Selective Laser Sintering of Polymers: Process Parameters, Machine Learning Approaches, and Future Directions

Abstract

Selective laser sintering (SLS) is a bed fusion additive manufacturing technology that facilitates rapid, versatile, intricate, and cost-effective prototype production across various applications. It supports a wide array of thermoplastics, such as polyamides, ABS, polycarbonates, and nylons. However, manufacturing plastic components using SLS poses significant challenges due to issues like low strength, dimensional inaccuracies, and rough surface finishes. The operational principle of SLS involves utilizing a high-power-density laser to fuse polymer or metallic powder surfaces. This paper presents a comprehensive analysis of the SLS process, emphasizing the impact of different processing variables on material properties and the quality of fabricated parts. Additionally, the study explores the application of machine learning (ML) techniques—supervised, unsupervised, and reinforcement learning—in optimizing processes, detecting defects, and ensuring quality control within SLS. The review addresses key challenges associated with integrating ML in SLS, including data availability, model interpretability, and leveraging domain knowledge. It underscores the potential benefits of coupling ML with in situ monitoring systems and closed-loop control strategies to enable real-time adjustments and defect mitigation during manufacturing. Finally, the review outlines future research directions, advocating for collaborative efforts among researchers, industry professionals, and domain experts to unlock ML’s full potential in SLS. This review provides valuable insights and guidance for researchers in regard to 3D printing, highlighting advanced techniques and charting the course for future investigations.

1. Introduction

Rapid advancements in 3D printing have been achieved within the last five years. This is attributed to various industries’ need for fast and accurate components fabricated by additive manufacturing (AM). While regulations and specifications for this manufacturing process are still being refined, its transformative impact across various fields has led to a rush to adopt and leverage its advantages [1]. Nearly three decades after its inception, this technology has reached a stage of maturity wherein it can produce high-quality components through advanced techniques [2]. Moreover, the costs associated with machinery and raw materials have significantly decreased, rendering their utilization economically viable. Three-dimensional printing is an AM technique, depicting any fabrication method that constructs items by gradually incorporating materials, typically in layers. To harness this technology, 3D graphics are essential, and they can be generated using various computer programs as the sole language understood by the machines. These graphics are further segmented into minute slices through anatomical software and transformed into a code transmitted to the printer to commence the production procedure [3,4,5,6,7,8]. Three-dimensional printing can create components ranging in sizes from millimeters to meters, utilizing diverse materials such as polymers (e.g., ABS, polycarbonate, nylon), metals (e.g., stainless steel, aluminum, titanium, precious metals), and even unconventional substances like concrete, sand, wax, and biological materials [8]. Notably, 3D printing possesses the unique ability of being able to efficiently manufacture small batches, irrespective of their complexity, within remarkably short timeframes. This stands in contrast to other manufacturing techniques like casting, which prove costly for limited production quantities due to the requirement of crafting specialized molds for each piece. Consequently, such techniques are primarily suited for mass production purposes [9]. AM encompasses a broad array of techniques, differing primarily in the materials employed and the method of layer integration to form objects [7]. Based on the state of the material used, 3D printing techniques can be classified into liquid-based (such as SLA and inkjet printing), solid-based (such as FDM), and powder-based (such as SLS) [10,11,12,13,14,15,16,17,18].

In liquid-based additive rapid prototyping, the primary raw printing material utilized is a liquid known as a photopolymer. When the photopolymer liquid is subjected to altered laser light via lenses, this liquid transitions from liquid to solid as the 3D printer traverses the three building directions X, Y, and Z. This process allows for the formation of the desired shape. Specifically, this type of printer is referred to as a Stereolithographic Apparatus (SLA). Within the realm of additive manufacturing technology, an SLA falls under the category of additive processes using liquid materials in 3D printing, which can be further categorized into three techniques based on the process of curing resin through sequential polymerization layer by layer. These techniques include (i) photopolymerization, (ii) material jetting, and (iii) DIW material extrusion, as shown in Figure 1.

In solid-based additive rapid prototyping, a raw material, known as a filament, is initially in a solid state. As the machine moves, the filament undergoes melting to achieve the desired design. Subsequently, the molten material cools down and solidifies, resulting in the final design. Typically, polymers are used as the filament material. This technology is known as Fused Filament Fabrication (FFF), and it is widely popular among everyday consumers due to its user-friendly nature and cost-effectiveness. Another type of printer that utilizes solid material is Sheet Lamination, where solid strips of material are cut and melted in layers to attain the intended form. Figure 1 depicts the various solid-based 3D printing techniques. Ongoing advancements in this field aim to reduce production time and enhance the surface smoothness of the fabricated objects.

Powder bed fusion, a common 3D printing process utilized in AM, comprises direct metal laser sintering (DMLS), electron beam melting (EBM), selective heat sintering (SHS), selective laser melting (SLM), and selective laser sintering SLS. In this technique, the raw material is in powder form and becomes solid when exposed to a specific type of laser beam, forming the desired shape as the machine moves along three axes [19,20,21,22,23]. SLS is one of the commonly used printers that utilizes this technology.

Among these methods, our study focuses on the study of SLS powder-based 3D printing technique parameters. As shown in Figure 1, there are two types of selective laser processes: selective laser sintering and selective laser melting. Although both methods operate on similar principles, they differ in the temperatures used. Selective laser sintering (SLS) is a polymer-based powder process in which a laser systematically sinters particles, building parts layer by layer, whereas selective laser melting (SLM) uses metal powder that is either pure or alloyed rather than polymer powder [24].

This paper aims to provide a comprehensive review of selective laser sintering (SLS) technology, focusing on its operational principles, challenges, and the potential of machine learning (ML) integration. The review delves into the impact of various processing variables on material properties and part quality, as well as the application of different ML techniques in optimizing SLS processes.

2. Principle of SLS

Selective laser sintering (SLS) is a highly popular AM technique. It involves sintering thin layers of powdered materials spread over the platform of a printing bed using a laser as the power source. It is commonly used for producing nylon-based products and utilizes an additive layering technique. The method utilizes a laser of high-energy, such as a carbon dioxide-type laser, to selectively merge fine particles of plastic powders into a solid object with a three-dimensional form. The laser scans cross-sections of the part, based on a 3D digital description (e.g., from a CAD file), onto the surface of a powder layer. After each cross-section is finished, the thickness of the powder layer is reduced by one layer, followed by the addition of a new layer of material. This SLS process is repeated until the part is completely formed, as shown in Figure 2. SLS is particularly useful for rapid prototyping and manufacturing metallic or polymer powder components with complex shapes and excellent mechanical properties. It can produce high-precision, durable components that are suitable for direct use in end-use applications, low-volume manufacturing, or rapid prototyping. A notable advantage of SLS is its ability to create bulk components without requiring support structures. However, the surface roughness of SLS-produced parts tends to be relatively high due to factors such as the orientation of the build tray, partial fusion of powder grains, and specific characteristics of the powder and manufacturing parameters [25].

The sintered density and surface roughness are strongly influenced by surface active elements, powder size and shape, and processing parameters such as laser power, scan rate and scan line spacing. Extensive research is needed to understand the effect of these factors on the SLS process.

3. Applications of SLS

Selective laser sintering (SLS) is a highly effective 3D printing technology that has greatly influenced various industries by producing strong, durable parts and complex geometries. Figure 3 illustrates different applications of the SLS. Here are some notable applications and innovations of SLS across different sectors [26,27,28,29,30]:

(i)

Aerospace

SLS is employed to create lightweight, complex parts such as turbine blades and brackets, which are essential for reducing aircraft weight and enhancing fuel efficiency. Additionally, aerospace companies utilize SLS for rapid prototyping to expedite the design and testing phases of new components. SLS is also used to produce tooling and fixtures for aerospace manufacturing, helping to reduce lead times and costs.

(ii)

Automotive

SLS enables the production of custom parts, such as bespoke interior components and performance-enhancing modifications, without the need for costly molds. It is also utilized to create functional prototypes for testing and validation, thereby accelerating the development cycle. Additionally, SLS can manufacture end-use parts, such as brackets and housings, which take advantage of the technology’s strength and durability.

(iii)

Healthcare

SLS is employed to create complex medical devices, including custom prosthetics and orthotic devices tailored to individual patients’ needs. The technology also allows for the production of precise and ergonomic surgical instruments. Additionally, SLS is being explored for bio-printing applications, such as creating scaffolds for tissue engineering.

(iv)

Consumer Goods

SLS enables the production of customized consumer products, including personalized jewelry, bespoke eyewear, and unique fashion items. It is also used to create functional prototypes for new consumer products, facilitating the design and iteration process.

(v)

Architecture and Construction

Architects use SLS to create detailed and accurate scale models of buildings and structures for visualization and presentation. Innovations in SLS enable the production of complex architectural elements and structural components that would be difficult or impossible to achieve with traditional methods.

(vi)

Education and Research

SLS is used to create educational models and teaching aids in fields such as engineering, biology, and design. Research institutions also utilize SLS for rapid prototyping of experimental designs and components across various scientific disciplines.

4. Materials Fabricated by SLS

Nylon 12 and Nylon 11 are the most common single-component powders used in SLS, and both can be reinforced with materials such as glass or carbon fiber to create composites with enhanced properties, like increased strength and rigidity. In addition to nylon, other popular SLS 3D printing materials include polypropylene (PP), polyetheretherketone (PEEK), and thermoplastic polyurethane (TPU) [26,32,33,34,35,36,37].

(i)

Nylon/PA 12

Nylon/PA 12 is known for its excellent balance of mechanical properties, including good impact resistance and flexibility. Additionally, it has low moisture absorption, which helps maintain dimensional stability. It is the most commonly used material in the SLS process. It has a density of 1.01 g/cm³ and melts between 178 and 180 °C.

(ii)

Polyamide 11 (PA11)

Nylon 11 offers similar characteristics to Nylon 12 but is derived from a renewable source—castor oil. It provides slightly better flexibility and toughness due to its chemical structure. It maintains its mechanical properties over a wide temperature range, providing good thermal stability.

(iii)

Polypropylene (PP)

Polypropylene is valued for its chemical resistance, low density 0.946 g/cm³ (crystalline), and high flexibility. However, it can be challenging to use in printing due to its tendency to warp and its lower surface quality compared to nylons. It melts between 130 and 171 °C.

(iv)

Polyetheretherketone (PEEK)

This high-performance thermoplastic offers exceptional strength, rigidity, and thermal stability. It is commonly used in demanding applications such as aerospace and medical implants because of its high resistance to chemicals and extreme temperatures. It has a 1.32 g/cm³ density and a 343 °C melting point.

(v)

Thermoplastic Polyurethane (TPU)

TPU is a flexible and elastic material, making it ideal for applications that require both durability and flexibility. It is commonly used in applications such as gaskets, seals, and flexible components.

5. Challenges Associated with Fabricating Polymers Using SLS Technology [26,33,38,39]

(i)

Powder Handling and Reuse:

Powders can degrade with each build cycle due to exposure to heat and laser energy. Degradation affects the mechanical properties of the parts and may result in a poor surface finish. Also, contamination from foreign particles or degraded powder can impact part quality and consistency.

(ii)

Thermal Management:

Polymer parts can experience warpage and shrinkage due to the high temperatures involved in the sintering process. Proper thermal management and control are essential in minimizing these issues. Moreover, the parts may suffer from residual stresses. Uneven heating and cooling can induce residual stresses in the parts, leading to deformation or reduced mechanical performance.

(iii)

Surface Finish:

Parts produced with SLS can have a rough surface finish which might require additional post-processing steps like sanding or coating to achieve the desired smoothness.

(iv)

Dimensional Accuracy:

Achieving high dimensional accuracy can be challenging due to the nature of the sintering process. Factors such as powder layer thickness and laser focus can affect the final part dimensions.

(v)

Part Complexity and Orientation:

While SLS does not require support structures, complex geometries with overhangs may still present challenges in terms of powder removal and part integrity. The orientation of parts during printing can impact mechanical properties and surface finish, and optimizing orientation requires careful planning.

(vi)

Cost and Material Properties:

Nylon/PA 12 powders can be relatively expensive, and managing material costs is important for large-scale or high-volume production.

Ensuring consistent mechanical properties, such as strength and flexibility, across different builds can be challenging due to variations in powder quality and processing conditions.

(vii)

Post-Processing:

Removing the un-sintered powder from the build chamber and the finished parts can be labor-intensive and requires careful handling to avoid damaging the parts. Additional heat treatments or annealing processes might be required to improve the mechanical properties of parts.

Addressing these challenges involves optimizing the SLS process parameters, investing in quality control measures, and employing effective post-processing techniques. Advances in technology and materials continue to improve the performance and reliability of SLS for materials.

6. SLS Parameters

Optimal quality in building parts within the SLS sinter-station necessitates the configuration of various fabrication parameters in the software that are customized to suit the characteristics of the powder and meet the needs of the application. Understanding the interplay between these parameters is crucial. Below, we outline several fabrication parameters linked to process and powder properties.

6.1. Laser Power

The laser beam strength employed in fusing powdered material is controlled by the laser power setting. It is crucial to set the laser power at an appropriate level to ensure adequate energy for effective fusion while avoiding excessive melting or overheating. The laser energy employed greatly influences the surface topography, durability, and resistance to corrosion of the components used in SLS. Higher laser energy values promote better cohesion between the grains and reduce the distance between them, resulting in improved surface roughness on the fabricated products. Additionally, enhanced particle adhesion contributes to a decrease in the percentage of pores, thereby enhancing the overall strength, hardness, and corrosion resistance. A study was conducted to examine the influence of laser power on the surface roughness (SR) of SLS polymer parts oriented at 0°, 45°, and 90° angles. It was observed that increasing the laser power within the three different orientations led to a narrow range of fluctuation in roughness. The lowest roughness of the parts was achieved when the laser power was raised from 57 W to 67 W. A higher laser power facilitated the sintering of powder layers, resulting in reduced surface roughness. Conversely, at lower laser power levels, the layers were not adequately sintered [27]. Moreover, the examination of the laser power’s influence on the mechanical characteristics of Duraform polyamide powder, featuring particles sized at 60 µm, was conducted. The research uncovered that employing a laser power of 40.71 W resulted in the production of components with optimal tensile strength (TS), elongation, and yield strength (YS) [40]. The examination of the effect of processing parameters on the microstructures and characteristics of aluminum powder in both single-layer and multiple-layer constructions was conducted to explore the SLS process. The densification mechanism and solidification process of the fabricated layers were significantly influenced by factors such as laser energy input, particle size, and the distribution of different elements. Through the analysis of single layers and initial exploration of aluminum powders, it was observed that particle shape, oxygen content, and the type and ratios of alloying elements determined the laser energy required for sintering and melting the material. Generally, increasing these factors led to higher laser energy input, resulting in increased amounts of molten metal and larger agglomerations [41].

The mechanical properties of the fabricated parts, specifically density and hardness, are influenced by the power of the laser sintering process. Increasing the laser power has been observed to lead to higher density and hardness values. This is attributed to the greater amount of heat generated, which permeates the machine bed, promoting a denser packing of powder particles [40,41,42]. Similar findings regarding laser power and its impact on density and hardness were reported by Gibson and Shi in their study on polymer (nylon)-based samples produced via SLS [43].

6.2. Scanning Speed

The scanning speed in the selective fusion processes refers to the rate at which the laser beam moves across the powder bed. This parameter plays a crucial role in determining the quality and accuracy of the fabricated parts by influencing energy distribution and heat transfer. Careful selection of the scanning speed is necessary to achieve the desired part characteristics. In the case of SLS, the scanning speed is an important factor that affects both productivity and part strength. Utilizing a high scanning speed may reduce particle adhesion, while using a low scanning speed can lead to excessive energy concentrations and particle diffusion. Therefore, it is essential to study and determine the appropriate range of scanning speeds that ensure the integrity and strength of the produced parts.

A study was conducted to investigate the effect of scan speeds (180, 200, and 220 mm/s) at different orientations (0°, 45°, and 90°) on the surface roughness of polymer parts fabricated via SLS. The surface roughness varied with different scan speeds and orientations, and the lowest surface roughness was achieved at a scan speed of 200 mm/s [27]. The authors demonstrated that, at high scan speeds, the printed layer was non-uniform, resulting in a lack of continuous molten material. Conversely, insufficient scanning speeds during powder sintering led to uneven melting, causing distortions or cavities in the fabricated parts.

6.3. Layer Thickness

The layer thickness determines the vertical height of each layer that is fused in the process. This parameter directly affects the resolution of the final part and is typically chosen based on the desired level of detail and the properties of the powder used. The thickness of the formed layers has a significant impact on the properties of products manufactured using SLS. When the layer thickness is increased, the adhesion between powder particles decreases, resulting in lower material properties. In research investigating the effect of layer thickness on SLS polymer powder, different orientations (0°, 45°, and 90°) were considered and layer thicknesses of 0.1 mm, 0.11 mm, and 0.12 mm were applied. It was found that increasing the layer thickness led to an increase in surface roughness in all orientations. The lowest roughness value was obtained with a layer thickness of 0.1 mm at 0° orientation. Additionally, it was observed that larger layer thicknesses required higher energies for grain fusion [27]. In another study by Vishal Sharma et al. [42], Duraform polyamide parts were fabricated using SLS. The researchers examined the mechanical properties of components manufactured with various layer thicknesses, including 0.09, 0.1, and 0.11 mm. The findings revealed that the lowest layer thickness, 0.09 mm, enhances the best mechanical properties.

6.4. Hatch Spacing

Hatch spacing refers to the distance between adjacent laser scan paths. It influences the level of overlap between scan lines and can impact the density and strength of the final part. Scan spacing is also known as hatch spacing or hatch distance. It is the separation between two successive laser beams. It is measured as the distance between the centers of the obtained beam to the center of the next beam, as shown Figure 4. The effect of scan spacing with 0.6, 0.20, and 0.24 mm at different fabrication orientations on the surface roughness of the SLS polymer was studied. The optimum surface roughness was obtained at 0.24 mm scan spacing for the 0 and 45° orientations. Proper overlapping of adjacent scan lines at an appropriate scan spacing ensures homogeneity in energy distribution and optimal powder blending. Scan spacing that is too small results in overheating and excessive energy concentration, adversely affecting the surface of the SLS components [27].

In a study conducted by Vishal Sharma et al. [42], the effect of SLS parameters on the mechanical properties of Duraform polyamide powder with a particle size of 60 µm was investigated. The scan spacing was identified as a crucial parameter that significantly influenced the mechanical properties of Duraform polyamide. Different scan spacing values of 0.15, 0.2, and 0.25 mm were applied, and it was found that the lowest scan spacing of 0.15 mm had the most pronounced effect on the mechanical properties, as mentioned earlier.

Experiments were performed to analyze the impact of SLS parameters on the density and hardness of polyamide parts [42]. The researchers employed the response surface methodology (RSM) and analysis of variance (ANOVA) approaches to determine the optimal working conditions and assess the significance of each process parameter on density and hardness. The scan spacing was varied, with values of 0.1, 0.2, and 0.3 mm. Notably, a scan spacing of 0.1 mm exhibited the most substantial effect on both density and hardness. The experimental data revealed that increasing the scan spacing led to a decrease in both density and hardness. This observation can be attributed to the inadequate packing of particles and the presence of un-sintered particles when larger scan spacing values were used. Consequently, the density and hardness were negatively affected. It is worth noting that the scan spacing should not exceed the diameter of the laser beam [42].

6.5. Preheating Temperature

In the SLS process, the preheating temperature is the temperature at which the powder bed is maintained before and during the sintering process. This temperature is crucial to the SLS process. Preheating the powder bed to a temperature close to the polymer’s melting point reduces thermal gradients and minimizes the risk of warping or distortion in the final part. A uniform preheating temperature enhances mechanical properties and dimensional accuracy by reducing the viscosity of the polymer powder, which improves the flow and compaction of the powder layers. This leads to better layer bonding and overall part quality. Additionally, a consistent preheating temperature lowers the energy required by the laser for sintering, bringing the powder closer to its melting point, which improves the efficiency of the sintering process and reduces the risk of incomplete sintering [26,45,46,47,48]. It also helps alleviate thermal stresses within the part, which is particularly important for complex geometries and helps to prevent defects or failures. For common polymers used in SLS, such as Nylon (PA), the preheating temperature is typically set at around 10–20 °C below the material’s melting point. For example, Nylon 12, with a melting point of approximately 178 °C, is generally preheated to around 165–170 °C [45].

6.6. Powder Particle Size Distribution

In the SLS process, the powder particle size distribution (PSD) greatly influences the printing process and the quality of the final parts. A narrow PSD, with uniformly sized particles, enhances powder flowability, ensuring a more even distribution of powder layers. This results in smoother surfaces and consistent layer application. Uniform particle sizes also contribute to predictable thermal behavior during sintering, as consistent heating and sintering rates are achieved across the powder bed, leading to more uniform part properties. Conversely, variations in particle size can affect the energy required for sintering; larger particles may need more energy to melt, potentially causing inconsistent sintering if not managed correctly. A fine, narrow PSD generally yields a smoother surface finish, as finer particles fill the gaps between larger ones, while a wide PSD can result in a rougher finish due to the prominent texture created by larger particles. Parts made from powders with a narrow PSD typically exhibit more consistent mechanical properties due to the homogeneity of the powder bed. In contrast, a broad PSD can lead to varying sintering densities and mechanical properties within the same part. Additionally, PSD impacts the porosity of the sintered part; a well-controlled PSD usually results in lower porosity and higher density, whereas a broad PSD can increase porosity due to incomplete packing and variable sintering rates. Consistent particle sizes promote better layer-to-layer bonding, enhancing overall part integrity. A wide PSD may cause weak points where larger particles fail to bond effectively with smaller ones. Ultimately, a narrow PSD contributes to a more stable and repeatable SLS process, as the behavior of the powder during sintering is more predictable, whereas a wide PSD can lead to variability and inconsistent part quality [26,48,49,50,51]

6.7. Part Bed Temperature (T_b)

The temperature of the powder bed plays a critical role in the melting and sintering behavior of powdered materials during the SLS process. It is essential to maintain a controlled temperature within the optimal range specific to the powder being used. In the SLS process, the material is deposited layer by layer, and each new layer must fully coalesce and adhere to the previously sintered layer. This requirement is particularly important for semi-crystalline polymers, where maintaining crystallization in a few sintered layers is necessary. Therefore, the bed temperature of a polymer should be precisely controlled between its melting and crystallization temperatures [52,53].

Moreover, it is necessary to maintain a sufficient temperature difference between the print bed and the feed bed. If the feed bed temperature is too high and approaches or exceeds the print bed temperature, it can result in agglomeration and fusion of the material in the feed bed, leading to printing failures [54]. Therefore, in formulations, the feed bed temperature is typically set to be lower than the print bed temperature, with a temperature difference of around 20 °C [55,56,57].

In terms of the effects on density and hardness, observations have shown that increasing the bed temperature from 172 to 175 °C can lead to an increase in density [42,58]. However, further increasing of the bed temperature from 175 °C to 178 °C resulted in a reversal of this effect. Similarly, for hardness, an increase in bed temperature led to an increase when the scan count was 1 mm, but the effect reversed when the scan count was increased to 2 mm [42]. Higher bed temperatures generally contribute to a proper sintering process, resulting in higher densities and hardness. However, the observed reduction in density at bed temperatures of 175–178 °C may be attributed to polymer degradation, as suggested by Ho et al. [59].

It is important to carefully select the bed temperature during the fabrication of polyamide using SLS. According to reference [40], the optimal bed temperature was found to be 176 °C when the orientation of the bed was at 90 degrees and the laser power used was 40.71 W.

One of the challenges in SLS is the occurrence of the balling of metals and porous structures, which is caused by differences in wettability between the new powder bed and the sintered layers [41]. To mitigate this issue, it is crucial to choose an appropriate scan rate.

The determination of the part bed temperature is influenced by the glass transition temperature (Tg) of both amorphous and crystalline polymer materials. The stiffness and modulus of materials are affected by Tg and the melting temperature (Tm), as depicted in Figure 5.

To minimize the sticking together of powder particles at the surface of the part bed, it is recommended that the part bed temperature does not exceed the Tg of the amorphous material [43]. This helps prevent unwanted agglomeration or clumping of the powder particles during the sintering process.

By carefully controlling the bed temperature and considering the Tg of the materials used, it is possible to reduce temperature differentials between the in-sintering and after-sintering stages, leading to improved fabrication outcomes in SLS.

6.8. Thermal Distribution

The thermal behavior is a mechanism that links process parameters such as scan speed, hatch spacing, energy density, and laser power to product quality. This factor is related to heat distribution, radiation, and thermal convection within the SLS-built chamber [61]. It significantly influences the dimensional accuracy, surface finish, and the microstructural and mechanical strength of the printed parts [61,62,63]. Therefore, various aspects are studied to thoroughly understand how different process parameters and material characteristics affect the quality and surface roughness of 3D-printed parts. According to the SLS literature, scan speed and laser power are interrelated parameters [64], as they both influence the amount of energy transferred to the material, referred to as energy density. Therefore, these two parameters must be balanced to prevent pore formation due to either insufficient or excessive energy. Additionally, balancing these parameters can help reduce build times in industrial manufacturing. Lexow et al. [65] investigated powder bed fusion (PBF) at high scan speeds. In their study, they found that high-speed processing reduces the density of printed parts and affects particle fusion. Moreover, Peyre et al. [66] found, in their experimental analysis, which varying scan speeds during the sintering process can lead to local overheating and significant deformations, which in turn affect the mechanical performance of 3D-printed parts. Li et al. [67] concluded that high scan speeds coupled with low laser power lead to insufficient energy density for properly melting a single layer. Van et al. [68] observed that excessive energy could lead to undesirable results during the construction of the part and might impair the properties of the polymer.

6.9. Powder Morphology

The morphology of the powder, including particle size and shape, influences both flowability and powder bed density [69]. Powder size impacts the efficiency of laser melting. Larger particles typically reduce the packing density of the powder bed and require more energy to melt. In contrast, smaller particles are prone to agglomeration, which complicates the powder layering and coating processes. Typically, particle sizes range from a few microns to several hundred microns. The flowability of the powder is a crucial factor in determining the distribution of particles in the powder bed. Effective powder flow is essential for creating a flat powder surface and ensuring uniform layer thickness, both of which are necessary for consistent laser energy absorption. Researchers have reported that particle shape is a critical factor in the sub-functions of powder bed fusion (PBF), such as recoating and coalescence [70]. In this context, Chatham et al. [69] found that spherical particles enhance the dimensional accuracy of 3D-printed parts, leading to a denser powder bed. Conversely, Schmid et al. [71] demonstrated that non-spherical particles reduce the properties of 3D-printed parts. According to Chatham et al. [69], fine particles, being small enough to be influenced by surface forces, tend to agglomerate, which impedes flow and affects smooth layer coverage. This is in contrast to larger particles, which can disrupt the smooth surface of the powder bed and cause layer scarring. Such defects directly impact the quality of the manufactured parts.

Studies in

Table 1. A comparison of mechanical properties of polymer SLS.

SLS Specimen	Process Parameters	Hardness (HV or HRL)	Tensile Strength (MPa)	Impact Strength (KJ/m2)	Flexural Strength (MPa)	Specific or Young Modulus	Bend Strength (MPa)	Ref.
Pure Nylon 12	Bed temperature/°C Sintering window/°C		44	37.2	50.8		-	[72]
Recycled nylon elastomer			29.8	48.6	34.9		-
Carbon steel/nylon-12	Energy density (J/mm2)		-	-	-		3.1	[73]
PA-12 PEEK	Sintering time [min] Heating rate [°C/min] Cooling rate [°C/min]					1.02 (MPa*m3/kg) at (321 °C)		[74]
Polypropylene	Scan spacing (0.13 mm) Fill laser power (5 w) Scan size (65)	75 (coating)	40					[75]
Polyamide	Laser power (W) Scan spacing (mm) Bed temp. (°C) Hatch length (mm) Scan count	24 0.1 172 (98.5 HR) 120 2	-	-	-	-	-	[42]
PA	laser energy density building orientation				48 MPa at 0.02 j/mm2 and 90°			[75]
polycarbonate	Energy density	-	-	-	-	-	-	[76]
Polyetheretherketone (PEEK)	Relative density Temperature						0.4% increment at 89% relative density and 354 °C	[77]
PA11/nano Alumina	Scan speed mm/sec Laser Power (Watt)		0.1 MPa at 3%Al2O3 and 12 W					[78]
PA12-0.1wt %CNT	Powder bed temperature (°C) Layerthickness (mm) Laser power (W) Laser scan speed (mm/s) Laser scan spacing (mm)		53.45 MPa at 19 Watt	112 KJ/m2 at 25 Watt	85 MPa at 19 Watt	4000 MPa at 19 Watt		[79]
Polyamide 12/(0, 2.5, 5, and 10) Carbon Nanotube	Printing orientation 45° Layer thickness 0.2 mm Bed temperature 80 °C Nozzle temperature 255 °C Travel speed mm/s	30 HV (5 wt% MWCNT)	49 MPa	38 KJ/m2 (0 wt% MWCNT)	57 MPa (5 wt% MWCNT)			[80]
polyamide 12	Laser beam power 20 Watt Laser scan speed 3000 mm/s Layer thickness 100 µm Building platform temperature 160 °C		46.93 MPa					[81]
Polypropylene homopolymer and copolymer	Laser power p (W) Laser scan speed s (mm/s) Laser beam diameter (mm) Hatching distance h (mm) Layer thickness L (mm) Powder bed temperature Tb (°C) Chamber temperature (°C)		(15 MPa) iPP (19.1 MPa) CoPP					[82]
PP CP 75 Polypropylene	Chamber temperature (125 °C) Bed temperature (128 °C) Layer thickness (0.15 mm) Hatching distance (0.25 mm) Scanning speed fill (4500 mm/s) Laser power fill (20 W)		7.4 MPa Reused powder (3rd print cycle)					[83]
PA12	Laser power LP(% Changeable) Part orientation (XY plane) (°) (Changeable) Hatching was conduct axis) with the following Chamber temperature Moving plate temperature Hatching spacing Diameter of laser beam Infill Scanning speed Hatching orientation (XY plane) Changeable		26 MPa (at 0° orientation and 95% LP)			1170 MPa (at 0° orientation and 95% LP)		[84]
Polyamide 12	Wall thickness build direction		43.4 MPa (3 wall thickness and Transversal direction)					[85]

[72,73,74,75,76,77,78,79,80,81,82,83,84,85] elucidate the impact of various SLS parameters on the mechanical properties of diverse materials, underscoring the significance of optimizing these parameters to attain desired material characteristics and performances in SLS-fabricated components. Shi et al. [72] demonstrated that bed temperatures and sintering windows significantly affect the mechanical properties of pure Nylon 12 and nylon elastomer. Pure Nylon 12 exhibited superior tensile strength (44 MPa) and flexural strength (50.8 MPa) compared to nylon elastomer. Gibson and Shi [43] investigated the effects of scan spacing, fill laser power, scan size, and the number of layers on the hardness and tensile strength of polypropylene. Their results indicated that hardness decreased with an increase in the number of layers, whereas the highest tensile strength (40 MPa) was attained with two layers. Sharanjit et al. [42] optimized the parameters of SLS for polyamide, identifying the optimum levels of laser power, scan spacing, bed temperature, hatch length, and scan count to maximize density (1.028 g/cm³) and hardness (98.5 HRL). Warnakula and Singamneni [76] explored the SLS of nano-Al₂O₃-infused polyamide, finding that increasing laser power reduced porosity levels, achieving the highest density (79%) at 9 W and 1000 mm/sec, with the highest tensile stress (0.1 MPa) being recorded for the 3% Al₂O₃ composite at 12 W. Bai et al. [78] examined the fabrication of polyamide 12/carbon nanotube (CNT) nanocomposites, studying the effects of powder bed temperature, layer thickness, laser power, laser scan speed, and laser scan spacing on flexural, impact, and tensile properties. They reported the highest tensile strength (53.45 MPa), flexural strength (85 MPa), and Young’s modulus (4000 MPa) at 19 W, with the highest impact strength (112 KJ/m²) being 25 W. Rosso et al. [81] compared Multi Jet Fusion and SLS methods for fabricating polyamide 12, investigating SLS parameters such as laser beam power, laser scan speed, layer thickness, and building platform temperature. The polyamide 12 fabricated via SLS exhibited a tensile strength of 46.93 MPa under the specified conditions.

7. Porosity in SLS Parts

Figure 6 illustrates the various types of porosity detectable in SLS parts in association with processing parameters [86]. Porosity poses a significant challenge in the SLS process, as entrapped air between the matrix and reinforcement particles leads to its development. The void ratio is influenced by SLS processing parameters, including laser power and speed, particle shapes and reusability, as well as powder properties and morphology. Elevated or reduced levels of scan speed, laser power, layer thickness, or hatch distance can encourage pore formation [87,88,89]. Additionally, porosity may arise between different layers during the fabrication process, which is known as inter-layer porosity.

Porosity in SLS parts is influenced by several processing parameters. The laser power and scanning speed are crucial; if either is set too high or too low, it can lead to pore formation and affect the density of the final part. Additionally, the shape and characteristics of the powder particles, along with their reuse, can impact porosity. Irregularly shaped particles or reused powder may introduce voids, leading to increased porosity in the finished parts. Furthermore, the properties and morphology of the powder, such as particle size, distribution, and composition, also play a significant role. Proper optimization of these factors is essential in minimizing porosity and ensuring high-quality SLS parts.

8. Machine Learning ML for SLS

8.1. Overview

ML, a branch of artificial intelligence (AI), has become a powerful tool across various industries, including SLS for polymer materials. ML algorithms can be divided into three main categories: (1) Supervised Learning: This type of learning involves training the algorithm on labeled datasets to associate inputs with their corresponding outputs [90,91,92]. In the context of SLS for polymers, this involves using data on process parameters and resulting part properties to train the model. Supervised learning techniques include regression, which predicts continuous outcomes such as part strength or flexibility, and classification, which categorizes data into distinct classes, which is helpful for determining the suitability of parts for specific applications. (ii) Unsupervised Learning:

Unsupervised learning does not rely on labeled data but instead identifies patterns within the data independently [93,94,95]. In SLS, this can be useful for discovering similarities in materials based on their inherent properties, providing insights for further exploration or modification. Methods like data clustering are frequently used in unsupervised learning to detect grouped patterns. (iii) Reinforcement Learning: This approach utilizes an iterative process where the algorithm makes decisions and receives feedback, allowing it to efficiently tackle complex optimization challenges through a trial-and-error mechanism [96,97,98]. In the realm of SLS, reinforcement learning can be employed to optimize processes or formulations. For example, it might find the best sintering parameters for polymer materials by adjusting parameters based on feedback to improve part quality over time.

Each algorithm type is used based on the nature of the problem and the available data. While supervised learning requires labeled data, unsupervised learning offers insights from unlabeled data, albeit with fewer direct predictions. These ML techniques hold promise for advancing SLS processes, enabling precise control and optimization of polymer material properties and part quality.

8.2. ML-Based Monitoring Methods

Machine Learning (ML) has emerged as a powerful tool for enhancing process monitoring in selective laser sintering (SLS), offering unprecedented capabilities in defect detection, quality control, and process optimization. To comprehensively address this rapidly evolving field, we propose a multidimensional taxonomy that categorizes ML-based monitoring approaches in SLS. This taxonomy is structured along four key dimensions: the sensing method, underlying physical principle, nature of detected defects, and temporal resolution. Sensing methods encompass various technologies such as optical imaging, infrared thermography, acoustic sensing, and X-ray imaging, each tailored to capture specific aspects of the fabrication process. These methods differ not only in the type of data they collect but also in the physical principles they leverage—be it light reflection, thermal emission, mechanical wave propagation, or photon absorption.

Table 2. ML-based sensing methods for defect detection in AM.

Sensing Method	Physical Principle	Type of Defects Detected	Temporal Resolution	Example ML Techniques
Optical	Reflectance	Surface defects, Geometric deviations	Real-time within layer	Convolutional Neural Networks (CNN), Computer Vision [107,108]
Optical	Scattering	Powder bed anomalies, Surface roughness	Layer-by-layer	Image Segmentation, Support Vector Machines (SVMs) [109,110,111]
Infrared	Thermal emission	Internal porosity, Lack of fusion	Real-time within layer	Thermal Image Analysis, Deep Learning [112,113]
Infrared	Thermography	Temperature distribution, Cooling rates	Layer-by-layer	Time Series Analysis, Random Forests [114,115]
Acoustic	Ultrasound	Internal defects, Density variations	Post-build	Acoustic Signal Processing, Neural Networks [116,117]
Acoustic	Acoustic emission	Crack formation, Delamination	Real-time within layer	Spectral Analysis, Recurrent Neural Networks (RNN) [109,118]
X-ray	Absorption/transmission	Internal porosity, Inclusions	Post-build	3D Image Reconstruction, CNN for 3D data [119,120]
Laser Speckle	Interferometry	Surface deformations, Residual stress	Layer-by-layer	Pattern Recognition, Bayesian Networks [121,122]

provides a framework for categorizing various monitoring approaches. Additionally, Figure 7 shows some of the monitoring techniques that can be used in SLS. Structured light projection can detect defects during recoating and improve contour accuracy, emphasizing the importance of powder surface quality before laser exposure [99,100]. Additionally, in situ monitoring using laser profilometry ([101] or [102]) can identify curling and prevent process failures. These techniques aid in processing difficult materials and calibrating machines with challenging thermal uniformity. Additionally, using digital cameras to capture reflections on the melt surface is based on the hypothesis that uniformity indicates complete particle fusion [103].The nature of detected defects is another critical dimension, distinguishing between surface anomalies, internal porosity, geometric deviations, and other forms of defects. By categorizing approaches based on the type of defect they are designed to detect, this taxonomy aids in selecting the most appropriate monitoring techniques for specific quality control needs. Lastly, temporal resolution differentiates between layer-by-layer monitoring and real-time within-layer detection [104]. This dimension highlights the trade-offs between the richness of the data collected and the speed of processing, offering insights into how monitoring strategies can be optimized for different production requirements. Through this taxonomy, researchers and practitioners can more effectively navigate the complex landscape of SLS process monitoring, ultimately driving advancements in both academic research and industrial application. Through this taxonomy, researchers and practitioners can more effectively navigate the complex landscape of ML-based process monitoring in SLS, fostering advancements in both academic research and industrial applications.

Figure 7. Monitoring techniques used in SLS: (a) fringe projection [99], (b) laser profilometer [101], (c) thermal infrared camera [102], (d) acoustic sensing [105], (e) X-ray imaging [106].

Figure 7. Monitoring techniques used in SLS: (a) fringe projection [99], (b) laser profilometer [101], (c) thermal infrared camera [102], (d) acoustic sensing [105], (e) X-ray imaging [106].

8.3. Process Optimization

SLS involves numerous process parameters that influence the final part properties, such as laser power, scan speed, hatch spacing, and powder bed temperature. Optimizing these parameters to achieve the desired part characteristics is a complex task due to the intricate relationships and interactions between the parameters.

Czelusniak and Amorim [123] examined the impacts of specific laser sintering parameters on the density, mechanical, and dimensional attributes of carbon fiber-reinforced PA12 components produced via SLS. They employed a supervised learning-based Gaussian process to characterize the correlation between the sintering parameters and the quality attributes of the components. Furthermore, they utilized stochastic optimization via an evolutionary algorithm to uncover trade-off solutions for diverse multi-objective optimization challenges. The Gaussian process proved to be a highly effective modeling approach for most of the response variables analyzed. Their results suggested that the laser sintering parameters greatly influenced the physical and mechanical properties, demonstrating intricate and non-linear relationships. Furthermore, Kim and Zohdi [124] investigated the optimization of tool path selection in SLS processes through numerical simulations alongside the application of deep learning techniques to expedite simulations and ensure precise results. Effective process optimization in SLS not only enhances the overall manufacturing process but also enables the production of high-quality, customized parts tailored to specific applications. As the field of AM continues to advance, further research and improvements in process optimization will be essential for fully unlocking the potential of SLS and other AM technologies.

8.4. Defect Detections

Polymer-based SLS processes are susceptible to various defects like porosity, warpage, poor surface finish, and incomplete fusion, which can compromise the mechanical properties and dimensional accuracy of printed parts. ML algorithms, especially computer vision and deep learning techniques, have shown promise in detecting these defects. Generally, a significant issue at the mesoscale is the porosity of parts created through additive manufacturing (AM). In metal AM, achieving full density is crucial because porosity greatly impacts the mechanical performance of parts, particularly their fatigue properties [123,124]. Multilayer perceptrons (MLPs) can model complex non-linear relationships but often struggle to explain their predictions. Conversely, Gaussian processes (GPs) can estimate uncertainties in predictions but are more computationally intensive, with the same input data compared to MLPs [125]. Therefore, both MLPs [126,127,128], as well as GPs combined with Bayesian methods [129], were utilized to predict porosity based on various process parameters in selective laser melting (SLM). Additionally, open porosity is desirable in some applications, such as auxetic structures for energy absorption and porous structures for medical implants. In the SLS of PLA material, support vector machines (SVMs) and MLP techniques were used to predict open porosity [130]. Ensemble-based multi-gene genetic programming (MGGP) offers an advanced evolutionary method for automatically developing both model structure and its coefficients, addressing the limited generalization capability of traditional MGGP. This approach was successfully applied to optimize the selective laser sintering (SLS) process for a 58 wt% hydroxyapatite and 42 wt% polyamide powder blend, enabling precise control of open porosity through fine-tuning of process parameters [131].

Regarding the surface roughness, Koç et al. [132] developed a comprehensive surface roughness dataset for selective laser sintering (SLS) using polyamide 12 powders, producing specimens in various orientations on an EOS Formiga P110 machine. They then created a Deep Neural Network (DNN) model to classify surfaces as rough or smooth based on part orientation. The model achieved 95.8% accuracy on the training set and 100% on the test set. When benchmarked against Support Vector Machine (SVM) and Naïve Bayes (NB) models, the DNN demonstrated superior performance in predicting surface roughness for SLS-produced parts. In the context of polymer-based SLS, a common methodology is to utilize high-quality image data of the printed components as input to train convolutional neural networks or other deep learning models. These models are then able to learn and identify the patterns and characteristics associated with various types of defects by analyzing a labeled dataset consisting of both defective and non-defective samples. In situ monitoring data from polymer SLS machines, such as laser power, scan speed, and temperature readings, can also be leveraged. Unsupervised ML algorithms, like clustering or anomaly detection models, can be trained on this process data to identify deviations that may indicate the presence of defects.

By combining image data and process monitoring data as inputs, ML models can potentially achieve higher defect detection accuracy for polymer SLS parts. For instance, Westphal and Seitz [133] investigated the application of ML algorithms, particularly convolutional neural networks (CNNs), for automatic defect detection in SLS processes. They addressed the challenge of part defects and irregularities in additive manufacturing (AM) by proposing complex transfer learning (TL) methods. These methods utilized pretrained CNN models such as VGG16 and Xception with weights from the ImageNet dataset, adapted classifiers, and very small datasets to classify powder bed defects during SLS part manufacturing. Figure 8 showcases a flowchart detailing the complete transfer learning (TL) process, including all associated calculations. Widely recognized evaluation metrics, such as accuracy, precision, recall, the F1-score, receiver operator characteristic (ROC) curves, the area under the curve (AUC), and heat maps, were employed to assess, compare, and visualize the performance of the convolutional neural network (CNN) models. The VGG16 model architecture demonstrated the highest scores across various metrics, including the accuracy (0.958), precision (0.939), recall (0.980), F1-score (0.959), and AUC value (0.982), as depicted in Figure 9. These findings highlight the efficacy of convolutional neural networks (CNNs) for defect detection and suggest their potential as an alternative non-destructive quality control and manufacturing documentation approach for additively manufactured components. Furthermore, a study conducted by Klamert et al. [102] explored the use of deep learning, particularly convolutional neural networks (CNNs), for real-time quality control in SLS processes. Their approach demonstrated a high accuracy of 98.54% in detecting curling failures through the analysis of infrared thermography recordings, which could potentially improve the efficiency and sustainability of SLS manufacturing. Overall, the application of ML enables automated inline defect detection, leading to improved quality control, root cause analysis, and process parameter optimization for polymer SLS processes.

While ML holds significant potential in SLS, several challenges need to be addressed. These include data availability, model interpretability, and the incorporation of domain knowledge into ML models. Additionally, the integration of ML techniques with in situ monitoring systems and closed-loop control strategies is an area of active research.

While ML holds significant potential in the field of SLS, several challenges need to be addressed to fully harness its capabilities. These challenges can be quantified and discussed as follows: (1) Data Availability: ML models require large amounts of high-quality data for training and validation. In SLS, obtaining comprehensive datasets that cover a wide range of material properties, process parameters, and part geometries can be challenging and time-consuming. Quantitatively, many ML studies in SLS have relied on datasets with limited sample sizes, often ranging from a few hundred to a few thousand data points. Increasing the availability of diverse and well-annotated datasets is crucial for improving the accuracy and generalizability of ML models. (2) Model Interpretability: While ML models can effectively capture complex relationships between inputs and outputs, they often lack interpretability, making it difficult to understand the underlying reasoning behind their predictions. In SLS, where process understanding is essential for quality control and process optimization, interpretable ML models are highly desirable. Quantitatively, methods like local interpretable model-agnostic explanations (LIMEs) and SHapley Additive exPlanations (SHAPs) can offer insights into the relative significance of various input features. However, the effectiveness of these techniques in capturing the complex physics underlying SLS processes is an area that warrants further exploration. (3) Incorporation of Domain Knowledge: SLS is a complex process governed by various physical phenomena, including heat transfer, phase transformations, and powder–material interactions. Incorporating domain knowledge and physical principles into ML models can improve their performance and enhance their interpretability. Quantitatively, hybrid modeling approaches that combine physics-based models with data-driven ML techniques have shown promising results, but their application in SLS is still limited. (4) Integration with In Situ Monitoring and Closed-loop Control: In situ monitoring systems, such as thermal cameras and melt pool monitoring, provide valuable real-time data during the SLS process. Integrating ML techniques with these monitoring systems and developing closed-loop control strategies can enable real-time process adjustments and defect mitigation. Quantitatively, the development of robust and responsive closed-loop control systems remains an active area of research, with challenges in data processing, decision-making, and actuator response times. (5) Computational Resources: As ML models become more complex and datasets grow larger, the computational resources required for training and inference can become substantial. In SLS, where real-time or near-real-time predictions are often desired, the trade-off between model complexity and computational efficiency needs to be carefully balanced. Techniques such as model compression, quantization, and hardware acceleration can help address this challenge, but their impact on model performance in SLS applications requires further investigation.

Addressing these challenges through collaborative efforts among researchers, industry practitioners, and domain experts will be crucial for realizing the full potential of ML in SLS and advancing the state of the art in additive manufacturing.

9. Conclusions

The fabrication parameters of SLS play a crucial role in determining the mechanical properties of polymer materials. The optimization of these parameters is essential for obtaining high-quality parts that meet the requirements of specific applications. Based on the comprehensive analysis of SLS for polymers, the following conclusions can be drawn:

• The quality and properties of SLS-produced parts are heavily influenced by numerous process parameters. Key factors include laser power, scanning speed, layer thickness, hatch spacing, preheating temperatures, powder particle size distribution, and part bed temperatures. Careful optimization of these parameters is crucial for achieving desired part qualities such as density, strength, and surface finish.
• Laser power has a direct impact on density and hardness, with both properties increasing as laser power increases. Scan spacing is a crucial parameter that significantly impacts both the density and hardness of the printed parts. Experimental findings reveal that an increase in scan spacing leads to a decrease in both density and hardness. Consequently, to achieve optimal performance in these metrics, it is recommended to use a low scan spacing value, such as 0.1 mm.
• The relationship between bed temperature and hardness is linear, with hardness continuously increasing as bed temperatures increase. However, the effect of bed temperature on density is more complex, with density initially increasing and then decreasing as bed temperatures further increase.
• Scan count has a positive correlation with density, with higher scan counts resulting in increased density. However, the effect of scan count on hardness is dependent on bed temperature. At lower scan counts, increasing bed temperatures lead to increased hardness, while at higher scan counts (e.g., 2), the trend reverses, with hardness decreasing as bed temperatures increase.
• By leveraging the power of ML and optimizing the process parameters, SLS has the potential to revolutionize the manufacturing industry, enabling the production of high-quality, customized parts with improved efficiency and sustainability.
• ML techniques have demonstrated high accuracy in defect detection for SLS processes. For instance, the VGG16 convolutional neural network model achieved impressive metrics for powder bed defect classification, including accuracy (0.958), precision (0.939), recall (0.980), F1-score (0.959), and AUC value (0.982) [133].
• The challenges in applying ML to SLS are significant and quantifiable. Data availability is a major issue, with many ML studies on SLS relying on limited datasets, often ranging from a few hundred to a few thousand data points. This highlights the need for larger, more diverse datasets to improve model accuracy and generalizability.

Future perspectives include:

• • The development of high-performance polymers and composite materials.
  • Focusing on increasing print speed and building larger parts or assemblies.
  • The integration of SLS with other manufacturing technologies, such as robotic arms and automation systems, to enhance its capabilities and applications.