On the Influence of Alloy Composition on the Additive Manufacturability of Ni-Based Superalloys

Twelve different superalloy compositions have been studied in this work—five novel compositions and seven heritage grades, including the cast/wrought alloys IN625, IN718, and Waspaloy of moderate \(\gamma ^{\prime }\) content but then also IN713, IN738LC, IN939 and CM247LC which are widely employed for investment casting applications—these display excellent creep properties on account of their high \(\gamma '\) fraction. The alloys termed ABD850AM, ABD900AM, and AM-Dev are grades designed specifically for the AM process, with the objective of maintaining both performance and processability. Lastly, ExpAM and ABD850AM+CB are experimental compositions, with low and high carbon and boron contents respectively, aimed at assessing the influence of C and B, as well as a very high \(\gamma ^{\prime }\) volume fraction in the case of ExpAM. Feedstock powders were produced via argon gas atomization; their particle size distributions each had D10 and D90 values ranging from 19.2 to 56.1\(\,\mu \)m respectively and D50 between 32.3 and 36.5\(\,\mu \)m. The measured alloy compositions as determined by inductively coupled plasma-optical emission spectroscopy (ICP-OES) and ICP-combustion analysis are given in Table I.

Processing was carried out by the laser powder bed fusion (L-PBF) method in an argon atmosphere using a Renishaw AM 400 pulsed fiber laser system of wavelength 1075 nm and build plate size of 250\(\times \)250 mm\(^2\). The processing parameters employed were: laser power 200 W, laser spot size diameter 70\(\,\mu \)m, powder layer thickness 30\(\,\mu \)m, and pulse exposure time 60\(\,\mu \)s. A ‘meander’ laser scan path pattern was used with hatch spacing of 70\(\,\mu \)m and laser speed of \(0.875\,\)m/s, the path frame of reference was rotated by 67° with each layer.^[13] Consistent with Renishaw’s recommended practice, the speed on the borders was reduced to \(0.5\,\)m/s. To assess the composition-dependence of the severity of cracking, cubes of dimensions 10 \(\times \) 10 \(\times \) 10 mm\(^3\) were printed. Vertical bars of dimensions 10 \(\times \) 10\(\times \) 52 mm\(^3\) were also printed to assess the mechanical properties of the alloy variants. These were manufactured with 16 inverted pyramid legs to allow for easy removal from the baseplate. Cylinders of diameter 3 mm and height 1 mm were produced for differential scanning calorimetry (DSC). Each alloy and geometry was processed with identical parameters, which were selected on the basis of a pre-requisite study that optimized processing conditioning to minimizing cracking using this manufacturing system.^[14]

The printability was assessed by measurement of cracking severity and morphology. First, extensive quantitative stereological assessment was made with optical micrographs taken using a Zeiss Axio A1 optical microscope at 100\(\times \) magnification. For this purpose, the cube samples were sectioned on (i) planes with normal corresponding to the build direction (termed XY plane) at mid-height and (ii) planes with normal perpendicular to the build direction (termed XZ). The severity of cracking was determined by imaging the XY plane at the sample height midpoint and on the XZ plane through the center; 5 images were considered per alloy. In this analysis the bulk of the sample—400\(\,\mu \)m from the sample edge—was considered. The ImageJ software was used to compare the alloys by determining the crack count density (cracks/mm\(^2\)) and crack length density (mm/mm\(^2\))—note that the crack length was defined as the caliper diameter corresponding to the largest line length across each crack.

In the case of a pulsed laser, solidification occurs in discrete melting events; it follows that a solidification crack must be of length equal to or smaller than a single melt pool radius. Conversely, any crack of length greater than one melt pool radius must have propagated as a solid-state crack, although of course one cannot eliminate the possibility that the solid-state crack propagated from a solidification crack. In order to provide quantitative information, the perimeter to area ratio (units \(\mu \)m\(^{-1}\)) of cracks was assessed. Images were taken at a magnification of 100\(\times \), giving a resolution high enough to identify the jagged features of solidification cracks. A further 70 images were taken of each alloy to compare the contributions of each mechanism on a representative scale; where cracking occurred, many thousands of cracks were analyzed.

To garner further information concerning cracking mechanisms at higher spatial resolution, a Zeiss Merlin Gemini 2 field emission gun scanning electron microscope (FEG-SEM) was used equipped with an Oxford Instruments XMax 150 mm/mm\(^2\) energy-dispersive X-ray spectroscopy (EDX) detector. Line scans were performed to determine qualitatively the composition of carbides observed in the microstructure. Scans comprised 500 points at an accelerating voltage of 5 kV and probe current 500 pA resulting in an estimated interaction volume of width and depth of \(\sim \)100 nm. Samples for microstructural observation were polished to 0.04\(\,\mu \)m colloidal silica finish, the microstructure of the alloys after manufacturing always comprised \(\gamma \) and MC carbide with \(\gamma ^\prime \) absent; an example of the CM247LC as-printed microstructure compared to the heat-treated microstructure following electrolytic etching with 10 pct phosphoric acid at 3 V direct current is given in Figure 1. In what follows, it is emphasized that only material in the as-printed condition is considered.

Uniaxial tensile, isothermal stress relaxation and constrained cooling tests were performed with an Instron electro-thermal mechanical testing (ETMT) machine developed at the National Physical Laboratory (NPL). Specimens were machined with axis along the build direction using electrical discharge machining (EDM) with dimensions as given in Figure 2. To negate the influence of any machining-induced surface roughness, all surfaces were polished to a 4000 grit finish by hand. For each of the three types of test, joule heating (using a 400-A DC power supply) under free expansion conditions (\(\sigma = 0\) MPa) was employed to reach the test temperature using a heating rate of 200 K/s, thus avoiding any second phase precipitation. The test temperature was monitored and controlled with a type K thermocouple which was fusion-welded under argon gas to form a small bead and subsequently spot-welded under argon to the center of the specimen. Further insights into the workings of the ETMT system have been reported in Reference 15. Since it was needed for the interpretation of our results, an estimate of the average linear coefficient of thermal expansion between 300 °C and 800 °C was made during this pre-test sample heating as shown in Figure 3. The coefficient of thermal expansion is found to have a dependence on the Al content, as previously reported in Reference 16.

Isothermal uniaxial tensile tests were performed for all the alloys at room temperature and at 800 °C—this temperature being taken to be representative of where the ductility dip is likely to be severe.^[4] Specimens were strained rapidly at a rate of \(10^{-2}\,\)s\(^{-1}\) to reduce the influence of any dynamic precipitation during testing. An iMetrum non-contact digital image correlation system was employed for measuring the strain; 3 repeats were performed for each alloy. The elastic modulus of each alloy was determined between 150 and 300 MPa. The modulus determined at 800 °C correlated with increasing equilibrium \(\gamma ^\prime \) volume fraction, see Figure 3, though was likely due to the contributions of \(\gamma ^\prime \) formers in solid solution of \(\gamma \).

Stress relaxation tests were performed to quantify the capacity of the alloys to relieve the stress accumulated by processing. Tensile specimens were heated to 1000 °C allowing for free expansion to occur as shown in Figure 2, followed by a loading to 100 MPa at 25 MPa/s after which the sample grips were instantaneously locked into position, in order to observe the relaxation of the stress during cooling.^[17] The ability to stress relax was benchmarked by comparing the stress remaining after 60 s. Further stress relaxation tests at various supersolvus temperatures (1100 °C to 1200 °C) were performed on the ABD850AM+CB, Waspaloy, and IN939 alloys by the same procedure.

A constrained bar cooling test was performed to simulate the stress build up expected of additive manufacturing. Tensile specimens were heated under free expansion conditions to 1100 °C then left for 20 seconds to ensure thermal-mechanical equilibrium. The displacement of the ETMT grips was then fixed, and the sample was cooled to 300 °C. The accumulated stress during cooling was measured for the various alloys at cooling rates of 100 K/s, 80 K/s, 50 K/s, and 25 K/s.

A NETZSCH 404 F1 Pegasus DSC was used to compare solidification behavior of the as-processed microstructure, albeit at cooling rates far lower than those experienced during AM processing. The alloys were subjected to a thermal history consisting of (i) heating to \(700\,^\circ \)C at 20 °C/min, (ii) heating from 700 °C to 1450 °C at 10 °C/min (iii) cooling to 700 °C at the same rate, and (iv) cooling to room temperature at 20 °C/min. The DSC signal from the first heat causes the transformation from a non-equilibrium additively manufactured microstructure to a homogeneous liquid. The first cool was indicative not of AM conditions, but of the solidification path from homogeneous liquid to solid at 10 °C/min. The thermophysical properties of interest derived from the DSC signal upon heating were the onset of incipient melting \(T_{\text{S'}}\), and the liquidus \(T_{\text{L}}\), as illustrated in Figure 16(a). From cooling the onset of solidification \(T_{\text{L'}}\) and the solidus temperature \(T_{\text{S}}\) were extracted as shown in Figure 16(b). The solidus and onset of incipient melting as well as liquidus and onset of solid nucleation temperatures were estimated following the procedures of Boettinger, Chapman, and Quested.^[18‐20] In order to evaluate legacy crack criteria on the basis of an experimentally determined solidification path, the solid fraction as a function of temperature was estimated by considering it proportional to the ratio of the enthalpy of the transformation at T to the total enthalpy of the phase transformation. This estimate is subject to possible error because other phase transformations may have occurred throughout the freezing range.

Modeling of solidification under equilibrium and Scheil–Gulliver assumptions was implemented using Thermo-Calc employing the TCNI8 and TTNi8 databases to simulate the alloy solidification path.^[21] In the case of the Scheil models, solidification was considered to have ended at a liquid fraction remaining of 0.01. The propensity for backdiffusion of C to occur was deemed negligible in accordance with the dimensionless back diffusion parameter \(\alpha \).^[22]where D is the chemical diffusion coefficient of carbon in \(\gamma \) at temperature,^[23] \(\lambda _{\text{cell}}\) is the cell spacing, and \(\Delta T_{\text{freezing}}\) is the magnitude of the freezing range. Given an experimentally determined \(\lambda _{\text{cell}}\) of 1\(\,\mu \)m,^[4] and typical cooling rates reported in the literature—which indicate the cooling rates in the solidifying material range between 10\(^{4}\)K/s and 10\(^{6}\,\)K/s^[24‐27]—the value of \(\alpha \) is substantially less than unity. Hence, backdiffusion of C was deemed negligible and modeling of solidification under Scheil assumptions was considered a reasonable first approximation.

Our experimentation involving extensive characterization of cracking provides proof of the significant effect of composition on processability. This is apparent in the optical micrographs of the XY transverse section—normal corresponding to build direction—in Figure 4. Cracking is prevalent in the legacy alloys IN713, IN738LC, IN939, and CM247LC as well as the two experimental compositions—ExpAM and ABD850AM+CB. The other alloys examined here process largely crack-free. The distribution of cracks across the XY plane is uniform in IN713 and ExpAM, whereas the others cracked to a greater extent at the edges, where the effective laser scan speed was reduced. If one puts aside the prevalence for surface cracking for one moment and considers the cracking information for the bulk, the data for different measures of crack severity developed here—count density and length density—are consistent with each other, see Figure 5. When examined on the longitudinal XZ plane, there is a clear disposition for the cracks to be aligned along the build direction, see Figure 6, with SEM images indicating that cracks of both solid-state type and solidification type are present in all cracked alloys. Solid-state cracks tend to exhibit a long and straight morphology, whereas solidification cracks appear more jagged.

In order to strengthen our findings further, detailed quantitative stereological analysis has been employed to quantify the extent to which solidification and solid-state cracks occur in each alloy, and to allow for distinguishing between them. We emphasize first the quantification of cracks at depth in the bulk, since these are the greatest in number and since it may be possible to improve the surface cracking susceptibility by tailoring the processing conditions at the surface. The morphology of cracks in different alloys is summarized by the kernel density estimate of the probability distribution of the perimeter to area ratio \(\mu \)m\(^{-1}\) of cracks in each of the 6 cracked alloys, see Figure 7. Solidification cracks, distinguished by their characteristic jagged morphology and lower perimeter to area ratio, arise to greater extent in the IN738LC, IN939, CM247LC and ABD850AM+CB. This is proven by the deviation of their perimeter to area ratio distributions, which possess a smaller median, interquartile range, and lower/upper adjacent values relative to IN713 and ExpAM. Solid-state cracking is the dominant mechanism in IN713 and ExpAM; in these alloys, cracks are longer and straighter, exemplified by the shift of their perimeter to area distributions towards greater values. Nevertheless, overlap between the distributions for some alloys is apparent for approximately 0.5 < perimeter/area < 0.7, suggesting that the two mechanisms are both operating simultaneously and competing against each other. Cracks in this range are of mixed mode, indicating they began as solidification cracks and then propagated further in the solid-state. The probability distribution of ABD850AM+CB is cut off at 0.19 and 0.67 as the distribution is comprised of only 24 cracks observed.

Finally, consider again the propensity for some alloys—IN738LC, IN939, CM247LC and ABD850AM+CB—to exhibit increased cracking at sample edges. In these alloys, cracks near edges observed on the XY plane have uniform spacing of \(\approx \) 80\(\,\mu \)m—this periodicity is attributed to the preferential nucleation of cracks to minimize strain energy.^[28] The slower scan speed used at the sample edge resulted in coarser microstructure less able to accommodate strain, with larger grains with increased texture component as compared to the bulk.^[29] Furthermore, the sample edge was processed with greater energy density and larger melt pools, both of which have been observed in the literature to exacerbate solidification cracking in particular.^[30] ExpAM and IN713 crack at the edges, though not comparatively more than in the bulk. However, the 4 alloys mentioned above which are prone to more solidification cracking in the bulk exhibit more cracking at the edge (relative to their bulk) as they are more sensitive to the local microstructure and processing conditions.

Figure 12 confirms that the stress experienced during cooling in the constrained-bar test depends upon alloy composition. But various factors will contribute to this to varying extents—the composition-dependence of elastic modulus and thermal expansion coefficient of Figure 3, but also a possible stress relaxation effect—and it is helpful to deconvolute these in order to determine which is the more important. Further experimentation has indicated that cooling rate does indeed influence somewhat the build up of stress, see Figure 17 which includes data for ABD850AM+CB, Waspaloy and IN939. Clearly, slower cooling provides more time for stress relaxation. Particularly for the stronger IN939 alloy, the initial rate of change of stress with temperature shows no dependence upon cooling rate, implying the need for interpretation based upon thermal-elastic-plastic behavior with no creep relaxation playing any role in that initial regime. Figure 18 provides further insight into what is happening in these tests—consider the case of the least strong and most strong alloys considered here, IN625 and CM247LC respectfully. The variation of the measured stress with temperature for the two cases matches closely the product of elastic modulus, thermal expansion efficient and temperature difference \(E\alpha \Delta T\) for several hundred degrees below the initial temperature of 1100 °C. Particularly for the stronger alloy CM247LC, the product \(E\alpha \Delta T\) lies below the measured uniaxial flow stress initially but surpasses it in the ductility-dip regime. For the weaker alloy IN625, the measured stress in the constrained bar test determined at low temperature matches well the measured flow stress consistent with only a weak work-hardening effect.

Modeling of the constrained bar test puts the thermal-mechanics of this situation on a quantitative basis. The net rate of straining is zero; therefore the rates of elastic, plastic, creep and thermal straining \(\dot{\varepsilon }_{\text{elastic}}\), \(\dot{\varepsilon }_{\text{plastic}}\), \(\dot{\varepsilon }_{\text{creep}}\) and \(\dot{\varepsilon }_{\text{thermal}}\) must satisfyIn this expression, \(\dot{\varepsilon }_{\text{thermal}} = \alpha \dot{T}\) is determined by the cooling rate, we find that as the specimen is cooled one of the mechanical strain-rates (\(\dot{\varepsilon }_{\text{elastic}}\), \(\dot{\varepsilon }_{\text{plastic}}\) or \(\dot{\varepsilon }_{\text{creep}}\)) dominates, with the range of dominance of each mechanism and therefore the detailed response depending on cooling rate and the mechanical properties of the material. Initially, the stress \(\sigma \) equals zero. As the sample is cooled the response is elastic andGiving, \(\sigma = \sigma _{\text{elastic}} = E \alpha \Delta T\), as shown in Figure 18 for IN625 and CM247LC. As the stress increases eventually creep takes over as the dominant mechanism. For a material that creeps according to Norton’s Law we findwhere R is the ideal gas constant and the parameters approximated by fitting are the creep exponent n (4.8 for both Waspaloy and CM247LC), the activation energy for creep Q (800 kJ/mol for Waspaloy and 1000 kJ/mol for CM247LC) and the pre-exponential constant in Norton’s Law A (\(6.1\times 10^{-7}\,\)MPa\(^{-n}\,\)s\(^{-1}\) for both alloys). The actual stress is the smallest of those given by the elastic response and Eq. [4]. The initial response is elastic, but as the sample is cooled the stress increases, creep takes over as the dominant mechanism and the stress is given by Eq. [4]. For Waspaloy, which has a low creep strength at elevated temperatures, the initial elastic range is very short and creep becomes the dominant process at low stresses, as shown in Figure 19(a), As the temperature is decreased further the creep strength and therefore the stress given by Eq. [5] increase. In the case of Waspaloy, we find that eventually elastic deformation dominates again. This transition occurs whenSubstituting [4] into [5] gives a relationship between the stress and temperature at the creep-elastic transitionwhich we have designated using the subscript ce. The actual stress and temperature a the transition is given by the intersection of Eqs. [4] and [6] on a plot of \(\sigma \) against T. When \(T < T_{\text{ce}}\) the response is elastic the stress is given byuntil the onset of plastic deformation. For quantitative calculations one needs a plasticity law, taken here to be the Ramberg-Osgood relationship with \(\sigma _{\text{y}}\) the material’s yield stress and m a material-dependent constant, fitted to be 3.4 and 3.6 for Waspaloy and CM247LC respectively, consistent withwhere for simplicity we have assumed that \(\sigma _{\text{y}}\) does not vary with temperature. The transition to plastic deformation occurs when \(\sigma \) given by Eqs. [4] and [7] equals \(\sigma _{\text{y}}\). We designate the temperature at which this occurs as \(T_{\text{y}}\). It follows that subsequent response is given byand we can describe the stress as follows for \(T < T_{\text{y}}\)To summarize, the stress development during cooling can be represented by the equation set:It has been found that the above formulation can largely rationalize the temperature-dependence of the build of stress with decreasing temperature, see Figure 19 for Waspaloy and CM247LC. All the above regimes of behavior are not exhibited by all materials. In the case of Waspaloy, the initial elastic response is small and creep dominates for the slow cooling rate considered in Figure 19 until approximately 900 °C when the behavior is elastic; and it becomes thermal-plastic at \(\sim \)700 °C. For CM247LC which displays much greater creep resistance, the deformation is initially linear thermal-elastic over a wide temperature regime until the onset of rate-insensitive plasticity again at \(\sim \)700 °C. In each case the temperature range over which creep occurs is small and therefore limited creep deformation occurs during cooldown. In order to compare the plastic strain accumulation to the determined composition-dependent ductility exhaustion limit, a strain-rate independent first order approximation can be made of the plastic strain as followsOur calculations indicate a critical minimum ductility in the range 0.5 to 1 pct is likely to be needed for the avoidance of ductility-dip cracking in the critical temperature range of 700 °C to 900 °C. These findings rationalize the poor resistance of the likes of IN713 and ExpAM to solid-state cracking.

To summarize, the analysis presented in this section allows a number of critical points to be confirmed. First, the balance of evidence suggests that solid-state cracking—in susceptible alloys—will occur in the ductility-dip regime due to the brittleness which is prevalent there. Second, since the cooling rates of the AM process are extremely rapid,^[24‐27] the development of stress will be thermal-elastic-plastic in origin, such that for the highest strength alloys there will be little influence of stress relaxation on the composition-dependence of cracking susceptibility. Third, nonetheless the alloy composition is important: the greater the extent of alloying, the greater the stress build up to drive cracking and the greater the risk of the yield stress being exceeded in the ductility-dip regime. Thus, the cracking risk is exacerbated by greater alloy strength but for any given strength, the risk is ameliorated if the low-temperature ductility is greater. These two factors are strongly correlated via the strength/ductility relationship. Finally, the critical ductility needed to avoid solid-state cracking appears to be in the range 0.5 to 1 pct in the ductility dip regime. Obviously, the above applies only to solid-state cracking and ignores the solidification cracking effect which is now considered in the following section.

So far, the occurrence of solidification cracking has been correlated with the magnitude of the freezing range and presence of continuous films of solute enrichment at crack tips. But a number of composition-dependent solidification cracking criteria should be considered, for example those due to Clyne & Davies (CD),^[44] Rappaz, Drezet, & Gremaud (RDG)^[45] and Kou.^[46] In each case, a parameter \(\Phi \) can be identified, which is predicted to scale with cracking susceptibility. The CD approach for instance considers the dimensionless ratio of the time during solidification during which the alloy is vulnerable to cracking \(\Delta t_{\text{vulnerable}}\) to the time during which it can relieve the accumulated stress \(\Delta t_{\text{stress relief}}\). Assuming a constant cooling rate this ratio is proportional to each respective temperature range, consistent withThe CD criterion can be criticized on account of its rather arbitrary definition of the range during which the material is vulnerable; however, keeping this constant for the 12 alloys allows for comparison. The alternative RDG criterion is based on a balance of pressure, such that the formation of hot tears depends upon the maximum strain rate \(\dot{\varepsilon }_{\text{max}}\) before a cavitation pressure \(\Delta p_{\text{c}}\) results in the formation of a void. Both the mechanical contribution and the ability for liquid to back fill are dictated by integrals of the solid fraction vs temperature \(f_{\text{s}}(T)\), A, B, and C^[45] consistent withwhereandwhereFor the calculations carried out here, the viscosity \(\mu \), the thermal gradient G, the velocity of the isotherms \(v_{\text{T}}\), and the shrinkage factor \(\beta \) have been taken from Reference [45], the secondary arm spacing \(\lambda _2\) from Reference [4]. The grain coalescence temperature \(T_{\text{cg}}\) is taken to correspond to \(T_{f_{s}=0.99}\). The hot cracking susceptibility is proportional to \(\dot{\varepsilon }^{-1}_{\text{max}}\) and values are normalized between zero and unity. The RDG criterion suffers from a similar criticism to the CD criterion, as the grain coalescence temperature—critical to evaluating the integrals A, B and C—is not well understood. Furthermore, approximating liquid filling by the flow of fluid through a packed bed may not be the most appropriate for the cellular AM microstructures. Finally, the Kou solidification cracking index is formulated around the balance of increasing spatial dimensions associated with solid formation and volumetric liquid flow within a differential control volume between two growing grains.^[46] Here, the cracking susceptibility is taken to be proportional to the maximal value of the ratio of change in temperature \({\text{d}}T\) relative to the change in square root in the solid fraction \({\text{d}}(\sqrt{f_{\text{s}}})\) near the final stages of solidification, consistent withFigure 20 summarizes the predictions of the different criteria considered. One can see that, with the exception of the CD index evaluated with the experimentally-evaluated DSC traces (which probably gives an underestimate of the solidification range)—the indices are in line with the findings of our experimentally-determined solidification cracking susceptibility. However, no single index predicts correctly the composition-dependence of solidification cracking for all methods of determining \(f_{\text{s}}\{T\}\). Moreover, no one method of estimating \(f_{\text{s}}\{T\}\) correctly predicts solidification cracking across all the indices. The strength of applying these legacy criteria to multi-component Ni-superalloys is their sensitivity to the final stages of solidification whereby the final liquid is solute enriched and stable at lower temperatures. This in turn presents a problem, as \(f_{\text{s}}\{T\}\) is discontinuous when a new phase becomes stable, making the exact numerical procedure by which the Kou criterion is evaluated highly sensitive to these discontinuities.

One can argue that a modified criterion is needed to take account of contributions of mechanical and stress relaxation properties. A very detailed analysis is beyond the scope of this paper, but approximate calculations can be made as follows. Assume the contribution of stress relaxation is more pronounced at near solidus temperatures as is suggested by the strong temperature dependence of stress relaxation of Figure 11. A modified criterion could consider the solidification cracking susceptibility to be proportional to the ratio of strain accumulated during the vulnerable regime \(\alpha \Delta T_{\text{vulnerable}}\) to the strain relieved in the stress relief regime. We apply this as shown in Eq. [19] using experimentally determined values of E, and \(\alpha \), and the \({\text{d}}\sigma /\text{d}t\) derived from the secant line of the stress relaxation behavior from 0 to 30 s at 1000\(\,^{\circ }\)C, according to:This is in essence a mechanically-weighted CD index. The modified index increases the relative predicted susceptibility of the solidification cracking alloys IN738LC, IN939, and CM247LC appreciably, see Figure 20, by accounting for thermal expansion, stiffness and the capacity of the material to stress relax.