1 Introduction

Laser-sintered 17–4 PH stainless steel (SS) is a material showing a very good corrosion resistance and high strength guaranteed by its martensitic structure [1]. The high strength is due to the precipitation of fine cuprum-rich face-centred cubic phase. A dendritic structure with body-centred cubic martensite and about 50% of reserved austenite is favoured by the laser-based powder bed fusion process. According to [2], the mechanical properties of additive manufactured materials are comparable with those produced with traditional technologies such as casting and machining. Nevertheless, the ultimate tensile (UTS) and yield (\(\sigma_{Y}\)) strengths of laser-based powder bed fusion-produced 17–4 PH SS are quite lower than those of wrought material in H900 condition (age-hardening treatment: air quenching at 900 °C for 4 h) since the additive manufactured 17–4 PH SS is not completely martensitic, thus showing retained austenite. The high speeds of solidification during the laser-based powder bed fusion process, indeed, prevents the formation of the martensite phase in the as-built material leading to a metastable austenitic microstructure.

Another feature to be pointed out is that the characteristic porosity which comes from the production process can impact the mechanical properties of the material [3]. The voids can serve as crack nucleation sites. Some authors [4, 5] suggested applying hot isostatic pressing (HIP) to reduce the presence of porosities. For this reason, this material is generally hardened. Nevertheless, in the literature, several works can be found in which this type of material shows a low high-cycle-fatigue resistance. Internal and surface defects such as porosities, inclusions and a high roughness in the as-built condition can act as crack nucleation sites [3, 6, 7]. Moreover, contradictory results regarding the effectiveness of the HIP process on AM 17–4 PH SS are present in literature [8, 9]. For this reason, having specific and reliable high-fatigue data are fundamental to identify the limits of AM materials.

Nevertheless, another peculiar advantage of additive manufacturing (AM) is its capability to produce lattice/reticular structures characterized by small sections and complex geometries. It is well known that those structures are often subjected to high plastic deformations [10]–[12]. The knowledge of the low-cycle-fatigue (LCF) performances is therefore a key aspect in the design of new structures and systems. Considering that the complex shape of reticula impedes surface finishing, it is also fundamental to understand the reduction of the fatigue properties related to the characteristic surface properties obtained with the laser-based powder bed fusion process.

With these premises, this work aims at the characterization of the mechanical properties (HCF and LCF) of a 17–4 PH SS produced via AM. In addition to a first series of as-built samples in which the surface roughness is the one that can be obtained from the laser-based powder bed fusion process, a second series of samples was produced with the same technology but successively machined. This is to assess the material fatigue life, accounting for the effect of the surface roughness typical of the AM technology that, as already pointed out, cannot be modified with finishing operations (e.g. in lattice/reticular structures) in many actual applications.

Moreover, a third series of AM parts was heat treated via hot isostatic pressing (HIP) provided that several authors claim this treatment to be beneficial in terms of fatigue performances thanks to the reduction of the internal porosities induced by the AM production process.

Finally, a last series of samples having the same geometry was produces from wrought (NO AM) 17–4 PH SS material to better quantify the performance differences of AM powders with respect to traditional production alloys.

2 Chemical composition and microstructure

Standard EOS M280 parameter settings were used to produce all the AM samples employed in the experimental HCF and LCF campaigns. Specimens were printed in the building (vertical) direction. The machined AM samples were obtained starting from as-built geometries with stock added to permit machining to final dimensions. A manually controlled lathe was used, equipped with carbide turning inserts suitable for 17–4PH stainless steel grade type.

The chemical composition of 17–4 PH SS is shown in Table 1. The particles of the powder have a mean size of 44.454 µm. Chromium (Cr) interacts with carbon (C), forming chromium carbides along grains [13].

Table 1 Chemical composition of the 17–4 PH SS
Full size table

To prevent the creation of Cr23C6 [14] and to increase strength, niobium (Nb) is used. It promotes the formation of NbC. Silicon (Si) decreases the mechanical properties promoting the formation of ferrite, but it is useful for fluidizing the material during the casting. To compensate this effect, chromium and nickel (Ni) are used. They reduce the formation of ferrite in favour of austenite. Nickel and manganese (Mn) are helpful to avoid the formation of Cr23C6 and FeS phases, respectively, forming Cr2N and MnS instead. Cuprum helps increasing the overall strength of the material, forming precipitates.

In addition, chemical analysis and scanning electron microscopy (SEM) measurements, performed at Politecnico di Milano, were made to highlight the possible presence of defects close to the surfaces, which could affect the fatigue behaviour. Defects were found to be up to hundreds of micrometers. This evidence was confirmed by microcomputer tomography (µ-CT) measurements.

A first scouting measure was made with a voxel size of 2.25 µm and a second detailed measure with a voxel size of 1 µm. Measures were made at the University of Kassel.

From Fig. 1 it can be appreciated that the propagation direction is not influenced by the melt pool boundaries, but it is instead completely inside the pools. The chemical composition of the melt pools was investigated by energy-dispersive X-ray spectroscopy (EDS) analysis. No appreciable difference was detected among the pool centre and boundaries: Al (0.28–0.30%), Si (0.47–0.51%), Cr (17.1–17.4%), Fe (73, 15–74.6%), Ni (3.76–3.79%), Cu (3.27–3.30%) and Nb (0.24–0.25%).

Fig. 1
figure 1

SEM: melt pools in the transversal (a) and in the build (b) directions, defect (c) and crack path (d) and µ-CT (mm) (e) analyses

Full size image

Finally, micro-hardness tests were performed on the four series of samples. Measurements were made for ten radial positions starting from the surface to the axis. No relevant differences were observed among the NO–AM, as-built and hipped series with a slightly higher hardness for the hipped samples, while the machined ones exhibited higher values. This phenomenon could be ascribed to the machining process that might have removed or reduced the initial surface or subsurface porosities, and modified the surface material state due to the heat generated. Also, the residual stress state resulting from the cutting operations could have had an influence on the increased hardness (Fig. 2).

Fig. 2
figure 2

Vickers micro-hardness measurements along the radial direction

Full size image

Additional data regarding the 17–4 PH alloy, data regarding the tooth root bending strength and the ductile behaviour and fracture are available [15, 16], which were published in the framework of the same research project.

3 Experimental tests

3.1 HCF

High-cycle-fatigue (HCF) tests were performed on cylindrical samples (Fig. 3, on the right) with a geometry designed according to the ASTM E466 standard [17]. The experimental facility used for testing is a STEPLab-UD04 fatigue testing machine capable of applying up to 5 kN at maximum frequency of 35 Hz [18].

Fig. 3
figure 3

Standard specimen for quasi-static (left) and HCF (right) tests according to ASTM E466

Full size image

In addition to the HCF tests, a quasi-static tensile characterization was performed. The geometry of the adopted samples is shown in Fig. 3 (on the left). Tensile tests were performed at room temperature with a crosshead speed set to 0.1 mm/min.

The tested samples present the cone-cup-shaped fracture surface, typical of the ductile failure. Plastic deformation was also confirmed by the presence of a fibrous and irregular surface (necking region).

Preliminary static tests were also performed to determine the yielding point (Fig. 4). These data (Table 2) have been used for the calibration of the Basquin parameters, as described in the following.

Fig. 4
figure 4

17-4PH stress–strain curves of as built, machined, hipped and no-am samples, from tensile tests

Full size image
Table 2 Yielding stress of as built, machined, hipped and no-am samples, from tensile tests
Full size table

HFC tests were performed according to the ASTM E466 [17]. The fatigue limit \(\sigma_{F}\) was calculated with the short staircase approach [19,20,21,22]. More in detail, several tests at different load levels were performed. The staircase method prescribes to define a force increment \(\Delta F\) (that is directly related to the accuracy of the results that can be achieved). The actual \(\Delta F\) was set to 50 N, corresponding to a stress interval Δσ equal to 10 MPa.

If the initial test, performed at a force level \(F_{i}\), reaches the run-out condition (50,00,000 cycles without failure), the next test has to be performed on a force level increased by \(\Delta F\) i.e. \(F_{i + 1} = F_{i} + \Delta F\).

Otherwise, if the specimen broke, the next force level must be decreased by \(\Delta F\), i.e. \(F_{i + 1} = F_{i} - \Delta F\).

Figure 4 shows the short staircase sequences (6 tests—maximum value according to Dixon [23]) for the four series of samples: as-built, machined, hipped and NO AM (wrought). The Dixon statistical approach considers, in addition to the number of run-outs (O) and failures (X) as the traditional staircase method, also the sequence order. The statistical significance of the results is ensured by a statistical correction coefficient \(k\). The fatigue limit can be calculated by means of the following relation:

$$ \sigma_{F} = x_{f} + k \cdot \Delta \sigma $$
(1)

in which \(\sigma_{F}\) is the fatigue limit at a 50% probability, \(x_{f}\) is the stress level of the last test (ID = 6 with reference to Fig. 5) and \(k\) a coefficient depending on the sequence order (Table 3).

Fig. 5
figure 5

Short stair-case sequences for the four series of tested samples

Full size image
Table 3 Results of the HCF tests
Full size table

For the as-built testing campaign, the tests order results in O–X–O–X–X–X (\(x_{f} = 265\) MPa). Having a \(k\) factor equal to 0.661, the fatigue limit results \(\sigma_{F} = 271\) MPa.

For the machined samples, the test order was O–X–O–O–X–O (\(x_{f} = 336\) MPa). The Dixon coefficient \(k\) was 0.372, resulting in a fatigue limit equal to \(\sigma_{F} = 340\) MPa.

For the hipped series, the test order was O–O–O–O–X–O (\(x_{f} = 264\) MPa). The Dixon coefficient \(k\) was 0.894 resulting in a fatigue limit equal to \(\sigma_{F} = 243\) MPa.

For the wrought (NO AM) series, the test order was O–X–X–O–X–O (\(x_{f} = 346\) MPa). The Dixon coefficient \(k\) was 0.831 resulting in a fatigue limit equal to \(\sigma_{F} = 355\) MPa.

Additional tests were performed for the four groups at higher levels of stress to obtain the left part of the Wöhler diagram, corresponding to finite life.

Figure 6 shows the S–N curves. The continuous line refers to the as-built condition, the dashed line to the machined one, the dots represent the hipped condition and the dot-dashed line the wrought material.

Fig. 6
figure 6

HCF diagrams

Full size image

3.2 LCF

Low-cycle-fatigue tests were performed on cylindrical samples with standard dimensions according to ASTM E606 [26], as shown in Fig. 7 (on the left). As for the HCF, four series of samples were tested.

Fig. 7
figure 7

Testing machine (left), Standard specimen for the LCF tests according to ASTM (right)

Full size image

The comparison with data from literature [24], [6], [25] shows a good agreement of the results.

Strain-controlled fatigue tests were performed on the STEPLab UD04 testing machine (Fig. 7, on the right) according to [26]. Tests were performed at a strain ratio \(R_{\varepsilon } = - 1\). The testing frequency was set to 0.1 Hz [27, 28]. This value, the lowest prescribed by the ASTM E606, was kept constant during the tests. Higher speeds may affect the results promoting a slight increment in the temperature of the sample and a modification of the mechanical properties.

The stabilized cycles for each tested deformation level were interpolated by means of the Ramberg–Osgood (RO) equation [18] (Eq. 2) to obtain the stabilized curves (Fig. 8).

$$ \varepsilon_{a} = \varepsilon_{ae} + \varepsilon_{ap} = \frac{{\sigma_{a} }}{E} + \left( {\frac{{\sigma_{a} }}{{K^{^{\prime}} }}} \right)^{{\frac{1}{{n^{^{\prime}} }}}} $$
(2)
Fig. 8
figure 8

Cyclic RO curves vs. constitutive QS ones

Full size image

Here, \({\varepsilon }_{a}\) is the total strain amplitude, \({\varepsilon }_{ae}\) and \({\varepsilon }_{ap}\) are the elastic and the plastic portions of \({\varepsilon }_{a}\), \({\sigma }_{a}\) is the stress amplitude,\(E\) is the elastic module, while \({K}^{^{\prime}}\) and \({n}^{^{\prime}}\) are constants that depend on the material being considered [29]. In this form, K and n are not the same as the constants commonly seen in the Hollomon equation [30].

The parameters of the RO equations are reported in Table 4.

Table 4 Parameters of the Ramberg–Osgood equation
Full size table

The stabilized curves are shown in Fig. 8. The as-built series shows a slight hardening behaviour [31], which is more evident for the machined series. On the other side, the hipped series show, at least in the low-strain range (\({\varepsilon }_{a}\) < 0.06), a marked hardening. Finally, the NO AM series shows a mixed behaviour with softening just after the yielding and hardening for higher strains, with a transition at \({\varepsilon }_{a}\) = 0.05.

Even after the stabilization, tests were continued up to rupture of the sample to obtain data for the \(\varepsilon - N\) fatigue curves, better known as the Basquin–Coffin–Manson (BCM) diagrams (Eq. 3)

$$ \varepsilon_{a} = \frac{\Delta \varepsilon }{2} = \frac{{\sigma_{f}^{^{\prime}} }}{E}\left( {2N} \right)^{b} + \varepsilon_{f}^{^{\prime}} \left( {2N} \right)^{c} $$
(3)

where \({\Delta }\sigma_{f}^{^{\prime}} ,b,\varepsilon_{f}^{^{\prime}} ,c \) are the four material constants to be identified.

Figure 9 shows the experimental data for the four series of samples and the fitting curves using the BCM equation.

Fig. 9
figure 9

LCF diagrams

Full size image

The calculation procedure prescribes to firstly determine the Basquin coefficients (\(\sigma_{f}^{^{\prime}}\) and \(b\)) starting from the yield stress \(\sigma_{Y}\) and the high-cycle-fatigue limit \(\sigma_{F}\) with a linear interpolation.

The Basquin equation (describing the relation between the elastic deformation \(\varepsilon_{ae}\) and the number of cycles \(N\)) can be used to separate the plastic deformation \(\varepsilon_{ap}\) from the total one.

$$ \varepsilon_{ap} = \varepsilon_{a} - \varepsilon_{ae} = \varepsilon_{f}^{^{\prime}} \left( {2N} \right)^{c} $$
(4)

With the aim of determining the maximum likelihood estimators for \(\varepsilon_{f}^{^{\prime}}\) and \(c\) according to Eq. 4, the ASTM 739 standard [28] was used.The equation of a line fitting the experimental points can be rewritten as

$$ \log (N) = \widehat{A} + \widehat{B}\log (\varepsilon_{ap} ) $$
(5)

The maximum likelihood estimators of A and B are the following:

$$ \widehat{A} = \overline{Y} - \widehat{B}\overline{X} $$
(6)
$$ \widehat{B} = \frac{{\sum {(x_{i} - \overline{X} )(y_{i} - \overline{Y} )} }}{{\sum {(x_{i} - \overline{X} )^{2} } }} $$
(7)

where \(\overline{X}\) and \(\overline{Y}\) are the averaged values of the dependent variable \(x_{i}\) (\(\log (\varepsilon_{ap} )\)) and independent variable \(y_{i}\) (\(\log (N)\)).

Equation 5 can be transformed as follows:

$$ \varepsilon_{ap} = 10^{{ - \frac{{\widehat{A}}}{{\widehat{B}}}}} \left( \frac{1}{2} \right)^{{\frac{1}{{\widehat{B}}}}} \left( {2N} \right)^{{\frac{1}{{\widehat{B}}}}} $$
(8)

Therefore, the maximum likelihood estimators of the Coffin–Manson coefficients \(\varepsilon_{f}^{^{\prime}}\) and \(c\) (Eq. 4) can be calculated as

$$ \varepsilon_{f}^{^{\prime}} = 10^{{ - \frac{{\widehat{A}}}{{\widehat{B}}}}} \left( \frac{1}{2} \right)^{{\frac{1}{{\widehat{B}}}}} $$
(9)

and

$$ c = 1/\widehat{B} $$
(10)

The final values are reported in Table 4.

4 Discussion

Figure 10 compares the material properties of the four series. Concerning the HCF, as expected, the machining process significantly increases the fatigue performances with respect to the as-built condition, due to a superior surface finishing that reduces the local stress intensification thus improving fatigue life. The fatigue limit for the as-built samples is 25.4% lower with respect to the fatigue limit of the AM machined samples. In the same way, wrought material (NO AM) performs better in comparison with the AM machined samples. In this case, while the surface roughness is the same, the internal porosities affect the overall performance, with differences of about 4% (Table 5).

Fig. 10
figure 10

Comparison between the four different series of samples

Full size image
Table 5 Parameters of the Basquin-Coffin-Manson equation
Full size table

While the above-mentioned results were expected, less trivial is the effect of the hipping process. While several scholars claim that this treatment can improve the fatigue performances by reducing the internal porosities, the actual data show a reduction of the fatigue limit of about 12% in the presence of the HIP process, instead. This effect could be justified considering that, as demonstrated by du Plessis et al. [32], the near-surface pores are not always closed and hipped and sometimes become over-pressurized after the treatment itself. Especially for small samples subjected to axial loads, like in the present case, this effect could lead to a premature failure of the hipped material with respect to the as-built one.

On the other hand, the LCF tests show that while the wrought (NO AM) material performs better for HCF loads, the AM material seems to have a better performance in terms of LCF. The comparison between the AM machined and the wrought (NO AM) samples shows that the first can survive longer for the same applied cyclic deformation. This might be explained by the different microstructure of the material which contains austenite.

The effect of the surface finishing is evident, and the as-built samples are less resistant with respect to the AM machined ones, even though the two BCM curves have a similar shape. The HIP process produces again the worst performance.

5 Conclusions and future work

In this experimental study, four series of samples made of 17–4PHSS were produced, tested and characterized in the high- and in the low-cycle-fatigue range. The series of samples differ in terms of surface finishing, heat treatment and manufacturing technology. A first series of AM as-built samples was used as reference. A second series, machined, was used to study the effect of the surface finishing. The third series was treated by means of HIP. The fourth series was manufactured from NO AM material for comparison.

The HCF limit was determined using the short staircase approach according to Dixon. For the LCF, first the RO equations were tuned. Successively, based on the results of strain-controlled tests, the BCM fatigue curves were calibrated.

The surface finishing (machined vs. as-built) seems to significantly improve the HFC properties (+ 25%). The comparison between the machined and the NO AM series which have the same surface finishing shows that the wrought material is better performing (+ 31.0%) in terms of fatigue limit.

The hipping process is typically adopted on AM materials to reduce the porosities and, therefore, to improve the mechanical behaviour. Indeed, this effect was confirmed by the static tests (the hipped samples show mechanical properties aligned with the machined and NO AM—wrought material), while unlike that expected, the process seems to not improve the fatigue limit but reduces it (−10.3%, hipped vs. as-built).

On the other side, in the LFC range, the four series perform differently. The machining promotes a constant increase of the maximum possible strain amplitude (about + 0.085). In the low range of cycles, the as-built series performs better with respect to the hipped and the NO AM ones. After about 800 cycles, both series (hipped and NO AM) overtake the as-built curve.

While small samples were used because they better represent the typical dimensions of reticular structures, a future work will be to extend this study to bigger diameters to understand if the actual finding could be generalized.