*Article* **Li-Distribution in Compounds of the Li2O-MgO-Al2O3-SiO2-CaO System—A First Survey**

#### **Thomas Schirmer 1,\*, Hao Qiu 2, Haojie Li 3, Daniel Goldmann <sup>2</sup> and Michael Fischlschweiger <sup>3</sup>**


**\*** Correspondence: thomas.schirmer@tu-clausthal.de; Tel.: +49-5323-722917

Received: 20 October 2020; Accepted: 30 November 2020; Published: 4 December 2020

**Abstract:** The recovery of critical elements in recycling processes of complex high-tech products is often limited when applying only mechanical separation methods. A possible route is the pyrometallurgical processing that allows transferring of important critical elements into an alloy melt. Chemical rather ignoble elements will report in slag or dust. Valuable ignoble elements such as lithium should be recovered out of that material stream. A novel approach to accomplish this is enrichment in engineered artificial minerals (EnAM). An application with a high potential for resource efficient solutions is the pyrometallurgical processing of Li ion batteries. Starting from comparatively simple slag compositions such as the Li-Al-Si-Ca-O system, the next level of complexity is reached when addingMg, derived from slag builders or other sources. Every additional component will change the distribution of Li between the compounds generated in the slag. Investigations with powder X-Ray diffraction (PXRD) and electron probe microanalysis (EPMA) of solidified melt of the five-compound system Li2O-MgO-Al2O3- SiO2-CaO reveal that Li can occur in various compounds from beginning to the end of the crystallization. Among these compounds are Li1−x(Al1−xSix)O2, Li1−xMgy(Al)(Al3/2y<sup>+</sup>xSi2−x−3/2y)O6, solid solutions of Mg1−(3/2y)Al2+yO4/LiAl5O8 and Ca-alumosilicate (melilite). There are indications of segregation processes of Al-rich and Si(Ca)-rich melts. The experimental results were compared with solidification curves via thermodynamic calculations of the systems MgO-Al2O3 and Li2O-SiO2-Al2O3.

**Keywords:** lithium; thermodynamic modeling; engineered artificial minerals (EnAM); melt experiments; PXRD; EPMA

#### **1. Introduction**

With respect to the development in electromobility as well as to the changes in circular energy systems, Li-ion batteries are of central importance. To safeguard raw material sources especially for critical elements such as Co, Ni and Li as key components of this technology, efficient recycling processes are essential. One of the most important routes to recycle these battery types is the pyrometallurgical processing, which can deal with a broad range of input material. While Co, Ni and other valuable heavy metals such as Cu report into the alloy melt, Li is transferred at least in major amounts into the slag phase of this process.

A resource and energy efficient recovery of Li from the slag could be accomplished, if Li were concentrated in specific Li-rich artificial mineral phases, which could then be separated after crystallization and cooling of the slag by means of mineral processing technologies, generating concentrates for following hydrometallurgical processing.

Previous research has shown that Li can be recovered in the form of the LiAlO2 crystals through flotation from a remaining silicate slag matrix [1]. The hydrometallurgical processing of Li enriched silicate slag has also shown that Li recovery can reach 80–95% [2].

As long as the complete system, based on a Li2O-Al2O3-SiO2-CaO mixture, does not contain any other element, the results are promising. As soon as other elements are added, new phases start to crystallize.

Besides Li and Al, reporting from the Li-ion battery input into the slag, Si, Ca and often Mg (at least partly from dolomite as slag building component) are introduced as slag builders to ensure an optimized split between metal alloy melt, slag and dust phase in the pyrometallurgical process.

Until now, the thermodynamics of the overall process have not been investigated sufficiently and therefore for extended systems such as Li2O-MgO-Al2O3-SiO2-CaO this work serves as a starting point. Consequently, this should allow understanding some basic principles and giving further insights into these slag-systems. Additionally, a solid ground should be provided for further research on these slag systems, because in the future more complex slag systems, e.g., Mn-containing mixtures, should be investigated since they will represent future inputs to this recycling route especially for the NCM-type batteries.

The Umicore Battery Recycling Process is a vital pyrometallurgical process developed for the recovery of NiMH and spent lithium-ion batteries [3]. From the composition of a slag with the compounds Li2O-MgO-Al2O3-SiO2-CaO and high aluminum content, it is observed that Li is present in the slag in the form of the LiAlO2 [2], which would facilitate subsequent recovery by flotation. At the same time, the spinel phase appears in all three Umicore slags, and, in one of the Umicore slags, Li is even partially dispersed in the spinel phase [3].

Even though spinel phases appear in different slags if bivalent ions such as those of Mg are present, there is little published research on the impact of spinel on the formation of separate LiAlO2 crystals because of the scavenging of Al from the melt and the formation of Mg1−(3/2y)Al2+yO4/LiAl5O8 solid solution.

In this study, three synthetic slags with different contents of MgO based on the Li2O-MgO-Al2O3-SiO2-CaO oxide system were prepared using pure chemical reagents. The degree of supercooling was then reduced by controlling and cooling the melt slowly to obtain thoroughly crystallized synthetic slags for research. The synthetic slags were then analyzed by X'Ray powder diffraction (PXRD) and electron probe microanalysis (EPMA) for mineralogical studies and finally compared to the solidification curves obtained by thermodynamic calculation. This served as a starting point for studying the influence of spinel formation and understanding important phase reactions in the five-component oxide system Li–Mg–Al–Si–Ca.

To increase the knowledge on the behavior of slag systems and the options to predict and stimulate the creation of artificial mineral phases, an interdisciplinary approach was taken, comprising thermodynamical modeling, pyrometallurgical processing, mineralogical analysis and prediction and testing of mineral processing technologies. In this paper, the focus is put on mineralogical analysis in connection with thermodynamic modeling.

#### **2. Background**

To better understand the results presented in this article, the existing information about the compounds of important binary and ternary systems containing Li2O, MgO, Al2O3, SiO2 and CaO is summarized. This information serves as the starting point to analyze and improve the existing data and develop respective thermodynamic modeling strategies.

#### *2.1. Important Binary Phase Systems Containing Li*

In the systems Li2O-CaO and Li2O-MgO, except for limited solid solution, no explicit phase reactions are reported (e.g., Konar et al. [4]).

In the system Li2O-Al2O3, several stable lithium aluminate compounds are described: Li5AlO4, LiAlO2 and LiAl5O8 [5,6]. Additionally, a high temperature compound LiAl11O17 at 0.8 < Al2O3 < 0.92 and >2200 ◦C is mentioned [5]. The compounds Li2Al4O7 synthesized by Kale et al. [7] and Li3AlO3 were found to be instable by Kale et al. [7] and are not part of the data published by Konar et al. [5]. In this phase diagram, there is also a thermal barrier at the mole fraction of Al2O3 = 0.5 (LiAlO2), so that at 0.18 < Al2O3 < 0.5 the resulting mixture is Li5AlO4/LiAlO2 and at 0.5 < Al2O3 < 0.82 the resulting mixture is LiAlO2/LiAl5O8. The two compounds important for this work, LiAlO2 and LiAl5O8, both have polymorphs. According to Konar et al. [5], LiAlO2 comprises a tetragonal γ-phase (high temperature) and a cubic α-phase (low temperature) modification and LiAl5O8 generally crystallizes in a spinel (high temperature) and a low temperature primitive cubic form [8]. According to Li et al. [9], LiAlO2 comprises four polymorphs: a tetragonal γ-phase, a rombohedral α-phase, an orthorhombic β-phase and two phases of high temperature.

In the system MgO-Al2O3, the only binary compound is cubic MgAl2O4 (spinel) with the idealized composition at a mole fraction of Al2O3 = 0.5. At this ratio, there is also a thermal barrier. In the area of mole fraction 0 < MgO < 0.05 in the temperature range 1900–2800 ◦C, solid MgO can form a solid solution with Al2O3 [5]. The region of mole fraction 0.5 < Al2O3 < 0.96, particular important for this study, comprises a complete solid solution, so that an Al-rich melt can be in equilibrium with a spinel relatively enriched in Mg [10].

The system Li2O-SiO2 comprises the binary compounds Li8SiO6, Li4SiO4, Li6Si2O7, Li2SiO3 and Li2Si2O5 [11]. Additionally, the prediction shows two thermal barriers at the composition Li4SiO4 and Li2SiO3.

#### *2.2. Important Ternary Phase Systems Containing Li*

In the system Li2O-MgO-Al2O3, three important primary crystallization fields can be predicted [5]: spinel (MgAl2O4), MgO and γ-LiAlO2. Interesting isopleths are spinel-LiAl5O8, spinel-LiAlO2, spinel-Li2O and MgO-LiAlO2. From this intersects, it can be concluded that a limited amount (i.e., maximum mole fraction = 0.31) of LiAlO2 can be dissolved in MgO. Additionally, the compounds LiAl5O8 and MgAl2O4 can be combined to an ideal spinel solid solution [5].

The system Li2O-Al2O3-SiO2 contains the Li-bearing binary systems Li2O-SiO2 and Li2O-Al2O3, as described in Section 2.1 [12]. With respect to the present work, the primary crystallization fields of LiAlO2, LiAl5O8, eucryptite (LiAlSiO4) and spinel are of particular interest. The compound LiAlO2, described in Section 2.1, can additionally incorporate Si according to an substitution of Li<sup>+</sup> + Al3<sup>+</sup> = Si4<sup>+</sup> + v (vacancy) so that the general formula is α (LiAl4<sup>+</sup>, vSi4<sup>+</sup>])O2 and γ (Li, v)Li[Al3<sup>+</sup>, Si4<sup>+</sup>] MO2 [12]. The compound eucryptite can be derived from SiO2 via a substitution of Li<sup>+</sup> + Al3<sup>+</sup> = Si4<sup>+</sup> + v [12] and crystallizes as quartz in the trigonal system, whereas a low temperature α-polymorph is disordered and a β-polymorph is ordered. Additionally, this compound can incorporate Mg and be broken up into the compounds LiAlO2, Mg0.5AlO2 and SiO2 [13]. The spinel crystallizes in a cubic system and can have a very variable chemistry with respect to the Al/Mg ratio and the solid solution with LiAl5O8 (see Section 2.1).

#### **3. Materials and Methods**

#### *3.1. Materials*

#### Chemicals

The chemicals used for producing synthetic slags are lithium carbonate (Merck, purum), calcium oxide (Sigma-Aldrich, reagent grade, St. Louis, MO, United States), silicon dioxide (Sigma-Aldrich, purum p.a., St. Louis, MO, United States), aluminum oxide (Merck) and magnesium oxide (98% wt.%, Roth, Karlsruhe, Germany). All chemicals ordered via Merck KGaA, Darmstadt, Germany.

#### *3.2. Methods*

#### 3.2.1. Experiments

The chemical compositions of input materials for the synthesis of slags are listed in Table 1. The chemicals were manually mixed in a mortar and grinded in a disc mill for 5 min. Each sample was placed in a Pt-Rh crucible and heated in a chamber furnace (Nabertherm HT16/17, Nabertherm GmbH, Lilienthal, Germany) in an air atmosphere. The heating regime is shown in Figure 1. A heating rate of 2.89 ◦C/min was first employed to reach 720 ◦C, which is the melting temperature of Li2CO3, and then a heating rate of 1.54 ◦C/min was used to aid in the decomposition of Li2CO3 and to reach the target temperature. Samples were kept at 1600 ◦C for 2 h. Thereafter, the samples were cooled to 500 ◦C at a cooling rate of 0.38 ◦C/min and quenched in water.

**Table 1.** Calculated Theoretical Chemical Bulk Composition of the Samples According to theWeighed Quantities.


**Figure 1.** Schematic diagram of temperature regime.

#### 3.2.2. Chemical Bulk Analysis

The element content was determined with ICP optical emission spectrometry (ICP-OES 5100, Agilent, Agilent Technologies Germany GmbH & Co. KG, Waldbronn, Germany). Samples were melted with lithium tetra borate in a platinum crucible at 1050 ◦C, and then the samples were leached with dilute hydrochloric acid to measure the content of Al, Ca, Mg and Si. To measure other elements, the samples were mixed with nitric acid and digested at 250 ◦C and under a pressure of 80 bar in an autoclave (TurboWAVE, MLS, Leutkirch im Allgäu, Germany).

#### 3.2.3. Mineralogical Investigation

An overview of the mineralogical composition was provided by powder X-Ray diffraction (PXRD), using a PANalytical X-Pert Pro diffractometer, equipped with a Co-X-Ray tube (Malvern Panalytical GmbH, Kassel, Germany). For identification of the compounds, the pdf-2 ICCD XRD database, the American Mineralogist Crystal Structure Database [14] and the RRUFF-Structure database [15] were assessed.

The analysis of single crystals and grains was carried out with electron probe microanalysis (EPMA). EPMA is a standard method to characterize the chemical composition in terms of single spot analysis or element distribution patterns, accompanied by electron backscattered Z (ordinal number) contrast (BSE(Z)) or secondary electron (SE) micrographs. To carry out EPMA measurements, the sample was prepared as polished block in epoxy resin, coated with carbon and characterized using a Cameca SXFIVE FE Field Emission) electron probe, equipped with five wavelength dispersive (WDX) spectrometers (CAMECA SAS, Gennevilliers Cedex, France). The following elements/(lines) were used to quantify the measurement points: Na (Kα), Mg (Kα), Al (Kα), Si (Kα), K (Kα), Ca (Kα), Ti (Kα), Mn (Kα) and Fe (Kα). To calibrate the wavelength dispersive X-ray fluorescence spectrometers (WDRFA), an appropriate suite of standards and analyzing crystals was used. The reference materials were provided by P&H Developments Ltd. (Glossop, Derbyshire, UK) and Astimex Standards Ltd. (Toronto, ON, Canada). The beam size was set to 0, leading to a beam diameter of substantially below 1 μm (100–600 nm with field emitters of Schottky-type, e.g., Jercinovic et al. [16]). To evaluate the measured intensities, the X-PHI-Model was applied [17].

Lithium, one of the key elements in this study, cannot be directly analyzed since EPMA uses X-ray fluorescence to detect the elements in the sample and the extremely low fluorescence yield and long wavelength of Li Kα makes the direct determination of this element nearly impossible. With the reasonable assumption that other refractory light elements such as Be and B are not present in the investigated material and volatile elements and compounds such as F, H2O, CO2 or NO3 − are effectively eliminated during the melt experiment, Li can be calculated using virtual compounds, as depicted in as described in Section 4.3.1. If necessary, the balanced Li concentration was included into the matrix correction calculation. To access the analytical accuracy with respect to Li-containing compounds, the international reference material spodumene (Astimex) and the in-house standard LiAlO2 were used (Table 2). Additionally, Li containing crystalline phases identified by X-ray diffraction (PXRD) could be referenced to the EPMA result.


**Table 2.** Repeated Measurements on Two Li-Compounds. Spod, Spodumene; %StdDev, Percentage Standard Deviation of the Measured Points (pepeats: *n* = 5); R, Recovery; LiAl, LiAlO2.

#### 3.2.4. Thermodynamic Modeling

For a better understanding of the experimental mechanisms investigated in the Li2O-MgO-Al2O3- SiO2-CaO system, the thermodynamic modeling of the phase behavior and the solidification in subsystems is of high relevance and hence applied in this work. Especially the knowledge of the phase behavior of the MgO-Al2O3 subsystem and the phases solidified at respective temperatures of certain component concentrations of the Li2O-Al2O3-SiO2 subsystem is important and contributes to the clarification and understanding of primary crystallization mechanisms figured out by the mineralogical characterization. On principal, based on already existing experimental data and thermodynamic studies, which are stated below, an optimized database for the subsystem was completed and applied to calculate the respective phase and solidification behavior. Generally, all calculations, i.e., for the binary MgO-Al2O3 and the ternary Li2O-Al2O3-SiO2 subsystems, were performed with the modified quasi-chemical model (MQM) [18–20] and the compound-energy formalism (CEF) [21], implemented in Factsage [22].

Specific insights into the database adaption regarding the two subsystems are presented subsequently. The thermodynamic database for the oxides such as MgO and Al2O3 comes from the FT oxide database [22] without any modification. Regarding the ternary subsystem Li2O-Al2O3-SiO2, the thermodynamic properties of SiO2, Al2O3 and the mullite solid solution were used from the FT oxide [22] database without any modification. However, for the Gibbs energy of the Li2O, the optimized value from [11] was integrated into the database. For compounds such as Li2SiO3, Li4SiO4, Li6Si2O7, Li2Si2O5-LT (low-temperature form) and Li2Si2O5-HT (high-temperature form), the thermodynamic data were taken from [11]. The standard formation enthalpy of Li8SiO6 was optimized in this work with a value of 3, 521, 499.2 J/mol. Furthermore, for the binary compounds in the Li2O-Al2O3, the standard formation enthalpy of the Li5AlO4 was optimized to 2,389,980 J/mol. The standard entropy of LiAl11O17 was optimized to a value of 350.55 Jmol<sup>−</sup>1K−1. The ternary compounds including the α- and β-eucryptite solid solutions, β-spodumene solid solution and α-LiAlO2 solid solution were obtained from [12] without any modification. However, the Gibbs energy of the end member *G*0 *VaAlO*<sup>2</sup> in the <sup>γ</sup>-LiAlO2 solid solution was calculated with the assumption that the reciprocal energy of endmember is zero, while the other three endmembers were obtained directly from [12].

Based on these data, the CALPHAD calculations were performed for the subsystems, which are used for further explanations and discussions in connection with the new experimental findings in the next section.

#### **4. Results**

This section presents the measurement results of the melt experiments from PXRD and EPMA. First, three PXRD measurements from experiments with different Mg-concentration are compared (Section 4.2). In Section 4.3, an overview of the material with BSE(Z) micrographs and detailed spatially resolved quantitative point measurements and element distribution profiles recorded with EPMA are presented. Additionally, in Section 4.4, experimental findings are compared with thermodynamic model predictions for the relevant subsystems.

#### *4.1. Bulk Chemistry of the Melt Experiments*

The measurement results presented in Table 3 show that 14–20% of Li is lost during the melting and cooling of the material. The same applies for Na, which always appears as contaminant in open systems due to the overall availability (air, dust, skin, clothing, etc.).


**Table 3.** Comparison of the Bulk Chemical Composition Measured with ICP-OES of the Four Melt Experiments, Given in Mole Percent. The bold emphazises the Li-loss which is important to see (Li is volatile).

#### *4.2. PXRD Comparison of the Three Melt Experiments*

The results of the PXRD measurements are presented in Figure 2, showing an overview of the diffractograms of all experiments (above) and three enlarged sections, showing important spinel and lithium aluminate diffraction peaks.

**Figure 2.** (**A**) PXRD of the solidified melt. G, gehlenite; S, spinel; L, LiAlO2; E, eucryptite. (**B**) Enlarged section of the main spinel peak, \* 1,: position of the main peak of MgAl2O4 from the ICCD-PDF2 No. 00-021-1152; \* 2, position of the main peak of an Al-rich spinel from the ICCD-PDF2 No. 00-048-0528, peaks; Mg0.52Al2.32O4, average composition of a spinel grain of the melt experiment with 5.32 Mol% Mg, determined with EPMA (see Section 4.2). (**C**) Enlarged sections of the first two main LiAlO2 peaks.

The overview of all XRD measurements show the compounds gehlenite, spinel, LiAlO2 and eucryptite (Figure 2A), whereas eucryptite is at the detection limit (<2–5 wt.%). The enlarged section of the 2-theta region of the main spinel peaks gives an indication of the changing composition of the spinel with the change of the Mg content (Figure 2B). Because of the high Al-concentration, the main (100%) spinel peaks of all experiments lie between those of the standard spinel MgAl2O4 and an aluminum-dominated Mg1−(3/2y)Al2+yO4. Additionally, there is an indication of increasing spinel content with rising Mg concentrations. The Li-Al-oxide peaks are best explained with the diffraction pattern of LiAlO2 (ICCD PDF2 No. 00-038-1464). The comparison of the two main peaks of the three experiments gives a hint that the amount of crystalline LiAlO2 could be negatively correlated with the amount of Mg in the melt because the highest main peak intensities were measured in the sample with the lowest Mg concentration (Figure 2C).

#### *4.3. EPMA Results*

The main compounds of the melt experiments, determined with EPMA, were:


The compound (GCAS) is an end member of the melilite-like calcium-alumosilicate (MCAS), which is used for this phase with higher amounts of ions in addition to Ca:

• Melilite-like calcium-alumosilicate (MCAS): (Na,Ca,Li)2(Al,Mg,Li)(Al,Si)2O7, which according to the calculations (Section 4.3.3) can also be a potential host for Li

An overview recorded with BSE(Z) shows a matrix of bright Ca-alumosilicate (GCAS/MCAS) interspersed with dendritic or massive dark LiAl. Within this mixture, large idiomorphic or hypidiomorphic crystals of spinel can be identified (Figures 3 and 4).

**Figure 3.** Electron micrograph (BSE(Z) of the solidified melt. Medium grey grains, spinel; dark gray sections and dendrites, LiAl surrounded by Ca-alumosilicate (GCAS, light grey sections); black, pores or preparation damage.

**Figure 4.** (**A**): Overview of the solidified melt (backscattered electron micrographs BSE(Z)). Medium grey grains, spinel; dark gray sections and dendrites, LiAl surrounded by Ca-alumosilicate (GCAS, MCAS, light grey sections); black, pores or preparation damage; red square, detail presented in (**B**). (**B**) Enlarged section from the red square in (**A**): the blue rim marks the grain of ELAS where the scan of Line 3 (red line) was measured. (**C**) Quantitative line scan of Line 3 (red line in (**B**)).

4.3.1. Lithium Aluminate (LiAl) and Lithium-Alumosilicate (ELAS)

The LiAl can be classified into two morphologic forms: massive and dendrite-like (Figures 3 and 4A). A closer look into the massive LiAl reveals thin lath-shaped grains of ELAS or a corresponding melt (Figure 4B). A line scan over such a lath-shaped grain reveals a quite homogeneous composition with more or less sharp borders to the surrounding LiAl (Figure 4C). The ELAS can be described as a mixture of the virtual compounds LiAlO2, Mg0.5AlO2 and SiO2, as listed in Table 4.

Mult. depicts a factor to multiply the three components to generate an optimized ELAS or Li1−xMgy(Al)(Al3/2y<sup>+</sup>xSi2−x−3/2y)O6 similar to the average measured concentrations (except for Li) on Line 3 (Figure 4). The Li value results from the multiplications and this was used to calculate the formula of the analyzed ELAS in the sample. In a similar manner, the Si-containing LiAl with the general formula Li1−x(Al1−xSix)O2 can be calculated as a mixture of SiO2 and LiAlO2. The calculated formulas of the ELAS and the LiAl are:

ELAS: (Li0.96Mg0.24)(Al)(Al0.45Si1.55)O6

LiAl: (Li0.94)(Al0.94Si0.06)O2

The Si concentration in the dendritic LiAl is distinctively lower as in the massive crystals (compare Tables 4 and 5). The calculated formula of the LiAl in this case is:

$$\text{LiAl: (Li}\_{0.97}) (\text{Al}\_{0.97}\text{Si}\_{0.03}) \text{O}\_2$$

**Table 4.** Calculation of virtual compound ratios and average composition of the ELAS and the LiAl on Line 3, shown in Figure 4. Opt., calculated ideal composition; Meas., measured average; Mult., factor for multiplication of the virtual compounds; (Calc.), calculated values (Li, O); %StdDev, percentage standard deviation of the measured points (LiAl (Meas.), repeats, *n* = 23).


**Table 5.** Calculation of Virtual Compound Ratios and Average Composition of the LiAl in the Dendrites (Dend.) Shown in the BSE(Z) Micrograph of Figure 3. Opt., Calculated Ideal Composition; Meas., Measured Average; Mult., Factor for Multiplication of the Virtual Compounds; (Calc.), Calculated Values (Li, O); %StdDev, Percentage Standard Deviation of the Measured Points (LiAl (Dend.) (Meas.), Repeats, *n* = 4).


#### 4.3.2. Spinel

Spinel as the first crystallizing compound obeys the crystallization equilibrium inasmuch as the composition of the spinel with the highest Al content is connected with the corresponding Al-rich melt. The Mg concentrations in the measured profile (Figure 5B) are increasing from the center to the rim of the crystal. A look at the ratio of Mg/Al in a line scan through a spinel crystal starting at the center

of the grain in the direction to the rim shows no increase within a first region. After this first region, the ratio increases. Closer to the rim, the ratio decreases sharply and directly at the rim (a few μm) the ratio development of the two elements is reversed again (Figure 5B).

**Figure 5.** (**A**): Electron micrograph of a spinel crystal (medium grey), partly with a thin coat of LiAl (dark grey sections) surrounded by GCAS (light grey sections). Black, pores or preparation damage. (**B**) Development of the Mg/Al ratio from the center to the rim of a spinel grain along the red line in Figure 5A.

#### 4.3.3. Ca-Alumosilicate (GCAS/MCAS)

The matrix component of the melt experiments (e.g., Figure 3 or Figure 4A light grey sections) can be generally expressed as X2YZ2O7, where X can be Na<sup>+</sup> and Ca2+; Y can be Al3+, Mg2<sup>+</sup> and Fe2+; and Z can be Al3<sup>+</sup> and Si4+. The coordination of X is 8, and Y and Z are tetrahedral [23]. This Ca-alumosilicate compound generally is known as melilite. The investigated material comprises two types of Ca-alumosilicate:


These two types are difficult to distinguish in the BSE(Z)-micrograph (Figure 4A) because of the almost same light grey shade (very similar mean atomic number). Because Na is not part of the initial materials (impurity), the concentration of the MCAS compound can be considered rather low and represents the eutectic residual melt. Nevertheless, this compound is interesting to assess a potential Li incorporation into the matrix of Ca-alumosilicate. In theory, Li<sup>+</sup> can be present in 4 or 8 coordination, whereas the ionic radius is very similar to Mg (4-coordination) or Na (8-coordination) (e.g., the ionic radii are published by Shannon (1976) [24]). The MCAS possesses a lower total sum of the measured concentrations and excess Si when calculating the chemical formula of the MCAS using the general melilite based on seven oxygen atoms.

Table 6a depicts how a calculation of virtual Ca-alumosilicate (CAS) can be conducted using five virtual compounds, namely Li2Si3O7, Na2Si3O7, Ca2Al2SiO7 Ca2MgSi2O7 and Ca3Si2O7, assuming (limited) solid solution between those compounds. The Li value resulting from the multiplications was used to calculate a chemical formula of the analyzed MCAS. According to this calculation, the Li2Si3O7 makes up about 1 wt.% of the total composition (see Table 6b).


**Table 6.** (**a**) Multiplication Factors (Mult.) for Calculation of an Optimized GCAS and MCAS. (**b**) Average Composition of the GCAS and MCAS Ca-Alumosilicate Solid Solution in Single Point Analysis, Compared with the Optimized Compounds Calculated with the Factors of Table 6a. Opt., Calculated Ideal Composition; Meas., Measured Average; (Calc.), Calculated Values (Li, O); %StdDev, Percentage Standard Deviation of the Measured Points (Repeats, *n* = 7 (GCAS), *n* = 6 (MCAS)).

(**b**)

The calculated formulas of the GCAS and the MCAS are:

GCAS: Ca2.02(Al1.96Mg0.03)(Al1.96Si)O7

MCAS: (Na0.18Ca1.67Li0.15)(Al0.52Mg0.32Li0.08)(Al0.17Si1.82)O7

#### *4.4. Comparison of Experimental Findings with Thermodynamically Modeled Subsystems*

Based on the respective phases of interest, the relevant subsystems are MgO-Al2O3 and Li2O-Al2O3-SiO2. The modeled phase diagrams are presented in Figures 6 and 7, respectively. In Figure 6, a comparison between the modeled phase equilibria and the experimental data is made. In Figure 6, the composition of the initial melt and the composition of different spinel grains from two line scans and several single spot measurements, analyzed experimentally at room temperature (RT), are presented in an overlay with the thermodynamic phase equilibrium data for the subsystem MgO-Al2O3. The composition of the initial melt is the starting point of the spinel crystallization. It can be seen that all measured spinel grains show a significant higher Mg concentration compared to the initial Mg concentration in the melt.

**Figure 6.** The calculated MgO-Al2O3 phase diagram at 1 atm total based on [22]. Lq, Liquid; spinel (s.s), spinel solid solution; Al2O3(Cor), corundum. In this diagram, the composition of the initial melt and the composition of different spinel grains from two line scans and several single spot measurements, analyzed at room temperature (RT), are presented. The composition of the initial melt is the starting point of the spinel crystallization.

**Figure 7.** Calculated Li2O-Al2O3-SiO2 liquidus projection at 1 atm total pressure based on [22] is shown. Red line, equilibrium solidification paths starting at the initial point of the "product" (**A**) and the "raw mix" (**B**). The initial points represent the respective component concentrations in the liquid phase. Isothermal lines are drawn in Kelvin at every 100 K. In this diagram, the average compositions of the single compounds, analyzed with EPMA at room temperature (see Tables 4, 5 and 6b), and the bulk chemistry of the "raw mix" and the "product" are presented.

Figure 7 shows the thermodynamic calculated Li2O-Al2O3-SiO2 subsystem. The equilibrium solidification paths for the "raw mix" (Figure 7B) and the "product" (Figure 7A) composition are calculated and presented in the respective ternary phase diagram. Additionally, the average compositions of the single compounds, analyzed with EPMA at room temperature (see Tables 4, 5 and 6b), and the bulk chemistry of the "raw mix" and "product" are visualized in an overlay with the modeling results in Figure 7.

The "raw mix" and the "product" composition is in the spinel solid solution area, which is concluded by the thermodynamic modeling results based on the subsystem. Hence, the thermodynamic modeled solidification predicts spinel as the primary crystallizing phase (see Figure 8). After decreasing the temperature continuously under assumed equilibrium conditions, the solidifications of different phases are shown in Figure 8, for the "product" (Figure 8A) and "raw mix" (Figure 8B) initial concentrations, respectively. The thermodynamic prediction of the subsystem solidification shows that spinel as primary crystal is formed in solid solution with high temperature LiAl5O8 for both initial compositions. With progressing solidification, low temperature LiAl5O8 is also crystallizing out of solution. This finding holds true for both initial compositions. With increasing solidification progress, LiAlO2 and Li2SiO3 are formed with a very low amount of eucryptite, for the "raw mix" configuration, while, for the "product" composition (Figure 8A), eucryptite is formed in a higher amount without any LiAlO2.

**Figure 8.** Calculated equilibrium solidification curves of the "product" (**A**) and the "raw mix" (**B**) for Li2O-Al2O3-SiO2 system. Calculated point interval is 5 K.

#### **5. Discussion**

The experimental investigation of solidified melt in connection with thermodynamic modeling of chemical reactions and solidification is an important tool to investigate how a slag system behaves and how it can be engineered. These investigations and the obtained results can serve as starting point for understanding efficient design of experiments to generate the desired phases. With a combination of thermodynamic calculation and mineralogical investigation, the probability that the artificial slag contains the desired phases can be maximized. Therefore, one purpose of the experiments carried out in this project was a first survey of the mineral compounds and the morphology of a solidified melt with the basic components Li2O, MgO, Al2O3, SiO2 and CaO with a melt composition in the primary crystallization field of spinel in the subsystem Li2O, Al2O3 and SiO2. Another purpose was to investigate the influence of the Mg content on the ratio of the mineral compounds. The results of these experiments are also intended to serve as a basis for further thermodynamic modeling. Additionally, the applicability of the combination of PXRD and EPMA to this research topic was assessed. This includes the calculation of the lithium containing mineral compounds on basis of the EPMA result without access to measured lithium concentrations. In the following, the different identified phases are discussed:

#### *5.1. Spinel-Like Oxides*

The experiments show idiomorphic phenocrysts of spinel as the first crystallizing compound with decreasing temperature. The spinel crystals are surrounded by massive hypidiomorphic crystallites of LiAl and melilite-like alumosilicate (GCAS/MCAS). Additionally, LiAl forms dendritic elongated structures of hypidiomorphic crystallites. The changes in the chemistry of a single spinel crystal (Figure 5B) can help to explain a part of the crystallization curve of the melt. This is also used to validate thermodynamic model predictions of the three-component subsystem. Starting in the primary crystallization field of spinel, an Al-rich Mg1−(3/2y)Al2+yO4 starts to form. These crystals are in equilibrium with a corresponding melt (Figure 5B (blue area) and Figure 6). The EPMA reveals that, compared with the Mg/Al ratio of the melted material, all measured spinel grains are enriched in Mg. This observation shows the complex spinel behavior, which cannot be explained with the simple binary phase diagram MgO-Al2O3. Nevertheless, the measured Mg/Al ratio increases (Figure 5B, yellow area). This can be explained with the composition of the melt reaching the phase boundary between spinel and LiAl. Through scavenging of Al from the melt during formation of LiAl, the Mg concentration in the melt increases and therefore the spinel–melt equilibrium changes. At a later stage of the crystallization process, the Al concentration in the crystal increases again, an indication that now a solid solution between Mg1−(3/2y)Al2+yO4 and Mg-free LiAl5O8 forms (Figure 5B, green area). At the end of the crystallization, the Mg concentration rises again (Figure 5B, red area). This is an indication that the crystallization leaves the crystallization path between LiAl5O8 and spinel in direction of the crystallization path between eucryptite (or ELAS) and spinel. Therefore, the crystallization of the spinel would no longer include the aluminum-rich LiAl5O8 and the relative Mg concentration of the crystallizing spinel would increase, although a part of the Mg is incorporated into the ELAS. The crystallization path concluded by experimental observations of the developing spinel composition is on principal in good correlation with the thermodynamically predicted solidification phases in the early stages (Figures 7 and 8). However, for lower temperatures, the solidification predictions deviate from the experimental findings, which is due to non-equilibrium cooling conditions and hence phase generation. The modeled results show that small deviations in the initial concentration in the spinel solid solution field can result in strong deviations regarding appearing solid phases and solidification path behavior.

The PXRD patterns of the investigated melts with increasing Mg concentration show a displacement of the spinel main peak. The angular position of the main peak of these spinel variations is between the simple MgO × Al2O3 compound and a pattern of an Al-rich spinel with the formula Mg0.39Al2.41O4 and weakly correlates with the Mg concentration as:

$$\mathbf{M} \mathbf{g}\_{\mathbf{Y}} = 20.198 \times \mathbf{d}\_{311} - 48.247 \tag{1}$$

#### *5.2. LiAl and ELAS*

The formation of dendrites is an indication for rapid crystallization of LiAl in a small temperature interval from a supercooled melt and/or (macro)segregation (for macro segregation, see, e.g., Ahmadein et al. [25]). Due to the rather long cooling cycle (two days, Figure 1), undercooling may be improbable but cannot completely be excluded. Nevertheless, it is plausible that a segregation of an Al-Li-rich melt occurs from which the first generation of LiAl crystals forms. The Si concentrations of the dendritic LiAl is lower than in the massive LiAl, indicating a different origin, thus a different melt as well (compare LiAl in Tables 4 and 5). A few parts of the massive LiAl contain small lath-shaped grains of ELAS. This compound can be derived from eucryptite and can contain up to 18 wt.% Mg0.5AlO2. These grains are an indication of segregation of Si- and Mg-rich phases (melt) from the Al-rich LiAl-melt, as described above. Interestingly, the representing point of this calculated compound is not located in the primary crystallization field of pure LiAlSiO4. This is due to the lower calculated Li content because of the Li-free Mg0.5AlO2 compound.

#### *5.3. Ca-Alumosilicate*

The matrix of the material consists of slightly hypidiomorphic grains of Ca-alumosilicate. The morphology of the crystals indicates that the formation starts together or slightly after the

beginning of the crystallization of LiAl, which itself often shows hypidiomorphic growth. The chemical composition of these Ca-alumosilicates starts with nearly ideal gehlenite (GCAS) with minute amounts of impurities such as Mg. The other type of Ca-alumosilicate (MCAS) incorporates higher amounts of impurities such as Na and Mg (Table 6b). Because of the presence of an alkaline element such as Na, the latter compound seems to represent the end of the crystallization, i.e., the residual eutectic melt. Interestingly, this compound delivers a total of distinctively less than 100 wt.% (element concentrations calculated as simple oxide compounds) and possesses an excess of Si after calculation of the melilite formula. This is an indication that another silicious component is present in the crystal structure. Because the sample contains no free SiO2 (like quartz) and the analysis shows no additional element for calculation of a silicious component, incorporation of Li into the crystal or glassy structure as shown in Table 6b is plausible. After incorporation of a virtual compound Li2Si3O7, a formula can be calculated indicating a consistent crystal-like chemistry or a stoichiometric glass.

The mineralogical characterization of a melt as presented above provides a basis for refining the thermodynamic model, showing the real assemblage of components and the real chemical composition of the compounds/phases. An example would be ELAS. The eucryptite compound, used for the thermodynamic calculation, is ideal LiAlSiO4. EPMA reveals that the real eucryptite-like alumosilicate (ELAS) can be expressed with (Li0.96Mg0.24)(Al)(Al0.45Si1.55)O6. This compound contains Mg, which has to be taken into account when using this compound to predict a crystallization curve. The same is valid for the lithium aluminate compound (LiAl, Li1−x(Al1−xSix)O2) that contains Si. Another important property is the inherent potential kinetic inhibition of the phase reactions in the system of interest. The morphology of the slag including structure and habitus corresponds to the crystallizing reactions during the cooling process. Additionally, the total chemistry and the spatial resolved development of element ratios can be used to explain the solidification process.

#### **6. Conclusions and Outlook**

In this work, an experimental investigation of a Li2O-MgO-Al2O3-SiO2-CaO system was carried out in combination with thermodynamic modeling of relevant subsystems. Based on bulk chemistry analysis, PXRD and EPMA, the crystallization paths of various phases were reconstructed and explained. It was shown that spinel is always the primary crystallizate. Furthermore, depending on minute variations in the chemistry of the melt, the result of the thermodynamically predicted further phase development can be substantially different. In this case, comparing the solidification of the raw material and the product the unpredictable loss of Li during the melt experiment seems to offer the possibility of a complete suppression of the LiAlO2 formation in favor of Mg1−(3/2y)Al2+yO4/LiAl5O8 solid solution, although this was not observed in the experiments. The eucryptite and Li-silicate compound are the ends of the solidification in both scenarios. Nevertheless, a knowledge and/or control of all reaction parameters such as partial pressures of all elements (particularly, Li here) and compounds, grain size distribution, morphology and chemistry of the raw material is crucial to develop an efficient and reproducible slag modification process. The solidification route of the system could be qualitatively predicted by thermodynamic modeling of the Li2O-Al2O3-SiO2 subsystem with the result that minute variations of the initial chemistry can lead to different solidification paths.

Additionally, a relative Mg enrichment of spinel grains could be observed experimentally. Furthermore, the development of the composition in single spinel grains during spinel grains give an indication of the existing solid solution Mg1−(3/2y)Al2+yO4/LiAl5O8, which was only predicted and not verified in the past.

The results presented in this article show that Li cannot be incorporated into a single early crystallizing compound in an easy way. The investigations show that Li is present in LiAl, ELAS and with higher uncertainty in spinel (as solid solution Mg1−(3/2y)Al2+yO4/LiAl5O8) and MCAS. To modify this complex multi-component system (oxides of Li, Mg, Al, Si and Ca) to gain desired mineral compounds requires, besides experimental work (melt experiments, component printing

and combinatorial thin film deposition), new thermodynamic modeling strategies even for higher component systems, especially with a good quantitative predictability for the phase fractions.

Furthermore, future research work will concentrate on the development of phase separation processes, predominantly by flotation for the main identified Li-bearing phases described in this paper (basic research on the way) and on the extension of component mixtures in the slag building process, advancing step by step into slag systems expected in melting processes of actual and future battery systems.

**Author Contributions:** T.S. conceived the paper. T.S., H.L. and M.F. conducted the literature review. All experiments were designed and performed by H.Q. and D.G. The chemical bulk analysis was executed by the analysis laboratory of the institute. The phase analysis (PXRD and EPMA) and the mineralogical investigation were conducted by T.S. Thermodynamic modeling was conducted by H.L. and M.F. Interpretation and discussion were conducted by all authors. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Clausthal University of Technology in the course of a joint research project "Engineering and processing of Artificial Minerals for an advanced circular economy approach for finely dispersed critical elements" (EnAM).

**Acknowledgments:** We acknowledge support by Open Access Publishing Fund of Clausthal University of Technology.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study, in the collection, analyses, or interpretation of data, in the writing of the manuscript, or in the decision to publish the results.

#### **References**


**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Optimization of Manganese Recovery from a Solution Based on Lithium-Ion Batteries by Solvent Extraction with D2EHPA**

**Nathália Vieceli 1,\*, Niclas Reinhardt 2, Christian Ekberg <sup>1</sup> and Martina Petranikova <sup>1</sup>**


**Abstract:** Manganese is a critical metal for the steelmaking industry, and it is expected that its world demand will be increasingly affected by the growing market of lithium-ion batteries. In addition to the increasing importance of manganese, its recycling is mainly determined by trends in the recycling of iron and steel. The recovery of manganese by solvent extraction has been widely investigated; however, the interaction of different variables affecting the process is generally not assessed. In this study, the solvent extraction of manganese from a solution based on lithium-ion batteries was modeled and optimized using factorial designs of experiments and the response surface methodology. Under optimized conditions (O:A of 1.25:1, pH 3.25, and 0.5 M bis(2-ethylhexyl) phosphoric acid (D2EHPA)), extractions above 70% Mn were reached in a single extraction stage with a coextraction of less than 5% Co, which was mostly removed in two scrubbing stages. A stripping product containing around 23 g/L Mn and around 0.3 g/L Co can be obtained under optimized conditions (O:A of 8:1,1MH2SO4 and around 13 min of contact time) in one stripping stage.

**Keywords:** lithium-ion battery; battery recycling; manganese recovery; solvent extraction; D2EHPA; factorial design of experiments

#### **1. Introduction**

Manganese is one of the most abundant metals in the Earth's crust; however, manganese is highly dispersed (low-grade), and minerals are widely distributed. The identified manganese resources are concentrated in a few countries—the main manganese mining areas are in China, South Africa, Australia, and Gabon [1–3].

The main end use of manganese is in the steel industry, which accounts for 90% of the world´s manganese demand. Manganese is also widely used in ironmaking and alloys with aluminum, magnesium, and copper [3–6]. Non-metallurgical applications account for only 5–10% of the manganese consumption, which is used in electrical systems, in the chemical industry, in the ceramic and glass production, and in the agricultural sector [7]. In electrical systems, manganese dioxide is used for cathodic depolarizer in dry cells, alkaline batteries, and lithium-ion batteries (LIBs) [4].

Natural manganese dioxide is used in dry cells, while high-grade synthetic manganese dioxide is produced chemically or by electrolysis to be used in alkaline batteries and LIBs [4]. Lithium manganese spinels (such as LiMn2O4) and layered lithium–nickel–manganese– cobalt (NMC) oxide systems have an important role in the development of advanced rechargeable lithium-ion batteries, with cost and environmental advantages [8]. Thus, nowadays, most automakers and some electronics makers use some version of NMC system in their LIBs [9].

In this context, the United States of America Department of Defense has recently classified manganese as one of the most critical mineral commodities for the United States because it is essential for important industrial sectors, has no substitutes, and has a

**Citation:** Vieceli, N.; Reinhardt, N.; Ekberg, C.; Petranikova, M. Optimization of Manganese Recovery from a Solution Based on Lithium-Ion Batteries by Solvent Extraction with D2EHPA. *Metals* **2021**, *11*, 54. https://doi.org/10.3390/met110 10054

Received: 1 December 2020 Accepted: 24 December 2020 Published: 29 December 2020

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

potential for supply disruptions, since the country is strongly dependent on imports [10]. Additionally, the United States included electrolytic manganese metal in the National Defense Stockpile in 2019 as a critical material for defense purposes [2].

Although it is expected that steel will continue leading the manganese demand, the consumption of manganese in batteries applications is projected to grow fast in the next decade, boosted by the rapid growth in the lithium-ion battery market, which is expected to increase from \$35 billion USD in 2020 and reach \$71 billion USD in 2025 [11,12]. Thus, electrolytic manganese dioxide (EMD) for the battery industry is expected to be the fastestgrowing segment of the manganese market [13], increasing the manganese production along with the global demand for batteries [14].

EMD is generally produced from high-grade manganese ores [15], and in general, converting manganese ores to EMD involves a high-temperature pyrometallurgical process, which has some drawbacks such as environmental impacts, high-energy consumption, and high costs. Furthermore, because the roasting process decreases the oxide content in the ore, EMD producers face competition from chemical and steel industry buyers of high-grade manganese ores [16]. In this context, the recovery of manganese from spent LIBs can help decrease supply risks and impacts linked to the primary production of manganese. However, although there is an increasing importance of manganese, its recycling is mainly determined by trends in the recycling of iron and steel, and in general, materials are not recycled specifically due to their manganese content [2,17,18]. Moreover, when it comes to LIBs recycling, the presence of manganese in the leaching solutions has been linked to a decrease in the selective separation of cobalt and nickel, and for this reason, manganese should be previously recovered [19,20].

The recovery of manganese from primary and secondary resources by solvent extraction has been investigated by several authors [14,20–26]. Table 1 (on the next page) summarizes the optimal extraction conditions described in some studies focused on the extraction of manganese from different feed solutions, including from leach solutions from spent LIBs. It is possible to highlight that bis(2-ethylhexyl) phosphoric acid (D2HEPA) is the most widely used extractant to recover Mn from liquors from LIBs as well as from other solutions.

Although several studies on the recovery of manganese by solvent extraction have been published, the effect of different variables affecting the process is generally approached using one-factor-at-a-time, which does not allow identifying interaction effects among them. In this context, the main goal of this study was to optimize the solvent extraction of manganese using the factorial design of experiments and response surface methodologies to assess and model the effects of the variables affecting the process. The optimization of the recovery of manganese was studied using a synthetic solution based on an acid leach from spent LIBs. The results can support further investigations focused on the recovery of manganese from spent LIBs, which can be considered an important secondary resource of a critical material for many important industrial sectors.

*Metals* **2021**, *11*, 54


*Metals* **2021**, *11*, 54


#### **2. Materials and Methods**

Bis(2-ethylhexyl) phosphoric acid (D2EHPA, 97%, Sigma Aldrich, Germany) was used as solvent extraction reagent as it was supplied, without any additional purification. Isopar L (Exxon Mobil, USA) was used as diluent. A synthetic solution was prepared based on the chemical composition of an original solution obtained through the acid leaching of spent lithium-ion batteries with sulfuric acid, which was investigated in detail in previous work (unpublished results). The synthetic solution was prepared using sulfates (NiSO4.6H2O, CoSO4.7H2O, MnSO4.H2O, Li2SO4, Sigma Aldrich, Germany) and Milli-Q water. Impurities typically present in acid leach solutions from LIBs such as Cu and Al were not included into the synthetic solution because they are generally removed using conventional purification processes, for example, cementation and purification, before the solvent extraction.

Preliminary extraction tests, scrubbing, and stripping tests were performed in glass vials (3.5 mL) using a shaking machine (IKA-Vibrax, Germany) operating with 1000 vibrations per min to promote the contact between phases. The experiments were performed at room temperature. Specific conditions used in the preliminary tests are reported in the Results section. The extraction and stripping of manganese and cobalt were optimized using factorial designs of experiments and response surfaces. These methodologies are explained in detail by Montgomery [35]. For the factorial design of experiments of the extraction phase, tests were carried out using plastic containers (50 mL), in which the stirrer from a mixer-settler device was coupled. The stirring speed was set at 1000 rpm, and the tests were also performed at room temperature.

The pH of the aqueous phase was measured using a pH meter (Metrohm 827 pH lab, Switzerland), and the electrode was regularly calibrated before and during the experimental procedures. The pH was adjusted whenever it was needed with 5 M or 10 M NaOH to minimize the dilution effect of the feed solution. Samples from the aqueous phase were taken 10 min after finishing the contact time at the established pH to obtain a complete separation of phases. Chemical analysis was performed by Inductively Coupled Plasma— Optical Emission Spectroscopy (ICP-OES, iCAP™ 6000 Series, USA) using samples from the aqueous phase, which were diluted in 0.5 M nitric acid. The extraction efficiency of metals was determined by Equation (1):

$$\%E = 100 \ast \frac{D\_X}{D\_X + \left(V\_{aq}/V\_{org}\right)}\tag{1}$$

where *Vaq* and *Vorg* represent the volume of the aqueous phase and the volume of the organic phase, respectively, and *DX* is the distribution ratio, which describes the ratio between the concentration of a certain metal (*X*) in the aqueous phase and in the organic phase and it can be determined by Equation (2). In some cases, the log *D* is used to assist the interpretation of results.

$$D\_{\mathbf{x}} = \mathbb{C}\_{X \text{ } argamic} / \mathbb{C}\_{X \text{ } adjcounts \text{ }}\tag{2}$$

The separation factors (*β*) between two elements (*X* and *Y*) can be calculated using Equation (3), and it is determined by the division of the distribution ratio of each element, being normally greater than one. This equation was used to determine the separation factor of manganese in preference to other metals.

$$
\beta = D\_X / D\_Y \tag{3}
$$

#### *Experimental Design*

A full 2k factorial design of experiments was used to fit a second-order linear regression model to the experimental results. To estimate the experimental uncertainty, four additional experiments were performed under the same conditions at the central level of the factors (*nC*, central point). The effects of three factors (k = 3), each one with two levels (23 factorial design), on the process response (*y*, manganese extraction or cobalt extraction) were studied. The factors and levels were selected based on results from preliminary tests and on the literature review.

Experimental design of the extraction stage: The factors investigated in the design of experiments to model the extraction stage were equilibrium pH (*x*1), organic to aqueous ratio, O:A (*x*2) and molar concentration of D2EHPA (*x*3). Each factor was varied in two levels.

Experimental design of the stripping stage: To model the stripping stage, the effect of the following three factors was evaluated: molar concentration of sulfuric acid (*x*1), organic to aqueous ratio, O:A (*x*2) and stripping time (*x*3). Each factor was varied in two levels.

Axial points were included (*2k* axial points) in both designs to estimate the quadratic terms of the models, setting up a central composite design. Tests were performed in random order. The distance of the axial points from the central point was α = 1 (face-centered central composite design). The standard, high, and low levels of the factors are presented in Table 2.



\* Equilibrium pH after a contact time of 10 min, with a maximum variation of ±0.05 from the value defined in the design.

The process response, *y*, was used to fit the coefficients of a linear second-order regression model, using the linear least squares method. Only statistically significant variables were considered in the models (*p*-value smaller than the significance level of 0.05). Analysis of variance (ANOVA) was used to assess the significance of the fitted model. The variance of the response accounted for the models was evaluated by the coefficient of determination (*R*2), and the existence of pure quadratic curvature was determined by hypothesis testing. Response surfaces and contour plots were used to assist the optimization of the processes.

#### **3. Results and Discussion**

#### *3.1. Preliminary Tests of Extraction*

Preliminary tests were performed to determine the best conditions to be further investigated in the factorial design of experiments. The extraction of Mn, Ni, Co, and Li at different contact times can be observed in Figure 1. The mechanism of extraction of manganese using D2HEPA is very fast. The extraction of Mn was about 60% after only 5 min of contact time, and after 10 min, the extraction achieves the maximum values (approximately 70%). The coextraction of Co, Ni, and Li is slightly higher after 5 min of contact time, but it is still lower than 20%. At 10 min of contact time, the increase in the extraction of Mn resulted in a decrease of the coextraction of the other metals. The coextraction of Co, Ni, and Li after 10 min of contact time was around 11, 5, and 3%, respectively. This is in accordance with the results reported in the literature. Chen et al. [24] studied the extraction of manganese from the leaching liquor of spent LIBs using cobaltloaded D2EHPA, and they reported that the equilibrium was achieved after only 3 min. Hossain et al. [28] also observed that the kinetics of the manganese extraction using Co-D2EHPA was fast, and the equilibrium was achieved in 5 min. Thus, low contact times are required for the extraction of manganese.

**Figure 1.** Extraction of metals at different leaching times. Conditions: O:A of 1:1; 0.5 M bis(2 ethylhexyl) phosphoric acid (D2EHPA) and pH of 3.5.

#### *3.2. Effect of the Concentration of Modifier (% Volume of TBP)*

Preliminary tests using TBP (tributyl phosphate, Sigma Aldrich, Germany) as a modifier were performed to evaluate its potential to increase the extraction of manganese as well as its separation from the other metals. The extraction of Mn, Ni, Co, and Li without using TBP and when volumetric concentrations of 2.5%, 5%, and 10% TBP were used can be seen in Figure 2, where the error bars represent the standard deviation of triplicates. The extraction of Mn had a slight increase when the concentration of TBP was increased until 5%. However, the coextraction of all other metals also increased when TBP was used as a modifier. For all evaluated metals, the extraction decreased when 10% of TBP was used. Considering that no formation of a third phase was observed, it was decided not to use TBP in the next tests.

**Figure 2.** Extraction of metals using different volumetric concentrations of TBP as a phase modifier. Conditions: contact time of 10 min, 0.5 M D2EHPA, equilibrium pH of 3.5, organic to aqueous ratio (O:A) of 1:1. Error bars represent the standard deviation of triplicates.

#### *3.3. Effect of the pH on the Extraction of Metals*

The extraction of Mn, Co, Li, and Ni for three different molar concentrations of D2HEPA (0.4, 0.5, and 0.6 M) at different pH values can be seen in Figure 3. Some tests were performed using 0.2 M D2EHPA, but in this case, the extraction of manganese never exceeded 30%, and since this concentration is lower than the ones usually reported in the literature, further tests using 0.2 M D2EHPA were not performed. The initial pH of the synthetic solution based on the composition of the LIBs leach liquor was 3.8. After contacting the synthetic solution with the extractant, the pH of the aqueous phase decreased to about 2. This behavior was expected, considering the mechanism of extraction of metals using D2EHPA (Equation (4)) described by Zhang and Cheng [14], which results in a decrease in the pH.

$$\overline{M}^{2+} + 2\overline{(HA)\_2} \leftrightharpoons \overline{MA\_4H\_2} + 2H^+ \tag{4}$$

where *M* represents the metal, (*HA*)<sup>2</sup> represents D2EHPA in the organic phase, and *MA*4*H*<sup>2</sup> represents the metal–organic complex [14].

**Figure 3.** Extraction of metals using different molar concentrations of D2EHPA: (**a**) 0.4 M D2EHPA, (**b**) 0.5 M D2EHPA, (**c**) 0.6 M D2EHPA. Conditions: O:A of 1:1, contact time of 10 min.

The extraction of manganese increased with the pH for the three different concentrations of D2EHPA, but when the pH was increased to about 4, the coextraction of other metals was also more pronounced, mainly of cobalt. The increase in the molar concentration of D2HEPA also promoted an increase in the extraction of manganese, which was more pronounced when 0.6 M D2EHPA was used.

#### *3.4. Effect of the Organic to Aqueous Ratio (O:A)*

Preliminary tests were performed to evaluate the effect of the O:A ratio on the extraction of metals (Figure 4). The extraction of manganese increased with the O:A ratio (Figure 4a); however, the coextraction of cobalt also increased with the O:A ratio. For this reason, O:A ratios from 0.5 to 2 were further investigated in the factorial design of experiments. The isotherm representing the distribution of manganese in the aqueous and organic phase can be seen in Figure 4b. The extraction of manganese can be theoretically achieved after two extraction stages using an O:A ratio of 1.25.

**Figure 4.** (**a**) Extraction of manganese and cobalt using different O:A ratios and (**b**) McCabe–Thiele diagram of the Mn extraction. Conditions: equilibrium pH of 3.5, 0.5 M D2EHPA, contact time of 10 min. Error bars represent the standard deviation of triplicates.

#### *3.5. Extraction Stage: Factorial Design of Experiments and Regression Model*

The conditions of the factorial design of experiments and respective responses (manganese and cobalt extraction) for each experiment are presented in Table 3. Tests from 1 to 8 correspond to the base 23 design. Tests from 9 to 12 are the replicates in the central point of the design and were used to determine the experimental error. Tests from 13 to 18 are the axial points added to the design. All the tests were performed at room temperature using a contact time of 10 min. The concentrations of metals in the raffinate and in the organic phase are reported in the Supplementary Materials (Table S1), as well as the extraction of Ni and Li, which in general remain at low values. The Supplementary Material (Table S2) also reports the distribution ratios (*D*) and separation factors (β).

**Table 3.** Conditions of the experimental design and results for the extraction of manganese and cobalt.


The adjusted regression model (*y*) for the extraction of manganese and the extraction of cobalt are represented by Equations (5) and (6), respectively. The models are only valid for the range of values tested in this study, and they only include factors with a statistically significant effect on the responses (α = 0.05).

$$\text{Mn } (\%) = 72.0 + 14.7 \,\text{x}\_1 + 22.3 \mathbf{x}\_2 + 5.7 \mathbf{x}\_3 - 7.4 \mathbf{x}\_2^2 \tag{5}$$

$$\text{Co } (\%) = 6.6 + 10.0 \,\text{x}\_1 + 5.2 \mathbf{x}\_2 + 6.0 \mathbf{x}\_1 \mathbf{x}\_2 + 3.7 \mathbf{x}\_1 \mathbf{x}\_2 \mathbf{x}\_3 + 4.7 \mathbf{x}\_1^2 \tag{6}$$

The results of the analysis of variance of the fitted models for the extraction of manganese and cobalt are presented in Table 4, which was adapted from the ANOVA table from the Regression Analysis tool of Excel (Analysis ToolPak add-in). The replicates in the central level of the design allow estimating the experimental pure error and decomposing the Residual Sum of Squares (RSS) into the Sum of Squares due to Pure Error (SSPE) and the Sum of Squares due to Lack of Fit (SSLOF). The presence of curvature was verified for both models using the pure curvature testing (*p*-value = 0.048 and 0.046 for manganese and cobalt, respectively). The significance of the fitted models is indicated by the results of the *F*-test. The model adequacy was assessed by the Lack of Fit (LOF) test, but the results were lower than the significance level (α = 0.05) for both models, given the low experimental error in the central point of the design and a small variance of the experimental error when compared to the residual error.


**Table 4.** Results of the analysis of variance of the fitted models for the extraction of manganese and cobalt.

Pareto charts of the standardized effects of the variables on the responses are presented in Figure 5a for the manganese extraction and in Figure 5b for the cobalt extraction. The standardized effects were calculated by dividing each coefficient by its standard error. The standardized effects correspond to the *t*-statistic values. A variable is considered statistically significant if its *p*-value is smaller than the defined significance level (0.05 for a confidence level of 95%). The significance level is identified in the graphs by dashed lines (2.36 at abscissa) and it corresponds to the 0.975 quartile in the Student´s distribution, with seven degrees of freedom (total number of estimated coefficients subtracted from the total number of experiments). Thus, the effect of variables and their interactions is more significant as they are to the right of the red dashed line.

**Figure 5.** Pareto charts of the absolute values of the standardized effects of the factors for the regression model for (**a**) manganese extraction and (**b**) cobalt extraction with a significance level α = 0.05. Legend: *x*1: pH, *x*2: O:A ratio, *x*3*:* molar concentration of D2EHPA, (Q): quadratic terms, (L): linear terms.

The variables with higher effects on the manganese extraction were *x*<sup>2</sup> (O:A ratio), *x*<sup>1</sup> (pH), and *x*<sup>3</sup> (molar concentration of D2EHPA). The quadratic effect of the factor *x*<sup>2</sup> is also significant in the extraction of manganese. Then, it can be concluded that the extraction of manganese increases with the increase of the pH, extractant concentration, and the O:A ratio. The quadratic terms *x*<sup>1</sup> <sup>2</sup> and *x*<sup>3</sup> 2, as well as all the interactions, did not present a significant effect on the manganese extraction in the range of values tested in this work (at a confidence level of 95%).

Regarding the extraction of cobalt (Figure 5b), the main effects were accounted for the variables *x*<sup>1</sup> (pH), *x*<sup>2</sup> (O:A ratio) and the interactions of *x*1*x*<sup>2</sup> and *x*1*x*2*x*3, with a positive effect on the response with the increase of their levels. The quadratic terms *x*<sup>1</sup> 2, *x*<sup>2</sup> 2, and *x*3 2, the factor *x*<sup>3</sup> (molar concentration of D2EHPA), as well as the interactions *x*1*x*<sup>3</sup> and *x*2*x*<sup>3</sup> did not present a significant effect on the extraction of cobalt in the range of values considered for a confidence level of 95%.

The coefficient of determination (*R*2*)* was used to assess the goodness of fit of the models. The model for the manganese extraction presented an *R*<sup>2</sup> = 0.98 and for the cobalt extraction an *R<sup>2</sup>* = 0.96. This coefficient indicates that 98% and 96% of the response variability is explained by the fitted models, respectively. The relation between the experimentally observed responses for the extraction of manganese (Figure 6a) and cobalt (Figure 6b) is represented in the scatter plots below. This relation demonstrates that the adjusted models can provide a good fit to the experimental results under the range of values considered in the study.

**Figure 6.** Responses predicted by the model versus experimentally observed: (**a**) manganese extraction and (**b**) cobalt extraction.

#### *3.6. Response Surfaces: Extraction of Manganese and Cobalt*

To help to understand the effect of the different factors on the extraction of manganese and cobalt, response surfaces were used. They were depicted using contour plots to show a clear representation of the surfaces. Contour plots are represented by a set of lines of constant response, being constructed in planes defined by pairs of variables. Therefore, each line represents a particular response of the fitted model.

The contour plots representing the manganese extraction when the factor *x*<sup>1</sup> (pH) was fixed at its low level (−1, pH = 2.5), standard level (0, pH = 3.2), and high level (+1, pH = 4) can be seen in Figure 7a–c, respectively. The responses for the extraction of cobalt under these same conditions are represented in Figure 7d–f. To construct the contour plots, the level of the factors *x*<sup>2</sup> (O:A ratio) and *x*<sup>3</sup> (molar concentration of D2EHPA) was changed from the low to the high level. The responses (*y* = % extraction) are represented by legends on the left of each graph. Results are only valid in the range of values considered in this study.

The extraction of manganese when the pH was set at 2.5 is represented in Figure 7a. High manganese extractions can be achieved for any level of concentration of D2EHPA provided that the O:A ratio is also at a high level, which is explained by the highest effect of the O:A ratio on the response. At the lowest pH, the lowest extraction of manganese was verified at the lowest level of the O:A ratio (0.5:1) and at the lowest concentration of extractant (0.4 M). On the other hand, when the pH was 2.5, the highest extraction of manganese was observed at the highest level of the O:A ratio (2:1) and at the highest level of concentration of D2EHPA (0.6 M). However, when the pH was 2.5, the extraction of manganese never exceeded 70–80%, which can be explained by the mechanism of the reaction of D2HPA, by which an increase in the concentration of H<sup>+</sup> ions will move the equilibrium to the left side, hiding the formation of products. When the pH was set at 2.5, it is possible to observe in Figure 7d that the extraction of cobalt was kept at a very low level and never exceeded 5%, which was reached only when high concentrations of D2EHPA or high O:A ratios were employed.

The behavior of the extraction of manganese when the pH was 2 was similar to the one when the pH was 3.25, as can be observed in Figure 7b. However, the increase in the pH resulted in an increase in the highest extraction of manganese, which was raised to 80–90%. The lowest extraction of manganese at pH 3.25 was also obtained when the concentration of D2EHPA and the O:A ratio were at their lowest levels (0.4 M and 0.5:1, respectively). The highest extraction of manganese at pH 3.25 was achieved when the other two factors were at the highest level (0.6 M and 2:1). Extractions of manganese above 70% can be obtained for the whole range of values tested for the concentration of D2EHPA, provided that the O:A ratio is at least 1.4:1. When the pH was set at the standard level (3.25), the extraction of cobalt is mainly dependent on the O:A ratio (Figure 7e). Thus, it is possible to keep the coextraction of cobalt below 8% provided that the O:A ratio does not exceed around 1.4:1.

Contour plots representing the extraction of manganese when the pH was set at 4 can be seen in Figure 7c. The extraction of manganese reached higher values when the other two factors were combined at a higher pH, which is explained by the significant effect of the pH on the response, as it was discussed in the regression analysis. At the highest pH, the extraction of manganese was always above 50%. The lowest extraction was obtained when the concentration of D2HEPA and the O:A ratio were at the lowest level (0.4 M and 0.5:1, respectively). When both factors were increased to the highest level, the extraction of manganese achieved the maximum results. It is important to highlight that for certain conditions, the fitted model slightly overestimated the responses (above 100%). The coextraction of cobalt also increased to higher values when the pH was set at the highest level (Figure 7f), which is also compatible with the significant effect of the pH on the cobalt response, which was observed in the regression analysis. The highest coextraction of cobalt was observed when the concentration of D2EHPA and the O:A ratio were at their highest levels (0.6 M and 2:1, respectively) and achieved around 35%. At pH 4, the coextraction of cobalt remained at lower levels when both the O:A ratio and concentration of D2EHPA were set at lower levels.

Considering the results using the fitted models, to keep the coextraction of cobalt low even though obtaining high extractions of manganese, the pH, O:A ratio, and concentration of D2EHPA should be kept at intermediate levels. For this reason, the next stages (scrubbing and stripping) were studied using a loaded organic obtained at the central level of the tested factors (pH of 3.25, O:A 1.25:1, and 0.5 M D2EHPA). The concentration of the loaded organic obtained at these conditions to be used in the next stages was compatible with the results of the factorial design of experiments.

#### *3.7. Scrubbing of the Loaded Organic*

According to Ritcey and Ashbrook [36], scrubbing usually refers to the removal of unwanted coextracted species in the loaded organic. The purpose of scrubbing the organic phase is to replace coextracted or mechanically entrained Co, Ni, or Li together with Mn [20]. Although it can be considered an important stage to purify the loaded organic and selectively remove some undesired metals, the scrubbing stage was not studied in detail in this work, and the scrubbing conditions proposed by Peng et al. [20] were used. Thus, the loaded organic obtained using the standard conditions of the factorial design of experiments was scrubbed twice with a pure solution containing 4 g/L Mn prepared using MnSO4.H2O, without pH adjustment (pH: 4.4) for 10 min at an O:A ratio of 10:1. The final composition of the scrubbing solutions (1 and 2) after contact with the loaded organic and the resultant organic phase is presented in Table 5.

**Table 5.** Composition of the scrubbing solutions and the resultant organic phase after two scrubbing stages with 4 g/L Mn (O:A of 10:1, contact time of 10 min).


#### *3.8. Stripping Stage: Factorial Design of Experiments and Regression Model*

The experimental conditions of the factorial design for the stripping of the loaded organic and respective responses are presented in Table 6. The final concentrations of manganese and cobalt (g/L) in the stripping product were considered as the process responses. All experiments were performed at room temperature after two scrubbing stages (detailed in Section 3.7).

**Table 6.** Conditions of the experimental design and results for the stripping of cobalt and manganese.


The regression models for the stripping of manganese and cobalt are represented by Equations (7) and (8), respectively, and only factors with a statistically significant effect on the responses were inserted in the models (α = 0.05). The models are only valid for the range of values tested in this study.

$$\text{Mn (g/L)} = 16.9 + 3.4\mathbf{x}\_1 + 6.8\mathbf{x}\_2 + 2.0\mathbf{x}\_3 + 3.3\mathbf{x}\_1\mathbf{x}\_2 - 4.0\mathbf{x}\_1^2 - 3.1\mathbf{x}\_2^2 \tag{7}$$

$$\text{Co} \left( \text{g/L} \right) = 0.25 + 0.14x\_2 + 0.04x\_3 + 0.03x\_2x\_3 \tag{8}$$

The results of the analysis of variance of the models are presented in Table 7. The presence of curvature was verified only for the model representing the manganese stripping with the pure curvature testing (*p*-value = 0.04). The results of the *F*-test can be related to the significance of the fitted models. The model adequacy was assessed by the LOF test, but the result for the manganese stripping was lower than the significance level (α = 0.05), which can be related to the low experimental error in the central point of the design.


**Table 7.** Results of the analysis of variance of the fitted models for the stripping of manganese and cobalt.

Pareto charts of the standardized effects of the variables on the responses are presented in Figure 8. A significant effect on the stripping of manganese (Figure 8b) was accounted for the three main variables: *x*<sup>1</sup> (concentration of H2SO4), *x*<sup>2</sup> (O:A ratio), and *x*<sup>3</sup> (stripping time). The interaction effect of *x*<sup>1</sup> and *x*<sup>2</sup> was also significant, as well as the effect of the quadratic term *x*<sup>1</sup> 2. Thus, the stripping of manganese will increase with the increase of the levels of these three variables. The quadratic terms *x*<sup>2</sup> <sup>2</sup> and *x*<sup>3</sup> 2, as well as all the other interactions, did not have a significant effect on the manganese stripping, considering the range of values tested at a confidence level of 95%. Only the variables *x*<sup>2</sup> (O:A ratio) and *x*<sup>3</sup> (stripping time) had a positive and significant effect on the stripping of cobalt (Figure 8b). Thus, the concentration of acid did not show a significant effect on the stripping of cobalt in the tested range nor did it have all the interactions and quadratic terms (at a confidence level of 95%).

**Figure 8.** Pareto charts of the absolute values of the standardized effects of the factors for the regression model for the (**a**) manganese stripping and (**b**) for the cobalt stripping. Significance level α = 0.05. Legend: *x*1: molar concentration of H2SO4, *x*2: O:A ratio, *x*3: stripping time, (Q): quadratic terms, (L): linear terms.

Both models presented an *R2* = 0.97, which is indicative that a large proportion of the variance of the response can be explained by the independent variables, considering the range of values tested in the experiments. The relation between the experimentally observed responses and those obtained using the fitted model for the stripping of manganese and cobalt are represented in Figure 9a,b, respectively, which illustrates how the models provide a good fit to the experimental results.

**Figure 9.** Responses predicted by the model versus experimentally observed: (**a**) manganese stripping and (**b**) cobalt stripping.

#### *3.9. Response Surfaces: Stripping of Manganese and Cobalt*

The contour plots in Figure 10a–c represent the response surfaces of the manganese stripping when the factor *x*<sup>2</sup> (O:A ratio) was set at its low level (−1, O:A = 1:1), standard level (0, O:A = 4.5:1), and high level (+1, O:A = 8:1), respectively. The stripping of cobalt for different combinations of O:A ratio and time is represented by the contour plots in Figure 10d, given that the concentration of sulfuric acid did not have a significant effect on it. The values of the response (*y*) are represented by legends on the left side of each graph. Results are only valid in the range of values considered in this study. The concentrations of metals remaining in the organic phase and in the stripping product for each test are reported in the Supplementary Materials (Table S3).

When the O:A used in the stripping was 1:1 (Figure 10a), a low concentration of manganese was obtained and never exceeded 10 g/L, which was expected given the larger volume of aqueous phase. At the lowest concentration of H2SO4 (0.05 M), the lowest concentration of manganese in the stripping product was verified at the lowest stripping time (2 min), being lower than 3 g/L Mn. With the increase in the concentration of H2SO4 and in the leaching time, a slight increase in the concentration of manganese was observed (maximum of 10 g/L).

The stripping behavior of manganese when the O:A ratio was set at 4.5:1 can be observed in Figure 10b. At this O:A ratio, the lowest concentration of manganese was around 8–10 g/L, and it was reached when the concentration of H2SO4 was the lowest (0.05 M) at the shortest stripping time (2 min). Increasing the concentration of acid from 0.9 to 2 M and the stripping time from 15 to 25 min promoted an increase in the concentration of manganese, which reached around 20 g/L.

The concentration of manganese was the highest when the O:A ratio was set at 8:1 (Figure 10c) and it was higher than 10 g/L for all tested conditions. The concentration of

manganese reached higher values when the other two variables (time and concentration of acid) were combined at the highest O:A ratio, which is related to the highly significant effect of the O:A ratio on the response, as previously discussed in the regression analysis. When the concentration of acid was at the lowest level (0.05 M) and the stripping time was also at the lowest level (2 min), the concentration of manganese was around 10 g/L. When both factors were increased to their highest levels, the concentration of manganese achieved the maximum results (23–25 g/L). In Figure 10a–c, it is also possible to observe how the concentration of acid (*x*1) has a more pronounced effect on the concentration of manganese in the stripped product, which was also represented by a quadratic term in the model, causing a curvature in the response surface. Thus, a slight increase in the concentration of acid can cause a higher effect on the concentration of manganese.

The stripping of cobalt (Figure 10d) was mainly affected by the O:A ratio and by the leaching time, while the concentrations of H2SO4 tested in this study did not have a significant effect on the concentration of cobalt in the stripped liquor. The concentration of cobalt increased along with the O:A ratio and the stripping time, but it never exceeded 0.5 g/L. Thus, it can be concluded that very high concentrations of manganese in the stripping product (>23 g/L) can be obtained using high O:A ratios and concentrations of sulfuric acid of around 1 M. However, the stripping time should not exceed around 13 min, in order to keep the concentration of cobalt at a low level (<0.3 g/L). Additionally, the fitted models can support the optimization of the stripping process.

**Figure 10.** Contour plots representing the (**a**–**c**) stripping of manganese (**a**) when the O:A was set at 8:1, (**b**) when the O:A ratio was set at 4.5:1, and (**c**) when the O:A ratio was 1:1. (**d**) represents the stripping of cobalt at different combinations of stripping time and O:A ratios.

The fitted models can help to optimize the solvent extraction of manganese and can also assist with the construction of distribution isotherms and McCabe–Thiele diagrams, which are very helpful to predict the distribution of metals in both phases of the system (aqueous and organic) and to theoretically determine the number of required stages. The distribution isotherms for the stripping of manganese and cobalt, whose results were determined using the fitted models, are presented in the Supplementary Materials (Figure S1).

#### **4. Conclusions**

The recovery of manganese from a solution based on lithium-ion batteries was investigated using the factorial design of experiments and the response surface methodologies in order to assess the effect of different factors on the solvent extraction of manganese. These methodologies were also used to optimize the extraction and stripping stages, aiming to minimize the coextraction of cobalt. Preliminary tests were performed to determine the experimental conditions to be further investigated in the factorial design of experiments. The use of a modifier (TBP) was tested, but the formation of a third phase was not observed, and for this reason, additional tests with a modifier were not performed. The extraction of manganese using D2EHPA was fast, and maximum results were achieved after 10 min of contact time.

The factors evaluated in the extraction stage were the equilibrium pH, the molar concentration of D2EHPA, and the organic to aqueous ratio. Under optimized conditions (O:A of 1.25:1, pH 3.25, and 0.5 M D2EHPA), extractions above 70% Mn were reached in a single extraction stage with a coextraction of around only 5% Co, which was mostly removed in two scrubbing stages. Other combinations of factors can also result in high extractions of manganese and low coextractions of cobalt. In general, the coextraction of lithium and nickel remained low. The variables considered for the optimization of the stripping stage were the concentration of sulfuric acid, the organic to aqueous ratio, and the stripping time. A stripping product containing around 23 g/L Mn and around 0.3 g/L Co can be obtained under optimized conditions (O:A of 8:1, 1 M H2SO4, and around 13 min of contact time) in a single stripping stage. Increasing the number of extraction stages can promote an increase in the concentration of manganese loaded in the organic phase and should be further investigated in up-scale tests using mixer-settlers. Moreover, the fitted models for the extraction and stripping stages can help optimize these processes and can also assist with the construction of McCabe–Thiele diagrams to predict the number of stages required to maximize the recovery of manganese.

The results obtained can support further investigations on the recovery of manganese from spent lithium-ion battery solutions, which are an important secondary resource of manganese, using solvent extraction with D2EHPA. Moreover, the use of methodologies to model and optimize the process can assist the process management, considering that multiple combinations of factors can result in high extractions of manganese and low coextractions of other metals. Knowing these alternatives can help to better design the process to reduce the consumption of energy and reagents, minimizing costs and environmental impacts.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/2075-4 701/11/1/54/s1,Table S1. Conditions of the experimental design and concentrations of metals in the raffinate and in the organic phase after one extraction stage. Contact time of 10 min. Legend: [aq]: concentration of metal in aqueous phase, [org] concentration of meta in organic phase, Table S2. Conditions of the experimental design, distribution ratios (*D*) and separation factors (β) after one extraction stage. Contact time of 10 min, Table S3. Conditions of the experimental design and concentrations of metals remaining in the organic phase and in the stripping product. Legend: [aq]: concentration of metal in aqueous; phase, [org] concentration of metal in organic phase, Figure S1. Distribution isotherms of (**a**) manganese stripping and (**b**) cobalt stripping obtained using the fitted models. Conditions used as input in the fitted models: stripping time: 13.5 min (coded variable: 0), O:A ratio: 8:1 (coded variable: +1), concentration of H2SO4: 1 M (coded variable: 0).

**Author Contributions:** Data curation, formal analysis, investigation, visualization, writing—original draft preparation, N.V.; Conceptualization, methodology, validation, writing—review and editing, N.V., N.R., C.E., M.P.; resources, N.R., C.E., M.P.; project administration and supervision, M.P.; funding acquisition, C.E., M.P.; All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was supported by VINNOVA (reference number of project: 2019-02069).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data presented in this study are available in supplementary material here.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


### *Article* **A Novel Pyrometallurgical Recycling Process for Lithium-Ion Batteries and Its Application to the Recycling of LCO and LFP**

**Alexandra Holzer \*, Stefan Windisch-Kern, Christoph Ponak and Harald Raupenstrauch**

Chair of Thermal Processing Technology, Montanuniversitaet Leoben, Franz-Josef-Strasse 18, 8700 Leoben, Austria; stefan.windisch-kern@unileoben.ac.at (S.W.-K.); christoph.ponak@unileoben.ac.at (C.P.); harald.raupenstrauch@unileoben.ac.at (H.R.)

**\*** Correspondence: alexandra.holzer@unileoben.ac.at; Tel.: +43-3842-402-5803

**Abstract:** The bottleneck of recycling chains for spent lithium-ion batteries (LIBs) is the recovery of valuable metals from the black matter that remains after dismantling and deactivation in pretreatment processes, which has to be treated in a subsequent step with pyrometallurgical and/or hydrometallurgical methods. In the course of this paper, investigations in a heating microscope were conducted to determine the high-temperature behavior of the cathode materials lithium cobalt oxide (LCO—chem., LiCoO2) and lithium iron phosphate (LFP—chem., LiFePO4) from LIB with carbon addition. For the purpose of continuous process development of a novel pyrometallurgical recycling process and adaptation of this to the requirements of the LIB material, two different reactor designs were examined. When treating LCO in an Al2O3 crucible, lithium could be removed at a rate of 76% via the gas stream, which is directly and purely available for further processing. In contrast, a removal rate of lithium of up to 97% was achieved in an MgO crucible. In addition, the basic capability of the concept for the treatment of LFP was investigated whereby a phosphorus removal rate of 64% with a simultaneous lithium removal rate of 68% was observed.

**Keywords:** lithium-ion batteries (LIBs); recycling; pyrometallurgy; critical raw materials; lithium removal; phosphorous removal; recovery of valuable metals

#### **1. Introduction**

The development of lithium-ion batteries (LIBs) has experienced an enormous upswing in recent years, which is, in addition to portable devices, mainly due to the steadily increasing demand in the electric vehicle (EV) sector. According to forecasts, this trend will continue in the coming years [1,2]. Further prognoses predict that sales of LIBs are expected to increase from 160 GWh in 2018 to over 1.2 TWh in 2030 [1]. Their use in electrical appliances, EVs and stationary storage is due to their advantages over other storage media, such as high energy density, long service life and high operating voltage [3,4]. Since consumed LIBs contain a large number of valuable metals, recycling has a considerable environmental impact in view of the conservation of valuable resources [5]. In addition to this idea of resource protection, waste reduction and the energy-efficient and economical treatment of hazardous substances are also driving recycling efforts [6]. The timeliness and necessity of recycling LIBs is further underlined by the 2020 list of critical raw materials published by the European Commission. Among others, cobalt, lithium and phosphorus can be found [7].

A major challenge with regard to recycling is posed by the strongly fluctuating waste stream. This is the product of the requirements of the countless applications for energy storage and the resulting multitude of electrode materials of LIBs [8]. In the respective literature there is a variety of different recycling processes, which can basically be divided into preparation for recycling, pre-treatment and main processing, including pyro- and hydro-metallurgy. In the first mentioned area, the processes of discharging and dismantling can be found [5]. The aim of the pre-treatment is to improve the recovery rate, to adapt

**Citation:** Holzer, A.; Windisch-Kern, S.; Ponak, C.; Raupenstrauch, H. A Novel Pyrometallurgical Recycling Process for Lithium-Ion Batteries and Its Application to the Recycling of LCO and LFP. *Metals* **2021**, *11*, 149. https://doi.org/10.3390/met11010149

Received: 11 December 2020 Accepted: 11 January 2021 Published: 14 January 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

the waste stream to the downstream process step and to reduce the energy consumption of the following pyro- or hydro-metallurgical process [6,9]. In Europe, there are several companies that already perform the preparation and pre-treatment of spent LIBs on a larger scale, like Accurec Recycling GmbH, Duesenfeld GmbH or Redux GmbH [10–12]. The latter starts the recycling process with collection and temporary storage, followed by manual sorting. As of this point in time there is still a considerable safety risk due to the residual charge of the LIBs. They are completely discharged, and the energy gained is fed back into the operating network. Subsequently, components such as electronics, cables, plastics, aluminum, and iron are dismantled and sorted. During the subsequent deactivation, the coating of the conductor foils is dissolved and the separator as well as the electrolyte are removed. During the mechanical treatment, the remaining components such as iron, aluminum, copper and the fine material (also called active material or black matter) of cathode and anode material are separated. The separation of the individual fractions is carried out with a magnetic separator, air separator and sieving [13]. The resulting black matter can be further treated in a pyro- or hydro-metallurgical process.

In pyrometallurgical treatment of LIBs, the physiochemical transformation temperatures above 1400 ◦C are used to recover the valuable metals [14]. As a partial step in an overall process, pyrometallurgy is a suitable instrument for purifying the feed stream of substances undesirable for hydrometallurgy. Fluorine, chlorine, graphite, phosphorus, etc., pose a particular challenge to hydrometallurgy. Pyrometallurgical processes are generally robust against impurities and organic contaminants, because volatile components can be evaporated [5]. Graphite from the anode can be used as a reducing agent and burned in various processes in the presence of oxygen, thus helping to maintain the process temperature. Since the reaction kinetics in pyrometallurgical processes increase extremely due to the high temperatures, productivity is higher compared to hydrometallurgy [15]. Although the large number of research activities in recent years has focused on hydrometallurgy [9], there is significant scientific output in the field of pyrometallurgy, some of which is already being applied on an industrial scale. Several recent reports claim that large-scale pyrometallurgical processes have greater potentials in terms of sustainability than their hydrometallurgical counterparts [16–21]. Industrial scale processes are those that have more than 1000 t/a recycling capacity. In Europe, the companies Umicore, Accurec and Nickelhütte Aue should be mentioned here, and outside the EU, for example, SungEel, Kyoei Seiko and Dowa. The overall processes usually lead via a mechanical and/or thermal step to pyro- and hydro-metallurgy [5]. The pyrometallurgical step is typically based on shaft furnaces or electric arc furnaces for melting this feedstock [22]. A direct comparison of the recycling efficiency of the individual processes is often very difficult, since the reference basis of the values given is usually not given or only partially given. However, it can be stated that recycling routes which include a pyrometallurgical step have the highest overall recycling efficiency, in some cases exceeding 50% [5]. Since pyrometallurgical processes are operated at high temperatures, their energy requirements are correspondingly high. In addition, large quantities of waste gas are produced which have to be treated. A disadvantage of current pyrometallurgical processes is the slagging of lithium, the recovery of which in turn requires an enormous hydrometallurgical effort [9,23]. The economic efficiency of lithium recovery depends on the concentration in the slag. As a rule, in the co-processing of LIBs in metallurgical plants, the lithium is diluted to such an extent that recovery is not economically feasible [24]. In recent years, a number of advances have been made in the field of slag post-treatment. These research ventures on a non-industrial scale focus, for example, on the concentration of Li in the slag by selective addition of slag-forming agents during the pyrometallurgical process and subsequent hydrometallurgical treatment [25,26]. Recent progress has also been made in the area of early-stage lithium extraction. In this process, sulphate roasting treatment was used to convert the cathode material from NMC batteries into a water-soluble lithium sulphate (Li2SO4) and a water-insoluble oxide (NiCoMn-oxide) [5]. However, depending on the price of lithium, processes specially developed for LIB recycling may in future be

quite economical in terms of lithium recovery [24]. Various advantages and disadvantages also result from the different interconnection types of the overall process. For example, the primary energy consumption via pyrometallurgical routes is higher, but the resulting additional costs are more than compensated by lower operating costs in the hydrometallurgical step [5]. The recycling of P from LIBs is described in the literature in very few publications. Most of them are related to the hydrometallurgical process route, other processes deal with the regeneration of the cathode material [27].

Hydrometallurgical processes are highly selective and can therefore achieve high purities [15]. Leaching is the key process in hydrometallurgy to convert the metals to ions in a solution. This can be divided into bio leaching with metabolic excrements of microorganisms or fungi and chemical leaching with organic or inorganic acids [28–30]. Subsequently, the valuable metals are separated and recovered from the leaching solution. Since the structure of the leaching solution is complicated, it is usually necessary to use several different methods from the portfolio of solvent extraction, chemical precipitation and electrochemical deposition [28]. Hydrometallurgical methods result in extremely good recycling rates of up to 100% [28,31]. They also require a high level of equipment and a large number of process steps, which usually results in a correspondingly high volume of polluted wastewater. In order to operate the process economically, it is very important to separate and concentrate as many metals and impurities as possible in advance. For each additional metal, at least 1–2 additional process steps would be required, which is only economical if the metal value or quantity is correspondingly high [15].

Especially with regard to the raw materials contained in LIBs, which are included in the list of critical raw materials of the European Commission [7], and from an ecological point of view, a sustainable handling of spent LIBs is essential. According to Elwert et al. [32], recycling processes specialized in LIBs will gain more and more importance in the future. This is due to the increasing rate of return of spent LIBs to the waste stream, more regulations by the authorities and also decreasing amounts of valuable nickel and cobalt for direct use in nickel and cobalt producing plants. Furthermore, the growing market for LFP and the increasing interest in lithium recovery also plays a major role. Of particular importance in terms of regulation is the recently published European Commission proposal to revise EU Directive 2006/66/EC, which sets recovery rates of up to 70% for Li and 95% for other valuable metals such as Co, Ni and Cu by 2030 [33], which forces recyclers to increase recovery rates and their process efficiency.

It can be summarized that there is a multitude of different recycling processes and methods, which are characterized by their positive properties in certain areas but also have individual disadvantages. In the field of pyrometallurgy, lithium slagging and in particular the absence of possibilities to recover from the slag with reasonable effort can be identified as a bottleneck.

The novel pyrometallurgical recycling process presented in this paper is characterized by the recovery of an alloy with a simultaneous utilization of lithium and phosphorus via the gas flow. The following points provide a more detailed insight into the theoretical considerations and practical implementations for the most efficient recovery of valuable metals from LIBs using this process. Initially, appropriate analyses were carried out to better understand the behavior of cathode materials in high-temperature applications under reducing conditions. To determine the lithium removal rate without the presence of phosphorus, the cathode material lithium cobalt oxide (LCO) was examined in an experiment. In addition to the successive optimization of the reactor concept and adaptation to the waste stream from spent LIBs, another experiment with LCO in a modified setup was performed and compared to the previous one. To verify the basic suitability of the pyrometallurgical apparatus for the simultaneous removal of phosphorus and lithium via the gas flow, experiments were carried out with the cathode material lithium iron phosphate (LFP).

#### **2. Process Concept and Methods**

#### *2.1. Used Materials*

In total, three different experiments were carried out with two types of feedstock. As Windisch-Kern et al. [34] have already described, experiments with black mass from a preprocessing step have already shown that lithium could be removed to a considerable extent. As this has raised additional questions, a detailed investigation of the pure cathode materials, i.e., LCO, lithium nickel manganese cobalt oxide (NMC—chem., LiNi0.33Mn0.33Co0.33O2), lithium nickel cobalt aluminum oxide (NCA—chem., LiNi0.8Co0.15Al0.05O2) and LFP, was indispensable. For the purpose of clarifying the questions dealt with in this paper, the materials LCO and LFP were used, which were produced by the Chinese company Gelon Energy Corporation. The appearance of this feedstock can be described as a fine, black powder. Since this carbo-thermal process requires a reducing agent and the graphite bed in the reactor is only used for energy input, graphite powder from coke pellets of a steel mill is added. The graphite cubes with a side length of 2.5 cm come from electrodes of a steel mill and have an average purity of 99% with an electrical resistance of 4–8 μ Ωm and a density of 1.55–1.75 g cm<sup>−</sup>3. [35] The amount of graphite powder required for the reduction was determined by stoichiometry of the respective cathode material. For this purpose, the weighed mass of the cathode material was multiplied by the molar ratio of LiCoO2 or LiFePO4. After determining the moles O by multiplying the mass O by the relative atomic mass of O, the necessary mass of C was calculated by using the relative mass of C and assuming that a conversion to CO takes place in the reactor. The corresponding percentage C requirement is finally obtained by a rule of three of the masses of O and C. Table 1 shows the composition of the input materials determined from their stoichiometric composition.

**Table 1.** Composition of the mixture of cathode material and graphite powder in wt.%.


The products obtained from the experiments were examined by ICP-OES and ICP-MS by means of aqua regia digestion according to ÖNORM EN 13657:2002-12.

#### *2.2. Material Specific Investigations*

Since the behavior of the individual cathode materials at high temperature applications is hardly or not at all described in the literature, detailed investigations were undertaken at the Chair of Thermal Processing Technology. These included analyses in a Hesse Instruments EM 201 with an HR18-1750/30 furnace heating microscope. The results should be used for planning the process control in the following experiments in the inductively heated reactor, which is presented in Section 2.3. Furthermore, a better understanding of the behavior of the cathode materials should be gained. To be able to simulate the planned process as detailed as possible, graphite powder was added to the cathode material. The addition of graphite powder was carried out to an extent of 10 wt.% under the assumption that C is converted to CO2 and transported away via the argon-purged atmosphere. The mixture of the corresponding cathode material and graphite powder was examined with at least one reproduction experiment. For better comparability a uniform heating rate was always set, which corresponds to the maximum possible with the heating microscope used. This is primarily to ensure the shortest possible residence time in the furnace chamber since interactions of LCO with the furnace material consisting of Al2O3 have been determined and damage to this should be prevented as far as possible. Up to a temperature of 1350 ◦C, a heating rate of 80 ◦C/min was selected, from 1350 to 1450 ◦C 50 ◦C/min and up to 1700 ◦C a heating rate of 10 ◦C/min with a holding time of 15 min at 1700 ◦C was dialed. To avoid oxidation with the ambient air, the reactor was flushed with argon at a flow rate

of 2 L/min. A maximum furnace temperature of 1700 ◦C was chosen, which allows for an approximate sample temperature of 1630 ◦C.

Figure 1 illustrates the standardized sample preparation. The material is centrally positioned in a cylinder with a diameter of 3.5 mm and a height of 2.5 mm on an Al2O3 platelet with an approximate weight of 0.1 g.

**Figure 1.** Structure of the sample on the white sample plate made of Al2O3 before the trial.

#### *2.3. Reactor Concept*

The novel reactor concept, which was constructed at the Chair of Thermal Processing Technology of the Montanuniversitaet Leoben, is based on the inductive heating of graphite pieces in a packed bed reactor. The cornerstone of knowledge generation in this field was laid at the chair already in 2012, by the EU subsidized project RecoPhos for the pyrometallurgical treatment of sewage sludge ash for simultaneous recovery of phosphorus and the contained valuable metals. Its results are described by Schönberg et al. [36] and Samiei et al. [37]. Based on this and corresponding follow-up projects, also in the field of basic oxygen furnace slag (BOFS) treatment, a batch operated post-lab-scale plant and a pilot-scale plant as a continuous process have been developed and built. This knowledge advantage was used to adapt the mechanism for pyrometallurgical recovery of valuable metals from processed LIB material in two ways. On the one hand, the theoretical idea of the continuous reactor, which should be conceptually similar to the set-up from research work in the field of sewage sludge ash and BOFS utilization, is applied. On the other hand, the post-lab-scale setup developed in the subject area mentioned above can initially be used for first experiments without further adaptations. In the long term, the realization of larger scales and corresponding throughput of recycled material as a continuous unit is planned. Intensive research activities on a small scale are indispensable for the most efficient implementation of gradual scale-ups to industrial maturity. In view of the process development as well as the knowledge gained about the input material, the previously mentioned apparatus in batch operation, the so-called InduMelt plant, is used for this purpose. These two process concepts and their respective challenges and developments are explained in detail below.

#### 2.3.1. Continuous Reactor Concept

In order to treat the expected future waste stream from used LIBs, a technology with correspondingly high throughput rates is required. The currently pursued approach at the Chair of Thermal Processing Technology is based on a continuous reactor concept which currently exists as a pilot plant with a material throughput rate of 10 kg/h. Even if, according to initial findings from investigations of the LIB black matter, the design must differ from that used for sewage sludge ash and BOFS, the basic principle remains the same. Ponak [38] describes the so-called InduRed reactor as an cylindrical arrangement of refractory materials filled with pieces of electrode graphite which allow a horizontal and radial homogeneous temperature distribution when heated by the induction coils, as seen in Figure 2.

**Figure 2.** Schematic illustration of the InduRed reactor for the continuous treatment of sewage sludge ashes and BOFS [39].

The process starts with the material feed from a feed vessel above the reactor via a screw conveyor and a low volume of argon. Inert gas purging is highly relevant at this position, especially at higher temperatures, as the graphite bed is protected against oxidation by possible false air and, mainly, to direct small ash or black matter particles directly onto the graphite surface. In the first zone, the melting zone, the fine-grained material inserted is heated to melting temperature without reaching the reduction temperature of the critical component phosphorus. The resulting molten film then moves through the reactor to the reduction zone. In this zone the corresponding energy is induced so that the reduction temperature is reached close to the implemented gas flue. At this point, it has to be mentioned that the graphite pieces are not supposed to participate in the reduction reactions and serve only as a susceptor material. Added carbon powder functions as a reductant. Through this reaction process, phosphorus is converted into the gaseous phase and can be removed directly from the reactor via the gas flue by means of a negative pressure-generating induced draft fan. Downstream there is a post-combustion chamber in which external air or oxygen are used to convert elemental phosphorus to P2O5. The subsequent hydrolysis finally enables the production of phosphoric acid. The remaining material in the reactor moves on to the discharge zone, where the third and last coil provides enough energy that the phosphorus-free material does not reach the solidification temperature and finally leaves the reactor via the reactor floor. The resulting material can be divided into a metal and a slag fraction, which, however, are not yet separated from each other in the current expansion stage and are collected in a vessel below the reactor output.

The advantages of this apparatus are manifold. In comparison with an electric arc furnace (EAF), no molten bath of metal is formed so that the P2 (g)–Fe (l) contact possibility and in further consequence the formation of iron phosphide can be decreased immensely. This fact is promoted by a thin molten film, which massively shortens the distance of mass transport. In this case it is particularly important for the diffusion of P and its removal as gas. The graphite pieces offer a large surface area for reactions and by coupling into the induction field, the heat for the endothermic reduction reactions is permanently provided directly at the respective particle surface. Even if the energy demand is increased, the main form of the reduction reaction is direct reduction, resulting in a lower carbon demand. [38] In the course of the reaction processes in the reactor, a very low oxygen partial pressure and a correspondingly high CO to CO2 ratio is established, which in turn promotes the reduction reactions [40].

In order to use this process also for the waste stream from spent LIBs, the reactor design and the corresponding post-treatment of the output streams must be adapted. The input material for the planned continuous process comes from a pre-treatment plant, which is a fine fraction as low in Cu and Al as possible consisting of a mixture of cathode and anode materials. After being fed into the pyrometallurgical reactor, the material should react according to the principle described above. The most important difference is that the idea of the treatment of this material is to remove not only phosphorus but also lithium from the reactor via the gas flow. An initial concept for the post-treatment of the liquid fraction, which leaves the reactor chamber via its bottom, provides for an oxygen inlet. Thus, in accordance with the different oxygen affinities, for example, the input stream of NMC, LCO, NCA and LFP should result in the purest possible CoNiFe alloy. Oxygen-affine elements such as Mn and Al, as well as the residues of P and Li that are not removed via the gas phase, are to be slagged. The resulting products can therefore either be sold on the market as raw materials as required, or further broken down into their constituent parts in further post-treatment steps, for example via the hydrometallurgical route. A further additional important step to be investigated is the post-treatment of the resulting gas fraction. In particular, it will be necessary to implement a corresponding process for the separation of Li and P and consequently to treat them further according to the resulting qualities. As the points just described show, a combination with other processes should be aimed for. With regard to the overall reactor design, an adaptation of the current development will be essential. This includes issues such as the optimal refractory material for the reactor wall or a possible need to expand the gas extraction system.

#### 2.3.2. Batch Reactor Concept

Based on the technology described in Section 2.3.1, the process design shall be adapted to the requirements of the black matter out of LIBs. For this purpose, in-depth tests were carried out for a better understanding of the material to be processed, which are partly described in Windisch-Kern et al. [34]. Since the scale of a continuous pilot plant for experiments of this kind would firstly be too complex and secondly would not correspond to the research status at the Chair of Thermal Processing Technology in the LIB field, the experiments were carried out on a post-laboratory scale. For this purpose, the reactor concept of the unit operating in batch mode was adopted from the developments in the field of sewage sludge ash and BOFS, as shown in Figure 3a and hereinafter referred to as Design 1. The system behind it is similar to the continuous concept, with the difference that the material to be investigated is already in the reactor at the beginning of the experiment and there is no material output via the ground. Most of the material melted during the experiment is accumulated and collected at the bottom of the reactor or adheres to the cube surface as spherical formations. In addition, the gas outlet is also not subjected to negative pressure, so that the resulting gases leave the reactor without constraint. For the construction of the reactor, an Al2O3 ring with a diameter of 20 cm, Al2O3 mortar and refractory concrete were used. The graphite bed provides a cube surface of 1725 cm2 for the transfer of the induced heat. An insulation around the reactor has the function of the protection of the induction coil, to reduce the heat losses and to enable as good a separation as possible from ambient air.

To enable a qualitative measurement of the exhaust gas flow, a gas scrubber was additionally installed at the outlet of the gas flue (Figure 3b). This was realized with a bubbling frit in which the exhaust gas is enriched in a 2.5 molar H2SO4 solution. The temperature was measured on the outside of the reactor by two category S-thermocouples and inside the reactor by two category K-thermocouples. Due to the expected breakage of the second mentioned thermocouples, they are only used to find a correlation between the outside temperature and the inside temperature.

Preliminary tests have shown that LCO, with its high cobalt content, is highly reactive to the crucible material of Al2O3. In addition, sampling proved to be particularly difficult because it was not possible to separate the black mass clearly from the mortar. This makes it almost impossible to close the mass balance in the future. Another disadvantageous fact of this reactor concept is that, due to its position, the highest induction of the current takes place in the upper part of the reactor. Because of the inevitable turbulence in the reactor during the experiment due to the gases, the material accumulates at the bottom of the reactor, so the energy supply position is suboptimal. To take into account the mentioned disadvantages, a new design was developed, which is shown in Figure 4 and hereinafter referred to as Design 2.

**Figure 3.** (**a**) Schematic representation of the original InduMelt plant (Design 1) [34]; (**b**) overall setup in test operation.

**Figure 4.** Schematic representation of the new reactor design (Design 2).

This is a cylindrical crucible with a half-arc bottom made of MgO. It was placed centrally on a refractory concrete structure in a way that only the lower part of the MgO crucible is within the induction coil. Appropriate insulation made of refractory matting should reduce the heat loss and thus the energy requirement and protect as far as possible against the ingress of false air from the environment. To be able to make a qualitative statement about the escaping gas during the test, an exhaust pipe made of Al2O3 was again implemented. The temperature was measured by a category S-thermocouple from below and in the reactor by two category K-thermocouples.

For a direct comparison of the different reactor concepts of the InduMelt plant, the same feedstock, the mixture LCO-C mentioned in Section 2.1 with a quantity of 550 g, was examined in both crucible concepts. To ensure that reproducible initial conditions prevailed in both designs, the charging of the cubes and the sample was also performed uniformly in all experiments. Thus, at the beginning, 15 cubes were positioned in the reactor and one third of the sample was charged onto them. After positioning a K-thermocouple, another 10 cubes were performed followed by addition of another third of the sample. This was repeated a second time to finally fill the reactor with 11 cubes after positioning the second K-thermocouple. This filling quantity also represents the maximum possible capacity of Design 1. The content of Design 2 is approximately 25% larger which, however, was not utilized due to the aforementioned comparability with Design 1. After the experiment, all components of the reactor are weighed. The adhesions to the graphite cubes are removed by light mechanical processing. These adhesions are consequently separated into fractions larger and smaller than 1 mm by means of a sieve tower, together with the remaining finer fraction that may be produced. With the aid of a magnet, these are further separated into magnetic and non-magnetic, with the former finally being assigned to metal and the latter referred to as slag. Larger pieces of metal are collected together after checking with a magnet. The same is done with larger non-magnetic pieces, which are again referred to as slag. The individual fractions are finally weighed and analyzed. The main difference in sampling between Design 1 and Design 2 is the collection of the diffused areas of the grout or reactor adhesions, which will be discussed in more detail in Section 3.2.1.

The aim is to determine the interaction between the cathode material respectively the reaction products of which and the corresponding crucible material and to compare them with each other. On the other hand, the individual transfer coefficients should provide information on whether the choice of the crucible material affects the recovery rates of the individual species.

In a third trial, the basic suitability of the overall reactor concept for the treatment of LFP with the aim of removing Li and P from the material was investigated. For this purpose, Design 1 from Figure 3a was selected again in which a quantity of 394.5 g of the mixture LFP-C from Section 2.1 was charged into the reactor.

#### **3. Results and Discussion**

#### *3.1. High-Temperature Properties of the Cathode Materials Used*

In a first step to determine the behavior of cathode materials from LIBs at temperatures above 1600 ◦C and under reducing conditions, experiments were performed in a heating microscope. Figure 5a provides a picture of the result of the experiment with the mixture with LCO. A strong dark blue coloration was observed on the platelet. This may be due to a reaction between the Al2O3 platelet and cobalt to form cobalt aluminate with its typical blue appearance [41]. The product of the melting process under reducing conditions is a metal structure, which can be classified as strongly magnetic after examination with a magnet. This magnetism could also be detected in experiments with LFP, which is shown in (b). In contrast to the experiment described above, there is no blue coloration, but a brown to reddish appearance.

**Figure 5.** Condition of the sample during examination under the heating microscope: (**a**) after analysis for LCO-C; (**b**) after analysis for LFP-C.

Figure 6 displays the recording of the replication experiment LCO-C via optical measurement in the heating microscope. In Figure 6 the cross-sectional area of the experiments with LCO-C (black dotted line) and its repetition LCO-C-Re (green dotted line) over temperature during the experiment is shown. This cross-sectional area is the size of the sample cylinder detected by the heating microscope respectively its change with temperature increase.

**Figure 6.** Results of the heating microscope of LCO-C: trend of the cross-sectional area of the experiments LCO-C and LCO-C-Re during heating and the median value of the both graphics.

It should be mentioned that by comparing the recorded images with the corresponding values of the cross-sectional area, faulty measurements caused by incorrect detection of the baseline by the heating microscope were removed from the data series. Since the basic behavior at the individual temperatures is nearly identical and differs only by different cross-sectional areas, a mean value was determined, which represents the red line. When looking at Figure 6, the first noteworthy surface changes can be observed from 675 ◦C onwards. From 675 ◦C to 845 ◦C a growth of the cross-sectional area was detected, which subsequently decreased again to 1054 ◦C with single deflections to about 80% of the original area. Up to a temperature of 1127 ◦C an increase in magnification was detected, which remained relatively constant with single deflections up to 1380 ◦C. From this temperature on, the cross-sectional area decreased continuously with a smaller slope in the range of 1393 ◦C to 1507 ◦C and a significant decrease up to 1525 ◦C. Up to the end there was a further decrease of the cross-sectional area, which, however, when looking at the single images from the heating microscope, can be traced back to the continuous distribution of the molten material on the platelet.

Figure 7 illustrates the results of the experiments in the heating microscope with the mixture LFP-C. The previously mentioned measurement error was particularly striking in the first experiment of LFP-C (black dotted line in Figure 7) in the range from 1163 ◦C to about 1400 ◦C. Nevertheless, a corresponding trend could be determined by correctly measuring individual values in some cases, which in turn could be confirmed by a repeated measurement of LFP-C-Re (green dotted line in Figure 7). Again, the mean value is shown in the red curve.

The results show that up to a temperature of approximately 920 ◦C the cross-sectional area first rises slightly and then falls back to just below 100% of the initial value. Afterwards, a pulsating enlargement of the surface takes place which decreases at about 1200 ◦C. After a further pulsating behavior between 1240 and 1310 ◦C the area is continuous again to remain relatively constant from about 1410 ◦C on.

**Figure 7.** Results of the heating microscope of LFP-C: trend of the cross-sectional area of the experiments LFP-C and LFP-C-Re during heating and the median value of the both graphics.

#### *3.2. InduMelt Experiments: Process Development and Suitability of the Different Reactor Concepts*

The main part of the experimental investigation of LIB cathode materials was examined in the InduMelt plant. For this purpose, three experiments were carried out, which will be considered separately below according to their purpose. According to the results of the analyses, as described in Section 3.1, the maximum necessary process temperatures for the tests in the InduMelt plant were set at 1525 ◦C for LCO-C and 1400 ◦C for LFP-C. This is due to the fact that the sample material should be completely liquid at this point in time according to the heating microscope.

#### 3.2.1. Results from Experiments with LCO-C in Both Reactor Designs

The first experiment, which is described in detail below, was carried out in Design 1. The entire experiment lasted nearly 8 h. Figure 8a shows the power input of the induction unit and the corresponding temperatures over the test time. Two type Sthermocouples (S-TC 1 and S-TC 2 in Figure 8a) were used on the reactor surface and two type K-thermocouples inside the reactor. The latter were initially installed at different heights in the reactor, one in the area of the first cube layer (K-TC bottom) and one in the upper area (K-TC top). In contrast to the S-thermocouples, the measurement results of the K-thermocouples are subject to considerable fluctuations due to melting of their insulation, influences of the material in the reactor, etc. However, an approximate temperature spread of the different thermocouple types of 500 ◦C could be determined up to the end.

During the process, noticeable anomalies were documented. Starting at 0.95 h and a K-TC bottom temperature of 450 ◦C a strong formation of condensate in the exhaust pipe to the gas scrubber was observed. This was attributed to the drying of the mortar. The temperature spread of the S-thermocouples at 1.75 h (124 ◦C S-TC 1) can be explained by a slight realignment of S-TC 1. Especially interesting was the continuously increasing white smoke from the exhaust pipe, which started at 5.10 h and 1180 ◦C internal temperature and stopped at 1341 ◦C. This resulted in a continuous white deposit in the exhaust pipe of the scrubber, as shown in Figure 8b. After the white smoke formation stopped, the acid in the scrubber gradually changed from transparent to a slightly yellow liquid. The assumption that the white deposits are Li or a corresponding compound could be confirmed after analysis of the liquid in the scrubber, which are summarized in Table 2.

**Figure 8.** Experimental performance of LCO-C in Design 1: (**a**) comparison of power and temperature over time; (**b**) deposits in the exhaust pipe of the gas scrubber.

**Table 2.** Results from the gas scrubber respectively from its frit liquid after suction of the exhaust gas in Design 1 LCO-C in mg/L.


Sampling after the experiment revealed a total of 4 fractions, the results of which are shown in Table 3.



The fraction defined therein as slag could be identified as dark to light grey nonmagnetic pieces smaller than 10 mm with minimal metallic inclusions, as illustrated in Figure 9a. The mortar shown represents the part into which the test material has diffused. This is optically visible by a dark discoloration of the originally white mortar. In Figure 9b the reactor is demonstrated from below after the concrete floor has been separated. During sampling, care was taken to find the clearest possible separation between the white mortar and the diffused areas, but this proved to be very difficult. The largest product of the experiment in terms of mass was the metal fraction, which could be obtained in pieces larger than 10 mm. The metal piece shown in Figure 9c serves as an example. The analysis showed an impressive purity of 100% Co and an impurity of only 0.01% Li. It should be noted that there may be some variation in sampling and digestion errors, resulting in the overall result not reaching exactly 100%. The fourth fraction was a powder with particles smaller than 1 mm, as can be seen in Figure 9d, which was mostly magnetic. This property is also confirmed by analyses with a cobalt content of 53.4%.

**Figure 9.** Products of the experiment LCO-C in Design 1: (**a**) slag; (**b**) ceramic ring and mortar seen from the bottom; (**c**) metal; (**d**) powder.

Taking into account the respective weighed masses and the analysis results, the transfer coefficients of the individual elements of the fractions were calculated. In detail, the analyses from ICP-MS and ICP-OES of the individual fractions were converted to mass percent and multiplied by the weighed mass at sampling. By adding the respective element masses, a total mass per element could be determined. This represents the amount that was still detectable in the fractions in the reactor after the test. Afterwards, a comparison of the masses before and after the experiment was carried out. The difference was assumed to be a transfer into the gas flow leaving the reactor during the experiment or a transfer into the individual solid fractions. The results of this calculation can be seen in Figure 10.

**Figure 10.** Transfer coefficients of the elements into the individual fractions in % of the experiment in Design 1.

At this point it should be mentioned that the transfer coefficients determined must be seen as initial guide values and internal comparison values and must be confirmed accordingly by repeated experiments in the optimum reactor setup. Nevertheless, the trend was also observed in experiments with NMC and NCA, as described in Windisch-Kern et al. [34]. The result of the transfer coefficients in Figure 10 shows a Li removal rate of over 76% into the gas stream from the cathode material used. Based on the thermokinetic consideration of LCO by Kwon et al. [16] and assuming that most of the Li has left the reactor during the phase of white smoke (approximately 1160–1340 ◦C), it can be assumed that it is Li2O. The transfer of over 21% Li into the mortar can be considered as an undesirable result. The percentage of Li in the slag is not negligible in the analyses (Table 2) with 5.6%, but due to the small quantity of slag it is insignificant for the total consideration with 0.6%. The small amount of Li in the metal (0.1%) is a great result with regard to the purest possible metal fraction. A further potential for improvement can be seen when considering the Li content of 1.8% in the powder, whereby this value can possibly be lowered with a longer

holding time of the final temperature. The result of Co can be interpreted as extremely promising. Only 7.5% is found in the powder and 95.2% in the metal, which can be directly transferred for further use in the corresponding metal industry. The resulting difference to 100% can be explained by the extremely difficult sampling, especially the identification of the individual fractions and the subsequent weighing. At this point, the proportion in the slag and mortar can also be neglected with less than 0.1%. The comparison of these results with other processes is difficult at this point because the composition of the input material differs significantly from a real waste stream of LIB. Nevertheless, by using pure cathode material, without impurities such as Al or Cu, a value of the theoretically maximum possible removal rate of Li can be determined. Vest [15] describes a Li2O transfer rate from the waste stream of LIB of 40.5% in their process based on an electric arc furnace. Even though a direct comparison with this value is not possible, the gap between Vest's result and the theoretically possible value in this method shows an enormous potential.

Much more remarkable in this context is the result of Design 2. The temperature record of the LCO-C experiment in Design 2, which can be viewed in Figure 11 in combination with the power input over time with the same naming as in the previous experiment described above, shows that the temperature is highest in the lower part of the reactor. Thus, the goal of the reactor design of a more targeted temperature provision in the lower area could be realized.

**Figure 11.** Experimental performance of LCO-C in Design 2, comparison of power and temperature over time.

The extreme fluctuations of the K-thermocouples between the test duration of approximately 1 to 3 h could be explained in retrospect in such a way that after the insulation around the thermocouple wires had melted, they reconnected at a higher point in the reactor. The initial theory could be confirmed when the reactor was opened after the experiment, because the thermocouples could be found in the upper part of the reactor free of cubes and no longer in the cube bed. Again, a white smoke formation with the same deposits in the exhaust pipe of the gas scrubber could be detected. This phenomenon occurred at an S-thermocouple temperature range from 1165 to 1340 ◦C. Again, as the smoke intensity decreased, a successive discoloration of the acid in the scrubber from transparent to a light yellow was to be determined. The results of the acid analysis from the gas scrubber can be taken from Table 4. The high value of Li confirms the impression, as already assumed in the experiment in Design 1, that Li can be removed from the reactor via the gas flow.


**Table 4.** Results from the frit liquid of the gas scrubber after suction of the exhaust gas in Design 2 LCO-C in mg/L.

The results of the investigation of the test material LCO-C in Design 2 can be seen in Table 5. Essentially, the analysis differs from the experiment in Design 1 only in that there was no mortar due to the construction. When the MgO crucible was weighed after the experiment, it was found to be 81.2 g heavier than the initial weight. This could be attributed to adhesions on the crucible, which were mechanically extracted as completely as possible. The result was a fine powder. Despite considerable mechanical effort, only 3.8 g could be removed from the crucible without damage, which will be referred to as crucible adhesion in the following.

**Table 5.** Results from the experiment LCO-C in Design 2.


The fractions did not differ in their appearance from those in Figure 9a,c,d. Only the metal pieces were larger, as shown in Figure 12. Analysis of the metal fraction revealed a purity of Co of 93.9% with a negligible amount of 0.12% Li.

**Figure 12.** Metal fraction from the LCO-C experiment in Design 2.

The transfer coefficients, which are illustrated in Figure 13, were determined from the results in Table 5, taking into account the corresponding weight changes of the individual fractions during the experiment.

This result represents a unique selling point in the pyrometallurgical processing of cathode material from LIB. Compared to Design 1, even though only 85.9% of the Co was transferred to an almost pure Co-metal phase, more than 97% of the Li were removed from the material and the reactor via the gas flow. Thus, a nearly 21% higher Li removal rate could be achieved in Design 2.

**Figure 13.** Transfer coefficients of the elements into the individual fractions in % of the experiment in Design 2.

Since the attribution of the Co not found to the gas phase is rather questionable to this extent and cannot be traced back exclusively to errors in sampling and analysis, a closer look at the results is necessary. In addition to the result display in Figure 13, another variant is possible. It is assumed that the difference of the weighed crucible adhesion (81.2 g) to the extracted amount (3.8 g) consists of the same composition as the extracted fraction. This results in a lithium removal rate of 81.7% with a Co value that was not found (i.e., attributed to the gas phase) of −3.12%. This variant cannot clarify the difference to 100%, but by combining the two methods, one obtains a range in which the transfer coefficients move.

Besides this result, Design 2 also turns out to be a better choice when considering the interactions between the sample material and the reactor. As shown in Figure 14a, a massive attack of the reactor wall was observed in Design 1 (Al2O3) with ring diameter reductions of up to 0.2 mm. On the other hand, Figure 14b shows the reactor in Design 2 (MgO) after the experiment. From the difference between the weighing before and after the test, it is known that the reactor was over 81 g heavier afterwards. The theory of adhesion to the reactor can also be seen in the illustration, whose boundary is marked with the red arrow.

**Figure 14.** Visual appearance of the crucibles after the experiments: (**a**) traces of attack on the Al2O3 crucible wall in Design 1; (**b**) appearance of the MgO crucible after experiment 2 with obvious adhesions.

This factor is particularly important for long-term experiments in a continuous setup. If the sample material and the Al2O3-ceramics are in contact for a longer period of time, this attack would lead to a destruction of the reactor. In addition to the advantages already mentioned, Design 2 also features a much simpler construction and therefore easier sampling. A drawback of Design 2 is the higher energy input required, which can be seen in the comparison of the power curve of Figures 8 and 11. However, this can be solved by an improved positioning of the induction coil. Furthermore, the consequences of adhesions in continuous operation must be investigated.

Even though the target temperatures were not reached in the experiments with LCO-C, it can still be assumed that a sufficiently high temperature was reached. Firstly, the temperature in the reactor was measured in the space between the refractory mat and the graphite cubes, as mentioned above, which implies that the temperature in the cube bed must have been even higher due to the heat input in it. In addition, if the temperature was too low, the appearance of the products would be different. An example of this is the metal from Co, as shown in Figures 9c and 12, whose melting point is known to be 1495 ◦C.

#### 3.2.2. Results from Experiments with LFP-C

The test with LFP-C was carried out in Design 1. Since the temperature measurement via the K-thermocouples was already faulty from the beginning of the experiment and a repair was no longer possible at this point in time, only the curves of the recordings from the S-thermocouples are visible in Figure 15. However, the delta value to the Kthermocouples should be similar to that in the LCO-C experiment in Design 1, whereby approximately 500 ◦C can be added to the value of the S-thermocouples to determine the internal temperature at higher temperatures.

**Figure 15.** Experimental performance of LFP-C in Design 1, comparison of power and temperature over time.

During the heating process, smoke development was particularly noticeable at an outside temperature of approximately 750 ◦C (approximate internal temperature of 1210 ◦C), which completely ignited after a short time. This flame, which is an indication of the reaction of phosphorus with oxygen [42], could finally be detected constantly up to an outside temperature of approximately 860 ◦C (approximate internal temperature of 1310 ◦C) as shown in Figure 16a. During this time the acid in the gas scrubber changed its color to a brownish liquid. The results of the exhaust gas analysis can be taken from Table 6.

**Figure 16.** Products of the experiment LFP-C in Design 1: (**a**) flame formation in exhaust gas flow; (**b**) slag; (**c**) ceramic ring and mortar seen from the bottom; (**d**) metal.

**Table 6.** Results from the frit liquid of the gas scrubber after suction of the exhaust gas in Design 1 for LFP-C in mg/L.


In the exhaust gas analysis only a small value for P and a very small amount of Li could be found, which is possibly due to the formation of a flame out of the exhaust pipe. In order to be able to make a statement about the efficiency of the reactor concept for the recycling of LFP, the results of the ICP-OES or ICP-MS must be examined more closely.

A total of 5 fractions could be detected during sampling. The appearance of the fraction classified as non-magnetic slag (Slag 1) differed significantly from that in Figure 9a. Individual spheres with a diameter of up to 5 mm were detected, as shown in Figure 16b. In comparison with the appearance of the magnetic metal fraction in Figure 16d, the difficulty of clearly classifying the individual fractions is obvious. The analysis showed that the Fe content in the slag was even higher than in the material identified as metal. In addition, however, a significant amount of Li (3.19%) and P (15.9%) was also analyzed. Slag 2 in Table 7 is a non-magnetic powder with particles smaller than 1 mm the appearance of which is the same as in Figure 9d. The same applies to the magnetic material identified as powder. In this experiment, care was again taken during sampling to separate as much diffused areas of the sample material from the mortar. These brownish areas are also visible in Figure 16c, which shows the ceramic ring with the mortar after removal of the refractory concrete. Analysis of the metal displayed in Figure 16d shows only 51 wt.% Fe and over 8 wt.% P. This low Fe content suggests that no complete reduction has occurred. A further indication for the correctness of this assumption is the fact that during mechanical processing of the fraction with a hammer, the spheres disintegrated into a powder already with a small amount of force.

**Table 7.** Results of the individual fractions after the experiment LFP-C in Design 1.


Referring to the respective masses of the individual fractions, the transfer coefficients can be taken from Figure 17.

**Figure 17.** Transfer coefficients of the elements into the individual fractions in % of the experiment in Design 1.

Of particular interest are the removal rates of Li (68.4%) and P (64.5%) over the gas flow. The remaining amount of Li here is distributed relatively evenly among the other fractions with a slight concentration on the remaining powder. As already mentioned, it is reasonable to assume that no complete reduction of Fe has occurred. This is also reflected in a considerable value of phosphorus in the metal fraction.

A direct comparison of the experiments in Design 1 shows that the Li removal rate for LFP-C of 68.4% is 8% lower than in the experiment with LCO-C. However, the parallel removal of phosphorus of 64.5% represents a respectable result. It also can be seen that in both the LCO-C test with a transfer coefficient of 21.1% (Figure 10) for Li and the LFP-C test with 6.1% (Figure 17), a significant amount of Li was transferred to the mortar. If the results of LCO-C in Design 2 are also included, it can be assumed that gasification rates for LFP-C are better in this construction method. Looking at the transfer coefficients in Figure 17, a significant portion of the Fe (11.5%) is attributed to the gas flow. De facto, this is the amount that was not recovered during sampling compared to the input amount. The correctness or falsification of this classification and the above mentioned assumptions must be investigated in further experiments. Particular attention must be paid to safety in the pyrometallurgical removal of phosphorus from LIB, consisting of the cathode material LFP, in the exhaust gas post-processing. In addition to oxidation in the air, as shown in Figure 16a, the high toxicity [43] is also of particular importance. These factors must be given special consideration in future developments of the reactor concept presented.

#### **4. Conclusions**

Within the scope of this paper, the suitability of a new pyrometallurgical recycling process associated with materials from LIBs for the recovery of valuable metals was investigated. With the background of a continuous adaptation of the reactor concept to the waste stream from spent LIBs, two different reactor designs were used, in each of which the cathode material LCO with carbon addition was examined for better comparability. In a third trial, the basic capability of the technology for the treatment of LFP was also examined. In addition, knowledge about the behavior of the examined cathode materials used in high-temperature applications was investigated in an upstream step in a heating microscope.

The transfer coefficients determined in the experiments of the novel pyrometallurgical recycling process serve exclusively as a comparison of the efficiency of the presented reactor concepts or as a first benchmark of the basic suitability of the process for the treatment of LFP.

From the experiments in the heating microscope the maximum necessary temperatures for the transformation into a molten phase could be determined. This state of aggregation is necessary in the long run to meet the requirements of the theoretically determined principle of a continuous process. For the experiments with LCO a temperature of 1525 ◦C and for LFP 1400 ◦C could be determined.

In the trial LCO-C in Design 1, 95.2% Co of the original input fraction was converted into the metal, with a purity of Co of 100%. Due to the high Li content in the mortar, only 76.4% of the Li could be transferred into the exhaust gas flow. This contrasts with the result of the experiment with LCO-C in Design 2. Its analysis shows a metal purity of 93.9% Co and a remarkable lithium removal rate in a range from 81.7% to 97.3%. In addition to this impressive lithium removal rate, Design 2 has also proven to be the better choice for future use due to its interaction with the feed material. For example, massive interactions and attacks on the Al2O3 crucible have been detected in Design 1, whereas there were no physical damages with the MgO crucible in Design 2. Although the danger of destruction of the reactor wall during long-term experiments in a future continuous process has been averted, the effect of the detected adhesions on the MgO crucible still needs to be investigated in further tests. However, the initially formulated goal of a more targeted heat supply in the lower part of the crucible was achieved by Design 2.

The experiment with LFP-C was performed in Design 1 and achieved a lithium removal rate of 68.4% with parallel phosphorus removal of 64.5%. Since the results of the LCO-C experiments showed that a higher lithium removal could be achieved when using Design 2, a repetition of the LFP-C experiment in this setup can be expected to result in a higher removal rate. In addition, since sampling in this experiment has proven to be particularly difficult due to the appearance of the fractions, and since detailed examination of the results has revealed questions that need to be clarified, such as the undetectable amount of Fe, further investigations are indispensable.

Nevertheless, it can be summarized that Design 2 with its MgO crucible has proven to be a better choice with regard to its suitability for pyrometallurgical treatment of material from LIBs. This is due to its inertness to the sample material as well as the higher Li removal rate determined. In addition, it was found that the use of the technology is also suitable for the cathode material LFP and that considerable P and Li removal rates have already been achieved. However, in order to be able to treat the fluctuating waste stream from spent LIBs with an appropriate efficiency, in-depth investigations are needed beforehand to gain knowledge of the behavior of all common cathode materials. In order to increase efficiency, the fraction referred to as slag must also be subjected to more detailed investigations in the future, for example with an XRD analysis. From this, knowledge of the phases present is to be generated and the formation of these is to be suppressed with targeted measures. In the current development phase, this has not yet been the focus of research. Furthermore, it is necessary to identify the influence of additional fractions such as Cu and Al from conductor foils on the process.

Compared to commercial techniques used today for recycling spent LIBs, the simultaneous recovery of lithium and phosphorus via the process presented in this paper is its most significant advantage. This first potential assessment for pyrometallurgical recovery of Li would also theoretically meet the requirements of the proposed amendment to the EU Directive 2006/66/EC of a Li recovery up to 70% by 2030. Aspects such as the economics, energy efficiency and environmental impact of this intermediate step in the overall recycling chain, as well as possible recovery rates of a waste stream of spent LIBs in the new reactor design, need to be determined in further studies.

**Author Contributions:** Conceptualization, A.H. and S.W.-K.; methodology, A.H.; investigation, A.H., S.W.-K. and C.P.; resources, A.H., S.W.-K. and C.P.; writing—original draft preparation, A.H.; writing—review and editing, A.H., S.W.-K., C.P. and H.R.; visualization, A.H.; supervision, C.P. and H.R.; project administration, C.P.; funding acquisition, H.R. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Zukunftsfonds Steiermark with funds from the province of Styria, Austria, grant number GZ: ABT08-189002/2020 PN:1305.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**

