Recycling of Rare Earth Elements

1 Recycling of rare earth elements

The development of efficient and economic strategies for RE recycling depends on different factors, for example, commodity prices, location and size of deposits, costs for disposal as well as hazard potential. Due to great worldwide deposits of RE containing ores and low commodity prices, only a few recycling processes are currently able to work on an industrial scale. However, the majority focuses only on production wastes or end-of-life (EoL) products with especially high RE contents [1–4]. Recycling of Fe₁₄Nd₂B and SmCo magnets from production waste has been state of the art for a couple of decades. Currently 10–30 % of the starting alloys still accumulate as different wastes during manufacturing, such as slurry, powder or chips [2]. Most of these chiefly pyrometallugical processes remove the plastic sockets, then grind the waste and blend the resulting powder with new starting alloys [4–6]. This strategy avoids an expensive and costly reduction of RE elements.

Regarding EoL products fluorescent lamp scraps and NiMH batteries currently serve as sources for efficient RE recycling. Although spent magnets often contain more than 30 wt.% of REM as well as there are many scientific articles dealing with this topic, yet none of the proposed strategies or processes has met the requirements for an upscale to industrial production [2, 3, 7–11]. The reasons are versatile: (i) Removing the small magnets from their sockets and housings in, for example, hard drives or electric motors often proves to be very complicated [7–9, 12]. (ii) Besides prices for RE oxides decreased in part by more than 98 % since 2011. For instance, CeO₂ dropped from 100 US$/kg [13] to 1.59 US$/kg [14] within the last 5 years. (iii) Since EoL products contain REs mostly in oxidized state, either because of their application (fluorescent powders, polishing agents, ceramics, glass) or due to corrosion (magnets, alloys, batteries), pyrometallurgical treatment must not be used for recycling. The same applies for impurities, accompanying elements (e.g. Ni, Co, Zn) and housings coming along with EoL waste. Disadvantageously, the expensive RE reduction using fused salt electrolysis cannot be avoided in a hydrometallurgical process.

Nevertheless, Umicore and Solvay Rare Earths have already developed a wet chemical EoL recycling for fluorescent lamps (2008) and NiMH batteries (2011) on an industrial scale [7, 15]. By now Osram has patented a process for RE recovery from fluorescent lamp scraps, too [16]. In case of permanent magnets the MORE project (2011–14), a joint project of Siemens, Mercedes, Umicore and several German scientific institutes and universities, examined three different strategies to recycle magnets from spent electric motors. Whereas both mechanical strategies, namely the reuse of undamaged magnets as well as blending the grinded magnets with starting alloys, missed the requirements for new permanent magnets, the third hydrometallurgical route yielded the REs as sellable oxides [7–9]. But neither the MORE project nor one of the wet chemical processes developed in scientific literature has succeeded in reducing the gap between laboratory bench and industrial process so far [2, 3, 7–11]. Against this background, unconventional recycling processes get more and more attractive. One approach, the so-called solid-state chlorination, shows numerous amazing effects for RE recycling and combines high yields with an extremely low demand for chemicals combined with minor costs for disposal. Within the SepSELSA project this method was adjusted and optimized for RE recovery from fluorescent lamp scraps [17–19]. Unexpected trends and side reactions required many different methods of analysis. In this context the optimization based on statistical designs, as they are used for development of analytic techniques, proved most helpful.

2 Recycling from fluorescent lamp scraps

Approximately 250–300 t of Hg-contaminated fluorescent lamp scraps with average RE contents of about 10 wt.% are deposed downhole in Germany every year [16]. Prior to a wet chemical recovery such as Solvay Rare Earth runs in La Rochelle, all the quicksilver has to be removed by distillation at >357 °C. Additionally to EoL lamp scraps there are Hg-free production wastes with RE contents up to 22 wt.%. In 2009, when the regulation EG 245 came into force, higher standards for light quality were applied so that the older and RE-free halophosphate fluorescent materials couldn’t achieve anymore [20]. As a result the amount of three-band fluorescent materials and the RE content are going to rise prospectively. In this regard EoL fluorescent lamp scraps and production waste were examined by the SepSELSA^[1] project in the course of process development.

2.1 Starting material

Both EoL scraps and production wastes always contain a mixture of different fluorescent materials whose composition varies in a wide range, due to joint repository and disposal. Alongside three-band materials and the older halophosphate these wastes embody residues of glass and metallic sockets, too. In general, the particular fluorescent materials are composed of a host lattice and one or more doped activator elements. RE elements serve mainly as activators, except for yttrium which forms a host lattice, too (Y₂O₃:Eu³⁺). Overall fluorescent lamp scraps may consist of up to 19 different elements. Table 1 summarizes the most common components of EoL wastes.

During the SepSELSA project the first batch of waste came directly from lamp production, whereas the second batch consisted of EoL lamp scraps (Table 2). To determine the composition of both batches X-ray fluorescent spectroscopy (XFS) was used after removing the quicksilver by heating up to 500 °C in a stream of nitrogen gas and maintaining this temperature for 4 h. Yet an exact quantification of particular fluorescent materials remained impossible, because some elements are part of more than one fluorescent material at the same time. Therefore, SEM-EDX spectroscopy was used to correlate each particle by its composition with the respective fluorescent material (Figure 1).

Figure 1:

Secondary electron images (SEI) of the used EoL fluorescent lamp scraps provided by SEM.

By developing an efficient recycling process all efforts were concentrated on yttrium and europium representing the main part of the six RE elements with a share of 70–80 %. Yttrium solely occurs in the red-light-emitting fluorescent material together with Eu³⁺ as activator. Regarding all elements the Y concentration amounted to 5.4 wt.% for the EoL batch and 15.7 wt.% for the production waste.

In contrast to yttrium Eu²⁺ serves furthermore as activator in blue fluorescent materials such as BaMgAl₁₀O₁₇:Eu²⁺ (BAM) and (Ca, Sr, Ba)₅(PO₄)₃Cl:Eu²⁺ (ScAp). The remaining REs lanthanum, cerium, gadolinium and terbium are components of blue and green emitting materials, but with quite low concentrations of about 0.3–1.8 wt.%. Altogether the production waste contains almost three times as much RE elements (21.7 wt.%) as the EoL lamp scraps. As expected, the content of outdated halophosphate is two to three times higher in EoL waste, as it’s proved by the respective concentrations of calcium and phosphor. The same is applied for silicon coming from glass breakage produced during cutting the metal sockets.

2.2 Solid-state chlorination

In contrast to acid leaching gaseous HCl_(g) reacts with the fluorescent lamp scraps during solid-state chlorination. Initially both batches are mixed with NH₄Cl and heated above 250 °C. At 184 °C NH₄Cl changes modification, and thermal decomposition sets in at T > 220 °C releasing gaseous NH_3(g) and HCl_(g).

Change of modification at T = 184.3 °C

Thermal decomposition at T ≥ 220 °C

Reaction with HCl_(g) at 250 < T < 400 °C

The last reaction yields water-soluble and nonvolatile metal chlorides. Volatile chlorides (FeCl₃ at 120 °C) and metal chlorides with low melting temperatures (ZnCl₂ at 318 °C) were not present in the fluorescent lamp scraps. They interfere with the gas-phase chlorination by forming liquid films on the particle surfaces, thus shielding them from further chlorination. In addition, the thermal decomposition is reversible, which allows recovering unreacted NH₄Cl in a subsequent step by allowing the gaseous phase to cool down. Yet compared to conventional acid leaching, the scrap constituent exhibits different behaviour towards gaseous HCl. Whereas Y₂O₃:Eu³⁺ reacts very fast to YCl₃ and EuCl₃, all of the phosphate, aluminate and borate fluorescent materials remain almost unaltered. By leaching the chlorinated solid with 100 g of distilled water all formed metal chlorides dissolve. For conducting this solid-state chlorination several suitable reactor types are at hand. Initially, in exploratory experiments an array of three sublimation reactors served to determine relevant influence factors and their respective ranges for the upcoming simultaneous optimization. As depicted in Figure 2, the mixture of NH₄Cl and fluorescent lamp scraps is placed on the bottom of a quartz tube reactor. A nitrogen atmosphere is required in order to prevent water-insoluble metal oxides from forming. By directing the gas flow next to a cooling finger unreacted NH₄Cl separates from the gaseous phase by depositing on the surface.

Figure 2:

Sublimation reactor used for all attempts of orientation as well as the sequential pre-optimization.

For optimizing the reaction on the basis of statistical designs the reactor type was changed to a rotary kiln operating in batch mode. Here, the mixture of NH₄Cl and fluorescent scraps was placed between two tapers in the horizontal quartz tube (Figure 3). Compared to conventional fixed or fluidized bed reactors as well as sublimation reactors the rotary kiln is able to operate in continuous mode, from which a significantly better performance compared with batch-type reactors is expected.

Figure 3:

Rotary kiln used for simultaneous optimization.

In fact, first, orienting experiments showed great differences between solid-state chlorination and classic leaching with aqueous hydrochloric acid (Figure 4). For comparison, in experiments with EoL lamp scraps as well as with production waste, chlorination was conducted at T = 300 °C for 2 h with an NH₄Cl/fluorescent lamp scrap ratio (AFR) = 2.0 g/g, followed by leaching with distilled water. Acid leaching served as reference and was conducted in 3 M hydrochloric acid at 60 °C for 3 h. In the leachate the concentration of the solid amounted to 15 wt.%.

Figure 4:

Comparison of leaching with hydrochloric acid and solid-statechlorination (SepSELSA project) for production waste (a) and EoL fluorescentlamp scraps (b) [19].

Considering the overall yield for RE elements, there is one advantage of the wet acid leaching over solid-state chlorination: metal chlorides are dissolved from particle surfaces, thus continuously regenerating a particle surface and thus preventing surface passivation. Moreover, HCl_(aq) leaching also mobilizes gadolinium, terbium, lanthanum and cerium, resulting in a slightly higher RE yield of about 5–7 percentage points. However, in view of the low concentrations of these four RE elements the differences remain small. And, these slightly higher overall yields were bought at the price of a considerably lower discrimination power between the major and minor constituents.

Consequently, mobilizing all elements results in an obvious disadvantage when it comes to RE separation. Owing to their highly similar chemical properties REs are difficult to separate, and it often requires more than 100 stages of solvent extraction. Solid-state chlorination overcomes separation problem through its higher selectivity towards the RE phosphors, allowing for selectively separating the major constituents Y and Eu. As depicted in Figure 4 Y–Eu selectivity towards the remaining four RE elements reached 99.6 % (production) and 99.7 % (EoL), respectively. Accordingly, the wet acid approach furnished Y–Eu selectivities, not exceeding 97.2 % for EoL products and 91.2 % for fluorescent lamp scraps from production waste.

The main difference between both methods concerns the gaseous phase leaving the reactor while chlorinating with NH₄Cl. The hot aerosol consists of NH_3(g), H₂O_(g), HCl_(g) and entrained solid particles. Cooling the gaseous phase below 220 °C leads to a quantitative recombination of HCl_(g) and NH_3(g) to NH₄Cl, because of the reversible decomposition reaction. Both NH₄Cl and entrained particles can easily be separated from the gas flow by bag filters and recirculated afterwards. Referring to sublimation reactors this happens on the cooling fingers. In contrast, recovering excess acid from wet acid leaching processes, for example, via diffusion dialysis, is more difficult and always incomplete. After separation of NH₄Cl the remaining gaseous phase consists of H₂O_(g) and NH_3(g). When transferred to a scrubber, NH₃ even yields a sellable by-product, if the resulting ammonia solution meets the requirements for purity.

2.3 Optimization of the solid-state chlorination

2.3.1 Choice of optimization method

There are two fundamental approaches for optimization: (a) sequential and (b) simultaneous.

Most commonly the optimum is determined by varying one influence factor after another while keeping all remaining reaction conditions constant. This one-factor-at-the-time strategy requires less experiments to obtain an optimum, but massive problems occur, when two or more factors correlate with each other (Figure 5). Moreover, this error may go unnoticed.

Figure 5:

Response surfaces for sequential optimization without (a) and with correlation (b) between both factors (modified according to Ref. [21]).

Without any correlation the optimum at 100 % is achieved accurately (Figure 5(A)). But if one factor cannot be varied independently without changing the second factor as well, the depicted one-factor-at-the-time strategy yields only an ostensible optimum instead (Figure 5(B)) [21]. Especially in complex reaction systems avoiding all correlations is often impossible. So a change of optimization methods is absolutely necessary. In this regard, the sequential simplex method using a geometric algorithm suits to determine the global optimum, but with a previously unknown count of experiments. However, the simplex method relies on prompt analyses, because every next experiment within this algorithm requires an evaluation of the previous one. If there is no prompt analysis at hand, statistical designs provide an alternative. These designs alter all factors at the same time on the basis of a symmetric plan. The evaluation takes place after conducting all experiments [21, 22]. By generating a model equation from the results covering all influences, it becomes possible to estimate the target value (e.g. the yield) at every point of the examined area. For solid-state chlorination a three-stage Box–Behnken plan was applied for optimization involving a polynomial of the second degree as model equation:

(4)y=b0+∑i=1kbixi+∑1≤i≤jkbijxixj+∑i=1kbiixi2

with	y	–	target (Y–Eu yield in the permeate)
	xi	–	factors (temperature, time, AFR)
	k	–	number of factors (here: 3)
	b0	–	ordinate intercept (here: 0)
	bi, bij, bii	–	regression parameters covering linear and squared influences as well as correlations between different factors

Determining all regression parameters is a mathematical problem of multiple linear regressions. Already the number of three factors renders the required calculations rather laborios. They are therefore commonly conducted by evaluation software such as Statgraphics (Statpoint Technologies, Inc.). By knowing the model equation the optimum can be calculated within the examined area. Despite the chosen optimization method three preconditions must be fulfilled: (i) the experiments have to be repeatable with a preferably low distribution of measurement values, (ii) all factors must have an influence on the target value and (iii) have to be varied in an appropriate range, where the influence exceeds the experimental error. Therefore, attempts of orientation are necessarily conducted during a sequential pre-optimization of the solid-state chlorination.

2.3.2 Attempts of orientation

The overall yield of yttrium and europium calculated from their concentrations in the permeate was chosen as target value (simplified: Y–Eu yield). The percentage of each element dissolved in the permeate and analysed by ICP-AES refers directly to the metal chlorides produced by solid-state chlorination, because all original components of the fluorescent lamp scraps are insoluble in water. At first AFR and temperature were varied between 1.0 and 4.0 g/g and 250–400 °C, respectively.

The experiments with altered temperatures were conducted at an AFR of 3.0 g/g and with a constant reaction time of 3 h. Starting at 250 °C Y–Eu yield increased constantly achieving maximum at 300 °C but drops at higher temperatures (Figure 6). This trend is dominated by yttrium as the major component amongst REs, whereas the yields of the residual RE elements either remain unchanged (B) or even slightly increase with temperature (A).

Figure 6:

Temperature dependency of solid-state chlorination for production waste (a) and EoL fluorescent lamp scraps (b) at a constantAFR of 3.0.

The reason for this relates to the very different composition of each fluorescent material. While HCl_(g) reacts preferably with oxidic Y₂O₃:Eu³⁺ at any temperature the more acid-resistant lanthanum, cerium, gadolinium and terbium containing aluminate/borate/phosphate fluorescent materials only react at higher temperatures to a limited extent. Similar behaviour was observed for the comparative leaching with hydrochloric acid (Figure 4). Of these four elements gadolinium produced the highest yield with 9.3 % at 400 °C. Furthermore, temperatures beyond 300 °C lead to an accelerated decomposition of NH₄Cl resulting in a fast-expanding gaseous phase that transports unreacted HCl_(g) out of the reaction zone. As a result, Y yield decreased at higher temperatures. In contrast to yttrium, Eu yields ranged about 71 % at >300 °C (Figure 6), because europium is a component of both the more acid-labile Y₂O₃:Eu³⁺ as well as the more acid-resistant fluorescent materials (Ca, Sr, Ba)₅(PO₄)₃Cl:Eu²⁺ and BaMgAl₁₀O₁₇:Eu²⁺. Therefore, temperatures from 275 °C to 325 °C are preferred for simultaneous optimization, wherein highest Y–Eu yields are expected.

As second influence on solid-state chlorination the AFR was examined at 350 °C by varying the ratio between 1.0 and 4.0 g/g. Both batches of fluorescent lamp scraps showed the same trend. Initially Y yield rose till AFR reached 2.0 g/g and decreased afterwards as the NH₄Cl amount accumulates further (Figure 7). This trend turns against all expectations whereby Y yields should rise while more HCl_(g) is available for chlorination. Eu yields always stay above 70 %, but there is no real trend to observe again. Concerning lanthanum till terbium, yields follow expectations and increase slightly at higher AFR to a maximum of 4.1 % for gadolinium at 4.0 g/g (Figure 7(A)).

Figure 7:

Dependency of RE yields as function of AFR for production waste (a)and EoL fluorescent lamp scraps (b) at 350 °C.

Due to this unusual trend of both fluorescent lamp scraps a side reaction was predicted, influencing Y yields at high NH₄Cl amounts in a negative way. A count of 19 different elements was made proving a real challenge. For clarification, solid filtration residues of the aqueous leachate were analysed via XFS to identify elements behaving contrary to yttrium (Figure 8). Because filtration residues contain the elements remaining unchlorinated, Y amount of substance shows a reverse trend and increases with rising AFR (Figure 8). Only two of the examined elements follow a distinct trend contrary to yttrium: calcium and barium. The first refers to outdated halophosphate, while barium is a component of the blue emitting BaMgAl₁₀O₁₇:Eu²⁺. As XFS was measured without standards by using an internal calibration the results had to be verified by other means.

Figure 8:

Amount of substance of selected elements in the retentate atdifferent AFR (EoL fluorescent lamp scraps chlorinated at 350 °C).

Therefore, solid-state chlorination was conducted with model substances to avoid potential side effects of other elements. At first, equal masses of Y₂O₃ and Ca₃(PO₄)₂ were mixed with NH₄Cl in different ratios and chlorinated at 350 °C (Figure 9). While Ca yield increases constantly till 43 % Y yield meets an optimum at 78 % (NH₄Cl/Y₂O₃ ratio 2.0 g/g) and thereafter drops stepwise to 43 % at 10 g/g. Actually Ca²⁺ doesn’t drop Y yields by itself, but along with Ca²⁺ mobilized phosphate ions precipitate dissolved Y³⁺ as YPO₄. Since solubility of YPO₄ (4.2 × 10⁻¹³ mol/L at 25 °C [23]) is much lower than for Ca₃(PO₄)₂ (1.1 × 10⁻⁷ mol/L at 25 °C [24]) precipitation of Y³⁺ increases with rising amounts of calcium entering the solution.

Figure 9:

Solid-state chlorination of (a) Y₂O₃/Ca₃(PO₄)₂ and (b) Y₂O₃/BaMgAl₁₀O₁₇:Eu²⁺at 350 °C and different NH₄Cl/Y₂O₃ ratios.

Repeating these experiments with BaMgAl₁₀O₁₇:Eu²⁺ instead of Ca₃(PO₄)₂ showed no interaction with yttrium. At 4.0 g/g Y yield achieved approximately 100 %, whereas barium remains almost unreacted with yields <0.5 % (Figure 9). Barium had no influence on Y yields.

Concerning simultaneous optimization the global optimum is estimated for an AFR from 1.0 to 3.0 g/g, because at 2.0 g/g both fluorescent lamp scraps achieved their highest Y yields.

But before commencing further optimization, experimental procedure has to prove repeatability. The less distribution caused by experimental error the more accurate information can be gathered about influences of factors and their correlations. Insufficient repeatability of experiments heavily reduces the information value of statistical designs. Thus, at least one experiment of each design is a three-fold determination. During pre-optimization solid-state chlorination was conducted at the estimated pre-optimum of 300 °C and 2.0 g/g twice. For example, for production waste the gap between the Y–Eu yields was quite low while both experiments achieved the highest average yields by then (90.3 % and 89.9 %). The particular yields of RE elements are depicted in Figure 4. Conclusively by proving repeatability and determining the proper range for the AFR all preconditions for optimization are fulfilled.

2.3.3 Simultaneous optimization

On the basis of the former sequential pre-optimization the target value (Y–Eu yield) and the ranges of all factors were defined. Suitable statistical designs for three factors (AFR, temperature and time) are central composite design, complete factorial design or the Box–Behnken plan. If the supposed optimum isn’t located at the borders of the examined area Box–Behnken plan needs the fewest count of experiments to reliably determine the optimum. Since that’s the case simultaneous optimization was conducted for production waste via 3³-Box–Behnken plan (Figure 10).

Figure 10:

Statistical design used for optimizing the solid-state chlorination of fluorescent lamp scraps (production waste) showing all experiments ().

Such a plan of three factors and three stages to even examine non-linear dependencies needs 15 experiments with a three-fold determination of the centre. Time, as third factor, was altered from 20 to 150 min, taking into account that the rotary kiln heats up with ∼ 10 K/min. Lower times for chlorination would distort the results, since chlorination already starts during heating phase as soon as temperature rises above 220 °C.

Finishing all experiments, evaluation commences by looking at the resulting Pareto diagram (Figure 11). The diagram depicts all sized influences and compares them with the experimental error (black line). Where the bars cross the black line, there is a significant influence on Y–Eu yield.

Figure 11:

Pareto diagram with all linear (A, B, C), square (AA, BB, CC) and cross (AB, AC, BC) effects.

Remarkably, time shows no direct influence. Only the cross effect BC, where time and temperature correlate, has a small significant impact on Y–Eu yield within the examined range of 20–150 min. This indicates a very fast chlorination being almost finished at 20 min, especially when compared to conventional leaching with hydrochloric acid taking 3 h for accomplishment. Finally, the model equation (5) is formed by removing nonsignificant effects from polynomial (4) and calculating the regression parameters:

(5)y=−912,92−23,80⋅A+1,07⋅B+6,56⋅C−8,60⋅A2+0,23⋅AC−3,9⋅10−3⋅−11,4⋅10−3⋅C2

where y is the target value (Y–Eu yield); A is the mass of NH₄Cl in g (per g fluorescent lamp scraps); B is the time for chlorination in min; and C is the temperature in °C.

The estimated optimum at 86.9 % ± 3.3 % was predicted for 312 °C, 20 min and an AFR of 2.76 g/g. Henceforth, a double-fold determination proved the optimum (90.2 % and 87.0 %). Alongside the missing influence of time additional positive and negative effects occurred. First of all, the five-times-larger reactor volume of the rotary kiln turned out to be disadvantageous compared to sublimation reactors. Due to spacious distribution of the gaseous phase greater amounts of HCl_(g) and NH_3(g) recombined and deposited outside the heating zone and lowered Y–Eu yields by 5–10 percentage points. This disadvantage remains valid as long as the rotary kiln runs in batch processing, where actual chlorination only takes place within a small part of the heating zone (Fig. 3). However, applying a rotary kiln offers new options for temperature control. A closer look at temperature dependencies reveals that local and global optima always require temperatures below the actual decomposition temperature of NH₄Cl (338 °C).

Even the proved global optimum at 312 °C reveals a difference of 26 °C. Under this reaction condition NH₄Cl only decomposes partially and forms a dynamic equilibrium between gaseous phase and solid NH₄Cl. At least two contrary effects must influence Y–Eu yield to describe this trend. Higher temperature levels generally are in favour of Y–Eu yield; however, they prove disadvantageous as soon as the respective optimum has been exceeded (dark surface in Figure 12). The reason for this trend must be related to the thermodynamics of decomposition. According to eq. (1) every mole of NH₄Cl produces 2 moles of gas. So 54 g of NH₄Cl decomposes to 102 L of gaseous products at 350 °C. The faster the gaseous phase expands, the more unreacted HCl_(g) is driven out of the reaction zone. On the contrary, improving Y–Eu yields requires a reduction of the gas phase volume. However, reducing temperature only poorly serves to decrease the molar volume. Truly, cooling from 350 °C to 25 °C halves the molar volume from ∼ 51 L/mol to 24.4 L/mol. For NH₄Cl this trend overlaps with the decomposition equilibrium, reducing the gas volume massively, when temperature drops below 338 °C (Figure 13). This way it becomes feasible to almost halve the gas volume by simply reducing temperature from 350 °C to 312 °C (optimum) instead. Without the decomposition equilibrium this reduction would hardly achieve 6 % between 312 °C and 350 °C. Thus, endothermic decomposition on the one hand and expansion of gas volume on the other are the two contrary trends greatly affecting temperature dependence, which shift the optimum to quite low temperatures. Although lower temperatures generally demand an extended reaction time (cross effect BC), Y–Eu yield only showed a slight time dependence between 20 and 150 min. Due to the batch operation shorter reaction times <20 min cannot be examined with this rotary kiln. Therefore, continuous operation is required to reduce influences on Y–Eu yield by preheating.

Figure 12:

Y–Eu yield depending on temperature and AFR at a constant reaction time of 20 min.

Figure 13:

Comparative reduction of NH₃/HCl gaseous phase with decreasing temperature starting at 350 °C (=100 %).

2.4 Recycling process

Subsequent steps following solid-state chlorination and aqueous leaching are almost identical to wet chemical processes, yet with two exceptions. There is no strongly acidic leachate, further processing of which would require considerable amounts of NaOH. The reason is that pH values of the obtained leachate only vary between 3 and 5. In view of the expenses for the cost-intensive NaOH, it is evident that the revenues of the overall process will be rather small, if ever. The second fact concerns purity of the liquid concentrates, as solid-state chlorination separates Y/Eu from La/Ce/Gd/Tb with selectivities >99.7 %. This simplifies the laborious RE separation to separating yttrium from europium by cementation. Further purification was achieved by adding small amounts of H₂SO₄ (till pH = 3) to precipitate Ca²⁺ as CaSO₄. Potentially excess sulphate ions were subsequently removed by adding Ba²⁺. The remaining solution was funnelled through a fix bed filled with Zn granules to reduce Eu³⁺ to Eu²⁺.

Thereafter, Eu²⁺ was easily precipitated as EuSO₄ analogous to alkaline earth metals. Up to 95 % of Eu²⁺ were separated per pass. Yet EuSO₄ is no sellable product and requires further treatment, for example, treatment with NaOH_(aq) and drying to yield Eu₂O₃. When Eu³⁺ had been removed, dissolved Y³⁺ was precipitated conventionally as oxalate. Accordingly, the process yields two different concentrates for yttrium and europium while lanthanum, cerium, gadolinium and terbium remain within the solid residue. Of course, recovery of lanthanum and cerium by leaching with concentrated acids like HNO₃ is possible but uneconomic due to low commodity prices for rare earth metals. Nevertheless, current experiments provide another option simplifying RE separation. It makes use of the high magnetic susceptibilities of gadolinium and terbium. In fact, applying magnetic fields at different temperatures not only allows for easily separating Gd³⁺ and Tb³⁺ from La³⁺ and Ce³⁺. The method also efficiently separates Gd³⁺ from Tb³⁺ – in aqueous solution [25]. Since prices for La₂O₃ and CeO₂ are still very low producing respective concentrates of lanthanum and cerium is not economic at present. Figure 14 depicts how yttrium and europium may be recovered economically. Solid residues containing residual REs are stockpiled until economic recycling is feasible.

Figure 14:

Scheme of the recycling process for fluorescent lamp scraps.

Treatment of the gaseous phase provides another feature affecting overall process economy: unreacted NH₄Cl is recovered by cooling, and NH_3(g) is removed by scrubbing with water. If NH₄Cl separation is complete, an ammonia solution is obtained, the chloride concentration of which is <250 mg/L. This way the process yields a market-established by-product, which to some extent contributes to compensate for production costs [26]. Compared to conventional leaching, recycling excess acid from the digestion medium is always incomplete and requires additional separation steps.

Concerning the number of process steps, both strategies developed within the SepSELSA project require fewer process steps to get the same raw concentrates compared with what has been published by Solvay and Osram. Solid-state chlorination even reduces the process effort to a half (Figure 15).

Figure 15:

Comparison of solid-state chlorination with industrial-applied recycling processes for fluorescent lamp scraps [16, 18, 19, 27, 28].

2.5 Summary

Solid-state chlorination provides a promising alternative for mobilizing RE elements from fluorescent lamp scraps. NH₄Cl as source for HCl_(g) offers many economic advantages compared to all applied wet chemical processes. Although NH₄Cl (∼110 €/t) roughly costs as much as 35 % hydrochloric acid (∼135 €/t), as a solid it contains almost twice as much the molar amount HCl per weight unit (68 wt.%) [29]. Due to high selectivities, less HCl is consumed, and unreacted HCl_(g) can be easily recovered from the gas stream leaving the reactor. Without consideration of potential revenues from selling the NH₃ solution as by-product, costs for chemicals are thus reduced by at least 50 %. Because there is no strongly acidic digestion medium, costs for downstream processing and waste disposal are lower, too. Depending on heavy metal content, disposing of a strongly acidic solution may cost up to 400 €/t. Waste water with low acidity, such as it results from solid-state chlorination cuts those costs to approx. a third [30]. Furthermore, chlorination profits from high selectivity yielding two pre-separated concentrates. Although cementation of Eu³⁺ is feasible in wet chemical processes, too, pH value has to be adjusted prior to Eu winning, thus incurring additional costs for NaOH. One has to realize that increasing pH from 0.7 to 3.0 equals neutralizing 99.5 % of the acid. Furthermore, hydrometallurgical treatment yields an yttrium-rich RE oxide of lower economic value than the pre-separated Y₂O₃ provided by solid-state chlorination.

Yet a general statement on process economy cannot be made without taking into account the commodity prices, location, desired production output, personnel requirement, etc. Within the framework of the SepSELSA project a feasibility study was conducted for Freiberg (Saxony, Germany), where there is a globally acting lamp producer in the near vicinity [30]. Whereas wet chemical treatment of fluorescent lamp scraps went uneconomic between 2014 and 2016 solid-state chlorination remained at least economic for higher concentrated production wastes, although commodity prices have dropped by another 80 % within the 2 years.^[2] At present, the main obstacle for scaling up solid-state chlorination is rotary kiln design, which implies with several different heating zones. Its development and construction are matter of current research activities at Freiberg University of Mining and Technology. Nevertheless, ∼ 25 tons of production waste have successfully been recycled with the SepSELSA process between 2014 and 2016.

3 RE metal recycling from Fe₁₄Nd₂B magnets

Permanent magnets containing RE metals belong to the strongest magnetic alloys currently available with magnetic energy densities up to 450 kJ/m³ [31, 32]. Because of versatile applications in, for example, hard drives, electric motors, wind power stations or MRI scanners the demand for neodymium, dysprosium, praseodymium, samarium and terbium substantially increased within last years. For instance, demand almost doubled from 2006 to 2012 reaching 42.000 t/a. This trend continues as it can be derived from the demand of 53.000 t/a, which has been prognosticated for 2017 [33–35]. Among RE alloys Fe₁₄Nd₂B magnets are the strongest ones and the most commonly used magnetic material. Consequently, a variety of processes have been developed for the treatment of production wastes, most of them pyrometallurgical ones [2–6, 12]. In the case of EoL magnets, it is corrosion, adhesives, plastics and accompanying elements, such as nickel, cobalt and zinc, which prevent the application of these well-established pyrometallurgical routes. As a consequence, hydrometallurgical treatment is almost exclusively preferred for EoL recycling [2, 3, 7–11]. Therefore, solid-state chlorination was regarded as a powerful alternative strategy for magnets, too.

3.1 Starting material

Prior to chemical or mechanical treatment magnet recycling starts with the systematic collection of EoL magnets and their separation from electronic scrap. Particularly the latter still represents a major challenge. It was in 2011 when the first automated process removing magnets from spent hard drives was applied by Hitachi on an industrial scale [12]. Although Fe₁₄Nd₂B magnets contain approximately as much RE (30–35 wt.%) as fluorescent lamp scraps they differ widely in structure. At first magnets are alloys containing RE metals. Because of sintering the starting materials during production, these alloys are not completely homogenous. Consequently, from grinding no distinct particles with homogeneous composition are obtained.

The EoL magnets examined by solid-state chlorination were derived from a rotor of a wind power plant. These 5 × 3 × 2 cm³-sized magnets were initially heated up to 350 °C for demagnetization before they were ground to particle size ≤ 100 μm. The grinding was conducted in two stages: magnets introduced to a horizontal impact crusher yielded small pieces applicable for the grinding in a vibrating cup mill providing a pyrophoric powder. For this reason grinding always has to be conducted under exclusion of oxygen. In SEM pictures the magnet powder appears as sharp-edged particles (Figure 16(A)).

Figure 16:

SEM-EDX analyses of magnet powders (a) before and (b) after solid-state chlorination.

Because of the ignoble character of Fe₁₄Nd₂B alloys they may easily be dissolved in 37 % hydrochloric acid and analysed via ICP-AES afterwards. The results (Table 4) show a quite simple composition. The iron amounts to 65.6 wt.%, and the share of RE elements only consists of neodymium and dysprosium with an overall concentration of 34.1 wt.%. Praseodymium, which serves as cheaper substitute for neodymium, and terbium, which enhances temperature stability analogously to dysprosium, were not found in the samples examined.

3.2 Preliminary tests

Applying solid-state chlorination for magnet recycling also has to meet requirements first, namely the forming of water soluble as well as nonvolatile metal chlorides. Therefore, iron is the main concern, as FeCl₃ already sublimes at T ≥ 120 °C:

Starting from metal alloys the redox potential of HCl_(g) is not high enough to oxidize iron to the third oxidation state (Fe^{3 +}). Instead nonvolatile FeCl₂ is formed. Corroded EoL magnets already containing Fe³⁺ yield small amounts of FeCl₃ that sublime during chlorination and are removed together with unreacted NH₄Cl from the gas phase. As long as these amounts remain small they will not interfere with the solid-state chlorination. In contrast to fluorescent lamp scraps all components but boron react with HCl_(g) to their respective metal chlorides. Iron as the most noble metal and main component defines the reaction conditions. Due to effective clogging of iron and RE particles in the sinter mass, iron needs to be chlorinated almost completely to achieve high yields for both RE elements. Metal chlorides form voluminous passivation layers on the particle surfaces, shielding them from further chlorination (Figure 16(B)). Owing to the absence of water, the metal chloride layer cannot be removed by solvation, like this would occur in the aqueous phase. It is for this reason, why solid-state chlorination suffers from lower yields. Again, the solution is provided by a rotary kiln, where the magnet powder is subjected to mechanical stress. Alternatively, unreacted magnet powder can be reused in a cycle, or the magnets are ground to even smaller particle sizes. Another important difference affects the leaching step following chlorination. This time the RE metal precipitation is highly pH sensitive, since Fe³⁺ competes for OH⁻ ions, thus causing undesired co-precipitation of mixed hydroxides. Consequently, the pH value needs to be finely adjusted in order to avoid precipitation of iron hydroxide species. Without considering the latter, leaching with distilled water leads to a partial precipitation within 3 h, thus lowering RE yields by >10 percentage points and complicating filtration due to formation of small particles with <2 μm in size (Figure 17(A)).

Figure 17:

H₂O leaching following solid-state chlorination (a) without and (b) with adjusting pH value to 4 by adding hydrochloric acid.

Adjusting the pH value with HCl_(aq) prior to dissolving the metal chlorides increases all yields instantly and reduces the extent of precipitation. The equilibrium is reached after 1 h, while pH value rises slowly to approximately 6 (Figure 17(B)). Fe yield decreases within the first 30 min, whereas yields of neodymium and dysprosium show a contrary trend. The reason is a redox reaction between dissolved Fe²⁺ and undissolved RE metals on the particle surfaces. Analogous to eq. (11) this cementation may enhance RE yield up to 7 percentage points:

From this reaction one can understand why solid-state chlorination does not necessarily have to be complete in order to retrieve the entire RE content of the magnet. Of course this occurs at the expense of Fe yield. In order to avoid any undesired precipitation, pH value has to be kept constant during leaching. Since diluted HCl_(aq) provides no significant buffering capacity at pH 3–4, acetic acid (1 M) and sodium acetate were used as buffer (pH = 3) for leaching the metal chlorides. In principle, adding diluted HCl_(aq) is sufficient to prevent any precipitation, though, but this way requires higher amounts of HCl_(aq), which in addition have to be added continuously, in order to keep the pH value at 3.

3.3 Optimization of the solid-state chlorination

Except for the leaching step simultaneous optimization was conducted in analogy to the fluorescent lamp scraps by using a 3³-Box–Behnken design. The target value was represented by the yield of all REs in the pregnant solution. All factors and temperature range (225–325 °C) were kept constant (Figure 10). Chlorination time was varied between 60 and 240 min. Since 1 g of magnet powder served as starting material the amount of NH₄Cl varied from 1.0 to 3.0 g.

After finishing all 15 experiments, the Pareto diagram was used to surprisingly show only temperature to have a significant influence on RE yield (Figure 18). Although 99 wt.% of the magnets react with HCl_(g), solid-state chlorination appeared to be finished after 60 min so that all effects involving time remain insignificant (C, CC, AC, BC). Yet all influences related to NH₄Cl (A, AA, AB, AC) need a different explanation.

Figure 18:

Pareto diagram with all examined effects influencing solid-state chlorination of Fe₁₄Nd₂B magnets.

For chlorinating iron, neodymium and dysprosium stoichiometrically as depicted in eqs (8–10) 1.0 g of magnet powder requires approximately 1.4 g NH₄Cl. Thus, four experiments were conducted with only 70 % of the stoichiometric NH₄Cl amount. However, the RE yields range between 76 % and 84 %. So at least four experiments gave yields which appear unrealistic, although all centre points showed excellent reproducibility (RE yield at the centre: 83.5 % ± 1.1 %). The predicted optimum of 84.2 % at 289 °C, 161 min and 1.83 g NH₄Cl could be confirmed as well by two-fold determination (83.8 % and 85.3 %). It should be noted that time and NH₄Cl amount can be chosen freely within the estimated area without affecting RE yield significantly (cf. Figure 18). On the basis of a correct choice of factors and ranges the missing influence of NH₄Cl must refer to at least one of two causes: (i) either leaching with the acetate buffer dissolves parts of the magnet powder at pH = 3 and distorts RE yields that way, or (ii) there is a great difference in how the several elements react with HCl_(g). In fact, the latter factor appears to apply for iron and RE metals being not associated with each other and therefore reacting separately. Both options were examined subsequently.

In case RE yields actually profit from adapting the pH value by buffering, the acetate buffer solution must be able to patially dissolve the magnet powder without solid-state chlorination. In order to check this hypothesis non-chlorinated magnet powder was subjected to leaching with acetate buffer as well as diluted HCl_(aq). Both leaching media were investigated at different pH values (Figure 19). Whereas diluted HCl_(aq) did not dissolve the magnets between pH 3 and 5, a substantial fraction was dissolved in 1 M acetic acid buffer: at pH = 3 almost 20 % of iron and dysprosium as well as 24 % of the neodymium were dissolved, respectively (Figure 19(A)). On the other hand diluted HCl_(aq) mobilized less than 0.5 % of the elements at the same pH value. As soon as pH drops below 3, HCl_(aq) contributes significantly to RE yield, too. Although pH value is the same, the concentrations of the acids may explain this different behaviour. While 37 % HCl_(aq) has to be diluted 1:10.000 (1.2 mM HCl) in order to increase pH value to 3, the 1 M buffer solution provides an 800-times-higher concentration of acetic acid. Due to the low concentration of hydrogen chloride, this solution is not suited for dissolving appreciable amounts of iron, neodymium and dysprosium.

Figure 19:

Contribution of the leaching step to RE yield by using (a) a buffer with 1 M acetic acid or (b) diluted hydrochloric acid for pH adjustment.

From these findings there emerged two options for process control with the aim to avoid Fe hydroxides from precipitating: (a) either diluted HCl_(aq) is used for pH adjustment, which results in lower costs for chemicals or (b) acetate buffer contributes to RE yield and reduces henceforth the demand of NH₄Cl. It should be noted, though, that using acetic acid to completely dissolve the magnets is far too expensive, since the price of acetic acid is three times higher than hydrochloric acid (∼ 450 €/t) [29].

For process control reducing the gas volume was most essential. Therefore, acetate buffer was chosen for further optimization. A lower NH₄Cl demand means less volume of the gas phase and consequently more magnet powder can be chlorinated within a certain time frame.

However, although the process becomes more efficient by dividing RE solubilization into two steps, namely solid-state chlorination as initial step and acid-buffered cementation as completing step, process optimization suffers from the increased amounts of parameters. The one-factor-at-the-time strategy is not applicable here, because buffer capacity and NH₄Cl amount cannot be varied independently. Further, in addition to the three factors of the Box–Behnken design buffer capacity has to be taken into account. But the range needs to be adjusted first, as previous experiments were conducted with an excess of 3.0 g acetic acid per gram magnet powder. Preferentially the range for optimization overlaps with the area of greatest changes in leaching behaviour. In order to determine this range, acetic acid concentration of the buffer was reduced stepwise from 1 M to zero. As depicted in Figure 20, a range between 0.025 and 0.25 mol/L was chosen for further optimization.

Figure 20:

Leaching of unchlorinated magnet powder with different buffer concentrations at pH = 3 (- - adjusted range for optimization).

However, the influence of the buffer solution on RE yields does not exclude a partial selectivity during chlorination as additional reason for the missing influence of NH₄Cl. To verify a potential selectivity, buffer concentration was reduced from 1 to 0.15 M. Hereinafter two experiments were conducted under same reaction conditions at 325 °C, where only the NH₄Cl amount was altered. The first experiment was done with 0.5 g NH₄Cl equalling 35 % of the stoichiometric amount. In contrast, an excess of 1.75 g NH₄Cl (125 %) was used for the second experiment (Figure 21). Although leaching with a buffer enhances RE yields again, it becomes obvious that Fe yield changes exclusively and drops by 30 percentage points between both experiments. The yields of neodymium and dysprosium remain at the same levels. Seemingly NH₄Cl only affects Fe yields.

Figure 21:

Comparison of two chlorination experiments at 325 °C (85 min) conducted with different amounts of NH₄Cl to prove partial selectivity.

These results imply a partial selectivity. Consequently, neodymium and dysprosium are chlorinated first and form a RE-rich layer on the particle surface by their respective metal chlorides. The less ignoble iron reacts much slower with HCl_(aq) so that reduced NH₄Cl amounts affect Fe yield by far stronger than RE yields. It has to be noted, though, that analysing liquid permeates only represents an indirect proof. For conclusive verification the chlorinated magnet powder has to be examined instead.

The mineral liberation analyser (MLA from FEI, USA) combines SEM and EDX with automated measurement systems for quantitative examination of rocks, minerals and synthetic materials. Sample preparation was decisive to prove the supposed partial selectivity. 3 g of the powder samples were mixed with the same volume of graphite to separate agglomerated particles from each other. The resulting graphite/sample mixture was casted with epoxy resin to a cylinder of 25–30 mm in diameter and hardened for 12 h. Subsequent grinding with SiC and polishing with a diamond suspension uncovered a smooth surface showing different cross sections of the particles. Coating this surface with carbon to conduct electrons represents the last step of preparation.

By electrons backscattered on a gold standard the SEM was calibrated on greyscales to enable the distinction between particles and background (e.g. epoxy resins, cracks). On this basis the measuring range of EDX is determined as well. Typical are the measurements of tens to hundreds of thousands of particles leading to millions of spectra per sample. All these spectra are compared with a reference list, formerly created for the material. This list also contains the element concentrations calculated from the spectra. Since magnets do not consist of classical minerals, no pre-recorded spectra were available and comparative spectra were collected at different locations of the sample and added to the reference list instead (Table 5). However, a pinpoint measurement is limited to an accuracy of 0.1–2 μm, due to the pear-shaped excitation. Therefore, mainly mixed spectra are obtained during measurement of small-sized particles or locations where compositions change strongly. Thus, assigning each measuring point to one of the references from the list is not conducted strictly, but with a previously defined tolerance. For the magnet powders, this tolerance was set to 70%. Since every element concentration of the reference list is coloured differently, it becomes possible to assign the respective colour to each measuring point for graphic illustration of the entire examined surface (Figure 22).

Figure 22:

MLA examined samples of (a) untreated as well as (b) under- and (C) over- stoichiometric chlorinated magnet powders.

Table 5:

Composition key used for the analysis of magnets by MLA.

With this in mind, both experiments (Figure 21) were reproduced without leaching. The chlorinated powders were examined with an MLA and compared with the starting material. Since particles were cut and polished during sample preparation, it became possible to analyse the cross section to determine the composition depending on particle radius. Spectra of similar composition were combined to coloured material phases, representing the heterogeneity of particles as well as the collection of mixed spectra on spots with changing composition (Table 5).

As is well known, during magnet manufacturing different alloys are sintered at 1,550 °C so that magnet powders could not establish a strict homogeneous composition (Figure 22(A)). In accordance with Table 4 these magnets contain 65.6 wt.% iron. The major blue domains show the same relation, as they occupy ∼ 90 % of the surface, whereas the residual 10 % of green-coloured domains predominantly consist of neodymium and dysprosium.

Substoichiometric amounts of NH₄Cl during solid-state chlorination turned the picture upside down (Figure 22(B)). Agglomerates were formed consisting of several small grains. For instance the depicted particle consists of at least 16 blue-coloured grains in which iron content increased compared to the original powder. Now green RE-rich layers occupy the space between all grains and on the particle surface. Some green spots even show Fe contents <1 wt.%. In fact, this proves directly the assumed partial selectivity.

Since iron has a shielding effect it has to be chlorinated almost completely to achieve high RE yields as well. Therefore, over-stoichiometric amounts of NH₄Cl are necessary to disintegrate the formed agglomerates (Figure 22(C)). However, the resulting mechanically instable layer of metal chlorides crumbled during sample preparation. The consequences are vast cracks in the epoxy resin and uneven surfaces of pink colour, which are inaccessible for analysis. Only few grains remained, forming the retentate of the filtration and being introduced to solid-state chlorination another time.

In summary, cases (B) and (C) offer two options for process control. On the one hand, substoichiometric amounts of NH₄Cl already furnish high RE yields ≥ 76 %. Unreacted residues considerably enrich iron. This option should be chosen, if economic reasons require a reduction of NH₄Cl amounts and waiving the recirculation of unreacted residues. This procedure allows for introducing larger amounts of magnet powder to the rotary kiln, thus enhancing reaction performance. On the other hand, excess amounts of NH₄Cl are necessary to enable a quantitative magnet recycling. For these purposes, Fe contents of the retentates need to be equal to or less than the starting material in order to quantitatively chlorinate all metals within a few cycles, typically 2–3 cycles.

Apart from buffer leaching influencing RE yields, the range of NH₄Cl amounts also has to be expanded to cover both scenarios. However, further optimization with adjusted ranges and factors is currently continued in the MagnetoRec project^[3], which also comprises the development of the first rotary kiln exclusively designed for solid-state chlorination (kilogram scale). MagnetoRec has started in August 2016.

3.4 Recycling process

Except for some few differences the desired process resembles the one developed for fluorescent lamp scraps (Figure 23). At first EoL magnets need to be demagnetized and ground to particle sizes <100 μm prior to chlorination to their metal chlorides. In the course of aqueous work-up, acetate buffer (pH = 3) prevents hydroxides from precipitating and enhances RE yields by additionally leaching the magnets through cementation. Due to comparatively high pH values, any further adjustment of pH values is redundant [36–38]. The same applies for the recovery of excess acids. For precipitation oxalic acid or hydrofluoric acid can be added to the permeate after filtration with the aim of retrieving RE oxalates or fluorides. Whereas calcining the RE oxalates leads to sellable RE oxides, RE fluorides may serve as starting material for the reduction to RE metals. Nevertheless, both options for precipitation achieve quantitative yields at pH ≥ 1.

Figure 23:

Scheme of the process developed for recycling Fe₁₄Nd₂B magnets.

In general, designing an entire process is not necessarily the aim of developing recycling strategies. Instead extending an industrial process by one or at most a few steps facilitates upscaling. Since at present there is no industrial recycling technology available, EoL recycling has to be implemented in primary RE production from their minerals and ores (Figure 24). This situation may be changed by transferring the solid-state chlorination on an industrial scale. It has to be noted, however, that mixed RE qualities, Nd/Dy, are obtained.

Figure 24:

Applied industrial processes for primary RE production from different minerals [32, 36–38].

Hitherto, all processes finally furnish a mostly acidic RE concentrate, which is used for further purification and RE separation (Figure 24). At this point permeates coming from solid-state chlorination could be fed into primary production. Applying solid-state chlorination for primary production is conceivable, but only for carbonatic bastnäsite (Ce,La,Nd,Y)[(F,OH)|CO₃]. Chlorinating monazite (La,Ce,Nd)[PO₄], xenotime (Y,Yb)[PO₄] or clays leads nowhere, since solid-state chlorination is not suited to mobilize REs from phosphates, borates or aluminates. Even industrial processes need two steps to dissolve RE elements from both phosphatic minerals. Initially either concentrated H₂SO₄ or NaOH_(aq) transfers all RE phosphates to sulphates and hydroxides, respectively. After phosphate ions were removed all RE salts are dissolved by adding acids or simply by diluting with water. For ion adsorbing clays treatment with a solution of NaCl and NH₄Cl already suffices to desorb RE ions.

3.5 Summary

As was shown, the solid-state chlorination of EoL magnets is saving costs for both chemicals and disposal. The chlorination protocol using NH₄Cl provides all the benefits already discussed for fluorescent lamp scraps (cf. 2.5). The partial selectivity of the chlorination step allows for designing processes very flexibly. In case of low commodity prices for rare earth metals, NH₄Cl demand can be reduced by more than 71 %, while concomitantly reactor performance is doubled without reducing RE yields upon magnets passing the reactor the first time (Figure 21). This can be achieved by using an acetate buffer which partially leaches the magnet powder and additionally enhances RE yields by a supporting cementation reaction. So there are three different effects, which allow for mobilizing RE, namely chlorination with HCl_(g), cementation of Fe²⁺ ions and leaching with small amounts of acetic acid. In this regard great expectations are placed on the MagnetoRec project, in which simultaneous optimization will finally combine solid-state chlorination and buffer leaching to recycle EoL magnets in the most efficient way.