Fatigue-resistant high-performance elastocaloric materials made by additive manufacturing

See allHide authors and affiliations

Science  29 Nov 2019:
Vol. 366, Issue 6469, pp. 1116-1121
DOI: 10.1126/science.aax7616

A million times cooler

Elastocaloric materials can be used for solid-state cooling applications because they can pump heat out of a system using a reversible phase transformation. However, many such materials fail after a small number of cycles. Hou et al. found that laser melting of elastocaloric metals can create fatigue-resistant microstructures. A nickel-titanium–based alloy could be cycled a million times and still produce a cooling of about 4 kelvin. This processing method could improve elastocaloric performance and move us closer to using these materials more widely for solid-state cooling applications.

Science, this issue p. 1116


Elastocaloric cooling, a solid-state cooling technology, exploits the latent heat released and absorbed by stress-induced phase transformations. Hysteresis associated with transformation, however, is detrimental to efficient energy conversion and functional durability. We have created thermodynamically efficient, low-hysteresis elastocaloric cooling materials by means of additive manufacturing of nickel-titanium. The use of a localized molten environment and near-eutectic mixing of elemental powders has led to the formation of nanocomposite microstructures composed of a nickel-rich intermetallic compound interspersed among a binary alloy matrix. The microstructure allowed extremely small hysteresis in quasi-linear stress-strain behaviors—enhancing the materials efficiency by a factor of four to seven—and repeatable elastocaloric performance over 1 million cycles. Implementing additive manufacturing to elastocaloric cooling materials enables distinct microstructure control of high-performance metallic refrigerants with long fatigue life.

Space cooling and refrigeration consume about one-fifth of the entire electricity demand worldwide (1). Vapor compression is a mature technology that dominates the market, but its efficiency has plateaued, and the high global-warming potential of the refrigerants remains a major concern. Solid-state cooling technologies, including thermoelectric (2) and passive radiative cooling (3), represent attractive green alternatives to vapor compression. Particularly promising are caloric (magnetocaloric, mechanocaloric, and electrocaloric) cooling techniques (47), which have the potential to surpass the efficiency of vapor compression. Because caloric materials enlist active heat-pumping through manipulation of their functional properties (magnetization, stress-state, and electric polarization) (7), they can be exploited for giant cooling effects. However, hysteresis in the properties is caloric materials’ Achilles heel because hysteresis represents work lost in every heat-pumping cycle as dissipated heat. Hysteresis also ultimately leads to material fatigue and failure. Although the long-term fatigue properties are critical for developing applications for caloric materials, evaluating these properties over a large number of cycles is not common.

The martensitic phase transformation of shape memory alloys (SMAs) may manifest in dramatic temperature-triggered shape change, enabling solid-state actuation technologies (8). The martensitic transformation can also be triggered by stress, resulting in superelasticity (9). During superelastic cycles, latent heat is released or absorbed upon loading or unloading, respectively, because of the exothermic-endothermic nature of the phase transformation. This stress-induced heat-pumping of SMAs is elastocaloric cooling (10, 11), one type of mechanocaloric cooling. Adiabatic temperature change (ΔTad) of elastocaloric materials can be as large as 31.5 K (12), and the largest reported isothermal entropy change (ΔS) of an SMA is 70.7 J kg−1 K−1 (13). Such unparalleled cooling potential has made elastocaloric cooling a front runner among the crowded field of alternative cooling technologies (14). Functioning elastocaloric cooling prototypes with capacity over 100 W have been developed (15), and elastocaloric regenerative heat pumps with ΔTad as large as 19 K have been demonstrated by using commercially available nickel-titanium (Ni–Ti) materials (16, 17). However, thermomechanical hysteresis of elastocaloric SMAs can limit the efficiency of cooling devices, and its impact on long-term performance has not been addressed.

We synthesized elastocaloric materials consisting of alloy matrix and intermetallic phases arranged in nanocomposite microstructures using the powder-feed laser-directed-energy deposition (L-DED) technique. This method results in a local melting of metal powders followed by rapid solidification (18, 19). When a Ni-rich blend of elemental Ni and Ti powders are mixed during this process, nanocomposite microstructures—composed of transforming, elastocaloric binary NiTi alloy matrix and a nontransforming Ni3Ti intermetallic phase—form in a two-phase mixture of comparable volume fractions, with intricate dendritic structures. This configuration enlists the nontransforming intermetallic phase for biasing the phase transformation, leading to considerable improvement in elastocaloric efficiency and transformation reversibility through work hysteresis minimization.

The L-DED–produced Ni–Ti nanocomposite exhibits substantially reduced hysteresis with a quasi-linear stress-strain behavior, resulting in a multifold increase in the materials efficiency, which is defined as the ratio of materials coefficient of performance (COPmaterials) to Carnot COP. We show that the elastocaloric thermodynamic cycle of these materials is stable over more than a million cycles. In contrast to rate-dependent hysteresis commonly observed in traditionally processed SMAs (20, 21), the hysteresis of the L-DED Ni–Ti nanocomposite is nearly rate-independent (from 0.0002 to 0.2 s−1), facilitating high-frequency elastocaloric operations. We used a constitutive model and in situ synchrotron x-ray diffraction experiments to confirm that their properties originate from kinematics of load transfer between transforming and nontransforming phases.

The key features of the L-DED process (Fig. 1A) are a millimeter-scale molten pool of mixed powders and a rapid cooling rate of greater than 103 K s−1 (22). Metal nanocomposites made by, for example, melt-casting (23) can display a stress-transfer mechanism responsible for high strength, a desirable attribute of functional alloys. Because eutectic solidification can naturally lead to the formation of composites, we used the eutectic point in the Ni-rich composition range (Fig. 1B) of binary Ni–Ti to obtain elastocaloric nanocomposite using L-DED (24). Optimization of processing parameters (such as layer thickness and hatching space) was guided by use of a normalized processing map (25) for high denseness (≈99%) and mechanical integrity, and the molten pool temperature in operation was maintained to be 1973 to 2173 K, as measured in situ with a ThermaViz pyrometer. We adjusted the ratio of the flow rate of elemental Ni and Ti powders to print different compositions of Ni–Ti materials in a range of geometries (Fig. 1, C to H).

Fig. 1 Design of elastocaloric Ni–Ti nanocomposite by directed energy deposition.

(A) Schematic representation of an L-DED process. Flows of Ni and Ti powders are individually controlled. The Ni and Ti powders are mixed and then fed to the laser beam. An induced molten pool moves to build materials layer by layer (fig. S1). (B) Phase diagram of Ni–Ti [adapted with permission from (53)] highlighting in blue the Ni-rich composition near a eutectic point and the molten pool temperature used in this work. (C to H) Photographs of L-DED–produced Ni–Ti nanocomposite rods, tubes, and honeycombs in [(C), (E), and (G)] top and [(D), (F), and (H)] front views, respectively. (I to K) Scanning electron microscopy (SEM) image (I), bright-field transmission electron microscopy (TEM) image (J), and high-resolution high-angle annular dark-field scanning TEM (HAADF-STEM) image (K) of as-built Ni51.5Ti48.5/Ni3Ti nanocomposite. In (I), the regions with different contrasts are crystallographically identified to be NiTi and Ni3Ti phases (figs. S2 and S3). In (J), typical curved interfaces are delineated. In (K) (a zoomed-in view of a curved interface), NiTi and Ni3Ti phases have an orientation relationship of NiTi[111]||Ni3Ti[112¯0], although each is slightly off the zone axis because of lattice strains within the interface (fig. S4). (L) Inverse fast Fourier transform (IFFT) image from the circled spot in the FFT image [inset; generated from (K)]. Interfacial dislocations are identified and marked with “T” symbols.

Rapid cooling of the molten pool during L-DED enables precipitation from off-eutectic compositions in a volume fraction comparable with that of eutectic structures, as predicted by the Scheil model (22, 26). We observed a substantial fraction of precipitates (up to ≈50%) in a wide compositional range of the Ni–Ti produced with L-DED (Fig. 1B). Curved microstructures can nucleate and grow because the temperature gradient (highest at the center and lowest at the periphery) of the molten pool leads to circulation of mass and heat within the pool driven by Marangoni shear stress (27), creating local perturbations of solute concentration and equilibrium temperature (28) on solid–liquid interfaces and breaking up the plane front in growth of steady-state eutectics. As a result of nonequilibrium conditions, a typical microstructure of the L-DED–produced Ni–Ti nanocomposite consists of transforming NiTi and nontransforming Ni3Ti phases with large aspect ratios, curved interfaces, and comparable volume fractions (Fig. 1I). The size scale of the microstructure is inversely proportional to the cooling rate (26), which is at least two orders of magnitude higher in L-DED than that of casting (≈0.1 K s−1), leading to a mixture of two phases at a submicrometer scale (Fig. 1J).

Large interfacial curvatures between the cubic B2-ordered NiTi phase and the hexagonal D024-ordered Ni3Ti phase (Fig. 1J) can be naturally accommodated with small lattice mismatches to make their interfaces semicoherent. An atomic-scale view of the adjacent regions displays strained boundaries (Fig. 1K) where interfacial dislocations are located (Fig. 1L). Preexisting sites of high nucleation potency such as dislocations have been reported to trigger atomic shearing for nucleation of martensite (29), in which a nucleation energy barrier is lowered [or completely suppressed in the case of spontaneous growth (30)]. These interfacial dislocations inherent to the curvatures and additional dislocations induced through mechanical pretreatment (fig. S5) therefore serve as preexisting nucleation sites to reduce energy barriers for martensite during the forward transformation and for austenite during the reverse transformation. In addition, these same nucleation sites can act as “micropockets” to accommodate remnant austenite and martensite after forward and reverse transformations, respectively, eliminating the necessity of a barrier-overcoming stage for nucleation during cyclic loading. After proper self-organization, prestraining, and prestressing (fig. S5, shakedown state), the intricate nanoscale network of connected microstructure suppresses the dislocation motion (31) and limits transformation dissipation, resulting in enhanced cyclic stability.

The L-DED–produced Ni–Ti nanocomposite exhibits quasi-linear behaviors and substantially reduced hysteresis (Fig. 2A). The full strain recovery upon unloading is accompanied by a cooling ΔTad (Fig. 2B), a signature of martensitic transformation, which reaches 4.1 K. In a caloric cooling system, ΔTad of caloric materials can be boosted into a large temperature span across a device by use of active regeneration schemes (16, 32, 33). The quasi-linear recovery behavior arises from the load transfer between the nontransforming intermetallic phase and the transforming non–load-bearing phase and has previously been also observed in melt-casted alloys after aging and/or cold-working (23, 34, 35). The effective modulus of the L-DED–produced Ni–Ti nanocomposite (~80–90 GPa) is higher than the typical austenite (~50 to 60 GPa), which indicates the effect of the nontransforming intermetallic Ni3Ti phase in the nanocomposite. As a result, as the austenite transforms to martensite, the intermetallic phase continues to carry the load elastically, and the resulting overall behavior is quasi-linear. To confirm this mechanism, we simulated the crossover from a regular superelastic to quasi-linear behavior by varying the volume fraction of nontransforming intermetallic phase and observed the appearance of quasi-linear behavior at a level of 40, 50, and 60% (Fig. 2C).

Fig. 2 Recoverable behaviors and elastocaloric properties of Ni–Ti nanocomposite.

(A and B) The stress-strain curves (A) and corresponding elastocaloric cooling at room temperature (B) of L-DED–produced Ni51.5Ti48.5/Ni3Ti nanocomposite aged at 923 K for 3 hours. The single arrows in (A) denote loading, and the double arrows in (A) and (B) correspond to unloading. (C) Simulated stress-strain curves from a micromechanics model that accounts for the volume fraction of nontransforming phase (insets). (D and E) Synchrotron x-ray diffraction patterns (D) during in situ loading and unloading and the (E) determined volume fraction of primary phases at different stress levels during the cycle. (F and G) Comparison of stress-strain curves for L-DED–produced Ni51.5Ti48.5/Ni3Ti nanocomposite and melt-casted Ni50.8Ti49.2 and Cu68Zn16Al16 alloys at the strain rate of (F) 0.0002 s−1 for isothermal loading and unloading and (G) 0.2 s−1 for adiabatic loading and unloading. In (F) and (G), the area enclosed by the loading and unloading curves represents total dissipation energy per unit volume associated with hysteresis. (H) Comparison of hysteresis area under isothermal and adiabatic loading and unloading as well as the ratio of COPmaterials to Carnot COP for L-DED nanocomposite and melt-casted alloys at maximum transformations. The color code for each material is common for (F) to (H).

The small hysteresis we observed is due to the topology- and defect-controlled kinematics of numerous nucleation events and coalescence, in which spatially dispersed preexisting nucleation sites (Fig. 1L) favor continual, heterogeneous nucleation of a new martensite followed by the sites’ coalescence. The resulting volumetric densities of obstacles that austenite-martensite transformation fronts meet in the course of transformation are reduced and require a decreased amount of frictional work to overcome, as observed in Cu–Zn–Al alloys (36). Additionally, the intermetallic phase has a large volume fraction (≈50%) and effectively guides the transformation process through elastic interaction with the transforming phase. This process in turn tempers multiple instabilities that occur during traditional nucleation and fast growth and reduces energy dissipation and effective interfacial friction. We captured the progression in in situ synchrotron x-ray diffraction measurements (Fig. 2, D and E).

We attributed the commonly observed rate-dependent hysteresis (for example, the difference in hysteresis curves between Fig. 2, F and G) to transformation-related heat in SMAs in which surface convection dominates heat transfer. From an explicit integral equation of the specific dissipated energy ΔE (which is equal to the generated heat) (37), we can approximate ΔE asΔEEfr + ΔTad ⸱ Δs(1)where Efr is the irreversible specific energy that is the generated heat through interface friction and Δs is the specific entropy change associated with the phase transformation. The ΔE during a stress-strain cycle manifests itself as the hysteresis area (divided by density) and increases with enlarged hysteresis. This relation can also explain the nearly rate-independent hysteresis we observed in the Ni–Ti nanocomposite (Fig. 2H) in which thermal conduction (thermal conductivity ≈18 W m−1 K−1) through a large-volume fraction of nontransforming phase and surface convection (with convective heat transfer coefficient ≈4 W m−2 K−1) collectively facilitate effective heat transfer and rejection in a transformation cycle. In this example, the second term on the right of Eq. 1 becomes small owing to the rate of heat dissipation approaching the rate of heat generation.

Decreasing Efr contributes to additional reduction in ΔE. Efr consists of two components: Efr = Ef + Ep (38), where Ef is the heat dissipated from frictional work in a transformation cycle, and Ep is the heat dissipated by plastic work within austenite-martensite interfaces because of their coherency loss. Although friction is ubiquitous in the propagation of austenite-martensite interfaces (39), reducing extended interfacial motions by having uniformly distributed sites for nucleation and coalescence can substantially curtail frictions, leading to reduced Ef. The resultant minimization of Ef accounts for the substantial reduction in Efr (Fig. 2H). In other alloy systems, relaxing local strain energy associated with phase transformation through improving lattice compatibility was found to lead to considerable reduction in Ep (4042). Efr remains constant at different rates and plays a role in the rate independence of hysteresis in the Ni–Ti nanocomposite.

Thermodynamics of cooling devices dictates that small hysteresis during isothermal loading and unloading in Stirling-like cycles leads to high efficiencies (15, 43). However, under the same heat-exchange conditions as that of the Brayton-like cycle, Stirling-like operation cycles require much longer time per cycle (leading to reduced output wattage) and additional system components for effective heat transfer (43). In comparison, adiabatic loading and unloading in Brayton-like cycles (44) can operate much faster with relatively simple heat-exchange systems, albeit suffering from lower intrinsic efficiency because of the larger hysteresis (Fig. 2G). COPmaterials in Brayton-like cycles are governed by the directly measured ΔTad with the adiabatic hysteresis, and COPmaterials in Stirling-like cycles are regulated by the latent heat with the isothermal hysteresis, on the basis of thermodynamically derived equations with full work recovery (24). The hysteresis of the L-DED–produced Ni–Ti nanocomposite is extremely small in both cycles and has a negligible difference (indicating rate independence). With a Carnot COP = 37.5 for hot heat-exchanger temperature (Th) = 308 K and cold heat-exchanger temperature (Tc) = 300 K, the ratio of COPmaterials to Carnot COP of the L-DED–produced Ni–Ti nanocomposite is ≈7 times that of melt-casted Ni–Ti for adiabatic Brayton-like cycles (and ≈4 times that for isothermal Stirling-like cycles) at maximum transformations (Fig. 2H).

We studied the long-term stability of the L-DED–produced Ni–Ti nanocomposite. We found that the Ni–Ti nanocomposite is stable in its mechanical behavior and elastocaloric response for more than 1 million cycles (Fig. 3, A and B), indicating its potential for use in regular commercial products with a typical 10-year life [operating at <1 Hz (15)]. This performance is in contrast to other cycled 3D-printed Ni–Ti materials (45, 46). Here, small hysteresis enabled by the nanostructured Ni3Ti reinforcements is one important factor responsible for long-term stability. We previously showed that by tuning the lattice compatibility using stoichiometry in ternary alloys, we could minimize hysteresis of martensitic transformation and improve its reversibility to extended numbers of cycles (40, 47). However, a comparison of different SMA materials revealed that the absolute value of hysteresis is not the only determining factor. Magnetic SMAs such as polycrystalline Ni–Mn–In and Ni–Fe–Ga (table S1) seem to deteriorate quickly after a small number of cycles (~100), even with a hysteresis area as small as 1.2 MJ m−3. For stress-induced fatigue, the endurance limit (that is, the stress amplitude able to attain a prescribed number of cycles, usually 107, at zero mean stress) is proportional to the ultimate strength of materials by a factor of ≈0.33 (48). Across a spectrum of elastocaloric materials, it is the ratio of hysteresis area, ΔE, to the input work, E, that ultimately determines the number of cycles over which the materials can sustain their performance (Fig. 3C).

Fig. 3 Stability of Ni–Ti nanocomposite over 1 million compression cycles and comparison with other reported bulk elastocaloric materials.

(A and B) Compressive stress-strain curves (A) and elastocaloric cooling (B) of L-DED–produced Ni51.5Ti48.5/Ni3Ti nanocomposite aged at 923 K for 3 hours before and after 1 million cycles. (C) Log-log plot of the dissipated fraction of input energy, ΔE/E, versus sustained compressive cycles for bulk elastocaloric materials in this work as well as those reported in the literature. A dissipated fraction of energy is the ratio of hysteresis area ΔE in a transformation cycle to the input energy E. “Lattice-compatible” refers to the alloy in which the lattice parameters of transformed and untransformed phases exhibit exceptional lattice compatibility (42). The straight line is a linear fit. The data from both polycrystalline and single-crystal materials are included. The numerical values used in this plot, as well as strain amplitude and references, are listed in table S1.

To understand this trend, we considered an analogy to the well-known S–N concept conceived by Wöhler in 1858 (49) that connects the stress amplitude (S) to the cycles to failure (N) in the structural fatigue of materials and obtained a correlation of ΔE/E (hysteresis as a fraction of input energy) to the cycles to “functional failure,” N, (which we define as the number of cycles at the onset of loss of their functionality) (Fig. 3C). In an ideal case of ΔE/E = 0 (transformation with no hysteresis), the number of cycles to functional failure would asymptotically approach infinity (or extremely large numbers). SMAs typically exhibit hysteresis in superelastic cycles; the best ΔE/E hitherto reported for cycling are from Zn45Au30Cu25 alloys optimized through tuning the lattice parameters (42) and a Ni51.5Ti48.5/Ni3Ti nanocomposite with friction-limited kinematics in this work, both of which possess a ΔE/E less than 10%. Because of similarity in the hysteresis behavior associated with input work among different materials, the energy-based (ΔE/E)–N correlation observed here for elastocaloric materials could in principle apply to other caloric materials (magnetocaloric and electrocaloric materials). Even though the data on fatigue behavior of other caloric materials are somewhat limited (table S1), our preliminary analysis indicates that the same correlation holds for them as well. Caloric materials based on first-order transitions with reported low cyclability (such as <10,000 cycles) can potentially have their functional fatigue lives extended if their ΔE/E can be decreased through, for example, materials processing.

The conventional wisdom in the SMA community is that presence of nonequiatomic phases such as Ni3Ti in the NiTi matrix is detrimental to materials integrity because the presence of brittle phases precipitated along grain boundaries can lead to fracture from local stress concentration (50) and mismatch stress generated from transformation-induced shape distortions in neighboring grains (51). The nonequiatomic phases have also plagued the self-propagating high-temperature synthesis used for porous Ni–Ti materials for decades because they occur inevitably and produce chemical inhomogeneity in porous implants (52). We created a Ni–Ti–based elastocaloric material whose exceptional stability and unusual operational efficiency are derived from their distinct and intricate nanocomposite structures made by means of additive manufacturing. This demonstration shows the potential for using additive manufacturing to optimize caloric cooling by providing a highly desirable topology flexibility into materials components that serve as both refrigerants and heat exchangers.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S8

Table S1

References (5485)

References and Notes

  1. Materials and methods are available as supplementary materials.
Acknowledgments: We thank N. M. Wereley and T. Pillsbury of the University of Maryland for assistance in cyclic experiments and J.-S. Park of the Argonne National Laboratory for support in configuring and aligning the x-ray diffraction experiments. Funding: Advanced Research Projects Agency-Energy (ARPA-E) of the U.S. Department of Energy (DOE) supported the original characterization of shape memory alloys at the University of Maryland under grant ARPA-E DEAR0000131. The use of the laser-engineered net shaping (LENS) equipment was supported by the Critical Materials Institute, an Energy Innovation Hub funded by the Advanced Manufacturing Office of the Office of Energy Efficiency and Renewable Energy of the DOE. The work at Ames Laboratory was also supported by the Division of Materials Science and Engineering of the Basic Energy Sciences Programs of the Office of Science of the DOE under contract DE-AC02-07CH11358 with Iowa State University. A.P.S. and C.C. acknowledge the funding from the National Science Foundation (Career Award CMMI-1454668), and N.S.J. acknowledges a Los Alamos National Laboratory Additive Manufacturing Graduate Fellowship given to the Alliance for the Development of Additive Processing Technologies (ADAPT). V.I.L. acknowledges the funding from the National Science Foundation (MMN-1904830) and Army Research Office (W911NF-17-1-0225). The in situ x-ray diffraction experiments were performed at the 1-ID-E beamline of the Advanced Photon Source, a DOE Office of Science User Facility operated for the DOE Office of Science by Argonne National Laboratory under contract DE-AC02-06CH11357. Author contributions: I.T. initiated and supervised the research. H.H., J.C., and I.T. planned the experiments and designed the samples. E.S. and R.T.O. prepared the materials using the LENS. D.S. and N.A.H. characterized the composition of materials. H.H. carried out the experiments—including heat treatment, DSC measurement, superelastic tests, elastocaloric cooling measurements, and long-cycle tests—and analyzed the data. T.M., L.Z., and M.J.K. conducted SEM and TEM analysis. N.S.J., C.C., and A.P.S. performed and analyzed the in situ x-ray diffraction experiments, and C.C., M.A.Z., and A.P.S. performed the finite element modeling. S.Q., Y.H., and R.R. discussed the thermodynamic cycles. V.I.L. discussed the mechanism involving interfacial dislocations and shakedown. H.H., A.P.S., J.C., and I.T. wrote the paper with substantial input from other authors. All authors contributed to the discussion of the results. Competing interests: I.T. is a founder of Maryland Energy & Sensor Technologies, a company that works on elastocaloric technologies. Data and materials availability: All data are available in the manuscript or the supplementary materials.

Stay Connected to Science


Navigate This Article