1. Introduction
Samarium cobalt (SmCo) magnets are high energy density permanent magnets suitable for applications such as high-speed motors and capable of operating at temperatures up to 550 °C. This is due to their excellent magnetic thermal coefficients, high Curie temperature, and a protective oxide (Sm
2O
3) layer, making them ideal for extreme environments [
1]. One type of SmCo magnet is the 2:17 type, often referred to as Sm
2TM
17, where TM stands for transition metals (Co, Fe, Cu, Zr). The name 2:17 refers to the atomic ratio of Sm to transition metals in the hard-magnetic matrix phase. This nomenclature, as well as a capital letter denoting the crystal structure, will be used to refer to phases throughout this paper (see
Table 1).
Sm
2TM
17 magnets are highly dependent on a very particular phase structure, chemical composition, and heat treatment regime utilised during the manufacturing process [
2]. This is a sub-granular nano-cellular microstructure that is comprised of FeCo-rich 2:17R rhombohedral Th
2Ni
17-type phase cells, providing the high energy product and remanence of the magnet [
3]. These are surrounded by a Cu-rich boundary 1:5H hexagonal phase. The Cu concentration gradient between the cells and the boundary phase provides domain wall pinning, which is the dominant coercivity mechanism in 2:17 magnets [
4]. A Zr-rich phase forms a superimposed lamella and acts as a diffusion pathway during processing, allowing enrichment of the 2:17R cells with Fe and Co and the 1:5H boundary phase with Cu [
5]. To form this structure, Sm
2TM
17 magnets undergo long, high temperature heat treatments: firstly, a solution heat treatment at 1100–1200 °C for 4–10 h to form a metastable phase (1:7H), which is stabilised at room temperature by rapid quenching. This phase is described as disordered 2:17 in some literature and adopts a TbCu
7 structure [
6,
7] which decomposes during subsequent heat treatment to develop a uniform, optimum micro/nano-structure [
8,
9,
10]. This is followed by ageing at 750–850 °C for up to 12 h which allows the precipitation of the fine 2:17R cells and the formation and enrichment of the boundary phase from the precursor 1:7H phase. Further ageing is achieved by slow cooling to 400 °C and holding for 2–10 h to finesse the nanostructure of the magnet [
11]. It should be noted that times and temperatures vary between manufacturers depending on specific elemental composition, desired grain size, desired cellular structure, and target magnetic properties.
In traditional manufacturing routes, the alloy is cast into water-cooled steel/copper moulds, causing variable cooling rate during solidification. The edges experience an extremely high cooling rate which diminishes towards the centre of the ingot. The variation in cooling rate causes an extremely inhomogeneous microstructure, segregated with large fractions of undesirable phases. Furthermore, these alloys are difficult to process, requiring high-energy jaw crushing and milling to reduce the particle size significantly. The composition and microstructure remain inhomogeneous, and hence heat treatment is crucial to achieving the desired magnetic properties [
12]. The macrosegregation within the alloy is characterised by large regions rich in Fe and Co, with the other regions becoming enriched with Sm and Cu. Small amounts of the desirable metastable 1:7H phase are present in cast ingots, at the edges where the cooling rate is highest [
13]. Theoretically, if the material could be cast with a sufficiently high cooling rate, the homogenisation step of the heat treatment could be skipped to save energy and thus cost.
Strip casting is a technique used in the production of other rare earth magnets that allows for rapid non-equilibrium cooling, potentially producing more desirable microstructures [
14,
15,
16,
17]. Strip casting works by pouring the melt onto a spinning water-cooled copper wheel, allowing for rapid solidification at cooling rates of around 10
4 K/s and producing long, thin flakes of the alloy. The wheel rotation speed is directly proportional to the cooling rate of the alloy but inversely proportional to the thickness of the flake it produces. The flake thickness of the alloy is typically between 100 and 1000 μm compared to ingots, which are typically >10 mm [
13,
17]. Strip cast flakes generally have a microstructure that varies along the direction of solidification across the thickness of the flake. Where the alloy initially contacts the wheel, a nucleation zone of rapidly solidified alloy forms. From this zone, directional columnar growth begins, consisting of fine needle-like grains with widths of only a few microns. As they grow along the direction of cooling towards the free side, eventually the rate of cooling is not sufficient to sustain their directional growth, and they lose their orientation, with their tips acting as nucleation sites for more randomly oriented grains. These grains are still relatively fine; however, they coarsen towards the free side of the strip. Long, thin flakes could also give a significant advantage during crushing and milling, potentially requiring lower energy techniques to reduce particle size for magnet manufacture due to the highly brittle nature of the flakes.
Strip casting of Sm
2TM
17 has only been explored in a few published papers, with only one paper seeking to obtain the 1:7H phase directly. Liu et al. found that casting at a wheel speed of 5 m/s produced a homogenous 1:7H phase structure with superficial nanoscale grains [
18]. However, this was performed using a melt spinner, which is not truly representative of strip casting. The melt is ejected from the crucible onto the wheel using high pressure gas, giving cooling rates up to 10
6 K/s and producing ribbons that are typically 20–50 μm thick, compared to strip casting, where the alloy is gravity fed onto the wheel, leading to higher thicknesses. This work presented by Liu et al. led to non-directional growth as the cooling rate was too high to allow for columnar, directional growth along a cooling gradient, which is synonymous with strip casting [
18]. Meng et al. also investigated strip casting of Sm(CoFeCuZr)
z alloys for magnet manufacture but did not detail the strip casting method or parameters used [
19]. They found that the fine grains developed formed the majority of the strip cast alloy and were very detrimental to the remanence and energy product of sintered magnets. When the strip cast alloy was milled, the grain size in these regions was smaller than the average particle size, which led to polycrystalline particles with poor alignment in the final magnet. Therefore, an alloy consisting of columnar grains would likely be more beneficial to magnet production than the fine, equiaxed grains produced in this work [
19]. Yang et al. [
20] investigated the effect of strip casting wheel speed as a part of a larger study on magnet manufacture. They found that slower wheel speeds were more conducive to forming a microstructure suitable for magnet manufacture; however, they were not aiming for direct production of the 1:7H phase and were completing the entire manufacturing route as would be used for cast ingots containing heavy elemental segregation. Whilst this showed very promising results from a magnetic point of view, it did not focus on direct production of the metastable 1:7H phase [
20]. Yang et al. [
21] investigated hydrogen decrepitation of strip cast alloys similar to those presented in [
20]. However, the micrograph showing the full cross-section of the alloy showed a rough surface that solidified in contact with the wheel, with a small nucleation zone and predominantly non-directional or equiaxed growth. There was very little evidence of the characteristic columnar growth associated with strip casting. Zheng et al. [
22] also investigated the hydrogen absorption of strip cast Sm
2TM
17-type alloys; however, in this work, there was no attempt to optimise the strip casting process. The authors produced ~300 μm thick flakes with some columnar growth, although there was a large amount of variation in the directionality of the columnar grains. However, they did report up to ~30–32 wt% 1:7H phase in their alloys, which shows great promise for the work presented in this paper [
22].
These studies have shown that non-equilibrium casting has the potential for tailoring the microstructure to avoid the need for long term heat treatment. This project instead aims to investigate the viability of strip casting Sm2TM17-type alloys to circumvent energy intensive homogenisation heat treatment in the production of sintered magnets.
4. Discussion
The as-received cast alloys consist predominantly of the 2:17H phase with small amounts of the 2:17R, 1:5H, and 1:7H phases, and the microstructure demonstrated significant elemental segregation as a result of slow cooling in a water-cooled mould. The macrosegregation of Fe and Co to the dark phase and Cu and Sm to the light phase would require a homogenisation treatment to be removed before further processing into a magnet. For all strip cast alloys, however, the phase distribution changed substantially to consist predominantly of the metastable 1:7H phase with small amounts of 1:5H phase and trace 2:17R phase. Due to the substantially increased cooling rate, the microstructure typically consisted of a nucleation zone where the melt started to solidify in contact with the wheel and columnar growth along the thermal gradient from the wheel side of the flake towards the free side, similar to that demonstrated by Meng et al. [
19] and Yang et al. [
20]. However, in a number of flakes, there were regions where the cooling rate was not sufficient to maintain columnar growth, resulting in non-directional growth towards the free side of the flake and elemental segregation similar to that observed in the as-received cast alloy.
It was expected that as wheel speed increased, the flake thickness would be reduced as the molten alloy was pulled onto the wheel more quickly, which should result in thinner flakes consisting of columnar grains and the absence of non-directional growth. However, the faster wheel speed of 3.0 m/s produced a strip cast alloy with minimal columnar growth and predominantly non-directional growth and elemental segregation. Further analysis of
Figure 4 showed that the contact surface between the wheel and the molten alloy was inconsistent as turbulence was introduced due to the high wheel speed. The sample that produced the most ideal microstructure was actually produced at the lowest wheel speed of 1.1 m/s in the thinnest flakes from the batch. The flake thickness was 252 μm, demonstrating a uniform microstructure consisting of a nucleation zone 16 μm thick, a columnar region of 222–235 μm, and non-directional growth of 3–25 μm. In terms of maximum columnar growth region, the thinnest flakes produced with a wheel speed of 2.1 m/s were 300 μm thick, with a 19 μm nucleation zone, 225–251 μm columnar region, and 24–46 μm non-directional growth. This would suggest that the maximum flake thickness should be 270 μm thick to include the nucleation zone and columnar region whilst avoiding non-directional growth where the cooling rate cannot be sustained. Considering the microstructure across all samples suggests that control of the melt flowing onto the wheel and the contact surface between the melt and the wheel is more important for obtaining the desired microstructure than the rotational speed of the wheel. This was further highlighted by the microstructural anomalies presented in
Figure 5, where turbulent flow led to inconsistent flake thickness and dual-microstructures within a flake. In a commercial system where >300 kg of alloy is produced in a single casting, small amounts of alloy with undesirable microstructures are unlikely to be detected and will be mixed in with the optimised material. However, in pilot-scale strip casters, there is more likely to be differential contact and cooling within a batch due to the low amount of alloy in the system (<5 kg) and sampling of the flakes for analysis after casting, resulting in a higher proportion of undesirable alloy. This has not been identified in most literature studies, as they utilise melt spinners that use pressurised gas to consistently eject the material onto the wheel rather than gravity feed through a series of tundishes. This is therefore not representative of commercial strip casting systems.
The XRD analysis presented in
Figure 6 shows that the cooling rate was largely sufficient to form 1:7H directly during strip casting, although evidence of the remaining minority phases highlights that the casting parameters were not quite ideal as some heat treatment would be required to develop the 1:7H metastable phase throughout the entire structure.
Figure 7 showed a difference in key peak intensities between the wheel side of the flake and the free side, which confirmed that the change in microstructure from the nucleation zone to the columnar region to non-directional growth is accompanied by a significant change in phase balance between the desired and detrimental phases. This was confirmed by the SEM and EDX analyses presented in
Figure 8. Overall, this would suggest that further optimisation of the tundish design to better control alloy flow onto the wheel would likely result in the desired columnar microstructure and phase balance of 1:7H and 1:5H without non-directional growth or unwanted minority phases. Alternatively, increasing the surface roughness of the wheel through shot blasting may increase adhesion between the wheel and the melt and result in more consistent cooling throughout the alloy.
When comparing these findings to those of Liu et al. [
16] and Liu et al. [
18], the flakes in this work show lower homogeneity in terms of phase balance and microstructure. Liu found that their strip cast ribbons contained the majority 1:7H phase with small traces of 1:5H, whereas this work found the majority 1:7H phase with fractions of 1:5H, 2:17H, and 2:17R. As Liu used the melt spinning technique to replicate strip casting, it is very likely that uniform contact was observed between the melt and the wheel as a result of the force applied by pressurised gas. This means that the cooling rate will be much more uniform than in gravity fed strip casting, as presented in this paper. However, homogeneity in the cooling rate means that the cross-sectional microstructure presented non-directional growth rather than columnar growth [
16,
18]. The fine-grained strip cast ribbons also led to lower remanence in fully processed magnets as it was not possible to produce single crystal particles by milling, which may not be a problem with the strip cast alloys presented in this work. Meng et al. [
19] demonstrated strip cast flakes that were 575 μm thick, where fine grains dominated 445 μm of the thickness and columnar grains the remaining 130 μm. This also led to a reduction in the magnetic properties of processed magnets compared to ingot starting materials, which was attributed to the high proportion of fine-grained material leading to polycrystalline particles after milling. As the strip casting parameters were not given and limited microstructural analysis was presented, no direct comparisons can be made regarding the optimisation of the strip casting process. Yang et al. and Yang et al. [
20,
21] demonstrated very similar findings to Meng et al. [
19], showing that slower wheel speeds in strip casting led to better magnetic properties; however, as Meng et al., were utilising the full manufacturing route for magnet processing and were not aiming to circumvent the long heat treatment as would be the aim of this work. Likewise, direct comparison with Zheng et al. [
22] is limited as they did not optimise the strip casting process; however, they obtained a similar combination of phases of the alloy but were not able to demonstrate directional columnar growth from the wheel side to the free side of the flake as presented in this work.