On the printability and transformation behavior of nickel-titanium shape memory alloys fabricated using laser powder-bed fusion additive manufacturing

Journal of Manufacturing Processes 35 (2018) 672–680

Contents lists available at ScienceDirect

Journal of Manufacturing Processes journal homepage: www.elsevier.com/locate/manpro

Technical Paper

On the printability and transformation behavior of nickel-titanium shape memory alloys fabricated using laser powder-bed fusion additive manufacturing M. Mahmoudia, G. Tapiaa, B. Francob, J. Mab, R. Arroyaveb, I. Karamanb, A. Elwanya, a b

T

⁎

Department of Industrial & Systems Engineering, Texas A&M University, College Station, TX, United States Department of Materials Science & Engineering, Texas A&M University, College Station, TX, United States

A R T I C LE I N FO

A B S T R A C T

Keywords: Additive manufacturing Laser powder-bed fusion Nickel-titanium Transformation temperature Printability Shape memory alloys Energy density Delamination Selective laser melting

Laser powder-bed fusion (L-PBF) additive manufacturing is regarded as an attractive alternative for producing parts with complex geometries for nickel-titanium shape memory alloys (NiTi SMAs). These alloys are known to pose challenges when processed using traditional subtractive or formative manufacturing technologies. Although L-PBF of NiTi has been investigated in some previous research efforts, very little emphasis has been placed on the manufacturability (or printability) of NiTi, where we use printability to refer to the capability of producing parts free of macroscopic defects. The current study elucidates challenges related to the printability of NiTi SMAs using L-PBF, and its interaction with their phase transformation behavior, responsible for their functional properties. More specifically, we conduct experiments and employ machine learning classification techniques to identify an adequate design parameter and an empirical rule for determining the printability of NiTi. Our results indicate that the linear energy density EL is a better design parameter for identifying satisfactory printability, while volumetric energy density, EV, is more relevant in controlling the transformation behavior of the processed material.

1. Introduction Since their discovery in the 1960s, nickel-titanium Shape Memory Alloys (NiTi SMAs) have found many applications in the automotive, aerospace, robotic, and biomedical industries [1]. SMAs are characterized by the shape memory effect (SME) and superelasticity (SE), which are the results of temperature-induced and deformation-induced reversible solid-to-solid phase transformations, respectively, enabling the part to recover its original shape after plastic deformation [2]. NiTi is a popular class of SMAs due to their biocompatibility [3], high transformation strain, high corrosion resistance, and high ductility [1]. Although NiTi SMAs have existed for approximately five decades and been used in various applications, the majority of manufactured NiTi parts have been limited to simple geometries such as wires, tubes, and sheets [4,5]. This is primarily because fabricating NiTi using conventional manufacturing methods such as machining, casting, or powder metallurgy and scaling up the production are difficult due to the high reactivity, and poor machinability of NiTi [6,7,3,8,9]. Furthermore, controlling the transformation behavior of NiTi SMAs is challenging

due to their high sensitivity to compositional variations, which is exacerbated by the loss of Ti to the formation of oxides and carbides, and Ni evaporation during melting practices or fabrication using conventional powder metallurgy techniques. For example, less than 0.5 at. % composition change on the Ni side of the stoichiometry results in changes in transformation temperatures in excess of 100 °C [10,11]. The challenges noted above have inspired the investigation of new manufacturing technologies that can address some of these challenges. Metal Additive Manufacturing (AM) techniques have been proposed as viable candidates [3,4,12]. First, metal AM processes enable the production of NiTi parts with complex geometries due to the layerwise nature of the process. Moreover, modulating the manufacturing process parameters during metal AM can potentially be used to achieve control on composition and microstructure of processed NiTi, allowing for tailoring its functional response [13,14]. Among existing metal AM technologies, laser powder bed fusion (L-PBF) processes, commercially known as Selective Laser Melting (SLM) or Direct Metal Laser Sintering, is the most frequently investigated technique in the AM of NiTi due to its capability of producing parts with high feature resolution and

⁎

Corresponding author. E-mail addresses: [email protected] (M. Mahmoudi), [email protected] (G. Tapia), [email protected] (B. Franco), [email protected] (J. Ma), [email protected] (R. Arroyave), [email protected] (I. Karaman), [email protected] (A. Elwany). https://doi.org/10.1016/j.jmapro.2018.08.037 Received 22 February 2018; Received in revised form 30 August 2018; Accepted 31 August 2018 1526-6125/ © 2018 Published by Elsevier Ltd on behalf of The Society of Manufacturing Engineers.

Journal of Manufacturing Processes 35 (2018) 672–680

M. Mahmoudi et al.

Table 2 Experimental design matrix taking three process parameters into account as the variables.

Fig. 1. SEM image of the initial NiTi powder of the process parameters used.

Table 1 Lower and upper bounds. Parameter

P

v

h

Lowest level Highest level

35 W 50 W

70 mm/s 450 mm/s

35 μm 120 μm

comparatively lower surface roughness than other AM processes like Directed Energy Deposition (DED). Due to the large number of process variables and parameters involved in L-PBF (e.g. laser power, scan speed, and hatch spacing) [15], successfully fabricating fully dense metal parts and controlling their properties require significant efforts in process planning and optimization, see for example [16–18]. Elahinia et al. has provided a recent review on the fabrication of NiTi alloys using AM, see [4]. The objective of the current study is to investigate the aspects related to the manufacturability of NiTi parts using L-PBF, and study the effect of different process parameters on the properties of these parts. More specifically, we investigate the effects of laser processing parameters (laser power, scan speed, and hatch spacing) on: (1) the ability of successfully producing macro defect-free parts, and (2) the transformation behavior of the fabricated parts. We discuss the effectiveness of previously proposed design parameters, such as volumetric energy density EV, in planning successful fabrication of NiTi using manufacturing experiments, characterization techniques, and machine learning tools. Ultimately, we propose the linear energy density, EL, as a more reliable design parameter for L-PBF of NiTi, and introduce a window of EL within which NiTi parts are more likely to be successfully printed. Prior works in the literature have studied various aspects related to the L-PBF of NiTi, including the effects of process parameters on structural and mechanical properties, density, impurity content, phase transformation, SME, and SE [12,19–24]. In most of these works, a maximum laser power of 100 W was employed, and energy density was defined as

ωV =

P , ρr ·d·t·v

Run No.

P(W)

v (mm/ s )

h (μm)

EV (J/ mm3)

ES (J/ mm2)

EL (J/ mm)

Building result

1 2 3

35.5 50 50

307 450 80

66 120 35

58.4 30.9 595.2

1.75 0.93 17.86

0.116 0.111 0.625

4 5 6 7 8

35 35 50 35 50

450 80 450 450 80

35 120 35 120 120

74.1 121.5 105.8 21.6 173.6

2.22 3.65 3.17 0.65 5.21

0.078 0.438 0.111 0.078 0.625

9 10 11 12 13 14 15 16 17 18 19 20

49.5 45.5 47.5 48 48.5 43.5 43 38 41 42.5 36.5 41

279 270 122 288 353 427 177 205 196 223 149 85

85 119 81 113 79 49 117 62 83 98 91 40

69.6 47.2 160.2 49.2 58.0 69.3 69.2 99.7 84.0 64.8 89.7 402.0

2.09 1.42 4.81 1.47 1.74 2.08 2.08 2.99 2.52 1.94 2.69 12.06

0.177 0.169 0.389 0.167 0.137 0.102 0.243 0.185 0.209 0.191 0.245 0.482

21 22 23 24 25 26 27 28 29 30 31 32

44 35 39 42 39.5 44 40 36.5 48.5 44.5 45 45.5

418 297 390 436 334 131 445 316 381 251 325 94

106 76 64 57 102 89 93 42 59 72 53 68

33.1 51.7 52.1 56.3 38.6 125.8 32.2 91.7 71.9 82.1 87.1 237.3

0.99 1.55 1.56 1.69 1.16 3.77 0.97 2.75 2.16 2.46 2.61 7.12

0.105 0.118 0.100 0.096 0.118 0.336 0.090 0.116 0.127 0.177 0.138 0.484

33 34 35 36 37 38 39 40

46 47 41.5 36 40.5 48.5 50 50

399 362 103 371 214 110 103 70

51 104 55 96 70 77 116 120

75.4 41.6 244.2 33.7 90.1 190.9 139.5 198.4

2.26 1.25 7.33 1.01 2.70 5.73 4.18 5.95

0.115 0.130 0.403 0.097 0.189 0.441 0.485 0.714

41 42

45.5 49

115 85

39 106

338.2 181.3

10.14 5.44

0.396 0.576

43 44

41 48

83 89

95 113

173.3 159.1

5.20 4.77

0.494 0.539

45 46

44.5 48.5

120 109

96 62

128.8 239.2

3.86 7.18

0.371 0.445

47

49

91

47

381.9

11.46

0.538

Defective Defective Nondefective Defective Defective Defective Defective Nondefective Defective Defective Defective Defective Defective Defective Defective Defective Defective Defective Defective Nondefective Defective Defective Defective Defective Defective Defective Defective Defective Defective Defective Defective Nondefective Defective Defective Defective Defective Defective Defective Defective Nondefective Defective Nondefective Defective Nondefective Defective Nondefective Nondefective

parameters was suggested with P = 77 W, v = 200 mm/s, h = 50 μm, corresponding to an energy density of ωV = 234 J/mm3 [22]. In another study [25] by the same group, the volumetric energy density was introduced as

EV =

(1)

where P is the effective laser power (measured on the surface of the powder bed), ρr is the relative density of the powder bed, d is the laser beam diameter, t is the powder layer thickness, and v is the laser scan speed. To minimize variations in the phase transformation temperatures and minimize impurity content, an optimal set of process

P , h·t·v

(2)

where h is the laser hatch spacing; the distance between two adjacent passes of the laser beam within the same layer. It was shown that both austenitic and martensitic transformation temperatures for the fabricated parts are positively correlated with volumetric energy density since higher energy input results in: (1) preferential evaporation of 673

Journal of Manufacturing Processes 35 (2018) 672–680

M. Mahmoudi et al.

Fig. 2. (a) Defective coupons built using three different process parameter settings exhibiting different levels of delamination, and (b) non-defective coupons.

gun, dual beam scanning electron microscope showing predominantly spherical particle morphology. Specimens were fabricated on a ProX 100 TM DMP commercial L-PBF system by 3D Systems, equipped with a fiber laser beam having a Gaussian profile with wavelength λ = 1070 nm, beam spot size of approximately 100 μm-diameter, and a maximum power of 50 W. Argon was used as inert protective atmosphere during fabrication. The laser selectively fuses the NiTi powder, guided by the digital CAD model, to form one layer, and then the process is repeated in subsequent layers. Coupons were built on a substrate made from bulk NiTi to ensure compatibility with the processed materials. Cubic test coupons of size 10 mm × 10 mm × 10 mm were fabricated under different manufacturing process parameter settings in order to investigate two aspects: (1) printability, and (2) transformation behavior. The powder layer thickness was maintained at a constant value of t = 30 μm. The laser power, scan speed, and hatch spacing for different coupons were selected following a 47-run Latin hypercube design for 3 factors. The ranges of the 3 parameters used are shown in Table 1. The Latin hypercube design is relevant because it spreads the design points uniformly over the parameters space [28]. The laser scanning pattern within the build plane consisted of parallel hatches (or rasters) at 45° and −45° angles for odd and even numbered layers, respectively. Phase transformation temperatures of fabricated parts were then measured using Differential Scanning Calorimetry (DSC) with samples of size 3 × 3 ×1mm3 were cut from the fabricated coupons using wire electric discharge machining (EDM), and then performed according to ASTM F2004-05 standard using a TA Instruments Q2000 calorimeter to measure Ms (austenite-to-martensite transformation start temperature), Mf (austenite-to-martensite transformation finish temperature), As (martensite-to-austenite transformation start temperature), and Af (martensite-to-austenite transformation finish temperature). In order to avoid having confounding factors in the experiments, the EDM-induced heat affected zone was cut from the DSC samples using precision saw. The heating and cooling rates were set to 10 K/min, and the phase transformation temperatures were determined using the intersecting tangents method following the ASTM standard. Wavelength-Dispersed Spectroscopy (WDS) experiments were done using a Cameca SXFive at with a LaB6 source. In addition to measuring the transformation temperatures, we conducted microstructural characterization of the NiTi coupons in longitudinal (parallel to the building z-direction) and transverse (perpendicular to the longitudinal plane) cross sections. The two surfaces were prepared using mechanical grinding, polishing and etching. The grinding was done using abrasive SiC paper to a grit size of 1200. Polishing was done using a colloidal silica suspension. Following the recommendation in [29], electro-polishing with the electrolyte

nickel, and (2) formation of Ni-rich phases. Walker et al. [26] later reported that lower values of volumetric energy density can minimize transformation temperatures and impurity content. In their work, they considered the effects of process parameters on part density, impurity content, transformation behavior, and shape memory behavior by printing single tracks with power levels up to 300 W. It was shown that laser power has a positive linear correlation with track width. In addition, the authors reported that too low or too high energy density are undesirable. On one hand, high energy inputs lead to wider melt tracks with irregular and wavy surfaces, which results in corrupting subsequent layers, interruption of fabrication, formation of cavities, and powder deposition failures. On the other hand, low energy inputs can result in discontinuous melt lines, hindering the fabrication of fully dense defect-free parts. They suggested power P = 250 W, laser scan speed v = 1250 mm/s, powder layer thickness t = 30 μm, and scan hatch spacing h = 120 μm as the optimal set of parameters corresponding to a volumetric energy density of Ev = 55.5 J/mm3. Some authors have tried using other design parameters: Li et al. [27] assessed the influence of L-PBF process parameters on the microstructural and structural integrity of Ti-rich NiTi, as well as the impact of the postprocess homogenization treatment on the microstructure and phase transformations. The authors used hatch spacing h and linear energy P density (defined as EL = v ) as design parameters and tried to simultaneously minimize porosity and stress-induced cracks. The authors concluded that the linear energy density was a useful tool in identifying the process window for L-PBF of Ti-rich NiTi. While previous works propose the volumetric energy density EV as an acceptable design parameter and use it as the basis for further analyses, we conduct experiments to evaluate the effectiveness of EV for manufacturability of NiTi and compare it with another design parameter EL. In addition, our experimental results shed light on how different process parameters affect the transformation behavior of NiTi parts fabricated using L-PBF. The rest of the paper is organized as follows: Section 2 describes the experimental setup and materials used for this work as well as sample preparation procedures and measurement instruments. Section 3 presents the data and results found from the experiments. Section 4 provides discussion and insight on the results. Section 5 summarizes the findings of the work and concludes the article with future research directions.

2. Experimental methods The Ni-rich NiTi powder used in this study was produced by Nanoval GmbH & Co. using gas atomization under argon with 50.9 at. % nickel content. Fig. 1 depicts the morphology of the powder obtained using scanning electron microscopy using a Tescan Lyra3 field emission 674

Journal of Manufacturing Processes 35 (2018) 672–680

M. Mahmoudi et al.

Fig. 4. Influence of volumetric energy density on martensite start (Ms) and austenite finish (Af) transformation temperatures of the non-defective AM coupons of the NiTi shape memory alloy. (For interpretation of the references to color in the text, the reader is referred to the web version of this article.)

consisting of 3 M H2SO4 in 1:1 ethanol–methanol solution was carried out at a temperature of 22 °C and a voltage of 20 V for a period of 30 s. For the optical micrographs, the final color-etching step of the surfaces was performed for a duration of 60 s according to the procedure described in [30]. The etching solution consisted of 120 ml pure water, 15 ml HCl, 15 g Na2S2O5, 10 g K2S2O5 and 2 g NH4HF. A microscope with linearly polarized light was used to acquire the micrographs of the polished and etched surfaces with the aim of identifying the grains. 3. Results The experimental design matrix with the 47 different process parameter combinations generated using the Latin hypercube sampling is shown in Table 2. Columns 2–4 show the laser power, scan speed, and hatch spacing for each process parameter setting, respectively. Similar P to EL and EV, we can define surface energy density as ES = h·v . The corresponding volumetric, surface, and linear energy density values are calculated in columns 5–7. The last column indicates the outcome of each experimental run (defective versus non-defective coupon). For the sake of this study, by defective we refer to a part that has visible macroscopic cracks, which indicates a failed fabrication process. It can be seen that in most cases the coupons were defective and only 9 out of 47 different parameter combinations resulted in non-defective parts. Note that in addition to the printability issues and corresponding macroscopic defects associated with L-PBF parts, other mesoscopic and microscopic defects can occur during L-PBF such as balling, geometric deviation, porosity, surface defects, and microstructural inhomogeneity. Although these issues might not affect the part printability, they can severely impair the part properties. The microscopic defects are out of the scope of this study. For a comprehensive review of

Fig. 3. Phase transformation behavior (a) for the NiTi powder and coupons with hatch spacing 35–47 μm; and (b) for the NiTi powder and for coupons with hatch spacing 62–120 μm.

Table 3 Transformation temperatures of non-defective AM coupons of NiTi as a function of process parameters and various energy densities. Run No.

P(W)

v (mm/ s )

h (μm)

EV (J/mm3)

ES (J/mm2)

EL (J/mm)

Ms (°C)

Mf (°C)

As (°C)

Af (°C)

47 46 44 42 40 3 8 20 32

49 48.5 48 49 50 50 50 41 45.5

91 109 89 85 70 80 80 85 94

47 62 113 106 120 35 120 40 68

381.9 239.2 159.1 181.3 198.4 595.2 173.6 402.0 237.3

11.46 7.18 4.77 5.44 5.95 17.86 5.21 12.06 7.12

0.538 0.445 0.539 0.576 0.714 0.625 0.625 0.482 0.484

61 47 12 39 34 65 61 64.2 66.6

18 −67 −65 −69 −64 28 −33 −1.5 −19.3

62 −51 −48 −49 −47 70 −6 67.8 −23

85 75 31 64 57 90 81 87.5 73.4

675

Journal of Manufacturing Processes 35 (2018) 672–680

M. Mahmoudi et al.

Fig. 5. Nickel content for coupon 3 (hatch spacing 35 μm) and coupon 8 (hatch spacing 120 μm) measure using WDS. Table 4 Lower and upper bounds. Parameter

P

v

h

Lowest level Highest level

35 W 50 W

70 mm/s 450 mm/s

35 μm 120 μm

these defects and state-of-the-art in situ monitoring techniques see [31]. Fig. 2(a) shows three as-built cubic coupons built using three different parameter sets, each one demonstrating some level of interlayer cracking caused by delamination. Delamination is the separation of adjacent layers within a part due to lack of bonding between layers. This may occur due to incomplete melting of powder or insufficient remelting of the underlying bulk material [32]. From the figure, it can be seen that higher energy densities lead to less interlayer cracking. Oppositely, Fig. 2(b) shows two coupons cut to a smaller size, which were successfully manufactured without delamination by utilizing higher energy. Surface decoloring due to oxidization can also be seen in the coupon on the right in Fig. 2(b) as a result of excessive energy input. Note that in addition to the macroscopic cracks that occur due to porosity or delamination, microscopic cracks can occur due to different mechanisms including solidification cracking and grain boundary cracking [33]. Whereas microstructural defects such as micro-cracks that are generally caused by lack-of-fusion or excessive energy input

Fig. 7. Scatter plot of martensite start (Ms) and austenite finish (Af) transformation temperatures as a function of linear energy density.

can be alleviated using post-processing (see [33]), the same does not hold true with macroscopic delamination [32]. Fig. 3 shows the DSC curves for the initial NiTi powder after a 900 °C solution heat treatment (black curves) and the nine nonFig. 6. Scatter plot of laser hatch spacing and linear energy density (EL) for the defective and non-defective L-PBF coupons of the NiTi shape memory alloys. The red dashed line is the decision boundary for the linear discriminant analysis LDA implemented to the data in Table 2. (For interpretation of the references to color in the text, the reader is referred to the web version of this article.)

676

Journal of Manufacturing Processes 35 (2018) 672–680

M. Mahmoudi et al.

4. Discussion 4.1. Effect of process parameters on the printability of NiTi We use the term printability to refer to the capability of successfully building a part, and concurrently avoiding or minimizing macrocracking defects. A common approach to enhance printability is based on volumetric energy density EV, which is a thermodynamic quantity [22,25–27]. From Table 2 we observe that EV ranges between 21.6 J/ mm3 and 338.2 J/mm3 for defective coupons and between 159.1 J/ mm3 and 595.2 J/mm3 for non-defective coupons. Hence, lower EV values, e.g. below 150 J/mm3, appear to increase the likelihood of an unsuccessful build. However, this is not readily generalizable, especially for the range of EV between 150 J/mm3 and 350 J/mm3 since the data shows both successful and unsuccessful builds within that range. This can also be observed in the literature where discrepancy exists between reported values of optimal EV: Walker et al [26] reports the optimal value for EV as 55.6 J/mm3 while Haberland [22] reports 234.4 J/mm3. It has been asserted by Bertoli et al. [37] that using the volumetric energy density as the sole design parameter might not be the best approach for fabricating parts using L-PBF. They conducted a series of experiments to investigate the viability of EV as a design parameter by comparing single tracks of 316L stainless steel printed with varying process parameters. Continuous and smooth tracks, rough and irregular tracks, and discontinuous tracks exhibiting balling phenomenon were all observed for a fixed value of EV with different power-speed combinations. Their results showed that both low (below 100 J/mm3) and high (above 242 J/mm3) values of EV are undesirable, resulting in the degradation of track shape, emergence of balling, and high porosity due to keyhole melting [38]. Based on the above discussion, we conclude that EV is not a viable design parameter for assessing printability in L-PBF processes. In this study, we assess the printability of NiTi using the linear energy density EL in contrast to the volumetric energy density EV as suggested by previous works in the literature. We start by providing a brief discussion to justify the choice of EL. Here, we reason why EL is a superior design parameter as follows. Recall that our definition of printability is more concerned with avoiding macroscopic flaws such as inter-layer cracking. Inter-layer cracking is caused by delamination, which in turn is directly influenced by the melt pool characteristics (temperature and geometry). We notice that melt pool temperature and size are mainly affected by process parameters such as laser power and speed. In contrast, other process parameters such as hatch spacing mainly change the part thermal history and microstructure homogeneity [14]; hence they have less impact on the part printability, as long as hatch spacing is less than the melt pool width. For example, if the energy input is insufficient and melt pool depth is not large enough to ensure bonding between newly melted layers and the underneath solidified layers, increase or decrease in hatch spacing will not avoid the delamination problem. We conclude that laser power and speed are parameters that directly affect printability, while hatch spacing and layer thickness are of less importance. Consequently, inclusion of hatch spacing and layer thickness in a design parameter is not recommended and EL might be a better choice compared to EV. To confirm this claim, we use machine learning techniques to find out whether EL and h are able to predict the printability given a set of process parameters. In other words, we want to see if EL and h are able to separate two different process conditions, namely, successful build and unsuccessful build. We conduct Linear Discriminant Analysis (LDA) to classify the data in Table 2 using EL and hatch spacing h. Fig. 6 shows the result of LDA for the coupons with parameter sets listed in Table 2. The red dashed line represents the decision boundary of the implemented LDA. We compute the apparent error rate by finding the proportion of the incorrectly classified points to the total classified data. The apparent error for the LDA shown in Fig. 6 is 8.5% which

Fig. 8. Schematic of the L-PBF process indicating the melting track, build plane, and the transverse plane.

defective coupons indicated in Table 2. Heat treatment was performed in order to remove the effect of residual stress and to dissolve precipitates that may have formed during gas atomization of the powder. We observe no transformation, represented by the flat curves, for the NiTi powder due to the high nickel content. On the other hand, all the nine L-PBF coupons show varying levels of phase transformation, driven by variations in process parameters. We notice that hatch spacing directly affects the transformation behavior of the parts. It can also be observed that coupons with smaller hatch spacing (35–47 μm) show sharp transformation peaks in contrast to coupons with larger hatch spacing (62–120 μm). This can be attributed to the difference in thermal history during processing of the material, since hatch spacing determines the amount of overlap between successive tracks and, in turn, the level of thermal cycling at a given location. It has been reported in a prior study by the co-authors that larger hatch spacing results in higher microstructural and compositional variability due to less overlap and hence less uniform thermal history [14]. Table 3 lists the phase transformation temperatures (Ms, Mf, As, and Af) for the nine non-defective coupons. We are particularly interested in Ms and Af values because Af − Ms is proportional to hysteresis, which is an important parameter for assessing the efficiency and fatigue response of shape memory alloys [11,34]. Fig. 4 shows a plot of Ms (blue curve) and Af (red curve) versus the volumetric energy density EV. The increasing trend observed in both Ms and Af is consistent with the effect of energy input on the phase transformation behavior reported in the literature [25,35]. This is likely due to higher nickel evaporation with increased energy input, which results in a higher titanium content in the matrix [36,19,25], and the formation of Ni-rich phases [11,21,25]. To confirm the higher Ni evaporation, we conducted WDS for the powder, coupon 3, and coupon 8. Among the measurements, the powder had the highest nickel content of 49.96 at. %. Measured nickel contents for coupon 3 and coupon 8 were 47.95 at. % and 49.09 at. %, respectively. The volumetric energy density and nickel content for these two coupons are compared using the plot in Fig. 5. From the figure, it is obvious that higher energy density results in lower nickel content (higher nickel evaporation) (Table 4).

677

Journal of Manufacturing Processes 35 (2018) 672–680

M. Mahmoudi et al.

Fig. 9. The optical micrographs of the NiTi shape memory alloy AM coupons fabricated using hatch spacings of (a), (b) 35 micrometers and (c), (d) 120 micrometers. The images were taken on (a), (c) transverse and (b), (d) longitudinal planes showing the anisotropic grain structure. For the remaining process parameters, refer to Table 3.

density EV is not a suitable design parameter for assessing printability of NiTi using L-PBF, Fig. 4 suggests that transformation temperatures and EV are strongly correlated. In the meantime, Fig. 7 shows a scatter plot of transformation temperatures against the linear energy density EL in which no significant trend can be seen. Hence, there is no significant correlation between transformation temperatures and EL. The reason for ineffectiveness of EL in explaining the transformation behavior is the fact that EL does not include hatch spacing and layer thickness. This means EL does not have enough information about the thermal history and hence microstructure and composition of the part. On the other hand, EV includes the hatch spacing and layer thickness, making it a better parameter than EL in assessing the part thermal history, and thus microstructure and transformation behavior. Notice, however, that solely relying on EV can be an oversimplifying assumption. For example, we notice variations in transformation temperatures for similar EV values, e.g. for EV ≈ 250 J/mm3 two different martensite start transformation values (47 °C and 67 °C) are measured, see Fig. 4. These variations are consistent with what has been reported in the literature

shows reasonable classification accuracy. Moreover, the small slope of the LDA line confirms that changing hatch spacing does not have a significant effect on printability (ideally, this line would be flat). On the other hand, EL can be regarded as an informative design parameter, with values below 0.4 J/mm will likely result in defective parts and values above 0.5 J/mm will result in non-defective parts. Manufacturing parts with 0.4 < EL < 0.5 might be risky and might result in unsuccessful builds.

4.2. Effect of process parameters on transformation behavior In this section we investigate process parameters that influence the transformation behavior, through focusing on the effects of these parameters on the part thermal history and microstructure. Thermal histories influence the amount of nickel evaporation during L-PBF, and consequently the matrix composition (Ni/Ti ratio) and transformation behavior [11,36]. Although we concluded in Section 4.1 that the volumetric energy 678

Journal of Manufacturing Processes 35 (2018) 672–680

M. Mahmoudi et al.

References

[20,22,29,13]. It is important to remember that although process parameters such as hatch spacing and layer thickness have less impact on part printability compared to laser power and speed, they directly affect the part thermal history. It has been shown that the part microstructure is directly influenced by the thermal histories during the metal AM [39]. Here, we investigate the effect of hatch spacing on part thermal history and transformation temperatures using OM images from the grain structure of the coupons. Two of the non-defective coupons (coupons 3 and 8) were selected to investigate the effects of hatch spacing on the microstructure. Laser power and speed are identical for these two coupons (see Table 2 for details), but coupon 3 features a short hatch spacing h = 35 μm, while h = 120 μm for the coupon 8. We are interested in the grain structure in two perpendicular planes: “build plane” and “transverse plane”. Fig. 8 schematically shows these two planes along with the melt pool and melting track. The arrow in the figure indicates the laser beam moving direction. Fig. 9 shows OM images of the two coupons for the two aforementioned perpendicular planes: build plane (images on the right side), and transverse plane (images on the left side). The linear scan strategy is reflected in the arrangement and morphology of the grains. The shorter hatch spacing in coupon 3 has resulted in a more regular grain structure than in coupon 8, because both the S-shaped grains in the transverse plane and the columnar grains in the build plane for coupon 3 are easier to identify compared to coupon 8. This observation is an example of the significance of hatch spacing on the part microstructure and hence thermal histories and transformation temperatures.

[1] Jani JM, Leary M, Subic A, Gibson MA. A review of shape memory alloy research, applications and opportunities. Mater Des 2014;56:1078–113. [2] Lagoudas DC. Shape memory alloys. Science and Business Media, LLC; 2008. [3] Elahinia MH, Hashemi M, Tabesh M, Bhaduri SB. Manufacturing and processing of niti implants: a review. Prog Mater Sci 2012;57:911–46. [4] Elahinia M, Moghaddam NS, Andani MT, Amerinatanzi A, Bimber BA, Hamilton RF. Fabrication of niti through additive manufacturing: a review. Prog Mater Sci 2016;83:630–63. [5] Chouhan P, Nath T, Lad B, Palani I. Investigation on actuation and thermo-mechanical behaviour of shape memory alloy spring using hot water. IOP Conference Series: Materials Science and Engineering, vol. 149. 2016. p. p012147. [6] Weinert K, Petzoldt V. Machining of niti based shape memory alloys. Mater Sci Eng: A 2004;378:180–4. [7] Wu S, Lin H, Chen C. A study on the machinability of a ti 49.6 ni 50.4 shape memory alloy. Mater Lett 1999;40:27–32. [8] Wu MH. Fabrication of nitinol materials and components. Materials science forum vol. 394. Trans Tech Publ; 2002. p. 285–92. [9] Xenos-Despina S, Gregory N. The unique properties, manufacturing processes and applications of near equatomic ni-ti alloys. 2016. [10] Frenzel J, Zhang Z, Somsen C, Neuking K, Eggeler G. Influence of carbon on martensitic phase transformations in niti shape memory alloys. Acta Mater 2007;55:1331–41. [11] Frenzel J, George EP, Dlouhy A, Somsen C, Wagner M-X, Eggeler G. Influence of Ni on martensitic phase transformations in niti shape memory alloys. Acta Mater 2010;58:3444–58. [12] Meier H, Haberland C, Frenzel J, Zarnetta R. Selective laser melting of niti shape memory components. Innovative developments in design and manufacturing: advanced research in virtual and rapid prototyping. 2009. p. 233–8. [13] Ma J, Franco B, Tapia G, Karayagiz K, Johnson L, Liu J, Arroyave R, Karaman I, Elwany A. Spatial control of functional response in 4d-printed active metallic structures. Sci Rep 2017;7:46707. [14] Franco B, Ma J, Loveall B, Tapia G, Karayagiz K, Liu J, Elwany A, Arroyave R, Karaman I. A sensory material approach for reducing variability in additively manufactured metal parts. Sci Rep 2017;7. [15] Yadroitsau I. Selective laser melting: direct manufacturing of 3D-objects by selective laser melting of metal powders. Lambert Academic Publishing; 2009. [16] Khanzadeh M, Rao P, Jafari-Marandi R, Smith BK, Tschopp MA, Bian L. Quantifying geometric accuracy with unsupervised machine learning: using self-organizing map on fused filament fabrication additive manufacturing parts. J Manuf Sci Eng 2018;140:031011. [17] Khanzadeh M, Chowdhury S, Tschopp MA, Doude HR, Marufuzzaman M, Bian L. Insitu monitoring of melt pool images for porosity prediction in directed energy deposition processes. IISE Trans 2017. [18] Khanzadeh M, Chowdhury S, Tschopp MA, Marufuzzaman M, Bian L. Porosity prediction: supervised-learning of thermal history for direct laser deposition. J Manuf Process 2018. [19] Meier H, Haberland C, Frenzel J. Structural and functional properties of niti shape memory alloys produced by selective laser melting. Innovative developments in design and manufacturing: advanced research in virtual and rapid prototyping. 2011. p. 291–6. [20] Bormann T, Schumacher R, Müller B, Mertmann M, de Wild M. Tailoring selective laser melting process parameters for niti implants. J Mater Eng Perform 2012;21:2519–24. [21] Haberland C, Meier H, Frenzel J. On the properties of Ni-rich niti shape memory parts produced by selective laser melting. ASME 2012 conference on smart materials, adaptive structures and intelligent systems. 2012. p. 97–104. [22] Haberland C, Elahinia M, Walker JM, Meier H, Frenzel J. On the development of high quality niti shape memory and pseudoelastic parts by additive manufacturing. Smart Mater Struct 2014;23:104002. [23] Dadbakhsh S, Speirs M, Kruth J-P, Van Humbeeck J. Influence of slm on shape memory and compression behaviour of niti scaffolds. CIRP Ann-Manuf Technol 2015;64:209–12. [24] Dadbakhsh S, Vrancken B, Kruth J-P, Luyten J, Van Humbeeck J. Texture and anisotropy in selective laser melting of niti alloy. Mater Sci Eng: A 2016;650:225–32. [25] Andani MT, Haberland C, Walker J, Elahinia M. An investigation of effective process parameters on phase transformation temperature of nitinol manufactured by selective laser melting. ASME 2014 conference on smart materials, adaptive structures and intelligent systems. 2014. V001T01A026. [26] Walker JM, Haberland C, Taheri Andani M, Karaca HE, Dean D, Elahinia M. Process development and characterization of additively manufactured nickel-titanium shape memory parts. J Intell Mater Syst Struct 2016;27:2653–60. [27] Li S, Hassanin H, Attallah MM, Adkins NJ, Essa K. The development of tini-based negative poisson's ratio structure using selective laser melting. Acta Mater 2016;105:75–83. [28] Montgomery DC. Design and analysis of experiments. John Wiley & Sons; 2008. [29] Bormann T, Müller B, Schinhammer M, Kessler A, Thalmann P, de Wild M. Microstructure of selective laser melted nickel-titanium. Mater Charact 2014;94:189–202. [30] Escher K, Huhner M. Metallographic preparation of niti shape memory alloys. Prak Metallogr 1990;27:231–5. [31] Grasso M, Colosimo BM. Process defects and in situ monitoring methods in metal powder bed fusion: a review. Meas Sci Technol 2017;28:044005. [32] Sames WJ, List F, Pannala S, Dehoff RR, Babu SS. The metallurgy and processing

5. Summary and conclusions In the present work, we investigate various design parameters in order to assess the printability of NiTi SMAs and the level of control on their transformation temperatures, using Laser Powder-Bed Fusion (LPBF) additive manufacturing technique. More specifically, we conduct experiments and employ machine learning classification techniques to identify adequate design parameters for determining the printability and controlling transformation temperatures of NiTi. Main findings and conclusions can be summarized as follows: 1. The field of Additive Manufacturing (AM) is in need of reliable design parameters to avoid fabricating defective parts; in particular interlayer cracks need to be avoided in L-PBF processes. This work shows how the conventional approach of using the volumetric energy density EV as the sole design parameter is too simplistic for achieving macro defect-free parts, particularly for Shape Memory Alloy (SMA) powders such as NiTi. 2. The experimental study and the machine learning techniques conducted in this work suggest that the linear energy density EL can be a better choice for predicting printability of NiTi given a specific combination of process parameters. This is mainly because laser power and speed are parameters that directly affect printability, while hatch spacing and layer thickness are of less importance. Consequently, inclusion of hatch spacing and layer thickness in a design parameter is not recommended. Hence, EL might be a better choice compared to EV when it comes to printability. 3. While EV is considered an overly simplistic design parameter for printability of NiTi in L-PBF, there is a remarkable correlation between EV and the martensitic transformation temperatures of the part. The DSC curves show that increased EV results in higher transformation temperatures in the fabricated parts. 4. Although parameters such as hatch spacing do not notably influence printability of NiTi, they can significantly affect the part thermal history and thus microstructure. This can be shown using OM images where shorter hatch spacing results in more regular grain structures compared with using longer hatch spacing.

679

Journal of Manufacturing Processes 35 (2018) 672–680

M. Mahmoudi et al.

[36] Khan MI, Pequegnat A, Zhou YN. Multiple memory shape memory alloys. Adv Eng Mater 2013;15:386–93. [37] Bertoli US, Wolfer AJ, Matthews MJ, Delplanque J-PR, Schoenung JM. On the limitations of volumetric energy density as a design parameter for selective laser melting. Mater Des 2017;113:331–40. [38] King WE, Barth HD, Castillo VM, Gallegos GF, Gibbs JW, Hahn DE, Kamath C, Rubenchik AM. Observation of keyhole-mode laser melting in laser powder-bed fusion additive manufacturing. J Mater Process Technol 2014;214:2915–25. [39] Bian L, Thompson SM, Shamsaei N. Mechanical properties and microstructural features of direct laser-deposited ti-6al-4v. JOM 2015;67:629–38.

science of metal additive manufacturing. Int Mater Rev 2016;61:315–60. [33] Carter LN, Attallah MM, Reed RC. Laser powder bed fabrication of nickel-base superalloys: influence of parameters; characterisation, quantification and mitigation of cracking. Superalloys 2012;2012:577–86. [34] Cui J, Chu YS, Famodu OO, Furuya Y, Hattrick-Simpers J, James RD, Ludwig A, Thienhaus S, Wuttig M, Zhang Z, et al. Combinatorial search of thermoelastic shapememory alloys with extremely small hysteresis width. Nat Mater 2006;5:286. [35] Haberland C, Elahinia M, Walker J, Meier H, Frenzel J. Additive manufacturing of shape memory devices and pseudoelastic components. ASME 2013 conference on smart materials, adaptive structures and intelligent systems. 2013. V001T01A005.

680