Recycling of NiTi Shape Memory Alloys – Fundamental and Technological Aspects of a Vacuum Induction Melting (VIM) Processing Route

1 Recycling of NiTi Shape Memory Alloys – Fundamental and Technological Aspects of a Vacuum Induction Melting (VIM) Processing Route S. Noorzayee1, S. T. Mai2, S. Menon2, B. Sánchez Ortega2, R. Drautz2, A. Kau<mann1, G. Eggeler1, J. Frenzel1* 1 Institute for Materials, Ruhr-Universität Bochum, Universitätsstr. 150, 44801 Bochum, Germany 2 Interdisciplinary Centre for Advanced Materials Simulation (ICAMS), Ruhr-Universität Bochum, Universitätsstr. 150, 44801 Bochum, Germany * Corresponding author email: [email protected] Abstract The present study explores recycling of NiTi shape memory alloys (SMAs) using vacuum induction melting (VIM). Recycling NiTi is considered as challenging due to unavoidable carbon and oxygen pickup, which a<ects structural and functional properties. A 1 kg high-purity NiTi ingot was prepared from elemental Ni pellets and Ti blocks using VIM with a graphite crucible. The resulting SMA ingot underwent three additional remelting cycles. Samples for chemical and microstructural analysis were taken from the original ingot and after each remelting step. The study analyzes how di<erent feedstocks - pure Ni and Ti versus NiTi SMAs - a<ect melt pool temperatures and VIM durations. It was found that the high heat of mixing during alloy formation serves as an internal heat source, allowing shorter VIM process durations. In contrast, remelting NiTi alloys, which lacks this heat release, relies entirely on external power, which increases the process duration and thus the time available for impurity pick-up. The VIM process is analyzed using CALPHAD-based computational thermodynamics, combined with novel atomistic simulations using machine-learning potentials to determine thermodynamic conditions. The study assesses energy balances and contributes to a better understanding of how VIM remelting a<ects the microstructures and functional properties of NiTi SMAs. Keywords: Shape memory alloys, Recycling, Vacuum induction melting, Impurity pick-up, Functional properties – Submitted for publication, 22.12.2025 – 1. Introduction 1.1 Background NiTi shape memory alloys (SMAs) can recover their original geometry after deformations significantly exceeding maximum elastic strains of other metals [1–4]. The e<ect arises from a reversible, martensitic transformation that can be triggered by either thermal or mechanical 2 driving forces [1–5]. Essentially, the transformation corresponds to a shear along a specific crystallographic plane [5–7]. Upon cooling, the high-temperature phase austenite transforms into the low-temperature phase martensite between martensite start and finish temperatures, MS and MF, or when su<iciently high mechanical stresses sAM are applied. The reverse transformation occurs during heating between austenite start and finish temperatures, AS and AF, or during mechanical unloading. NiTi SMAs can manifest two di<erent types of shape memory e<ects (SMEs), depending on composition and microstructure [1–5]. The thermal SME (one/two-way e<ect) allows the initial geometry to be recovered during heating, when the material transforms from martensite with a preferentially oriented microstructure back to austenite. In contrast, the mechanical SME (pseudo-/superelasticity) refers to large reversible deformations established by the formation of stress-induced martensite, which transforms back to austenite during unloading. NiTi SMAs are commercially the most successful SMA. They are routinely applied across a broad range of technological fields, e.g. in medical technology, industrial production, logistics and household devices [8–11]. New potential applications emerge in energy conversion [12–19] and high-temperature actuation [20–22]. NiTi-based SMAs outperform other types of SMAs (e.g. based on Cu and Fe) due to their superior functional and mechanical properties [23,24], low functional and structural fatigue [25–31], corrosion resistance and biocompatibility [32–34]. Furthermore, enormous scientific progress on the understanding of relationships between processing conditions, microstructures and properties has been achieved in the last decades, e.g. [35–39]. It has been documented that the functional properties of NiTi SMAs strongly depend on alloy composition and microstructure, e.g. [40–45], which both need to be precisely controlled during processing. The preparation of NiTi SMAs usually requires high-purity raw materials [31,40]. The extraction of Ni and Ti from their ores (Ni: Pentlandite, Garnierite, Limonite; Ti: Ilmenite, Rutile) is highly energyintensive, associated with a high global warming potential (GWP), and expensive [46–50]. Figure 1 compiles data on energy demand, ecological and economic aspects of Ni, Ti, and relevant engineering materials for reference. In Fig. 1, Ni* refers to extraction of Ni by flash furnace smelting and Sherritt-Gordon refining [51,52], and Ni** to extraction by pressure acid leeching and solvent extraction/electrowinning [53,54]. The gross energy requirement (GER) per kg of Ni in mining and refining at up to 194 MJ/kg outnumbers regular steel (23 MJ/kg) almost tenfold and stainless steel (75 MJ/kg) roughly threefold, Fig. 1a. Ti with a GER of 361 MJ/kg even overshadows the high GER of Al (211 MJ/kg) by a factor of 1.5. The resulting GWPs of Ni (16.1 kg CO₂ eq/kg) and Ti (35.7 kg CO₂ eq/kg) show a similar relation when compared to steel and Al, Fig. 1b. From an economic point of view, the production of both Ni and Ti is also associated with significantly higher costs than that of steel and Al, respectively. On average, Ni is more than 15 times as expensive as steel, while Ti prices are more than four times higher than those of Al [55]. We note that the data in Fig. 1 represent average values for industrial-grade metals. In case of NiTi SMAs, high-purity Ni and Ti are required to keep trace element concentrations, e.g. of O and C, low [31,40,56]. Therefore, the production of these raw materials is associated with additional purification e<orts and costs. 3 Figure 1: Ecological and economical aspects of NiTi recycling. (a) Gross energy requirements (GER) for the production and (b) associated global warming potential (GWP) [46] and (c) price comparison. [55] Ni* - generation through pyrometallurgy. Ni** - hydrometallurgical generation [46]. No diKerentiation is made for Ni refining in Fig. 1c. Data for conventional steel, stainless steel and Al are shown for reference. The field of sustainable materials has received significant attention during the last decade, with various approaches being pursued to improve material-related sustainability [57–64]. The most direct ones involve the selection of raw materials for material synthesis. Simple, abundant, and non-toxic alloy chemistries without critical elements o<er a convenient way to reduce both energy consumption and associated CO₂ emissions [59,61]. Highly durable, corrosion-resistant materials also improve sustainability by extending component lifetime [59]. Indirectly, simple designs that enable straight-forward disassembly improve the feasibility of recycling e<orts [63,64]. Establishing a closed material life cycle through recycling improves sustainability by reducing the amount of primary raw materials for production. In the case of NiTi SMAs, however, no systematic studies on recycling are available at present. Previous research on processing of NiTi-based SMAs, e.g. [35,65–67], allows us to forecast that it is demanding to precisely control alloy chemistry, microstructures and related functional properties during repeated remelting. The main challenge is that NiTi SMAs react very sensitively to small changes in composition and microstructures [40,68–71]. An increase in the Ni concentration by only 0.1 at% will cause a decrease in the MS temperature by more than 8K [40]. Increasing amounts of C and O, which are the most critical impurities in NiTi, provide a similar e<ect on transformation temperatures [40,68,69]. NiTi alloys have no solubility for both types of elements, and therefore, excess C and O are gettered in the form of TiC and Ti2NiOx inclusions, respectively. Both phases are Ti-rich and thus, their formation causes a local Ti depletion in the matrix, resulting in an increased Ni content 4 and in the reduction of transformation temperatures, e.g. [2,31,40,72,73]. Hence, it is key to keep impurity levels in NiTi at a minimum [31] as trace elements can hardly be removed. Ito et al. [74] and Miyamoto et al. [75] successfully managed to de-oxidize NiTi melts by adding Barium, a highly reactive alkaline-earth metal that is di<icult to handle in industrial processes. A study on applying electroslag remelting of NiTi failed in establishing su<iciently low O concentrations [76]. Recently, the application of electron beam remelting [77] for the purification of NiTi has been discussed, but no details are publicly available. Figure 2 prototypically shows the life cycle of a NiTi component, starting with melting, followed by thermomechanical processing, component assembly, usage and disposal. Recycling closes the loop by reintroducing the material to the melting step. Figure 2 also demonstrates how C and O can be introduced to the material. Blue circles indicate potential paths where C and O lead to internal impurification, whereas red circles are related to surface contaminations. C and O represent trace elements often contained in Ti raw materials used for melting. Small amounts of C are also found in Ni. Additional C and O amounts are introduced into the SMA during the melting process. Oxygen contamination is mainly caused by small-scale leakage of the melting system. Graphite crucibles, which are frequently used for induction melting of NiTi, slowly react with NiTi melts, resulting in additional C pick-up. In the later stages of the industrial life cycle, C and O appear as surface contaminations. Oxide layers form during thermomechanical treatments and are not always fully removed from the material. C-containing surface contaminations are caused by lubricants used for mechanical processing, e.g. wire drawing, and during usage. In general, recycling often requires a conditioning pre-treatment of the material, where surface contamination is minimized [78–80]. However, complete removal of all non-alloy species is physically impossible, even with significant cleaning e<orts. The corresponding elements can therefore be introduced into the material during remelting. Considering that a material passes through the industrial life cycle more than once, impurities accumulate and potentially reach concentrations that exceed specification tolerances. 5 Figure 2: Industrial life cycle of a NiTi component. DiKerent paths for the potential pick-up of C and O trace elements are indicated. For details see text. Induction melting and reactions between NiTi melts and graphite crucibles Industrial NiTi SMAs are often prepared by vacuum induction melting (VIM) in graphite crucibles [81,82]. Electromagnetic induction is exploited to induce eddy currents inside both the conductive crucible and feedstock materials [83]. VIM is associated with an electromagnetic stirring e<ect, providing excellent chemical homogeneity. The use of graphite crucibles for VIM preparation of NiTi SMAs has already been suggested in the 1960s by Drennen et al. [65] in their report on melting procedures for NiTi alloys. Graphite, unlike other crucible materials, only introduces small amounts of trace elements, in this case C, into the melt. Additionally, these crucibles are thermo-mechanically stable and can be used for a large number of melting cycles [35,67]. However, small C pick-up when using graphite is unavoidable. Previous studies investigated the e<ects of the type of graphite [67]. Moreover, a crucible-charging technique (“Ticladding”) was introduced to minimize C contamination of the alloy [35]. Zhang et al. [84] studied the reactions between graphite and NiTi melts in detail. In ref. [68], the e<ects of C on the martensitic transformations were analyzed. The following aspects are relevant for the present study: 1) C pick-up of the NiTi melt is a two-stage process. Both raw elements, pure Ni and Ti, can react with graphite [85,86] and dissolve C in the early stages of the melting process [35]. 2) After mixing of both elements, the NiTi melt still picks up C [84]. This process has been described as a thermally activated reaction where C atoms di<use through a TiC layer forming at the graphite/melt interface. Higher temperatures result in faster C contamination of the material. Even after reaching saturation, the C content of the melt still increases, as small TiC crystals, which detach from the graphite/melt interface, float into the liquid NiTi. 6 3) Carbides like TiC form during solidification of C-containing NiTi melts in intercellular/interdendritic regions [35,68], which alters the local alloy chemistry. The matrix surrounding a TiC particle is depleted in Ti, which locally results in lower transformation temperatures [68]. Figure 3 presents scanning electron microscopy (SEM) images of carbides in NiTi alloys. Materials with lower C concentrations contain fiber-like carbides, Figure 3a. Microstructures of C-richer SMAs are characterized by primary carbides, which can have dendritic morphologies, Figure 3b. Microstructures of C-containing NiTi SMAs can be rationalized on the basis of the ternary Ni-Ti-C phase diagram [87], Figure 3c, which reflects that binary NiTi (single-phase region highlighted by blue arrow) has no significant solubility for C, and TiC (red arrow) does not dissolve Ni. Therefore, alloys with small C concentrations are in a two-phase region of the phase diagram where NiTi and TiC are in equilibrium. The present study investigates the application of the VIM method for recycling of binary NiTi SMAs. The goal is to understand how the use of materials from primary and secondary sources a<ects VIM-operating conditions, C and O trace element concentrations, microstructures and the resulting functional properties. The results are used to assess sustainability-related aspects of NiTi recycling. An e<ort is made to invoke thermodynamic analysis for the assessment of the technical (re-)melting processes. As a conventional strategy, we employ thermodynamic calculations using the CALPHAD approach. Additionally, we perform atomistic simulations using machine learning potentials, a novel technique that could provide thermodynamic information for future analyses of less-studied material systems. Figure 3: Carbides of type TiC (red arrow) in NiTi SMAs (blue arrow). (a) Eutectic TiC. (b) Primary TiC. (c) Ternary phase diagram for the Ni-Ti-C system at 900°C [87]. 7 2. Experimental methods 2.1 Induction melting, heat treatments and chemical analysis A 1kg NiTi SMA ingot with equiatomic composition was prepared from elemental raw materials in a VIM furnace and subsequently subjected to three remelting cycles. After each melting step, small specimens were extracted to investigate how composition, microstructure and transformation temperatures are a<ected by VIM-based recycling procedures. A VIM furnace of type VSG 010 from PVA Tepla (Wettenberg, Germany) was used for melting. Figure 4 compiles photographs documenting the melting process. In a first step, elemental Ni pellets (>99.95 at%; supplier: Ampere GmbH, Dietzenbach, Germany), Ti blocks and Ti disks (99.95 wt.%; Hauner Metallische Werksto<e, Röttenbach, Germany) were placed in a graphite crucible from Tryba Stockum GmbH (Breuberg, Germany) for melting. Figure 4a shows how elemental raw materials (marked as 1) and the crucible (2) are mounted in the water-cooled coil (3) of the furnace. An Y2O3-coated steel mold (4), in which the molten material is poured for solidification, was placed below the coil/crucible assembly. The internal geometry of the mold yields an ingot of cylindrical shape (diameter 40mm, length 110mm) with a conical pouring cup. The melt pool temperature was measured with a dual-wavelength pyrometer of type ISR 12-LO from LumaSense Technologies GmbH (Frankfurt a.M., Germany), initially calibrated by determining the solidification temperature of the NiTi melt. Casting was done when a temperature of at least 40-50°C above the melting point was reached and all metal was in the molten state for a minimum of 10 s. To minimize C pick-up during the early stages of the melting process, the Ticladding technique described in ref. [35] was applied. Inductive heating was carried out under high-purity Ar atmosphere at a pressure of 500 mbar, after the VIM furnace was evacuated to a pressure of 3x10-5 mbar and subsequently filled with Ar (99.9999 vol.%). A constant power of 15 kW was applied throughout the complete inductive-heating procedure. All further details on VIM processing and as-cast NiTi microstructures are documented in the literature [35,67,68,88]. Figures 4b and c show early and late stages of inductive heating. Figure 4b shows the situation a few seconds prior to the first melting events. In Figure 4c, large parts of the Ni and Ti feedstock are liquid and remaining solid Ti blocks dissolve in the surrounding liquid. The blue color is caused by an optical filter blocking IR radiation and reducing the light intensity in the visible spectrum. After complete mixing, casting and cooling down for 2 h, the ingot was removed from the VIM furnace. Small material portions for subsequent analysis were extracted from its upper part (corresponding to the pouring cup) using an abrasive cutting machine of type Brillant 220 from QATM (Mammelzen, Germany). Prior to each remelting iteration, e<orts were made to minimize the contamination of the recycled material by establishing clean, metallic ingot surfaces. Small ceramic particles adjacent from the previous melting experiment were removed using a wire brush, followed by ultrasonic cleaning in acetone and ethanol for 3 min each. Figure 4d shows how parts of the initial ingot are placed in the crucible prior to the first remelting step. Early stages of inductive heating are shown in Figure 4e, and the later stages, where a small amount of NiTi is in the molten state, are presented in Figure 4f. We note that all VIM experiments were conducted using the same graphite crucible. This does not a<ect the purity of the resulting ingots [35], as only a very small amount of metal remains in the crucible after melting. 8 Figure 4: Induction melting of NiTi SMAs. Melting set-ups for the preparation of NiTi SMAs from (a,b,c) elemental feedstock (Ni pellets and Ti blocks), and (d,e,f) from pre-existing NiTi. (a,d) Initial stages prior to melting. (b,e) Early stages of inductive heating. (c,f) Melting. The blue color is caused by red/IR filters. In general, di<erent types of feedstock can a<ect the operating conditions during induction melting [83,89]. To analyze these aspects, the melting process was monitored in terms of temperature evolution and process durations. Figure 5 illustrates these aspects based on a simplified temperature T / time t profile. The total process duration is divided into three intervals: First, the time required to heat the solid feedstock until the first melting event occurs (tsolid); second, the duration between initial and complete melting (tmelting), resulting in a plateau of the T(t) curve due to the latent heat required to complete the melting; third, the final amount of time 9 (toverheating) needed to superheat the alloy by at least 40-50 K to prevent premature solidification when casting into the mold. The T(t) profile in Figure 5 also shows three characteristic temperatures: The feedstock melts at Tm, the temperature 10 s after complete melting is referred to as T10s, and casting is conducted at Tcasting. The parameters specified in Figure 5 were used to characterize how the VIM process di<ers when NiTi is prepared from pure elements or from an existing ingot. We note that true T(t) profiles usually deviate from what is shown in the illustration in Figure 5. Temperature distributions during inductive melting are inhomogeneous due to the skin e<ect, the divergent magnetic flux generated by the induction coil [83] and other e<ects related to heat and mass transport. Figure 5: Simplified schematic illustration describing the evolution of temperatures during melting. DiKerent times are required to reach the melting point (tsolid), to subsequently transform the material into the liquid state (tmelting), and to overheat the melt (toverheating). Three diKerent temperatures are important: Melting point Tm, the temperature of the melt 10s after complete melting, T10s, and the final casting temperature Tcasting. An e<ort was made to document how the masses of the ingot and the crucible evolve during the four (re-)melting experiments. Sample extraction resulted in an ingot material loss of around 50 g after each (re-)melting step. Figure 6 shows how the ingot geometry was altered by the material loss. Figure 6a shows the initial NiTi ingot resulting from the first melting experiment while Figure 6b presents the same material after three recycling iterations and repeated sample extractions. It is easy to identify the loss of material in Figure 6. Table 1 compiles data documenting how ingot and crucible masses evolve during our experiments. Table 1 also contains data describing the evolution of the melt volume VM, the contact area between melt and crucible AMC, as well as the ratio between both parameters. The corresponding data can be used in the rate equation published by Zhang et al. [84], which describes the evolution of the C concentration during melting: a large area AMC results in a large total C flux into the liquid, while a large melt pool volume slows down the increase of C concentration [84]. Masses listed in Table 1 were determined using a scale of type CP 4202 S by Sartorius AG (Göttingen, Germany). Melt volumes were calculated using temperature-dependent densities reported in literature [90]. The data show that melting and iterative sample extraction caused a 16 recycling and finally reaches a value of 25.5 °C after the third iteration. The largest reduction in MS occurs between the second and third recycling steps, where a drop of roughly -12 K was detected, while the change after the first recycling step is more moderate at about -6 K. Summarizing, the data presented in Figure 11 show scatter, and the average shift is close to -6.5 K per recycling step. Figure 11: EKect of recycling on phase transformation behavior. (a) DSC curves of the diKerent materials. (b) Drop of the martensite start temperature Ms related to recycling. 3.3 Thermodynamic analysis of melting and mixing reactions Synthesizing and recycling materials by cast metallurgy involves the following three steps: (i) heating the materials up to process temperature, (ii) melting of the raw elements (Ni and Ti) or of NiTi melts (requiring the latent heat ∆𝐻!) and, in the case of starting with the pure elements, (iii) mixing, which is associated with ∆𝐻!"# depending on the interaction energies involved in alloy formation [100]. A detailed assessment is complicated, because it is not clear which distinct reaction paths are followed. To obtain comparable quantities for the energy supply to heat up the material, we consider the casting temperature Tcasting as reference temperature. Heating of pure Ti, pure Ni and NiTi to Tcasting requires 934 J/g, 744 J/g and 1041 J/g, respectively. These enthalpy changes include the α-Ti to βTi transformation as well as the B19’ to B2 transformation [40,99] and the solid/liquid transition of B2 NiTi. Table 2 provides an overview on CAPLHAD data describing the amounts of heat required to heat the feedstock materials to their melting points / liquidus temperature and to the casting temperature. In case of NiTi, a value of 28 J/g was added to account for the transformation from B19’ martensite to B2 during heating. The enthalpy changes in the processes of melting and mixing are depicted in Figure 12a. Literature data from [101–105], CALPHAD calculations and the novel GRACE estimations are included. The latent heats (at melting temperature 𝑇 ! $"/𝑇 ! %" or liquidus temperature 𝑇& $"%") required to fully melt Ni, Ti or B2 NiTi are all close to 300 J/g. This reflects similarly large entropy changes during melting ∆𝑆!=∆(! )! and similar melting/liquidus temperatures. As solid NiTi is a B2 compound, Ni-Ti bonds are substantially preferred over Ni-Ni and Ti-Ti bonds. Therefore, the formation of NiTi is endothermic in nature [45]. During VIM of Ni and Ti, di<erent types of reactions can potentially occur, which all provide a release of heat, 17 represented by the negative enthalpies shown in Figure 12a. The first set of negative bars in Figure 12a reflects the amount of energy when Ni and Ti both mix in the liquid state at the melting point of pure Ti (1660 °C). This reaction is considered less important, as such high temperatures are unlikely to occur during VIM, see Figure 8. The second set represents the reaction where solid Ni and solid Ti form solid B2 NiTi. This reaction can occur when pure Ni and Ti are in contact during heating. As it requires slow di<usion in the solid state, this reaction only plays a minor role. The third set shows the energy release associated with the reaction where liquid Ni reacts with solid Ti to form a NiTi melt. This reaction governs the initial synthesis of NiTi. Melting of pure Ni could be directly observed when the initial ingot was prepared. As soon as a Ni-rich melt formed, the dissolution of solid Ti took place. Figure 12a documents that the heat release during phase formation is almost independent of mixing reactions. In all cases, the enthalpies of mixing range between -700 and -720 J/g. Thus, even when both processes of melting and mixing take place simultaneously, a net heat release of about 460 J/g results. An additional important finding is shown in Figure 12a: The latent heats and the heats of transformation determined by CALPHAD (traditional thermodynamics) or GRACE (novel atomistic calculations) are very close. They are moreover in excellent agreement with values reported in literature [101–105]. Figure 12b shows how the enthalpies of Ni, b-Ti and B2-NiTi depend on temperature. All enthalpy values increase with increasing temperature (slope ≈ 3R, Dulong-Petit [106]). Figure 12b also highlights the enthalpy change caused by NiTi formation via the reaction of liquid Ni with solid bTi (arrow pointing down). Figure 12c provides a more general view on the mixing enthalpies, which are relevant when a NiTi melt is formed by mixing of Ni and Ti atoms. Figure 12c contains data generated by CALPHAD and GRACE calculations, covering the complete compositional range in the liquid phase from 0 to 100 at.% Ti. The curves plotted in Figure 12c show that the mixing enthalpies are always negative, with a minimum close to 38 at.% Ti. The data show that both theoretical approaches again yield results in reasonably good agreement with one another. 18 Figure 12. Thermodynamic assessment of melting and mixing reactions. (a) Enthalpies involved in melting and mixing of Ni and Ti to form liquid NiTi or solid B2-NiTi. Literature data is collected from [101–105].(b) Dependence of enthalpies on temperature (CALPAD data; B19’ manually added using data from ref. [40] ). The central arrow indicates the reaction where liquid Ni and solid Ti form a NiTi melt. (c) Dependence of mixing heats on composition. We note 19 that CALPHAD, GRACE and literature sources yield slightly diKerent melting / liquidus temperatures. Therefore, results are reported for the model/origin-specific case. Table 2: Enthalpy changes (J/g) involved in heating diKerent feedstock from room temperature up to the final temperatures indicated, according to the thermodynamic calculations (CALPHAD). Values in italics are solely related to heating in the solid state. Melting enthalpies are indicated by an asterisk (*); enthalpies of solid-solid phase transformations are indicated by two asterisks (**). Feedstock\Final temperature 𝑇! "#$#+(1310+°C) 𝑇%&'(+(1360+°C) 𝑇) "#+(1457+°C) 𝑇) $#+(1668+°C) Ti (incl. α-β transformation) 900 934 1005 (1165 + 296*) Ni 711 744 (808 + 298*) 1261 NiTi (incl. B19’-B2 transformation) (728 + 249* + 28**) (1013 + 28**) (1093 + 28**) (1268 + 28**) 4. Discussion 4.1 General aspects of NiTi recycling and sustainable NiTi metallurgy The present work represents the first systematic fundamental study on recycling NiTi SMAs. Recycling these alloys has previously been considered di<icult. However, data such as those shown in Figure 11 demonstrate that remelting by VIM in graphite crucibles produces recycled NiTi SMAs that clearly exhibit a reversible forward/reverse martensitic transformation upon cooling and heating. We demonstrated that each recycling iteration results in a 6.5 K drop of MS, Figure 11b, caused by an increase of the C and O concentrations, Figure 9. It is clear that pick-up of C is related to the reaction between the NiTi melt and the crucible. The (smaller) increase in O levels, on the other hand, is caused by unavoidable small-scale leakage of the vacuum system and by oxides which cover the surfaces of the feedstock prior to remelting [35,84]. To compensate for changes in transformation temperatures, Figure 11, small amounts of additional Ti could be added during the remelting process. Also, it is likely that the process itself can still be optimized to minimize durations, required energies and impurity pick-up. Smaller blocks of NiTi for remelting can probably be heated more e<ectively than the cylindrical ingot used in the present work, Figure 6a. Additionally, a more intensive surface cleaning and/or etching of the material prior to remelting, and minimizing small scale leakage of the vacuum system, could further contribute to reducing O pick-up. The corresponding reactions are surface-controlled and therefore depend on the surface-to-volume ratio of semi-finished products or components considered for recycling. Thin wires, rods or tubes, or even cut-out parts from stent production may be regarded. In general, a large surface-to-volume ratio facilitates the introduction of trace elements, as has been documented in recycling studies for other materials, e.g. [78,107,108]. Our recycling approach was unable to provide NiTi SMAs with C concentrations below the strict limits for medical applications specified in ASTM-F2063-18 [56]. Therefore, one cannot use the corresponding material in the medical field. However, application fields which do not have to withstand large loading/unloading cycle numbers do not require such high purities. We also note that higher purities can be established when other remelting techniques such as arc melting, where NiTi melts do not react with graphite crucibles, are exploited. 20 4.2 Thermodynamic assessment of melting and mixing reactions This study represents one of the rare cases documenting how the type of feedstock a<ects processing conditions during induction melting. The data shown in Figure 8 demonstrate that the use of elemental Ni and Ti significantly shortens the overall process duration because less time is required for complete melting, and there is no need for additional overheating. In contrast, remelting bulk NiTi requires much longer durations, although the melting point of 1310 °C is lower than that of pure Ni (1457 °C) and Ti (1668 °C) [99]. While full melting requires the supply of the latent heat, the mixing and phase formation of Ni and Ti is associated with the release of a much larger enthalpy, as seen in Figure 12, acting as an internal heat source during alloying. The corresponding amount of heat is large enough to prepare NiTi alloys through a self-combustion synthesis [109], where pure Ni and Ti powders are mixed and subsequently heated. Accordingly, the assistance of the melting process by mixing and alloy formation is barely dependent on the exact reaction sequence, i.e. if the two processes take place in the liquid or solid phase. It is worth noting that the heat of mixing seems to play a general role when preparing NiTi SMAs from pure Ni and Ti. It allows process durations to be kept short and thus lowers C pick-up during VIM processing. In this context, the positive contribution of the heat of mixing has been overlooked in previous studies. While this is the case for the initial synthesis of NiTi with a net heat, the entire latent heat of about 300 J/g needs to be supplied during recycling. The enthalpies calculated using the foundational GRACE potential exhibit strong consistency with literature (experimental/thermodynamic) and CALPHAD data (Figures 12a, 12c). This result is significant because the potential, trained on a periodic table-spanning dataset, accurately predicted the Ni-Ti system without system-specific optimization. Thus, the GRACE potential provides robust and transferable thermodynamic predictions. 4.3. Realistic energy requirements for alloy formation and recycling by vacuum induction melting The enthalpies of the di<erent reactions considered in Figure 12 significantly a<ect the melting behavior during VIM processing in terms of durations and melt pool temperatures, Figure 8. However, the amounts of energy required for alloy formation and recycling in a technical process like VIM deviate from idealized thermodynamic considerations. A VIM furnace is not a calorimeterlike instrument with almost perfect thermal insulation. During VIM processing, significant energy losses occur due to uncontrolled heat transport (conduction and radiation) between feedstock, crucible, induction coil, and the water-cooled vacuum chamber. Therefore, an e<ort was made to assess the true amounts of energy for alloy formation and recycling in the VIM system under practical conditions. The data on VIM process durations, Figure 8a, can be used to calculate the process energies to melt 1kg-NiTi ingots. The total energy of a melting cycle can be calculated based on equation 1, 𝐸 = 𝜂 ∙ 𝑃 89: ∙ 𝑡;<;=> (1), where E represents the input energy for the melting process, h the e<iciency factor of the middlefrequency generator, Psec the electrical output power, and ttotal the total processing time (corresponding to the heights of the bars in Figure 8a). With an e<iciency factor of 0.95 (provided 21 by the manufacturer of the heating system) and a constant power of 15 kW, the calculation of the energies is straight forward, Figure 13. Figure 13a shows that close to 6.45 MJ are necessary to melt 1kg NiTi from the two pure elements in a laboratory-scale induction furnace. In contrast, a higher average amount of 11.7 MJ is needed for remelting of a NiTi ingot (of almost identical mass, Table 1), due to the lack of the heat of mixing acting as an additional heart source. The bar chart in Figure 13a also contains one additional data point labelled “Ideal process” which serves as a reference solely based on the enthalpy change required to heat 1kg NiTi from a temperature of 300 K to 1660 K. In fact, close to 1000 kJ represents the ideal/minimum amount of energy for ingot metallurgical processing. In the true technical process, however, almost 6 (for alloy formation) to 11 (remelting) times higher electrical input energies are required. The energy for VIM processing is not the only parameter relevant for the energetic assessment of recycling. Closing the material life cycle by recycling saves pure Ni and Ti, which would have to be extracted from ores. In fact, the generation of pure Ni and Ti for the synthesis of 1 kg NiTi requires close to 346 MJ (calculated using data published in [46]). This value considerably exceeds the amount of energy for the VIM melting process, Figure 13a, no matter which type of feedstock is used. Clearly, the higher energy requirement for recycling NiTi is almost irrelevant in view of the savings in pure Ni and Ti, as can be seen in Figure 13b where both energies (dark red: energies for melting process, same data as in Figure 13a; light red: energies for generation of pure Ni and Ti) are plotted for comparison. Figure 13: Energy consumption for the preparation of 1kg NiTi from elemental feedstock (recycling iteration 0) and for recycling of the resulting material (recycling iterations 1-3). (a) Input energy of the middle frequency generator used for VIM processing (red) and comparison with ideal loss-less process (orange). (b) Comparison of embodied energy (calculated from data presented in [46]) for the preparation of pure Ni and pure Ti from ores, and VIM melting energies. 4.4. EWects of recycling on microstructures In general, two di<erent TiC carbides can form in NiTi SMAs depending on the C levels in the material. TiC carbides with a fibre-like morphology form by a quasi-binary eutectic reaction during solidification of NiTi-C melts of hypoeutectic composition, Figure 3a. In contrast, large TiC particles with a dendritic morphology can be observed in NiTi SMAs with hypereutectic 22 concentrations, Figure 3b. The microstructural data in Figure 10 document that samples with C levels below 0.06 wt.% only feature small particles with circular cross sections, which correspond to the eutectic phases shown in Figure 3a. The NiTi sample that underwent 3 remelting cycles, however, additionally contained TiC particles with symmetrical dendritic shapes, with cross section diameters of several micrometers, Figures 7 and 10d. They also show crystallographic and morphological features which have been previously documented in ref. [110] for primary carbides. This allows us to conclude that the large particles in Figures 7 and 10d formed in the initial stages of solidification. The corresponding C concentration of this material state exceeds the quasi-binary eutectic composition, which has not been identified in experiments for a Ni50Ti50 SMAs so far. Note that the three-dimensional morphological details shown in Figure 3 were revealed by an electropolishing process, which fully dissolved the surrounding NiTi matrix. The results from flat metallographic cross sections naturally yield binary cuts through the 3D geometry and only allow to precisely measure cross sections of dendrites. The BSE-SEM micrographs in Figures 7 and 10 and thermodynamic data published in literature [87,99,111,112] can be used to generate a quasi-binary phase diagram NiTi/TiC, where the quasibinary eutectic point lies close to 0.26 at.% C / 1305°C as no primary carbides were observed for material states with lower C concentrations. The phase diagram in Figure 14 reveals that the solid NiTi and TiC phases cannot solve TiC and NiTi, respectively. The single phase stochiometric nature of the two phases is indicated by vertical thick lines at the outer edges of the phase diagram. Figure 14: Quasi-binary phase diagram of the NiTi-C system, constructed using SEM data obtained in the present study and thermodynamic data published in literature [87,99,111,112]. 4.5 Perspective of using recycled NiTi SMAs in application Recycling of NiTi SMAs represents an e<ective step towards sustainable metallurgy by significantly reducing the energy amounts required for pure Ni and Ti production used for alloy synthesis, Figure 13b. While an unavoidable chemical degradation will restrict usage of this material in the medical field, Figure 9, there is still a large variety of potential applications in other 23 technological areas, especially in energy conversion. In the last decade, solid state cooling applications (elastocaloric) have received remarkable scientific attention, and an already large number of prototypes has been designed, e.g. [12,13,15,16,113]. There also is an interest in the “opposite technology” of elastocaloric cooling, where low-temperature waste heat is converted into mechanical work using SMAs, e.g. [17,18,114]. Both technologies would not only benefit from lower costs of recycled SMAs. The production of corresponding devices would also benefit from a smaller ecological footprint when considering their complete service lives. Ecological aspects of using NiTi SMAs prepared from primary or secondary raw materials can in principle be compared to the situation of combustion and battery-electronic vehicles (BEVs). It is established that BEVs come with an initial ecological debt in terms of CO2, which is primarily related to the production of Li used for their batteries [115–118]. However, the total generation of CO2 during the complete life cycle of a BEV is supposed to significantly undermine the equivalent value for a fossil-fuel powered vehicle. In case of SMAs applied in energy conversion, “fresh” NiTi SMAs also have a high ecological footprint, meaning an ecological break-even point must be reached during their service life. Using recycled NiTi in these applications provides an attractive option to bypass this initial penalty and to improve the overall ecological footprint. 5. Summary and conclusions The present study explores the possibility of recycling NiTi shape memory alloys, considering a vacuum induction melting/remelting ingot metallurgy processing route. We analyze how repeated remelting steps alter VIM performance, chemical compositions, microstructures and functional properties. A 1kg NiTi ingot with equiatomic composition was prepared from elemental Ni and Ti feedstock and subsequently remelted three times. After each (re-)melting step, thermal analysis was performed and alloy chemistry and microstructure were analyzed. The experimental results were rationalized in the light of classical thermodynamic (CALPHAD) and novel atomistic simulations. From the results obtained in the present work, the following conclusions can be drawn: (1) It was experimentally observed that processing takes significantly longer when remelting a 1 kg ingot of an equiatomic NiTi alloy than when making the alloy starting with the pure metals. The phenomenon is related to the large enthalpy of mixing, which accompanies the alloying of binary NiTi melts from pure Ni and Ti feedstock. This favorably a<ects VIM processing: The enthalpy of mixing represents a significant internal heat source, which assists external heating and results in shorter processing times. This saves energy and shortens potential melt/crucible (C pick-up) and melt/atmosphere (O pick-up) interaction times. This positive e<ect of mixing on VIM processing is absent when NiTi alloys are remelted. In this case, longer processing times are required, associated with greater impurity pick-up. (2) An e<ort was made to calculate latent heats and enthalpies of mixing using conventional thermodynamics (CALPHAD) and atomistic calculations with novel machine learning potentials (GRACE). Both methods yield results in excellent agreement with each other and with literature data. They moreover fully rationalize the experimental findings. The fact that more energy is needed to melt NiTi than the same mass and ratio of pure Ni and Ti has so far not been appreciated in literature. However, from a global perspective, this process energy penalty is overcompensated 24 by the energy saved, which would otherwise be needed for mineral processing, extraction, and primary metallurgy to produce the pure metals. (3) Already after the first VIM remelting cycle, C levels of the recycled NiTi alloy exceed the maximum allowable impurity level for medical components as specified in ASTM 2063-18. Further work is required to refine the VIM processing route to minimize C contamination. Melting durations can potentially be reduced by applying higher heating power, optimizing induction frequencies or using optimized feedstock geometries and arrangements. Oxygen levels also increased, but this increase was less critical and did not exceed the maximum allowable levels. (4) The pick-up of C results in the formation of titanium carbides. This in turn reduces the Tiand increases the Ni-levels of the alloy. Consequently, the martensite start temperature slightly decreases in each remelting step, an e<ect which needs to be taken into account. A quasi binary NiTi-TiC phase diagram with an eutectic point at 0.26 at.% C rationalizes the observed microstructures. 6. Acknowledgement SN acknowledges financial support from the International Max Planck Research School for Sustainable Metallurgy (IMPRS-SusMet). JF and GE acknowledge funding by the Germany Research Foundation (Deutsche Forschungsgemeinschaft DFG) through projects 498172553 and 549600725. 7. Data availability Original scanning electron microscopy images, as well as data from chemical, thermal and thermodynamic analyses are accessible through the online data repository Zenodo (https://doi.org/10.5281/zenodo.17734505) 8. References [1] K. Otsuka, X. Ren, Physical metallurgy of Ti-Ni-based shape memory alloys, Prog. Mater. Sci. 50 (2005) 511–678. https://doi.org/10.1016/j.pmatsci.2004.10.001. [2] H. Funakubo, Shape Memory Alloys, Gordon and Breach, New York, 1987. [3] K. Otsuka, C.M. Waymann, Shape memory materials, Cambridge University Press, Cambridge, 1998. [4] D.C. Lagoudas, Shape memory alloys: modeling and engineering applications, Springer, New York, 2008. [5] K. Bhattacharya, Microstructure of Martensite: Why it forms and how it gives rise to the shape-memory eaect, Oxford University Press, Oxford, 2004. [6] A. Grünebohm, A. Hütten, A.E. Böhmer, J. Frenzel, I. Eremin, R. Drautz, I. Ennen, L. Caron, T. Kuschel, F. Lechermann, D. Anselmetti, T. Dahm, F. Weber, K. Rossnagel, G. Schierning, A Unifying Perspective of Common Motifs That Occur across Disparate Classes of Materials Harboring Displacive Phase Transitions, Adv. Energy Mater. 13 (2023) 2300754. https://doi.org/10.1002/aenm.202300754. 25 [7] S. Schwabe, R. Niemann, A. Backen, D. Wolf, C. Damm, T. Walter, H. Seiner, O. Heczko, K. Nielsch, S. Fähler, Building Hierarchical Martensite, Adv. Funct. Mater. 31 (2021) 2005715. https://doi.org/10.1002/adfm.202005715. [8] J. Mohd Jani, M. Leary, A. Subic, M.A. Gibson, A review of shape memory alloy research, applications and opportunities, Mater. Des. 1980-2015 56 (2014) 1078– 1113. https://doi.org/10.1016/j.matdes.2013.11.084. [9] R. Chaudhari, J.J. Vora, D.M. Parikh, A Review on Applications of Nitinol Shape Memory Alloy, in: A.K. Parwani, Pl. Ramkumar, K. Abhishek, S.K. Yadav (Eds.), Recent Adv. Mech. Infrastruct., Springer Singapore, Singapore, 2021: pp. 123–132. https://doi.org/10.1007/978-981-33-4176-0_10. [10] H. Sehitoglu, 10 Years of Shape Memory and Superelasticity, Shape Mem. Superelasticity 11 (2025) 309–310. https://doi.org/10.1007/s40830-025-00571-8. [11] C. Naresh, P.S.C. Bose, C.S.P. Rao, Shape memory alloys: a state of art review, IOP Conf. Ser. Mater. Sci. Eng. 149 (2016) 012054. https://doi.org/10.1088/1757899X/149/1/012054. [12] G. Zhou, L. Zhang, Z. Li, P. Hua, Q. Sun, S. Yao, Achieving kilowatt-scale elastocaloric cooling by a multi-cell architecture, Nature 639 (2025) 87–92. https://doi.org/10.1038/s41586-024-08549-9. [13] H. Mevada, B. Liu, L. Gao, Y. Hwang, I. Takeuchi, R. Radermacher, Elastocaloric cooling: A pathway towards future cooling technology, Int. J. Refrig. 162 (2024) 86– 98. https://doi.org/10.1016/j.ijrefrig.2024.03.014. [14] L. Manosa, A. Planes, E. Vives, E. Bonnot, R. Romero, The Use of Shape-Memory Alloys for Mechanical Refrigeration, Funct. Mater. Lett. 2 (2009) 73–78. https://doi.org/10.1142/S1793604709000594. [15] S. Fähler, U.K. Rößler, O. Kastner, J. Eckert, G. Eggeler, H. Emmerich, P. Entel, S. Müller, E. Quandt, K. Albe, Caloric eaects in ferroic materials: new concepts for cooling, Adv. Eng. Mater. 14 (2012) 10–19. https://doi.org/10.1002/adem.201100178. [16] J. Frenzel, G. Eggeler, E. Quandt, S. Seelecke, M. Kohl, High-performance elastocaloric materials for the engineering of bulkand micro-cooling devices, MRS Bull. 43 (2018) 280–284. https://doi.org/10.1557/mrs.2018.67. [17] B. Neumann, S. Fähler, Design guidelines for eaicient thermoelastic harvesting of low-grade waste heat, Energy Convers. Manag. X 27 (2025) 101099. https://doi.org/10.1016/j.ecmx.2025.101099. [18] P. Kumar, R.A. Kishore, D. Maurya, C.J. Stewart, R. Mirzaeifar, E. Quandt, S. Priya, Shape memory alloy engine for high eaiciency low-temperature gradient thermal to electrical conversion, Appl. Energy 251 (2019) 113277. https://doi.org/10.1016/j.apenergy.2019.05.080. [19] M. Nishida, E. Okunishi, T. Nishiura, H. Kawano, T. Inamura, S. Ii, T. Hara, Selfaccommodation of B19 ’ martensite in Ti-Ni shape memory alloys - Part II. Characteristic interface structures between habit plane variants, Philos. Mag. 92 (2012) 2234–2246. https://doi.org/10.1080/14786435.2012.669860. [20] P.E. Caltagirone, P.N. Ghezeljeh, O. Benafan, Shape Memory NiTiHf Machined Helical Springs: Balancing Displacement and Force Output for Actuation, Shape Mem. Superelasticity 10 (2024) 198–224. https://doi.org/10.1007/s40830-02400479-9. 32 [110] Z. Zhang, J. Frenzel, C. Somsen, J. Pesicka, J. Eggeler, On the Formation of TiC Crystals during Processing of NiTi Shape Memory Alloys, in: V. Karas (Ed.), Trends Cryst. Growth Res., Nova Science Publishers, New York, 2005: pp. 71–99. [111] J.C. Schuster, Y. Du, Die Konstitution des Dreistoasystems Ni-Ti-C, in: D. Hirschfeld (Ed.), Nichtmet. Met. 98, GDMB Gesellschaft für Bergbau, Metallurgie, Rohstoaund Umwelttechnik, Clausthal-Zellerfeld, 1998: pp. 73–77. [112] W.C. Ballamy, E.E. Hucke, The Liquid-Carbide Equilibrium in the Nickel-TitaniumCarbon System, J. Met. 22 (1970) 43–50. https://doi.org/10.1007/BF03355655. [113] L. Manosa, A. Planes, M. Acet, Advanced materials for solid-state refrigeration, J. Mater. Chem. A 1 (2013) 4925–4936. https://doi.org/10.1039/c3ta01289a. [114] A.B. Hegana, S.I. Hariharan, E.D. Engeberg, Electromechanical Conversion of Low-Temperature Waste Heat Via Helical Shape Memory Alloy Actuators, IEEEASME Trans. Mechatron. 21 (2016) 1434–1444. https://doi.org/10.1109/TMECH.2015.2481087. [115] D. Hall, N. Lutsey, Eaects of battery manufacturing on electric vehicle life-cycle greenhouse gas emissions, ICCT - The International Council on Clean Transportation, Beijin, Berlin, Brussels, San Francisco, Washington, 2018. https://theicct.org/wp-content/uploads/2021/06/EV-life-cycle-GHG_ICCTBriefing_09022018_vF.pdf. [116] T. Yuksel, M.-A.M. Tamayao, C. Hendrickson, I.M.L. Azevedo, J.J. Michalek, Eaect of regional grid mix, driving patterns and climate on the comparative carbon footprint of gasoline and plug-in electric vehicles in the United States, Environ. Res. Lett. 11 (2016) 044007. https://doi.org/10.1088/1748-9326/11/4/044007. [117] J. Liu, G. Santos, Decarbonizing the Road Transport Sector: Break-even Point and Consequent Potential Consumers’ Behavior for the U.S. Case, Int. J. Sustain. Transp. 9 (2015) 159–175. https://doi.org/10.1080/15568318.2012.749962. [118] M. Wietschel, M. Kühnbach, D. Rüdiger, Die aktuelle Treibhausgasemissionsbilanz von Elektrofahrzeugen in Deutschland, Fraunhofer, Karlsruhe, 2019. https://doi.org/10.24406/publica-fhg-299531.