Maraging steel powder recycling effect on the tensile and fatigue behavior of parts produced through the laser powder bed fusion (L-PBF) process

Additive manufacturing (AM) has advanced the manufacturing industry and has been employed in a wide range of industrial applications, including aerospace, automotive, medical, and die-casting equipment. To ensure the cost-effectiveness of the AM process, unfused powder must be recycled even if its characteristics may change after each cycle, making essential the validation of powder quality and component mechanical performances. Despite the research published to date, predicting the mechanical performance of printed parts issued from reused powder remains challenging since it is dependent on many AM process variables. Until now, no research has looked at the impact of powder recycling on the fatigue behavior of maraging steel components. This study investigates the impact of maraging steel powder reuse on powder characteristics, as well as on the tensile and fatigue properties of printed components. Our results indicate that the powder particle size distribution increased after eight powder reuses, particle morphology showed the presence of aggregates, broken particles, and shattered and deformed particles, while powder apparent density remained constant. Powder reusing had no significant impact on the surface roughness of as-built specimens. Although there was a slight decrease in mechanical properties over reuse cycles, tensile and fatigue performance remained globally stable, while the standard deviation of fatigue stress became narrower after eight cycles. Finally, fractography revealed that the fatigue fracture surfaces of components manufactured from an eight-time recycled powder have more fusion defects and carbon inclusions than the parts made from virgin powder.


Introduction
There has been a growing interest in metallic additive manufacturing (AM) due to the multiple benefits of printing sophisticated products that are lightweight and offer high accuracy, near-net-shape features, and excellent mechanical performance [1]. Among the most prevalent metallic additive manufacturing processes are laser powder bed fusion (L-PBF) and electron beam melting (EBM). Both technologies use the same manufacturing process: a recoater device spreads a fine layer of powder over a substrate, then an energy source (a laser in the case of L-PBF and an electron beam in the case of EBM) melts the powder layer by layer until the 3D part is formed. These processes make it possible to manufacture complex geometrical pieces with great dimension precision.
A drawback of L-PBF technology is the need to fill a larger volume than the volume of the printed parts. The unfused powder is exposed to different phenomena: a laser heat effect, a mechanical spreading process, a sieving process, and a storage process. Thus, its chemical and physical properties are altered, which could also affect the mechanical properties of the manufactured components. Because metallic powder is costly, unfused powder should be sieved and prepared for subsequent prints to ensure that the additive manufacturing process is cost efficient. In a case study, LPW Technology Ltd claimed a 92% decrease in material costs when powder was reused 15 times 1 3 (Rushton, 2019) [2]. Moreover, the new technology of plasma spheroidization allows for recovery of reused powder, which makes AM a green and clean process [2].
Recycling is necessary to keep costs as low as possible and to minimize the environmental impact. Although many studies have been conducted to date, the issue of how recycling powder impacts its chemical and physical features and the mechanical behavior of the printed component is still not completely understood [3]. Thus, engineers are still unsure about the performance of printed parts made from recycled powder, particularly when it comes to critical components [4]. According to recent studies [1,5], the spherical shape is the most appropriate shape for metal AM because it enhances powder flowability. The spherical shape also boosts other mechanical characteristics, notably ultimate tensile strength and yield strength, as well as surface roughness and dimension accuracy. Moreover, Bochuan Liu et al. [6] have noted that wide particle size distribution (PSD) delivers superior powder bed density, parts with higher density and smoother surface finish while powder with a narrower PSD has higher flowability and creates components with ideal tensile properties and hardness.
It is recognized that printing and sieving processes affect the chemical composition of the powder. Many researchers have reported a change in powder composition, especially the percentage of oxygen content. Tang et al. [7] observed a progressive increase in the oxygen content in the composition of Ti 6Al 4 V, from 0.08 to 0.19%, after 21 reuses. Consequently, ultimate strength and yield strength increased by 13% and 15%, respectively. In addition to the influence of the process, chemical composition can also be affected by storage techniques. Indeed, when the powder is not stored under inert gas, it is very susceptible to contamination by ambient air, which carries pollutants, fibers, and dust, and can also be influenced by humidity, corrosion, and oxidation [2]. Furthermore, Moghimian et al. [3] presented an overview of the factors affecting both the powder quality and the mechanical properties of different materials: powder production methods, L-PBF printing parameters, powder recycling, powder blending, and the sieving process.
Several researchers have recently studied the effect of reusing powder on chemical and physical powder characteristics and the mechanical performance of components produced with L-PBF and EBM technologies for different materials, such as titanium Ti-6Al-4 V [4,[8][9][10][11][12][13][14], stainless steel [15][16][17][18][19], aluminum AlSi10Mg [20,21], Inconel 718 [22][23][24], and maraging steel [25]. Because the different studies do not always arrive at the same results, even if they deal with the same alloy, it appears relevant to present an overview of the research works related to the effect of powder recycling. The following paragraphs summarize the information relative to each of these alloys.

Titanium Ti-6Al-4 V material
Carrion et al. [8] evaluated the powder properties and mechanical properties of Ti-6Al-4 V L-PBF specimens. After 15 reusing cycles, the particle size distribution became narrower, and the particles' diameter value D90 decreased by 18.7%. The powder flowability was improved significantly. The particles' morphology was not greatly modified. Furthermore, they observed that recycling had no effect on the tensile and fatigue behavior of specimens in the as-built surface state. However, when they machined the specimens made from used powder, the fatigue behavior in the high cycle regime of the specimen printed from recycled powder improved when compared with specimens made from new powder.
According to Quintana et al. [9], the size of powder particles on the bed and in the dispenser decreased with the number of reuses, and their shape remained generally spherical. Flowability was improved, while tap density decreased slightly as the number of reuses increased. A small increase in tensile strength was noted after 31 build cycles owing to the increased oxygen content.
However, Renishaw plc (2016) [10] noted that Ti-6Al-4 V powder PSD increased and became tighter, which improved flowability. Particle morphology remained mostly spherical over 38 cycles, with the occasional presence of aggregates and elongated particles. The ultimate tensile strength (UTS) and yield strength (0.2% YS) values increased slightly, by 100 MPa and 128 MPa, respectively, from virgin to build 38. These changes may be attributable to an increase in oxygen and nitrogen levels. The strain at break stayed almost constant.
Moreover, O'Leary [11] detected an increase in titanium particle size distribution (PSD), with a decrease in the number of smaller particles less than 15 µm and an increase in the number of larger particles greater than 45 µm, after five reuse cycles. Reused powder surfaces were rougher and less spherical. Like O'Leary, Seyda et al. [12] noted an increase in the PSD of L-PBF Ti-6Al-4 V powder and an improvement in apparent density and flowability after 12 recycling iterations. However, the surface roughness of printed parts was drastically impacted. That said, the parts produced from reused powder exhibited better hardness and tensile strength than specimens fabricated from virgin powder. According to the authors, the improvements were related to the increase in powder density and the presence of oxygen in the melt pool carried by the ambient atmosphere during the powder sieving process.
Additionally, Soundarapandiyan et al. [13] observed that particle size distribution increased and became wider on region near the melt pool because particles were partially melted, hard-sintered, and agglomerated. Away from 1 3 the melt zone, the PSD was narrower, and particle shape remained almost spherical. After 10 reuse cycles, the PSD moved slightly to the coarser side, the powder sphericity rose by 26% over the sphericity of fresh powder, and the number of satellites and finer particles decreased, resulting in enhanced flowability and density. The microstructure, Charpy impact energy, hardness, and elongation parameters did not change significantly. Because of the modest rise in recycled powder oxygen concentration, yield strength and tensile strength increased slightly. The fatigue life of recycled components was reduced due to an increase in lack of fusion defects and porosity.
In their study, Alamos et al. [14] concluded that there was no substantial change in the mechanical tensile properties of L-PBF Ti-6Al-4 V after eight reuse cycles. PSD narrowed slightly, but the diameter values D90, D50, and D10 were almost unchanged. The density of the printed components remained consistent.
Moreover, Popov et al. [4] reported that after 69 reuse cycles in EBM machines, the Ti-6Al-4 V particles were significantly damaged by the laser heat exposure and sieving process during powder recycling. The powder particle shapes were altered by the formation of elongated particles, "clip-clap," agglomerated, and broken particles. After the reuse cycles, UTS and YS rose slightly, while the elongation of the samples was significantly decreased. Samples produced from reused powder have exhibited a much lower fatigue life because of the existence of porosity, partially melted particles, and lack of fusion defects. However, the hot isostatic pressing (HIP) treatment enhanced fatigue life of specimens built from recycled powder.

Stainless steel material
Contaldi et al. [15] investigated the effect of powder reuse on two types of precipitation hardening stainless steel, the martensitic PH1 and the austenitic GP1. The PSD for both materials became narrower and showed a slight decrease of small particles over nine reuses. After recycling, the presence of oval-shaped particles and agglomeration in both materials differed from the normal spherical particle form. Furthermore, chemical composition, tap density, and apparent density of both materials remained relatively stable. The mechanical characteristics of PH1 were found to be insensitive to powder reuse, with only the elongation at break being slightly affected. However, samples printed from GP1 powder were more impacted, particularly yield strength, which rose by 190 MPa, and elongation at break, which decreased by 6%. The fatigue behavior of both materials remained generally constant over nine reuse cycles.
Ahmed et al. [16] noted a minor increase in the 17-4 PH stainless steel PSD over 10 reuse cycles. As irregular shaped particles grew in the feedstock, powder's flowability decreased. When compared with the first sample, the tenth specimen showed a 54% increase in pore size and a 17% increase in surface roughness, which led to a 7% drop in ductility. The UTS was not substantially changed.
Indeed, Arash Soltani-Tehrani et al. [17] remarked that the 17-4 PH stainless steel powder PSD got slightly narrower and shifted to the left, indicating a drop in the fraction of larger particles and a decrease in the proportion of fine particles (smaller than 15 μm) after 14 reuse cycles. Most of the particles remained spherical, which improved flowability. The researchers also observed that recycling and component location on the building plate had no influence on the tensile and fatigue behavior of specimens in their as-built surface condition. However, after machining the specimens built from recycled powder, their fatigue behavior in the high cycle regime improved remarkably, especially for the specimens located further away from the dispenser.
According to Jacob et al. [18], the hardness of the 17-4 PH stainless steel parts, tensile strength, and surface roughness showed no change throughout 11 reuse cycles. The recycled powder showed no change in either particle size distribution (PSD) or particle shape, while apparent and packed densities increased and flowability improved.
Moreover, Sutton et al. [19] reported that laser spatters, known as ejecta, compromised the morphological and chemical properties of AISI 304L stainless steel reused powder. After five cycles, the powder particles were coarser and their shapes became irregular. Powder-bed density showed a substantial increase since powder flowability was improved over reuse cycles. Tensile properties were marginally influenced throughout reuse cycles, which was linked to the existence of large pores on fracture surfaces. Additionally, it was noted that there was a declining trend in Charpy impact toughness due to the rise in oxygen content during printing iteration.

Aluminum alloy AlSi10Mg material
Del Re et al. [20] investigated on a L-PBF printer the effect of recycling AlSi10Mg powder on its physical and chemical properties as well as on the mechanical properties of printed parts. After eight reuse cycles, they discovered that particle's shape was less spherical with the presence of agglomerate particles. Furthermore, the PSD of the reused powder showed a slight leftward shift toward fine particles, indicating a progressive decrease of coarser particles owing to the sieving operation. Over the eight reuse cycles, apparent and tap densities and chemical composition stayed almost constant. However, yield and tensile strength declined by 10 MPa, while high cycle fatigue strength decreased from 160 to 145 MPa.
Asgari et al. [21] conducted a similar study for AlSi10Mg 200C parts. They concluded that the average particle size, microstructure, morphology and composition of the virgin and recycled powder were nearly identical. They differed only because of the presence of spatters, which showed an irregular shape, satellite particles, and rough powder particle surfaces. After printing cycles, there was no substantial change on the tensile properties of components.

Inconel 718 material
Yi et al. [22] reported that the Inconel 718 powder PSD increased, and particle morphology remained almost spherical after 14 reuse cycles. Furthermore, apparent density and flowability were significantly improved. There was a slight effect on the tensile properties of printed parts. Indeed, the UTS and YS stayed fairly constant, around 1025 MPa and 750 MPa, respectively, and the strain ranged from 27 to 30%.
Moreover, Rock et al. [23] studied the evolution of the physical and chemical characteristics of Inconel 718 powder after 10 building cycles. They observed that the PSD increased, the morphology of the particles was affected by the formation of agglomerates, and spatters with dendritic surfaces. Therefore, flowability significantly decreased after reusing.
Furthermore, Ardila et al. [24] analyzed the same material in an L-PBF machine and concluded that powder characteristics and mechanical parts performance did not significantly change after 14 cycles. Most particles remained spherical, with a slight increase in size distribution. Microstructure and porosity were very similar in all iterations thanks to their recycling strategy, which consisted of sieving the unfused powder to eliminate aggregation and drying it in the oven with air circulation to remove humidity. For mechanical characteristics, they applied the Charpy test to specimens and remarked that toughness was almost unchanged after printing cycles.

Maraging steel MS1 material
Sun et al. [25] evaluated the powder characteristics and mechanical behavior of components fabricated from virgin and 113-time recycled maraging steel powder. They observed that there was no significant change in PSD and particle shape throughout reuse cycles. However, spatter particles caused oxide inclusion on the top surfaces of printed parts. They also discovered that as-built specimens created from 113-time recycled powder had nearly identical microstructure and mechanical properties to those produced from new powder (940 MPa yield strength, 1127 MPa ultimate tensile strength, 11% elongation, and 47.5 J impact fracture toughness at room temperature). As reported in [26,27], mechanical properties depend on the heat treatment and the printing orientation. For example, after applying the heat treatment methodology suggested by EOS [28], the hardness of maraging steel parts increased substantially from 30 to 57 HRC.
Similarly, to investigate the effect of powder contamination on the mechanical properties of 18Ni-300 maraging steel components, Gatto et al. [28] printed specimens from two separate batches. Batch 1 had a cross-contaminated raw powder that contained Ti-Al oxides, whereas batch 2 contained virgin maraging powder. They discovered that the presence of contaminants had no effect on static tensile characteristics. However, fatigue endurance was dramatically compromised by contaminants (batch 1), while parts printed from clean powder (batch 2) had a fatigue life equivalent to forged specimens.
Finally, M. Horn et al. [29] discovered the presence of copper alloy CW106C foreign particles in the maraging steel powder feed stock, which happens during material changes inside the L-PBF machine or when using L-PBF multi-material machine. This cross-contamination decreased the parts' tensile properties and initiated cracks failure if the proportion of CW106C exceeded 2%.
The above-mentioned research reveals that, overall, powder reuse improves powder density and flowability, with a slight change in powder size and morphology. The parts' tensile properties remain generally constant, but fatigue performance is negatively impacted by powder reuse. Consequently, the physical and chemical powder characteristics evolve inevitably, although slowly, when the powder passes multiple times through the printing and sieving process, which could also influence the mechanical properties of the final part. However, powder behavior differs from one material to another, because each material has its specific chemical and physical properties and materials react differently to additive manufacturing, especially to the recycling process.
Despite all of the studies published to date, the impact of recycling on powder properties and on static and fatigue behavior remains unclear and ambiguous. Very few studies have addressed the subject, especially with respect to maraging steel powder. This paper aims to investigate the effect of powder recycling on particle size distribution (PSD), particle morphology, powder apparent density, and the tensile and fatigue properties of maraging steel parts. We also conducted a fractography study to observe the fracture surface and investigate the causes of parts failure. Through this research, an effort is made to link the mechanical testing results and changes in powder properties by applying a precise experimental methodology and by monitoring the important variables.

Material and powder characterization
The powder used was the EOS Maraging Steel MS1 (18% Ni Maraging 300). Its composition is detailed in Table 1 according to the material data sheet (EOS art-no.9011-0016) [30]. The study started with 20 kg of fresh powder. The quantity was chosen to allow eight printing cycles and all the specimens needed to reach the end of the experiments within optimal time and cost efficiency. The methodology was inspired from many of the above-mentioned articles [9,14,20,23,24,31,32]. After each cycle, all unfused powder was sieved manually with an 80-µm sieve.
The powder samples were analyzed with a Malvern Panalytical Mastersizer 3000 particle-size analyzer equipped with a Hydro LV module following Rayan et al. (submitted). The stirrer speed was set to 3000 rpm, a speed sufficient to keep the particles in suspension. Prior to the measurements, the sample was submitted for 60 s to Hydro LV ultrasounds at 25% power to help disintegrate aggregates. Three consecutive measurements of 30 s each were collected for each sample. The average statistics of the three measurements were computed (the coefficient of variation in Dv10, Dv50, and Dv90 of the three measurements was always < 1%, indicating a good sample dispersion). The Mastersizer general purpose optical model for nonspherical particles was employed with a refractive index of 2.757 and absorption 1.0 for stainless steel (values taken from the Malvern Panalytical database included with Mastersizer 3000 software). Instrument performance was confirmed using Malvern Panalytical's QAS4002 Quality Audit Standard.
Powder morphology was analyzed using a NanoImage SNE 4500 M scanning electron microscope (SEM). Apparent powder density was measured with an electronic balance and 25-ml plastic graduated cylinder.

Machine/printing parameters
The parts were printed on an EOS M290 400W machine in a nitrogen atmosphere with less than 1.3% oxygen concentration, the gas flow differential pressure is 0.7 mbar, and a chamber temperature of 40 °C. The MS1 040 performance M291 2.00 EOSPRINT template was used: laser power = 285 W, beam offset = 20 µm, laser speed = 960 mm/s, layer thickness = 40 µm, hatch space = 110 µm, and laser pattern = stripes rotated at a 47° angle with a 30° restriction angle at each next layer.

Recycling methodology
As in many other studies, for the recycling methodology we employed collective aging strategies to simulate the worst-case scenario of powder reuse [14,20,23,24,[31][32][33]. Indeed, our recycling approach maximized powder degradation over reuse cycles. To accomplish this, we began by filling the dispenser with 20 kg of new powder and, for each print, we designed a build plate that required almost all of the powder from the dispenser. This way, most of the powder was exposed to the L-PBF process at each printing cycle. After each cycle, we mixed the remaining powder from the dispenser with the unfused powder in the collector bin (rehomogenization). After removing the build plate from the machine, we sieved all unfused powder with an 80-µm mesh, and reloaded the dispenser with the sieved powder. At each cycle, we reduced the build height by lowering the height of the print. The height was based on the quantity of powder remaining in the dispenser. This cycle was repeated eight times until there was not enough powder in the dispenser to run the next print. The applied process is summarized in Fig. 1.

Build description and tracking parameters
Each build plate had four tensile specimens and 17 fatigue specimens printed horizontally at 0°. The horizontal direction was chosen over the vertical direction because it reduces the printing time and the volume of powder required to complete the study. We also considered that fatigue failures are more likely to occur in the central region (small volume) due to the shape of the fatigue samples. For this reason, the fatigue behavior should not be too strongly related to the printing direction. This assumption will have to be further validated in a future study.
The geometry of the printed specimens after machining conformed to ASTM E8-a16 and ASTM E606, respectively. The final thickness of the specimen was 3 mm for both types. According to Pay et al. [34], the mechanical properties of parts located near the collector bin are not as good as those placed near the dispenser. Moreover, Arash Soltani-Tehrani [17] reported that the impact of powder reuse on fatigue performance was especially significant for specimens printed near the collector bin. For these two reasons, the specimens were positioned close to the collector bin. We added four sacrificial lattice structures (diameter = 40 mm and cell size = 5 mm, height = adjustable) to speed up powder contamination and, at each print, use most of the powder from the dispenser bin (see Fig. 2 and Table 2).
We monitored three important factors that allowed to accurately describe the eight-cycle printing process and to gain a better understanding of the effect of powder recycling: 1. Repetition factor: N is the number of cycles.

Time factors: T las (h) is the laser operating time, and T mach
(h) is the machine operating time.

Quantity factors:
• Q con is the powder melted and wasted after each cycle. • H max is the build's maximum height.
After eight reuse cycles, the powder was exposed to the laser for 46 h over a total printing period of 81 h and the total amount of powder consumed was 9 kg.

Summary of powder consumption
From the 20 kg of virgin powder loaded in the dispenser, 5.63 kg was melted (28.2%), 0.78 kg (3.9%) was collected on the 80-µm sieve and discarded, 155 g (0.8%) was collected for analyzing powder morphology and PSD, and approximately 2.43 kg (12.2%) of the powder was loss during   cleaning and handling, leaving 11 kg (55%) of powder after eight printing cycles (Fig. 3).

Post-processing
After printing, the specimens were subjected to an aging heat treatment in an ambient atmosphere. According to D. Kim et al.'s methodology [28], the maraging steel parts printed horizontally only require aging treatment at 450 °C for 6 h, followed by air cooling to improve their mechanical properties without using solution treatment. The aged tensile and fatigue specimens were machined in two steps using a computer numerical control (CNC) machine. The tensile and fatigue specimens geometries satisfied ASTM E8-a16 [35] and ASTM E466-15 [36] requirements, respectively, and had a final thickness of 3 mm, as shown in Fig. 4 and Fig. 5. All other variables of build design, printing parameters, sieving process, heat treatment, and machining were carefully monitored to ensure consistency.

Mechanical testing
The surface roughness test was conducted using a Mitutoyo FORMATRACER SV-C3100/ 4100. The profile roughness arithmetical mean deviation Ra was measured on three different fatigue as-built specimens for each printing cycle except the fourth. Five 2.5-mm-long linear measurements were performed in parallel direction with the specimen axis according to the OLDMIX standard.
The tensile tests were conducted on three tensile specimens for each cycle, with a cross-head speed of 0.5 mm/ min at room temperature, according to the ASTM E8 Standard [35], using an MTS 810 servo-hydraulic machine, and employing an MTS extensometer model to measure tensile strain. Stress-controlled fatigue tests were performed using an MTS 810 machine in ambient atmosphere, applying a reversed sinusoidal load, with ratios K t = 1 and R = 0.1, adjusting the frequency to 25 Hz, and terminating at 2 × 10 6 cycles, according to ASTM E466-15 [36]. The mean fatigue stress amplitude S D was determined based on the staircase    method described in the French standard (Afnor) NF A03-405 [37] and Ekaputra et al. [38]. Lastly, the specimens' surface fractures were also analyzed using a NanoImage SNE 4500 M scanning electron microscope after cleaning the specimens with ethanol.
Elemental composition of printed specimens was determined using a Bruker Esprit Compact energy-dispersive X-ray spectrometer (EDS) coupled to the NanoImage scanning electron microscope. Each spectrum was acquired for 120 s of live time at an accelerating voltage of 20 kV. Elements were automatically identified and quantified by the Esprit Compact software, and results were normalized to 100%.

Powder particle size distribution
Over eight powder reuse cycles, Dv90 increased linearly with the number of cycles (y = 0.70x + 59.8; R2 = 0.87), resulting in an 11.4% increase of Dv90 (Fig. 6). There was no significant linear regression between Dv50 and Dv10 and the number of cycles (p > 0.05). Despite this, the maximum difference relative to virgin powder was an increase of 6.5% and 9.1% for Dv50 and Dv10, respectively (Fig. 6). No reason was found to explain why there is a significant linear regression for DV90 and not for DV10 and DV50. Moreover, Fig. 7 illustrates the evolution of powder particle size distribution (PSD). The powder had a log-normal size distribution with a median (Dv50) of 33.7-36.0 µm. A very small proportion of very fine particles < 9 µm (0.7-0.9% of total volume) was present in all samples. A small proportion of large particles in the diameter range 200-1000 µm (1.4-3.0% of total volume) were likely aggregates in the medium when the PSD was measured, since particles in this size range were not observed using the electron microscope. The PSD shifted slightly toward larger particles after eight reuse cycles (Fig. 7). The larger proportion of coarser particles after powder reuse could be attributed to the formation of elongated particles, satellites, aggregate particles, and spatters, which are produced by the lattices' cylinder and are small enough to pass the 80-µm sieve. The apparent density remained stable, at 4.64 ± 0.03 g/ml over eight reuse cycles.

Powder morphology
The morphology of virgin powder is globally spherical with the presence of some satellites and aggregate particles formed during the powder production process (Fig. 8A) [39,40]. The powder recycled five times (Fig. 8B) and eight times (Fig. 8C) exhibits shape degradation caused by the presence of aggregated and deformed particles, "clip-clap," elongated particles, broken particles, and shattered particles.  Figure 10 shows the results of the 3 measurement tensile properties for each specimen over eight reuse cycles. The tensile properties changed slightly (less than 5%): ultimate tensile strength (UTS) decreased by 72 MPa (3.6%), from 1985 to 1913 MPa, over eight print cycles, yield strength at 0.2% strain (0.2%YS) decreased by 58 MPa (3%), the Young module dropped by 4.5 GPa (2.4%), and elongation at break remained stable with a low average value of 1.5% ± 0.1%. These slight variations could not be only attributed to powder recycling. Such small variations are common in tensile testing. Figure 11 represents the evolution of the stress amplitude Sa applied to specimens printed in cycles 1 and 8 according to the staircase method. Five specimens for cycle 1 and seven specimens for cycle 8 reached 2 × 10 6 cycles (indicated by O). The number of failed specimens (indicated by X) is four for cycle 1 (44% failure) and seven for cycle 8 (50% failure). The parameters of the Dixon and Mood [37] approach are summarized in Tables 3 and 4.

Fatigue results
Even though the number of specimens was increased to 14 for cycle 8, the estimate of the stress standard deviation was still invalidated. According to the French standard (Afnor) NF A03-405 [37], the F BA was less than 0.3 ( F BA = MB−A 2 M 2 < 0.3). For analyzing the results of the 14 specimens in cycle 8, we used the simplified equation recommended by Ekaputra [38] and Snyder [41] to estimate standard deviation. According to the simplified equation, the standard deviation is equal to 0.53 × ΔS when F BA < 0.3, where ΔS is the step size. The fatigue test results are summarized in Table 5. The mean stress amplitude for cycle 1 is S D = 332 (MPa), while for cycle 8 the stress is S D = 323 (MPa). We noted that the mean stress amplitude was reduced by 2.7% from 332 to 323 (MPa), while the standard deviation decreased from 41 to 10.6 MPa after recycling the powder eight times.

Fractography and EDS analysis
SEM microscopy was employed to determine crack initiation sites for specimens of cycles 1 and 8. When applying the staircase method for specimens issued from virgin powder (cycle 1), we applied the same load 312 MPa for three specimens (1, 3, and 5) and noted that specimen 1 failed while specimens 3 and 5 survived 2 million cycles. As a result, we investigated the surface fracture of the first specimen. As seen in Fig. 12a, the final breaking surface seems to be brittle and localized near the specimen's edge  Fig. 12b. Figure 13a also demonstrates the surface fracture of the eighth specimen printed in cycle 1, which failed under the maximum load 372 MPa. Figure 13 b presents lack of fusion defects and cavities. The fracture surfaces seem to be clean and contain few LOF defects ( Fig. 12 and Fig. 13). As mentioned in the technical white paper [42]: "'river marks' are not always observed, and a brittle fracture may be relatively devoid of any features from which to identify locations of initiation or crack propagation direction." Using the staircase method for specimens produced from cycle 8, we applied the same load 322 MPa to two specimens: the first specimen failed early at 63,081 cycles, while the second specimen survived 2 million cycles. Subsequently, we chose to investigate the probable defect of the failed part. Figure 14a shows an overview of the fracture surface at × 36 magnification factor. The final breaking surface of the specimens is bounded by the red dotted polygon. The fracture surface of this specimen shows a substantial presence of defects that could lead to crack initiation and failure: Fig. 14b and Fig. 14c represents, respectively, a partially melted particle defect (indicated by a green circle) and inclusion defect (indicated by an orange circle). Figure 14d illustrates a partially melted particles defect and LOF defects (indicated by a blue circle). Figure 14e reveals the presence of an inclusion defect which has a spherical form and is highly agglomerated (indicated by an orange circle), whereas Fig. 14f illustrates this inclusion in the  2  4  8  16  322  1  3  3  3  302  0  0  0  0  Total  _  7  11 19   Table 6 Elemental composition of the inclusion presented in Fig. 14c and  backscattered electrons mode (BSE). The melted metal appears gray, while the inclusion is black (indicated by an orange circle), suggesting that this inclusion could be rich in carbon. Tables 6 and 7 present the results of the EDS analysis applied to this specimen. The chemical composition of inclusions shown in Fig. 14c and Fig. 14e and f are presented in Tables 6 and 7, respectively. As can be noted in Table 6 that is related to Fig. 4c, the composition of the inclusion is quite similar to those measured in the metal regions, which confirms that this inclusion is a metal particle.
Consequently, this metal inclusion could be a spatter that was previously present in the recycled powder or produced during the printing process.
However, the inclusion displayed in Fig. 14e and f shows a relevant presence of 76.4% of carbon and a small percentage of iron Fe 10.1%, while the metal disposed around contain less carbon C 24.5% and a substantial percentage of iron Fe 40.3% as reported in Table 7. Thus, we can conclude that this inclusion is a carbon inclusion.
Moreover, the fatigue specimens 4, 8, 10, and 12 printed in the eighth reuse cycle failed very early under the maximum stress load 342 MPa. The broken surface of specimen 10 was examined. As can be seen in Fig. 15, more internal defects, such as lack of fusion, partially melted particles, and gas pores (indicated by a purple circle) are apparent on the specimen's surface. This surface fracture has a high prevalence of fusion defects, generating many gaps and pores, and could be the main reason for fracture initiation and failure.

Discussion
The powder PSD was only slightly affected by powder recycling; the most significant effect was an 11% increase in Dv90 after 8 reuse cycles compared to the virgin powder. Recent studies have shown that particle size distribution PSD became narrower after reuse cycles, indicating a decrease in the proportion of fine particles involved in the melting process, followed by a drop in the quantity of larger particles, which were reduced progressively by the sieve device after reuse [7-9, 12, 14, 20, 24, 25]. Soundarapandiyan [13] and Richard and O'Leary [7] confirmed our findings by reporting the same behavior of the Ti-6Al-4 V powder PSD, which shifted rightward after five reuse cycles. Furthermore, Sutton et al. [17] found comparable results using AISI 304L stainless steel powder, demonstrating an increase in particle size after five printing cycles. V. Contaldi [14] discovered that the PSD of martensitic PH1 powder shifted to the right due to agglomeration as the smallest particles passed through the sieve device, resulting in an increase in the quantity of larger particles after nine times reuse. These findings align with our observations. However, the PSD of austenitic GP1 powder was narrowed and moved to the left, showing a decrease of fine and larger particle proportions. They concluded that the differences in PH1 and GP1 PSDs could be attributed to noise, handling, and manual sieving, all of which could induce uncertainty.
Powder particles shape was also drastically affected after eight reuse cycles with an increase in particle roughness and the presence of a large proportion of partly fused powder. These changes in recycled powder morphology are due to the presence of spatters (B. Szost et al. [43]; Renishaw 2016 [10]) and to laser heat exposure that sinters the particles and induces their mechanical deformation (Shanbhag [37], Renishaw 2016 [10] and Popov [4]). Even if most spatters are eliminated by sieving at 80 µm, smaller spatters pass through the sieve and influence the shape of recycled powder (Rock et al. [23]; Sutton et al. [31,42]).
Normally, the powder size should be reduced when the powder passes multiple times through the sieve device; however, in this study, the powder size increased mildly. A possible reason could be due to its shape alteration caused by laser heat and the formation of aggregates that could pass through the sieve device.
As Pal et al. [34] noted, the recoater deposits fine particles over the build plate and drags the larger ones over the collector. So, fine particles are more likely to be consumed during the melting process, which explains the reduction of their proportion, while the percentage of larger particles enhances over reusing cycles.
Although the alteration of powder size and morphology, the powder apparent density remained constant 4.64 ± 0.03 g/ml after eight printing cycles, as was also reported by V. Contaldi [15]. However, F. Del Re et al. [20] noted a slight increase in the tap density of aluminum AlSi10Mg powder over eight reuse cycles.
Regarding the surface roughness of as-built parts, Ra showed a mild variation around 7.62 and 8.74 µm over eight printing cycles. According to F. Del Re et al. [20], the surface roughness of AlSi10Mg parts remained constant over eight cycles. However, V. Seyda [12] discovered an important increase of 33.7% in the horizontal 10-point mean roughness Rz for titanium parts produced 12 times. The surface became rougher due to the increased proportion of larger particles after the reuse process.
As previously demonstrated, the tensile properties of maraging steel parts reduced slightly by 3% after 8 reuse cycles. Similar result was reported by Del Re et al. [20], they concluded that the yield and tensile strength of AlSi10Mg components dropped by 10 MPa, while elongation at break was around 18% after eight reuse cycles. This slight decrease was attributed to the powder storage and recycling methods (collective aging). Furthermore, Sun et al. [25] concluded that parts printed from maraging steel powder reused 113 times had identical static properties compared with ones produced from virgin powder.
However, Seyda et al. [12] reported a slight amelioration of tensile properties after recycling titanium powder 12 times due to the increase in the percentage of oxygen oxide within printed parts. Moreover, Popov et al. [4] found that both UTS and YS increased slightly after reusing titanium powder 69 times, while elongation was substantially reduced because of residual surface oxidation and surface defects of the recycled powder particles.
Many studies have confirmed our findings and reported the impact of powder reuse on the fatigue performance of printed parts. Del Re et al. [20] discovered a decrease of 10.3% on fatigue strength after reusing AlSi10Mg powder eight times, while Popov et al. [13] found that Ti-6Al-4 V part fatigue endurance was dramatically impacted after reuse due to porosity, partly melted particles, and lack of fusion (LOF) defects. However, Carrion et al. [6] and Arash Soltani-Tehrani [17] discovered that powder recycling enhanced, respectively, the fatigue properties of Ti-6Al-4 V and 17-4 PH stainless steel machined components in the high cycle regime, owing to enhanced powder flowability over powder reusing, which results in a more uniform powder bed layer and potentially fewer pores within parts. The unmachined specimens were not impacted by the reuse cycles. It was concluded that fatigue performance depends substantially on the surface condition of printed parts, and the effect of powder reuse is remarkable only on machined parts. Furthermore, the fractography and EDS analyses revealed that parts printed from new powder did not show a significant presence of internal imperfections. Significant pores, carbon inclusion, and LOF defects were detected on the surface fracture of specimens printed from reused powder, which could be the result of powder degradation. The spatters generated by the melting process are deposited in the powder bed, which will undoubtedly affect not only the powder quality but also creates pores and inclusions within the melted part [19,33]. Additionally, S. Pal et al. [34] reported that powder size and morphology affect the formation of melt pools during the melting process by creating more spatters. The proportion of porosity inside parts increased, which reduced the parts' mechanical properties. Moreover, Carrion et al. [8] reported that reusing powder led to the creation of smaller pores inside titanium parts. However, Soltani-Tehrani [17] reported that both 17-4 PH stainless steel parts printed from new and reused powder exhibited internal defects, such as LOF defects and gas pores, but powder reuse significantly decreased pore size, which ameliorated part fatigue performance. Furthermore, Popov et al. [4] reported that (HIP) heat treatment is an effective way to reduce gas pores; however, it does not eliminate LOF defects inside parts made from reused powder. Moreover, both A. Gatto [28] and H. Sun et al. [25] discovered that the fracture surfaces of maraging steel parts, produced from cross-contaminated powder and from recycled powder, respectively, contained a substantial amount of Ti-Al oxide inclusions, cavities and pores. The effect of this inclusion was more relevant with respect to part fatigue performance, while tensile mechanical properties seem to be less sensitive to these defects thanks to material ductility.
After powder reuse, mean fatigue strength decreased by 2.7%, from 332 to 323 (MPa), and the stress standard deviation became narrower (± 10.6 MPa). This would suggest that all parts printed in cycle 8 exhibit similar fatigue behavior, while parts made from fresh powder had a wide standard deviation (± 41 MPa), possibly caused by a higher probability of defects. The fractography and EDS analysis show that fusion defects are hugely present in parts made from eight times recycled powder, which could explain why the stress standard deviation of these components is narrower. Nevertheless, it is important to mention that fatigue behavior is intrinsically uncertain, since it depends on the surface condition, the existence of imperfections within the part, and the part's physiological properties, especially for AM components [20].
Indeed, as investigated above, parts produced from reused powder exhibit a relevant presence of pores, LOF, and carbon inclusion defects, which could also explain this slight decrease in the parts' mechanical performance. However, this small reduction of approximately 3% in tensile properties and fatigue performance could be ascribed to the rehomogenization of reused powder (mixing reused powder with dispenser's powder after each cycle): the reused powder that was spread over the build plate and over the collector bin was re-homogenized with approximately 3 kg of powder remaining in the dispenser after each print, which made it possible to slow down the degradation effect on its properties over reuse time, although there was a slight alteration of powder size and morphology caused by the L-PBF process and lattice parts.

Conclusion
Maraging steel powder was reused eight times in the L-PBF process in this study using a recycling approach which maximized powder degradation. The effect of powder reuse on particle size distribution, particle morphology and apparent density, as well as on the evolution of the surface roughness and mechanical properties of manufactured parts, were investigated in order to bridge the gap in knowledge regarding the impact of maraging steel powder reuse, particularly on part fatigue performance. The following conclusions can be made from the experimental results and observations: • The alterations of the powder properties that result from recycling are the particle size distribution and the morphology of powder particles. Particle size distribution (PSD) shifted slightly toward larger particles over eight printing cycles, with D90 increasing linearly by 11.4%. The electron microscope images confirmed the degradation of particle shapes over reuse cycles by the presence of aggregates, deformed and elongated particles, "clip-clap" particles, and broken and shattered particles, which could pass through the 80-µm sieve and contribute to an increase in the proportion of large particles. The slight increase in particle size had no effect on the apparent density of sieved powder which remained stable at 4.64 ± 0.03 g/ml through all eight cycles. • The arithmetic surface roughness Ra of as-built parts showed a slight variation of average 8.17 ± 0.42 µm over eight printing cycles, indicating that the impact of powder reuse was negligible on the surface roughness of maraging steel parts. • The 8 printing builds were consuming approximately 9 kg (45%) from 20 kg of maraging steel virgin powder, which was loaded initially in the feeder and passed through eight collective aging cycles, taking a total of 81 h of printing time and 46 h of laser operating time.
The tensile properties were mildly affected over reuse cycles: ultimate tensile strength (UTS) decreased by 3.6%, from 1985 to 1913 MPa, yield strength (0.2% YS) dropped by 3%, from 1917 to 1859 MPa, while the Young module was reduced by 2.6%, from 189 to 184 GPa, and elongation at break stayed stable around 1.58% ± 0.1%. This minor effect cannot be attributed to powder reuse, as variations of this magnitude are common in tensile testing. • Generally, fatigue behavior remained constant despite a minor decrease over reuse cycles. The mean stress amplitude decreased by 2.7%, from 332 to 323 (MPa), while the standard deviation decreased from 41 to 10.6 MPa after recycling the powder eight times. This could be attributed to the substantial presence of fusion imperfection defects inside parts produced from reused powder. • Fractography and EDS analysis proved that the fracture surface of specimens printed from virgin powder contained fewer internal defects compared with specimens printed from recycled powder, which had a relevant presence of pores, partially melted particle defects, gas pores, lack of fusion (LOF) defects and carbon inclusion defects. These defects were most likely formed as a result of the alterations in the powder's properties and the effect of the melting process during the reuse cycles.