution of FDM printer on the temperature field, and gradient variation with respect to space and time [100,113].

On structural performance
Statistical models have been used to establish a relationship between printing parameters (input variables) and mechanical behavior of FDM product (output response). Once established, it has been used as a tool to predict the mechanical properties based on input variables and, in turn, gain an optimization of FDM printing parameters. For instance, Ali and coworkers [114] proposed a twofactor-level Taguchi method for mechanical property optimization via design of experiments. Based on the analysis of mechanical properties derived from tests, the implementation of analysis of variance led to optimized printing variables for general use and for application/load-specific instances. Other statistical methods also have been applied for optimization of printing variables, such as polynomial regress model [115] and response surface methodology [116][117][118]. As being pointed out, the relationship of printing variables and their interactions and the manner in which they affect mechanical properties offers an insight to the feasibility of using FDM for rapid prototyping of products for practical implications.
For a part under real loads and fixtures, the common practice is to use finite element analysis for understanding how different printing parameters affect mechanical behavior of FDM products [18,119]. Yao and coworkers [120] proposed a method based on fracture mechanics for predicting the ultimate tensile strength (UTS) of FDM products. The factors influencing UTS included printing angle and layer thickness. By comparing the built micromechanical model with experimental data, it showed the relative residual sum of squares was all close to zero, indicating the model can accurately predict the UTS of FDM products for all angles and thickness.
In recent years, machine learning (ML) has been applied to aid solving the facing challenges from FDM, such as process control, process monitoring, and quality enhancement of FDM products [121,122]. For instance, Galatas and coworkers [39] applied the artificial neural network (ANN) to investigate the effect of FDM-printed-ABS core density and number of carbon fiber reinforced polymer layers on the mechanical properties. A nonlinear predictor based on ANN analysis was developed to predict the elastic modulus and specific strength of composite material. Due to good fast convergence, the three-term backpropagation was adopted to improve the elastic modulus and specific strength using far less training data and less computational effort. Yanamandra and coworkers [123] applied ML of microstructure to obtain counterfeiting and unauthorized production of highquality FDM parts for reverse engineers. With the help of the micro-CT (μCT) scan and scanning electronic microscope images, a refined recurrent neural network model achieved a high degree of accuracy in predicting fiber orientation within strands, which further was used to identify printing orientation at each layer of the FDM products.

Applications of FDM-printed devices
The FDM 3D devices are suitable for functional prototypes, tools, and low-volume manufacturing. FDM 3D printing technology has successfully provided some commercial solutions to industries, for example, customized cranial implant, customized kidney implant, COVID-19 breathing valve, customized mandibular implant, customized auto parts, flow pump body, industrial part model, customized running shoes, so on and so forth. Besides, Qi and coworkers [124] reported a multifunctional magnetorheological plastic containing PCL/TPU. This material has advantages in FDM printing, mechanical switching, shape memory, and self-healing. It has been used in the multifunctional magnetic sensitive materials, such as soft robots, medical care, and bionics and has broad application prospects. Sajadi and coworkers [125] reported tubulanes lattice polymer cubes using FDM technology. They made different kinds of cubes through computer simulation, and then used FDM technology to print polymer cubes by testing their structural properties. The results show that FDM technology can theoretically print crosslinked carbon nanotube-like tubulanes structures into physical parts, which have excellent strength. He and coworkers [126] reported a personalized triboelectric nanogenerator (TENG) layer based on FDM technology, which could be used to effectively harness vibration energy in environment. Although FDM has lower manufacturing accuracy compared with other 3D printing technologies, it has met the requirements of complex TENG devices.
The biomedical industry is already benefiting from FDM during recent years with the need for customizable, biocompatible, and sterilizable polymeric materials. It has to be admitted that every single patient or body has one's own unique needs. Given that, medical devices highly require the most customization of biomedical needs [127][128][129][130]. The traditional manufacturing makes the 3D products with high cost and most vital slow to produce. He and coworkers [131] reported the use of FDM technology to print PEEK nutcracker stents to treat nutcracker syndrome. Under laparoscopic surgery, his team implanted a PEEK porous venous extravascular stent "tailored" by FDM technology in advance for a patient with nutcracker syndrome. The patient recovered well after the operation. The stent did not cause any damage to the blood vessel during the review.
In terms of improving the mechanical properties of FDM devices, adding carbon nanotubes [132], graphene oxide [67], chopped carbon fibers [133], and other methods makes this technology extremely promising in transportation equipment, household goods, and medical industries. Topological optimization was also an efficient strategy for enhancing the mechanical properties [18,22,26,100]. In addition, FDM technology is used in the rapid manufacturing of polymer-based electronic devices [134], low-speed impact energy-absorbing structures [16], jewelry and fashion [135], optical devices [136], epidermal drug delivery polymerization microneedles [137], drone applications [39], bionic artificial bones [138], and other fields also have broad application prospects.

Conclusions
Except for the material itself, defects exist to limit the development of FDM. The faced challenges are to make FDM with more applicable materials, fast printing time, high dimensional accuracy, excellent surface quality, sufficient mechanical strength, so on and so forth. But the advantages of FDM technology in 3D printing are remarkable in new materials, new structures, new devices, and new functions. Indicated from the application of FDM 3D printing in cutting-edge fields, major technological breakthroughs have been achieved notably in research and development in the aerospace, commercial goods, and biomedical fields. With the development of 3D solid parts becoming more complex, precise, and multifunctional, FDM technology has shown a trend of high-resolution and fast printing.

Conflict of interest:
The authors declare no conflict of interest regarding the publication of this paper.