3.1. Pore Density Evaluation of the Model Pressure Vessel
The results of the defect analysis via computer tomography are displayed in
Figure 3. Distinct differences regarding pore distribution and the number of pores were encountered in the two model pressure vessels.
Although both vessels were produced in the same run, we encountered differences in pore distribution. A possible reason for this is the significant number of irregularly distributed accompanying samples on the installation space, which caused different laser dwell times.
The effect of pores on the total fatigue lifetime of a component was investigated in [
20]. It was hypothesized that crack growth increases abruptly when a pore is reached, whereas crack growth is retarded subsequently because the crack must be initiated again.
Therefore, the number of cycles until re-initiation of crack growth determines whether further crack growth is accelerated or retarded. In this context, pore distribution in the component is decisive (near or far from the surface, see
Figure 4). For this purpose, the impact of pores in an exemplary component section (transition region from the cylinder wall to the torispherical head) under internal pressure loading was evaluated first. This region is characterized by high tensile stresses at the inner surface and compressive stresses at the outer surface.
Therefore, three simple axially symmetric models were created using the measurement results from computer tomography (
Figure 5). Model I is a non-porous basic model for comparison, and model II features a pore near the surface, i.e., in a critical design area of high tensile stress. Model III presents a pore near the outer surface, i.e., under compressive stress. The stress distributions in the component wall through the pore positions are illustrated in
Figure 6.
An increase in stress at the pore edges only occurred if the pore was located in a critical design area of high stress concentration. In model II, representing a near-surface pore, significantly higher stresses occurred at the pore edges in comparison to the pore-free model I.
In the pore-free component (
Figure 6a), the highest stresses appeared at the inner surface of the intersection. If there was a pore nearby (
Figure 6b), the stresses at the pore edge were higher than at the inner surface. Therefore, the crack initiated from the pore. In the next step, the crack grew towards the highest stresses, which apart from the pore appeared at the inner surface. Only when the crack reached the inner surface did the crack growth direction turn from the pore towards the outer surface and the crack would grow into the depth of the wall [
20].
The analysis showed that pores acted as notches in critical design regions and thus may act as crack initiators. A crack retarding effect, as originally assumed, did not exist. In this context, it should be noted that the pores were idealized as spheres of equivalent volume in the FEM analysis. In reality, the surfaces of the detected pores were irregular, sometimes sharp-edged, which may have had an effect on crack initiation properties.
The fatigue analyses carried out showed that pores could significantly shorten the technical fatigue lifetime in areas of high stress. On the other hand, crack growth and thus residual lifetime itself was largely independent of the presence of pores. Small pore size caused only slightly accelerated crack growth. Consequently, crack growth corresponded approximately to that in a pore-free structure [
20].
These results demonstrate the enormous importance of fatigue crack growth for the design and lifetime evaluation of additively manufactured components. A criterion for safe components has to be low crack propagation rate or high crack growth resistance. Therefore, materials with low fatigue crack growth rate are of central importance for the tolerance to manufacturing defects.
3.2. Crack Propagation in Conventionally and Additively Manufactured 316L
In additive manufacturing of components, process defects (like pores, which may lead to crack initiation) in critical design regions cannot be completely avoided. For this reason, the investigation of crack propagation behavior is a prerequisite for the safety of components, for instance, for energy conversion processes.
Figure 7 shows a comparison of the crack growth behavior of the additively manufactured material in the different build directions in comparison to conventionally manufactured material.
In all additively manufactured specimens, the crack propagation rates per cycle (da/dN) were lower, regardless of build direction, with the only exception of a small ΔK range from ≈23–26 MPa√m in the first ZY specimen. The lowest da/dN was measured at the specimen from the ZX direction. For a given value of ΔK, the crack propagation rate was up to eight times lower than in the conventionally produced material. Furthermore, the highest ΔK value (≈11.5 MPa√m) for initiation of crack propagation (note that this value is not a threshold value, i.e., crack growth can occur even below this value) of all the additively manufactured specimens was measured in the ZX samples, too. However, this ΔK value for initiating crack propagation was about 2.3 MPa√m lower compared to conventionally produced material. In the XY and ZY specimens, lower ΔK values for crack propagation initiation were encountered (in the XY specimen: ΔK ≈9.6 MPa√m, in the ZY specimen: 9.4 MPa√m). In both directions, significantly higher porosity was observed (
Figure 8).
Pores can act as crack initiators (c.f.
Section 3.1). It is therefore hypothesized, that pores were responsible for the lower ΔK values for initiation of crack propagation.
Figure 9a displays the CT specimens after fracture.
Figure 9b depicts the crack paths in the LPBF-manufactured samples. In the XY direction specimen, the crack propagated perpendicular to the build direction, while in the ZY orientation, the crack propagated in parallel and in the ZX direction under an angle of 45° (cf.
Figure 10). The lowest da/dN value, accompanied by the highest ΔK value to initiate crack propagation, was measured at the ZX sample. The best reproducibility was found in the XY direction specimens (
Figure 7). However, only the ZY direction provided a valid fracture path (according to ASTM regulation [
21]). Regardless of build direction, the crack path was transcrystalline (cf.
Figure 9b,
Figure 10 and
Figure 11). Consequently, a weak connection of the individual melt pools can be excluded as a key factor in cracking behavior.
Figure 12 depicts the comparison of the fracture surfaces of conventionally and LPBF-manufactured specimens of different build directions. The fracture surface of conventionally produced material offered features of ductile fracture dimples. In contrast, a mixed fracture mode with ductile and cleavage areas, along with several pores, was found in the additively manufactured XY and ZY specimens (
Figure 12b,c). The same was observed in the case of the ZX sample, but the crack propagated along lattice planes in the grain interiors
(Figure 11d). In this specimen, preferred grain orientations into the {001}, {111}, and {110} directions were measured. With a percentage of 13.46%, the {110} direction was the dominating one (
Figure 10). In contrast, the XY and ZY specimens exhibited preferred grain orientations into the {001} and {111} directions. Furthermore, preferential orientation relations in the XY and ZY were less significant than in the ZX specimen (cf.
Figure 11).
The grains were mainly found to be oriented into the building direction (cf. EBSD mappings in
Figure 11). Corresponding to possible/preferred sliding systems in the fcc-lattice, the dislocation movement during cyclic loading took place in the direction of the oblonged grains (cf. local misorientation mappings in
Figure 13). In the case of the XY and ZX specimens, dislocations were oriented perpendicular to the crack path (
Figure 13a,c), resulting in an off-horizontal crack path (cf.
Figure 9b). In contrast, the main crack propagated horizontally and the dislocations were oriented parallel (
Figure 9b and
Figure 13b) in the ZY specimen. Crack branching occurred at pores (detailed description included in
Section 3.3) due to local stress concentration and changes in stress distribution. Secondary cracks grew partly perpendicular to the longer edges of the pores (
Figure 14) before cracks were arrested by dislocations, running parallel to the crack (
Figure 13b). Subsequently, the main crack continued to grow horizontally.
3.3. Manufacturing Defects and Their Impact on Fatigue Crack Propagation Behavior
Parallel to the vessel examinations of
Section 3.1, we investigated the fatigue crack growth behavior of conventionally and additively manufactured specimen material. Manufacturing defects were identified and the possible effects on crack propagation behavior were analyzed. The distribution, number density, and morphology of pores depending on build direction were investigated. Thereby, the lowest amount of circular pores (maximum diameter: 80 μm) was observed in the ZX specimen (
Figure 8c), which furthermore exhibited the lowest da/dN (
Figure 7). In the ZY specimen (
Figure 8b), circular and larger, elongated pores (aspect ratio ≈1.3–10) were observed, while in the XY (
Figure 8a) sample, predominantly small, circular pores (maximum diameter: 60 μm) occurred.
Nevertheless, several larger and elongated pores were found. Both ZY and XY specimens exhibited higher da/dN than the ZX sample (
Figure 7). Accordingly, the measured crack length increased considerably faster with increasing number of load cycles (
Figure 15).
However, because both the curves of the ZY and XY specimens increased monotonically, it cannot be concluded that the pores were decisive for the higher crack propagation rate when compared to the almost pore-free ZX sample. Furthermore, a tendency of crack branching at pores of an aspect ratio of ~1.3–2.8 (
Figure 14) was found in the ZY specimen. Material characterization thus confirmed the results of computational analysis of small pores, with a limited effect on crack growth (nevertheless, pores can act as crack initiators) [
20].
Crack branching generally leads to a decrease in the crack propagation rate—a further indication that pores may not constitute the decisive factor on differences in crack propagation rates.
Further processing-related issues, such as finely distributed SiO
2 particles, may constitute another influencing factor for crack propagation. These were observed in all LPBF-manufactured specimens (
Figure 16). On the other hand, such process-related, small, and finely distributed particles may have a strengthening effect, which was not further addressed in this study.
Moreover, traces of segregation (
Figure 17a) were found. EDX analyses (
Figure 17b) in the vicinity of the cracks confirmed enrichment in Ni and Mo along with depletion in Fe and Mn contents. The concentrations of the alloying elements Ni and Mo had a direct impact on solid solution hardening (and carbide precipitation) and consequently the strength of the material, which directly influenced resistance to crack growth and therefore crack propagation behavior. A plausible reason for segregation would be excessive cooling rate, resulting in non-equilibrium solidification in the LPBF process. Process optimization towards avoidance of segregation effects could offer further potential for the improvement of crack propagation resistance.
3.4. Novel Materials for Improved Structural Safety
In
Section 3.1 we demonstrated that processing defects, such as pores, may act as notches in critical design regions and thus act as preferred sites of crack initiation. Consequently, if a larger number of pores (i.e., crack starters) in additively manufactured components was present, materials of lower crack propagation rate would be required to at least maintain sufficient component lifetime and safe operation such as that known from conventionally manufactured materials.
Materials tailored for additive manufacturing processes would enable an additional improvement of component lifetime, safety, and cost savings. Of course, safety also depends on the maintenance and management of the structure, but here too, material development can contribute to increasing safety. Application of more robust materials increases the margins of inherently safe operation. The approach is not to minimize pores, but to develop a material with active crack obstruction mechanisms, which inhibit—or at least retard—both crack initiation and crack propagation. Two main mechanisms can be utilized to accomplish this. First is the option of taking advantage of in situ heat treatment by the so-called “temper bead effect” [
26,
27,
28,
29] during the AM process. The component will then benefit from potential time and cost savings over conventionally produced components. Rapid precipitation kinetics are the essential prerequisite to accomplish this. For this reason, the novel, stainless High Performance Ferritic (HiperFer) steel, developed at Forschungszentrum Juelich GmbH, is a high potential candidate. Strengthening of these steel grades is achieved by a combination of solid solution and intermetallic (Fe,Cr,Si)
2(Nb,W) Laves phase particle precipitation [
30]. The microstructural and mechanical properties of HiperFer steels are described in detail in [
31]. In the current paper, HiperFer-like steel was evaluated on the basis of its in situ heat treatment capability in laser metal deposition (LMD) processing. The second mechanism is to take advantage of “dynamic strengthening” under cyclic loading at high temperatures to actively obstruct crack initiation and propagation and thus enhance component lifetime and safety. In HiperFer, this can achieved by thermomechanically induced precipitation [
17] of the Laves phase under cyclic loading, or more generally speaking, under (cyclic) plastic deformation at temperatures above 600 °C [
18].
Figure 18 illustrates a comparison of thermomechanical fatigue curves (100% out-of-phase TMF experiments, i.e. Δε
mech. = -Δε
th., meaning that thermal expansion is fully obstructed by the testing machine during cycling through the temperature range of 50 to 650 °C) of 316L, Crofer22 H, and HiperFer 17Cr2. Crofer22 H is a predecessor of HiperFer 17Cr2 and was developed at Forschungszentrum Juelich GmbH for application in high temperature solid oxide fuel cell interconnectors. Crofer22 H cyclically hardens during the initial approximate 30 cycles (≈1% of technical lifetime) of fatigue testing by thermomechanically induced precipitation of Laves phase particles, preferentially at dislocations induced by cyclic plasticization. The end of the hardening range corresponds to the maximum of the fatigue curve (cf. “N
h”, Crofer22 H curve in
Figure 18). The technical lifetime was evaluated from the intersection of linear approximations to the “stable” and the “damage” curve sections of the TMF curves (cf. “N
f”, Crofer22 H curve in
Figure 18). The beginning of the damage range was evaluated by deviation of the fatigue curve from a linear approximation to the stable curve section (cf. “N
d”, Crofer22 H curve in
Figure 18). During the initial 30 cycles, Crofer22 H hardened by 205 MPa, which is equivalent to an increase in strength of 38% (
Table 6). After reaching the maximum stress range, Crofer22 H was characterized by a comparatively steep drop in stress range before it entered the stable range (
Figure 18) at around 380 cycles. The HiperFer 17Cr2 steel in contrast continuously hardened within the initial 70 cycles (≈5% of technical lifetime), avoiding the steep drop in stress range and directly entering the stable range on a higher stress range level (
Figure 18). The hardening range increased to 236 MPa, corresponding to an increase in strength of 42% (
Table 6). These differences in behavior were caused by differing precipitation kinetics and phase fractions of Laves phase precipitates. Dislocations induced by cyclic plasticization cause (temporary) hardening. Crofer22 H and HiperFer dislocations were immediately decorated and pinned by nucleating Laves phase particles, i.e., the temporary strengthening effect became quasi-permanent. The chemical composition of HiperFer 17Cr2 is optimized for high volume fraction of the Laves phase. For this reason, the precipitation process lasts longer (5%/1% of technical lifetime HiperFer/Crofer
®22 H) and creates more dislocation pinning particles in HiperFer, which causes increased strengthening and thus stabilization of the stress range on a higher level. In Crofer22 H, less dislocations were pinned by precipitates, i.e., a smaller extent of temporary dislocation strengthening became quasi-permanent, which caused a decrease in stress range after the maximum was reached.
Because the stable stress range was lower, damage occurred later in Crofer22 H. Despite the high level of strain, obstruction in the experiments the S-N curves of Crofer22 H and HiperFer decreased moderately, especially in comparison to 316L. The austenitic steel strengthened in the hardening range by only 47 MPa, which corresponded to an increase in strength of only 5.4% (
Table 6). 316L failed abruptly, with only a ≈13.8% damage range. With 29.9% of the lifetime in the damage range, HiperFer was found to be the superior material. Despite the fact that the effective strain range in the austenitic 316L steel (Δε
th., i.e. Δε
mech.
HiperFer = 0.71%, while Δε
mech.
316 L = 1.14%), i.e., dislocation hardening, was much higher, HiperFer reached a comparably high stable stress range level.
The TMF experiments demonstrated the high fatigue resistance potential of HiperFer steel. However, the total lifetime of a component is composed of the sum of the technical lifetime up to the initiation of detectable cracks (characterized for example by TMF life) and the residual lifetime (characterized by FCG). For this reason, it is necessary to develop a material that actively obstructs both the initiation and propagation of cracks in order to exploit the maximum material and component lifetime. HiperFer steels were developed with this objective in mind [
18]. A comparison of the ambient temperature FCG data (where thermomechanically induced precipitation in Laves phase-strengthened ferritic steel is not yet effective) of AM316L, conventionally produced 316L, and Crofer22 H in the precipitation annealed state gave an impression of the potential for the obstruction of fatigue cracks, inherent in these steel grades (
Figure 19).
In comparison to additively/conventionally manufactured austenitic 316L (cf. curves “AM316L-ZY, 2
nd”/“316L, 2
nd” in
Figure 19), it takes almost twice/1.5 times the stress intensity to initiate crack propagation in conventionally processed Crofer22 H. Over the whole stress intensity range, the crack propagation rate was lower than in conventional 316L. Crack propagation in the second, additively manufactured 316L specimen (AM316L-ZY) was slightly lower in comparison to Crofer22 H, but the specimen already hit the test termination criterion (a/W = 0.7) when Crofer22 H just reached the stable region of the cyclic crack propagation curve. However, HiperFer-like steel can only exploit its maximum potential at temperatures around 600 °C or above. As outlined in the previous section (cyclic), plastic deformation in fatigue loading in this temperature range results in “reactive precipitation hardening” [
32], which not only leads to higher strength, but to active crack obstruction [
17]. On the basis of this intrinsic mechanism, HiperFer 17Cr2 (conventionally manufactured variant) can reach up to an order of magnitude of higher residual fatigue lifetime than austenitic 316L (
Figure 20).
Crack propagation in HiperFer 17Cr2 remained at an almost constant velocity over a comparatively wide range of the cyclic crack propagation curve, despite increasing stress intensity ∆K.
In order to investigate the potential of the alloying system for AM, we developed a first model alloy (HiperFer
AM (Additive Manufacturing) for the additive manufacturing processes. The alloy composition was 22 Cr, 2 W, 1.5 Nb, and 1 Si (wt.-%). The increased Nb and Si contents in comparison to HiperFer 17Cr2 in combination with higher cooling rates in the AM process resulted in a highly supersaturated solid solution in the as-processed state, which in turn led to rapid precipitation kinetics, caused by high precipitation pressure in the direction of equilibrium. This resulted in a large amount of fine Laves phase particles (
Figure 21), even after the comparatively short high temperature periods of in situ heat treatment.
Figure 22 displays a typical microstructure of LMD-processed (by Fraunhofer ILT, Aachen) HiperFer
AM powder.
In addition, the grain boundaries were occupied by Laves phase particles (
Figure 21) in the as-manufactured state, stabilizing the grain structure without embrittling the material. As outlined in [
16,
17], this is an important issue to enable profit from dynamic strengthening. Pores contained in additively manufactured materials (cf.
Figure 21) acted as notches and led to accumulation of plastic strain in the surrounding material during cyclic loading. In HiperFer
AM, this led to thermomechanically induced precipitation [
18,
32] of Laves phase particles at these dislocations and strengthened the surroundings of the pores and actively delayed crack initiation. Further microstructural characteristics of HiperFer were particle-covered high angle grain boundaries and the particle-free zones (PFZs [
31,
33,
34,
35]) that formed along these (
Figure 21). Especially in higher tungsten-alloyed HiperFer, crack propagation through the densely particle populated grain interiors or to overcome the particle-covered grain boundaries is quite energy consuming.
For this reason, cracks did not necessarily follow fatigue typical intragranular paths, but to some extent followed the PFZs (
Figure 23). In consequence, crack growth could potentially be controlled by intelligent adjustment of PFZ width and grain size (i.e., particle-free volume).
A preliminary investigation into this issue covered a beam diameter variation in the LMD process (by Fraunhofer ILT, Aachen) from 3 to 0.66 mm and successfully provided differing microstructures. Three millimeters of beam diameter resulted in predominantly large, oriented, rod-shaped grains (
Figure 24a), while the 0.66 mm diameter tended to produce smaller, less rod-shaped, partly almost globular grains (
Figure 24b). Obviously grain size can be adjusted by changing the AM process parameters. By this, even microstructural adjustment to specific applications seems feasible. For example, the HiperFer
AM microstructure presented in
Figure 24a is suitable for application in the creep regime, while the microstructure displayed in
Figure 24b could be suitable for higher strength applications at low temperatures. Of course, mutual optimization of the chemical composition of HiperFer
AM to specific additive manufacturing techniques (LMD, LPBF, EBM (electron beam melting), WAAM (wire and arc additive manufacturing)) and vice versa is yet to come.