Variability in mechanical properties of additively manufactured 17-4 PH stainless steel produced by multiple vendors: insights for qualification

In applications where a combination of good strength and corrosion resistance is required, 17-4 precipitation hardenable (PH) stainless steel is a common material choice. This alloy is traditionally processed through a combination of casting, rolling, and machining. A variety of heat treatments are used to anneal and harden the material via precipitation strengthening. While additive manufacturing (AM) removes many geometric design constraints from these traditional forming processes, until recently, structures fabricated via laser powder bed fusion (L-PBF) were porous and contained undesirable columnar grain structures that contributed to unpredictable and anisotropic mechanical properties. Recent advances in L-PBF processing technology including improved gas flow, powder atomization, and print parameter optimization enable printing of high-quality 17-4 PH with properties that are comparable to traditionally processed material. However, it is yet to be determined whether these properties can be reliably reproduced across various machines and if machine-agnostic material property baselines can be established based on data available thus far. If baselines can be established, their implementation should allow for identification of machines that generate non-conforming material. In this work, we evaluate the consistency of mechanical properties in L-PBF 17-4 PH produced by six vendors with the ultimate goal of establishing mechanical property baselines, which is fundamental to modern qualification paradigms. We find non-conforming data from two vendors by examining anomalies in mechanical response (e.g., transformation induced plasticity effects) and demonstrate that typical sources of variation can be detected using qualification testing protocols. Ultimately, after standard solution annealing and heat treating, the microstructure and mechanical properties across vendors converged with very few, easily explainable exceptions. In particular, powder atomized in nitrogen promoted formation of retained austenite that lead to a yield point phenomenon in as-built conditions, and high surface roughness from as-built surfaces reduced the fatigue strength. However, with conventional post-processing heat treatments and surface polishing, AM 17-4 PH behaved comparably and consistently to conventionally processed material.


Introduction
Fueled by recent technological advancements, laser powder bed fusion (L-PBF) has matured substantially in the last 10 years and is capable of reliably producing high-quality materials that exhibit good mechanical, corrosion, and functional performance [1], and showing good process stability [2]. The L-PBF process offers many benefits over conventional manufacturing, including the ability to manufacture complex geometries, minimize tooling requirements, and enable rapid prototyping [3]. Altogether, this has motivated a variety of 1 3 industries and government agencies to explore L-PBF as an alternative processing route for the construction of application-specific components. Industry vendors, universities, and government facilities have acquired L-PBF machines and are fabricating materials and parts across the globe in ways once strictly reserved for a few, well-capitalized foundries and rolling mills. Fueled by quick demand for L-PBF parts and the decentralization of material fabrication, work in the field is focused on developing criteria for part and/or machine qualification, which is currently an expensive and burdensome process [4,5]. For instance, NASA released MSFC-STD-3716 and MSFC-STD-3717 in 2017 to set evaluation standards for AM parts in spaceflight applications, and Naval Sea Systems Command (NAVSEA) released Technical Publication S9074-A2-GIB-010_AM-PBF in 2020 to set requirements for parts processed by L-PBF for general applications. Other standards are currently in various states of draft and review, such as the Society of Automotive Engineers' AMS 7032 [6] and American Welding Society D20.1 [7], which sets requirements for AM machines to produce materials in compliance with materials specifications. Taking this further, traditionally the prior foundries and rolling mills were responsible for setting material specifications, but in the field of AM, such roles have shifted to the primary machine consumers, inherently introducing inconsistencies and uncertainties. Thus, similarly to conventional forming processes, AM requires standardization that is simple and cost-effective.
The requirements outlined in the aforementioned documents differ somewhat but generally include (1) obtaining a baseline set of mechanical properties for specific processes and alloys, (2) machine-specific qualification (carried out by manufacturers) to demonstrate that material meets the baseline requirements, (3) part-specific qualification for demonstrating production feasibility, and (4) ongoing quality assurance plans for production.
As a central component of qualification, property baselines are generated from large test matrices involving numerous powder suppliers, machines, vendors, part lots, and mechanical tests in a framework similar to MMPDS baselines [8]. As such, statistically meaningful mechanical property design values (e.g., A-basis [T 99 ], which refers to the 95% lower confidence bound on the 1st percentile of a property value) can be established that are relevant to parts despite their processing histories. During machine qualification, for instance, vendors must demonstrate that they can manufacture material that exceeds the T 99 requirement.
Fundamental to this qualification paradigm are several assumptions about the variability in the mechanical properties of a given material. First, the mechanical response of material produced using different powder suppliers, machines, vendors, and lots come from a common statistical family, such that the material baseline reflects the intrinsic behavior of the material rather than intricacies of specific manufacturing processes. Second, variation in mechanical properties is relatively limited, such that T 99 design values are not overly conservative and are conducive for the design of demanding load-bearing components. Third, outlier builds (which may occur due to poor quality control, build faults, material contamination, etc.) can be reliably identified in comparison to the baseline dataset during machine qualification and quality assurance plans. Similarly, sub-standard builds should occur with sufficient rarity that qualification (and quality assurance) does not become cost-burdensome to vendors.
As recently as 5 years ago, the aforementioned assumptions were not necessarily satisfied. For instance, several years have passed since the foundational study by Boyce et al. examining the variation in mechanical performance across 9 builds of 17-4 PH stainless steel using a large number of samples (n = 1065) [9]. Due to poor and inconsistent build quality, extreme variation in mechanical properties yielded unacceptably low T 99 design values that were smaller by a factor of 2 compared to the minimum allowable design values for cast 17-4 PH (AMS 5344) [9,10]. Since then, however, improvements in AM hardware [11] and process modeling [12,13], as well as a better understanding of the physics of defect formation [14,15], have led to improved build quality and consistency. For example, the once high porosity levels (~1%) in 17-4 PH [9,16] are now more typically 0.01-0.1% using modern L-PBF hardware [17]. In fact, as shown in Fig. 1, modern L-PBF processes can manufacture 17-4 PH with properties comparable to, or even in excess of, its certain conventionally processed, heat Year-to-year improvement in mechanical properties of 17-4 PH. Data is shown for specimens built using 17-4 PH in the non-heat treated condition. Improvements in L-PBF hardware and materials processing lead to increases in the maximum achievable mechanical properties, although continued use of legacy L-PBF equipment contributed to lower mechanical properties in more recent publications. Dashed line indicates minimum value for forgings heat treated to H1075 condition. Data from ref. [18][19][20][21][22][23][24][25][26][27][28] treated counterpart [18], instilling confidence in the reliability of L-PBF material for structural applications.
Despite this, recent advances in processing and mechanical performance motivate a re-evaluation of the feasibility of producing qualified, structural components using L-PBF. In this work, we characterized the microstructure, quasistatic tensile, and fatigue properties of 17-4 PH produced by six different vendors (including JHU/APL) in order to evaluate the variability in the quality and mechanical response of commercially available 17-4 PH. Testing was performed on each sample set using a protocol that resembled the machine qualification requirements outlined in currently available standards, thus providing insight into the viability of these qualification protocols. The material was manufactured on five different models of L-PBF machines, using vendor-defined laser parameter recipes, and included both argon-and nitrogen-atomized powder; altogether, this highlighted a diverse industrial base that is capable of producing AM material that could satisfy qualification requirements. Compared to data available in the literature, the processing history of all samples was well documented to the best of our ability based on proprietary restrictions and met sample size requirements to perform robust statistical analysis. Most importantly, we control differences in processing histories, sample geometry, and testing conditions, which are prevalent across publically available studies and hinder comparisons of mechanical properties across publically available studies [20,28,29]. In this work, we focus on 17-4 PH, which is a relatively mature L-PBF material system [30,31] with a desirable combination of mechanical properties and corrosion resistance, making it a candidate alloy for structural applications in naval, aerospace, and power generation industries [32].

Experiment
In order to address inconsistency in material lots, properties, and AM machines, material was sourced from five different vendors as well as fabricated by JHU/APL, as shown in Table 1. One challenge inherent in this investigation is obtaining a full description of processing conditions used by commercial vendors-as would be expected in an academic setting-due at times to safeguarding of proprietary information. However, this reinforces the importance of evaluating whether "stock" material from different vendors can yield consistent properties.
Vendors were asked to manufacture builds using (1) 17-4 PH powder from their typical supplier, (2) their standard AM machine, and (3) their standard laser parameters for 17-4 PH on the respective machine. Vendors were asked to refrain from post-processing treatments (e.g., surface modification, heat treatment) exclusive of de-powdering of the builds and removal from the build plate. Lastly, vendors were asked to use argon gas-atomized feedstock powder. Table 1 summarizes the sample processing history which includes details about the laser parameters used. As shown, samples from vendor 5 were returned having been processed with nitrogen-atomized powder (all other vendors were verified to have used argon). The laser power used was not provided by vendors 1-3 (data is displayed as maximum power offered by the machine in Table 1). Several vendors used machines from EOS; vendor 2 used an EOSINT M270 (model was introduced 2004) which predates by two generations the current M290 machine used by vendor 3 and JHU/APL.
Each vendor directly manufactured prismatic builds, from which tensile samples with their long axis in both the Z direction (vertically built) and XY direction (horizontally built) were machined in-house for testing. Vendors were instructed to provide samples in the non-heat treated condition. Half of the tensile samples built in the Z direction were heat treated in-house by solution annealing and aging to H900, while the other half were left nonheat treated for testing. The solution step was performed at 1040 °C for 30 min in an argon atmosphere followed by air cooling. The H900 aging step was performed at 482 °C for 1 h in an air atmosphere followed by air cooling. Blocks from vendor 4, however, were returned having been stress relieved prior to detachment from the substrate, then subsequently shot peened and H900 aged. Uniaxial tensile samples were tested in-house according to ASTM E8 [33]. Testing was performed on an MTS 30/G test frame with a 130 kN (30 kip) load cell and 12.5 mm gauge MTS extensometer to measure strain. Samples were tested with a controlled crosshead speed of 0.04 mm/min (nominal strain rate of 0.015 min −1 ) until yield, then the crosshead speed was increased to 0.12 mm/min (strain rate of 0.045 min −1 ) until failure. The extensometer accurately measured strain until necking. When the sample began necking, the extensometer measurements were invalid if extension occurred outside of the extensometer gauge section. Elongation after fracture was measured by fitting the two halves of the sample together and measuring the gauge length post-failure. Yield strength was calculated according to ASTM E8 [33]. Select samples were subjected to further uniaxial tensile testing using stereoscopic digital image correlation (DIC) for strain mapping and X-ray diffraction (XRD) analysis. For strain measurement via DIC, a uniform white coating was applied to the gauge section followed by black spray paint to form a random speckle pattern. Samples were tested with a servohydraulic Instron 5984 load frame equipped with a 50 kN load cell and a displacement rate of 0.381 mm/min (0.015 in/min). A GOM ARAMIS 3D DIC system with two 35 mm focal length lenses, a 150 mm camera separation, and 25° camera angle was used to acquire images at a frequency of 5 Hz. The images were post-processed in ARAMIS Professional software. The tracking facets were of size 0.5 mm, the facet spacing was 0.1 mm, and the interpolation size was 0.2 mm. The gauge length for the DIC virtual extensometer was chosen to match the 12.5 mm gauge length used in conventional tensile testing as mentioned above.
Following tensile testing, the grip regions (i.e., undeformed) were sectioned and mounted in the Z orientation to examine the microstructure via light optical microscopy in the build direction. Standard metallographic procedures were used to polish the mounts down to 20-nm colloidal silica, followed by chemical etching in Ralph's reagent for 20 s to reveal the grain boundaries. Density of selected samples was characterized using microfocus X-ray computed tomography (North Star Imaging X-50) with 12 μm/voxel spatial resolution, which found density of all samples to exceed 99.9%.
For fatigue testing, vendors manufactured cylindrical coupons in the Z direction, and surfaces were left unmachined for testing. Similarly to the tensile samples, while vendors were instructed to provide fatigue samples in the non-heat treated condition, fatigue samples from vendor 4 arrived stress relieved prior to detachment from the substrate, shot peened, and H900 aged.
JHU/APL printed additional fatigue samples in-house to investigate the effect of surface roughness, heat treatment, and build orientation (vertically built vs. 45° incline).
Samples were removed from the build plate using either wire electrical discharge machining or a band saw. Samples were left either unmachined, machined, or polished for testing. Only polished samples were heat treated with a solution anneal and H900 aging; all others were tested in the nonheat treated condition. Samples that were machined were lathe machined in the gauge section to a nominal surface roughness of R a = ~1 μm and contracted to Touchstone Testing Lab (Triadelphia, WV) for fatigue testing using ASTM E466. For the polished samples, Touchstone Testing Lab performed linear polishing in the gauge section using an 800 grit abrasive belt to remove residual circumferential tooling marks from the turning process. All fatigue tests were performed using an MTS 810 servo-hydraulic test frame with MTS 647 hydraulic wedge grips under load-control, a frequency of 20 Hz, a fully reversed stress ratio of R = − 1, and stress amplitudes ranging 10-80% of the yield strength as determined by prior quasi-static tensile testing, and concluded at 5,000,000 cycles (i.e., "run-out").
The surface roughness of as-built and machined fatigue samples was measured on a Keyence VK-X3000 series 3D surface profiler at 10× magnification based on a 12-mm linear evaluation in the gauge length of the fatigue sample, which is just under the 15-mm gauge length. The arithmetical mean height (R a ), defined as the average of the absolute value along the sampling length, was measured and averaged across three fatigue dog-bones for each sample condition. The surface roughness of polished samples was measured by Touchstone Testing Lab to be ≤0.15 μm on average using a Mitutoyo SJ-410 contact profilometer prior to fatigue testing.
For XRD analysis, a PANalytical Empyrean with a Cu-Kα sealed-tube source operated at 45 kV and 40 mA was used to collect XRD patterns in order to identify the crystalline phases in the samples. The Bragg-Brentano focusing geometry was used to obtain the diffraction patterns with a step size of 0.066°. Masks of 5-mm width were used to restrict the X-ray beam to either the grip or fracture region of the tensile samples. Figure 2 shows representative light optical micrographs of the grip regions of the tensile specimens in the non-heat treated condition (left column) and heat treated condition (right column). In general, most material in the non-heat treated state exhibited a characteristic columnar grain structure with clear melt pool boundaries as illustrated by the smaller red arrows. The size and morphology of the grains differed substantially between vendors due to differences in laser parameters used (e.g., layer thickness, raster pattern).

Microstructure
For instance, material from vendor 2 and JHU/APL were both manufactured using EOS M290 machines, but the grains in the latter were much more refined. Material from vendor 4 was heat treated prior to receipt, and thus showed a very different microstructure. After heat treatment to the H900 condition, the microstructures appeared to converge to a similar morphology. Thus, solutionizing and aging treatments can help combat microstructural differences arising from variable processing conditions used by vendors.
Material supplied by vendor 2 had noticeably more porosity than material from other vendors, but still achieved an overall density higher than 99.9%. Lower magnification images (not shown) found that pores in this material were irregular in shape and slightly larger in size (20-40 μm), consistent with lack of fusion porosity. We note that vendor 2 used an older-model EOS M270 machine (circa 2004), which has an older, non-optimized inert gas flow system compared to the EOS M290, that can drive process instability (e.g., larger vapor cloud size, inconsistent laser adsorption, spatter) and introduce more porosity [3].
Overall, all material showed high density that was indicative of good build quality, alluding to the relative maturity of L-PBF processing for 17-4 PH. Very few pores were evident in each of the micrographs and typically were ~10-40 μm in size. Additionally, microstructural differences in the non-heat treated state were largely erased after heat treatment. In total, the ability to achieve consistent material at the microstructural scale via L-PBF and heat treatment simplifies the baselining process.

Quasistatic tension testing
As shown in Fig. 3, quasistatic, uniaxial tensile testing revealed the mechanical response of specimens in the nonheat treated and heat treated conditions. Within the nonheat treated state, results are presented for specimens built in the XY (dashed blue lines) and Z (solid red lines) orientations, and additional stress-strain curves are presented for heat treated material built in the Z orientation (dashed gray lines). Among the samples with non-heat treated microstructures, the quasistatic mechanical properties were similar for vendors 1, 2, 3 and JHU/APL, with yield strengths between around 700 and 800 MPa, ultimate tensile strengths around 850-900 MPa, and strains to failure in excess of 10%. Material supplied by vendors 1, 2, and 3 showed limited anisotropy, such that the UTS in the Z orientation was within 36 MPa (~4%) of the UTS in the XY direction, while material built by JHU/APL showed slightly higher anisotropy with a UTS = 931 MPa in the Z direction and 1001 MPa in the XY orientation. The low level of anisotropy in these materials was consistent with publically sourced data sheets supplied by vendors 1, 3, and 5 as well as the machine manufacturer EOS itself for the EOS M290 machine used by vendor 3 and JHU/APL (Table 1) [26]. Material from vendors 4 and 5 showed conspicuously different mechanical responses, for reasons discussed in the following paragraphs.
As mentioned above, the material from vendor 4 was received stress-relieved with shot-peened surfaces and subsequent H900 aging, which yielded a much stronger material with YS = 1208 MPa and UTS = 1396 MPa, with a ductility of ~9%. The elevated strength was similar to that following heat treatment to H900 condition and also consistent with the ASTM A564 standard property values [34]. Though shot-peening has been shown to enhance YS by 40% relative to the untreated condition [35], machining is expected to have removed the damage layer such that the effect is attributed to prior heat treatment alone. For the purposes of machine qualification, this indicates that 17-4 PH material with unsuspected heat treatments can easily be distinguished from their mechanical response, as well as the metallographic analysis in Fig. 2.
In contrast to material provided by other vendors, material from vendor 5 exhibited a pronounced yield point behavior at a strength of 471 MPa, followed by significant work hardening, resulting in an ultimate tensile strength at around 1200 MPa until failure at an elongation of ~0. 18 for the Z orientation. The discontinuous yielding at the lower stress likely resulted from the strain-induced transformation of austenite to martensite, which is documented to occur in 17-4 PH [19,36] and resembles behavior in transformation induced plasticity (TRIP) toughened steels. The yield point behavior of this material was visualized using DIC full-field strain measurements (Fig. 4a), which revealed the formation and propagation of Lüders bands at engineering strains from 0.5 to 4%. Transformation of austenite to martensite was confirmed post-mortem using XRD (Fig. 4b), which revealed the presence of austenite (γ) in the undeformed grip region of the sample and elimination of γ in material near the highly deformed fracture surface.
The likely origin of this behavior was the use of nitrogen-atomized powder to build material by vendor 5. Nitrogen is a well-known austenite stabilizer and remains in the powder post-atomization. Prior studies have reported ~0.01 wt.% N in argon-atomized powder and 0.06-0.12 wt.% N in nitrogen-atomized powder [37]. After L-PBF, the levels of retained austenite were found to be <1% in the argonatomized powder but ranged widely from 20 to 97% in the nitrogen-atomized powder [36]. It should be noted that the material's tendency for discontinuous yielding was not detailed in the vendor-provided material data sheet for 17-4 PH. Still, this unusual behavior was easily identified in the tensile results. In general, the tensile data demonstrates consistency in behavior with two obvious outliers (vendors 4 Fig. 3 Uniaxial tensile stressstrain response of non-heat treated and heat treated samples from a-e vendors 1-5, respectively, and f JHU/APL. Non-heat treated samples are indicated by blue dashed lines for the XY build orientation and red solid lines for the Z build orientation. Heat treated samples are indicated by dashed gray lines (Z build orientation only). A representative curve (black dashed line) for samples from vendor 5 tested via DIC strain mapping is included in e to demonstrate correlation to conventional data. All samples were tested with machined surface finishes 1 3 and 5), indicating that proposed qualification frameworks could flag non-conforming material.
After heat treatment, the tensile properties of material from vendors 1, 3, 4 and JHU/APL mostly converged to a similar characteristic response. As shown in Table 2, the YS of these lots of material ranged from 1154 to 1311 MPa, the UTS 1369 to 1402 MPa, and elongation to failure was approximately 7-8%. Material from vendor 2 showed somewhat lower YS of 1036 MPa, UTS of 1185 MPa, and elongation to failure of ~3.5%; this was likely due to higher porosity (Fig. 2), which acts as stress concentrators and increases the susceptibility to premature failure. Material from vendor 5 showed an improved YS of 1123 MPa, UTS of 1403 MPa, and exceptionally high ductility of beyond 15%; this material still showed some minor yield point behavior, potentially indicative of a small fraction of retained austenite remaining in the microstructure [37,38]. The high ductility is associated in part with TRIP-related phenomenon.
In summary, the quasistatic mechanical testing demonstrated that (1) most lots of material showed similar behavior in the non-heat treated and heat treated conditions and (2) tension testing was capable of detecting non-conforming material due to unique powder atomization or heat treatment conditions.

Fatigue testing
Fatigue testing was conducted to explore vendor-to-vendor variability (Fig. 5a). Within Fig. 5a, all samples were tested in the Z orientation (vertically built), with as-built surfaces, and non-heat treated. Congruent data from JHU/APL-built  samples are included in Fig. 5a for comparison. In general, the S-N curves exhibited strong power law behavior but showed extensive variation in fatigue strength between materials provided by different vendors. For example, the fatigue strength at 1,000,000 cycles ranged from 105 MPa (vendor 2) to 320 MPa (vendors 4 and 5). Some of this variation was attributed to differences in processing conditions (e.g., vendor 5 achieved a high fatigue strength, possibly due to use of nitrogen-atomized powder), or surface condition/prior heat treatment (i.e., vendor 4 samples were shot peened and heat treated). Even by excluding these samples with unique pedigree, the fatigue strength at 1,000,000 cycles still ranged from 105 MPa (vendor 2) to 270 MPa (vendor 3) for samples with non-heat treated microstructure, as-built surfaces, and manufactured with argon-atomized powder.
Notably, vendor 5 exhibited very distinct S-N behavior compared to the other vendors. In particular, the material was weaker in the high-stress/low-cycle fatigue regime, yet stronger in the low-stress/high-cycle fatigue regime. This was possibly due to the strain-induced transformation of austenite to martensite. In particular, the deformation in samples loaded at stress amplitudes of 512 MPa and higher exceeded the known threshold for cyclic hardening due to transformation induced plasticity [16], which could therefore affect the low-cycle fatigue performance. Still, it remained unclear how much the retained austenite and its transformation to martensite could account for the distinct fatigue behavior. We leave this for future study beyond the scope of this work.
By comparing the fatigue performance against the measured roughness of the as-built surfaces, we found that the fatigue performance appeared to correlate with surface roughness, which is annotated to the right of the S-N curves in Fig. 5a and b. If one compares only conforming material (i.e., excluding shot-peened and nitrogen-atomized material from vendors 4 and 5, respectively), the material with for a the vendor-to-vendor comparison using unmachined and non-heat treated samples (wherein a vertically built JHU/ APL sample is included for comparison) and b surface finish/heat treatment/build orientation comparison using JHU/ APL-built samples. All data is non-heat treated unless otherwise annotated with "HT." For clarification, "as-built" refers to samples both unmachined and non-heat treated; the (45 + as-built) data are produced from unmachined samples. Tests were concluded at 5,000,000 cycles (i.e., "run-out"). Average surface roughness (R a ) values are displayed (TTL indicates roughness measured by Touchstone Testing Lab), along with upper and lower bounds of data from the Granta database for cast H900 (dashed black lines) and cast H1100 (solid black lines) 17-4 PH in b [39] the highest roughness (R a = 14.81 μm, vendor 2) showed the lowest fatigue strength, while remaining samples with moderate roughness R a = 5.25 to 6.56 μm showed improved fatigue performance. As discussed above, material from vendor 2 exhibited noticeably more porosity than that from the other vendors, which may simultaneously contribute to the lower fatigue strength.
To further investigate the observed trends, we assessed the fatigue performance of samples built at JHU/APL with different surface finishes, heat treatment, and build orientation (Fig. 5b). The data are compared to two sets of calculated fatigue strength models for cast 17-4 PH in the H900 and H1100 conditions, using data from Ansys Granta MI and MaterialUniverse™ [39]. Most notably, L-PBF 17-4 PH after heat treatment and surface polishing performed similarly to conventionally processed material, as evident in the (Z + polished + HT) and (45 + polished + HT) curves. Scatter in the (Z + polished + HT) curve was somewhat higher, with some points falling below the calculated H900 and H1100 curves, which may have been attributed to the unfavorable orientation of lack of fusion defects in the Z oriented, i.e., vertically built, material [29]. Fatigue strengths of 475 to 560 MPa were obtained at lives of 1,000,000 cycles.
Second, machined surfaces (purple asterisk data points) were inadequate at achieving "conventional" fatigue performance, which could be due to cold work on the surface and/or remaining surface imperfections [20,28]. A fatigue strength of approximately 300 MPa was obtained at 1,000,000 cycles, which was roughly 2× lower than the machined-and-polished state.
Third, similar knock-down factors were obtained for all samples with as-built surfaces (compared to the machined state), regardless of build orientation and heat treatment. Fatigue strengths at 1,000,000 cycles ranged from 110 to 175 MPa, which was roughly 4-5 times lower than the machined-and-polished state. Similar trends were reported by Witkin et al. for Inconel 718 [40], where samples with rough surfaces showed severely lower fatigue performance compared to samples with polished surfaces, regardless of build orientation. An older study of 17-4 PH showed similar fatigue performance of polished samples in the non-heat treated and heat treated conditions (though conclusions may be convoluted by pervasive lack of fusion porosity, which was ~0.55-0.75% in volume fraction) [16]. Overall, this highlights the overwhelming effect of surface roughness on fatigue failure for modern L-PBF processes.
In total, this analysis confirmed that baseline and qualification testing protocols should carefully specify the fatigue test conditions. The surface condition appeared to have the strongest effects on fatigue performance. Any anisotropic effects of build orientation were overwhelmed by surface condition, potentially allowing simplification of test matrices to only consider a single build orientation.

Conclusion
Recent advances in L-PBF technology have enabled production of high-quality 17-4 PH material with low porosity and good mechanical properties. In response, several government departments and standards agencies are actively developing qualification procedures to enable fielding of AM material in structural parts. In this study, we quantified the variability in microstructure, quasistatic tensile properties, and fatigue properties of material across six vendors (including JHU/APL) and evaluated their behavior using protocols emblematic of emerging qualification standards. These experiments support the following conclusions: • The 17-4 PH material provided by each vendor showed high relative density (in excess of 99.9% as evidenced by X-ray computed tomography data) but different microstructures depending on the processing routes. After heat treatment, the microstructures of all samples largely converged, suggesting a possible approach to improve consistency between vendors. The density of all samples was substantially improved compared to 17-4 PH material produced 5 years ago using prior generation machines. • The quasistatic tensile properties were largely consistent between vendors and comparable to conventionally processed material. In the non-heat treated condition, material from 4 out of 6 vendors showed yield strengths ranging roughly 700-800 MPa, ultimate tensile strengths 850-900 MPa, and strains to failure in excess of 10%, which is comparable to conventionally processed material in the H1100 or H1075 condition. After heat treatment to the H900 condition, the YS of these lots of material ranged 1154-1311 MPa, the UTS 1369-1402 MPa, and elongation to failure was approximately 7-8%. Obvious outliers due to use of nitrogenatomized powder (leading to TRIP-like effects) and undisclosed processing, e.g., prior heat treatment, were readily detected in the mechanical response and are likely to be identified during machine certification. • The fatigue response is dominated by surface roughness, and this effect overwhelms that of the build orientation and heat treatment for the same sample condition. Fatigue testing revealed that material after surface polishing could achieve performance comparable to conventionally processed material. Knock-down factors due to surface roughness are estimated to be ~4-5 for as-built surfaces. Fatigue strength at 1,000,000 cycles varied substantially between vendors, ranging from 105 to 270 MPa. Very different fatigue behavior was observed in samples built with nitrogen-atomized powder, or with shot-peened surfaces.
• Overall, extensive mechanical testing revealed that many vendors and models of L-PBF machines can produce "acceptable" 17-4 PH material with high density, uniform microstructures, and consistent quasistatic and fatigue properties. Prescribed build requirements were not met by all the vendors, which may prove to be a central challenge to establishment of MMPDS-style baseline property datasets. However, the proposed machine qualification frameworks reliably identified non-conforming manufacturing practices (e.g., unique powder atomization conditions, or heat treatment). Build conditions should be carefully specified and closely monitored during procurement where possible to ensure consistent material. • More broadly, this investigation alludes to challenges with establishing material property baselines. These difficulties emerge from the decentralized nature of the AM industrial base, across which uniform "best practices" regarding powder specifications and post-processing may not exist. This research suggests that variation in build conditions across different machines need not prohibit establishment of baseline datasets because build quality and mechanical properties are sufficiently consistent. Thus, MMPDS-style baseline databases can be assembled for L-PBF materials. In addition, this research highlights the value in establishing AM powder specifications (which do not currently exist) for further mitigating variability in material properties and facilitating development of property baselines.