*Review* **Challenges in Ecofriendly Battery Recycling and Closed Material Cycles: A Perspective on Future Lithium Battery Generations**

**Stefan Doose 1,2,\*, Julian K. Mayer 1,2, Peter Michalowski 1,2 and Arno Kwade 1,2**


**Abstract:** The global use of lithium-ion batteries of all types has been increasing at a rapid pace for many years. In order to achieve the goal of an economical and sustainable battery industry, the recycling and recirculation of materials is a central element on this path. As the achievement of high 95% recovery rates demanded by the European Union for some metals from today's lithium ion batteries is already very challenging, the question arises of how the process chains and safety of battery recycling as well as the achievement of closed material cycles are affected by the new lithium battery generations, which are supposed to enter the market in the next 5 to 10 years. Based on a survey of the potential development of battery technology in the next years, where a diversification between high-performance and cost-efficient batteries is expected, and today's knowledge on recycling, the challenges and chances of the new battery generations regarding the development of recycling processes, hazards in battery dismantling and recycling, as well as establishing a circular economy are discussed. It becomes clear that the diversification and new developments demand a proper separation of battery types before recycling, for example by a transnational network of dismantling and sorting locations, and flexible and high sophisticated recycling processes with case-wise higher safety standards than today. Moreover, for the low-cost batteries, recycling of the batteries becomes economically unattractive, so legal stipulations become important. However, in general, it must be still secured that closing the material cycle for all battery types with suitable processes is achieved to secure the supply of raw materials and also to further advance new developments.

**Keywords:** lithium-ion battery; LIB; battery recycling; mechanical recycling processes; hydrometallurgy; pyrometallurgy; battery generation; circular economy; solid state batteries

#### **1. Introduction**

The goal of economical and sustainable battery cell production remains a key element on the way to establishing electromobility as a green technology of the future [1]. Sustainable process management and development also includes the economic recycling and recirculation of materials used in cell production with a simultaneously low energy input, which leads to a reduction of the ecological CO2 footprint in battery cell production [2–5]. Therefore, the establishment and sustainable further development of an internationally leading, competitive battery cell production must go hand in hand with the development of appropriate recycling technologies [6–9]. The recycling technologies must be flexible and adaptable to future production technologies and especially materials that are processed in the future with regard to new battery generations [10].

Moreover, closing the material cycles for batteries on the basis of scalable production and recycling technologies is a central component for a CO2-reduced or CO2-neutral battery cell production and thus for electromobility (e.g., achieving the "Green Deal" goal of the

**Citation:** Doose, S.; Mayer, J.K.; Michalowski, P.; Kwade, A. Challenges in Ecofriendly Battery Recycling and Closed Material Cycles: A Perspective on Future Lithium Battery Generations. *Metals* **2021**, *11*, 291. https://doi.org/10.3390/ met11020291

Academic Editor: Yoshikazu Ito Received: 31 December 2020 Accepted: 3 February 2021 Published: 8 February 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/).

European Union (EU)). Only closed material cycles in batteries can enable a conversion from carbon-based energy sources to sustainably produced electrical energy in ecological, economic, and social terms [7,11,12]. To achieve closed loop material cycles, appropriate recycling technologies must be developed. For example, according to the new battery directive proposal of the EU Nr. 2019/1020, 95% of cobalt (Co), copper (Cu), and nickel (Ni) as well as 70% of lithium (Li) have to be recovered from spent lithium-ion traction batteries by 1 January 2030 [13]. Moreover, according to the report on the circular economy of traction batteries published by the Circular Economy Initiative Deutschland of acatech (National Academy of Science and Engineering in Germany) 90% of Co, Ni, and Cu as well as 85% of Li should be recycled from spent lithium battery systems by 2030. In addition, a recycling rate of 70% of the entire battery should be aimed for [14]. In a worldwide comparison, the EU sets very high requirements with the Battery Directive. In the USA, there are no generally applicable requirements for the return of lithium ion batteries (LIBs). However, voluntary consortia (e.g., the End of Live Vehicle (ELV) Solutions consortium) work together here to close material cycles. China follows a similar approach to the EU. Producers are encouraged to take back batteries that have been put into the market and to return them to the materials cycle [15]. Other countries in Asia (South Korea, Japan) are pursuing similar goals as the EU (South Korean RoHS/ELV/WEEE Act, 2007 and Japan's End-of-Life Vehicle Recycling Law).

For the recycling of lithium-ion batteries today, usually at first, a deactivation of the battery system is realized, which is followed often (but not compulsory) by a dismantling of the battery system down to the cell modules (or more seldom individual cells). The deactivation can be achieved by full electrical discharge and subsequent short circuiting, by treatment in a saline solution, or by pyrolysis (high heat treatment) of the battery systems at temperatures of more than 200 ◦C [16,17]. Afterwards, in general, three types of process technologies are used, which are combined in a different manner: mechanical, hydrometallurgical, and pyrometallurgical processes [7,17–22]. Figure 1 gives an overview of different possibilities to combine the different process types. The different process steps can be applied in different sequences and above all in different processing depths [19,23]. One lean way is deactivation by pyrolysis, pyrometallurgical processing, and slagging. Pyrometallurgical processes are well established for processing primary materials and can achieve high recovery yields concerning the metals cobalt, nickel, and copper, but they show challenges regarding the recovery of lithium. Therefore, to recover lithium and manganese, pyrometallurgical processes have to be combined with hydrometallurgical processes for processing the slag. Overall, a relatively small overall recovery of the batteries material can be expected in this case due to graphite, polymers, and electrolyte being burned, although a very high recovery of Ni, Co, and Cu is possible [24]. A more elaborate way puts together another combination of processes. Here, the battery cells or modules are discharged in the first step, for example, before they are mechanically crushed [25]. Subsequently, the black mass or the shredded battery material is pyrometallurgically processed before it goes into a final hydrometallurgical step. Here, for example, the degree of mechanical processing can be varied [18,24,26]. The amount of recovered materials increases, as polymer components as well as aluminum can also be recovered. Furthermore, after deactivation and mechanical processing, there is also the possibility of proceeding directly to hydrometallurgical processing (as it was proposed for the LithoRec process [20,27,28]). This route enables early recovery of the polymer battery components as well as Cu and Al. In addition, hydrometallurgical processing can also recover Mn that cannot be recovered by pyrometallurgical processes. In addition to the recovery of the individual substances such as Ni and Co, also, the direct reconditioning of the cathode material by hydrometallurgical treatment is carried out on an industrial scale. Avoiding pyrometallurgical processing during battery preparation theoretically reduces the energy requirement and thus improves the ecological footprint [29]. Furthermore, other process routes are theoretically possible, but they shall not further be discussed here; they are discussed in more detail in relevant literature.

**Figure 1.** General overview of some potential recycling process chains in different combinations.

Overall, the overview of the recycling processes shows that many different recycling process routes are possible and in industrial use or under development at pilot scale at least [17,28–30]. As a requirement for a future process chain, besides achieving high recovery rates of more than 90% or even 95% [13,14] and in parallel sufficient material purities for further usage as battery material, it is therefore to be set that it should be highly flexible in order to achieve the most energy-efficient multi-material recovery possible. To reach this goal, mechanical, thermal, and chemical process steps are to be used and combined in different ways. Future recycling processes for Li-ion batteries must not only be able to process new materials but should also replace energy-intensive processes currently used for classic Li-ion batteries with low-energy, environmentally friendly process steps.

#### **2. Lithium Battery Development**

Today, lithium-ion batteries with liquid electrolyte dominate the market for electromobility. They are also used in portable devices as well as in the field of stationary energy storage. On the cathode's side, particles of lithium transition metal oxides such as lithium nickel manganese cobalt oxide (NMC, LiNi*x*Mn*y*Co*z*O2) and lithium nickel cobalt aluminum oxide (NCA, LiNi*x*Co*y*Al*z*O2) or also lithium metal phosphates, especially lithium iron phosphate (LFP), are mainly used. In the case of the NMC and NCA materials, the nickel content increases steadily, and the cobalt content decreases continuously. Polyvinylidene fluoride (PVDF) is currently used as a standard binder on the cathode side. On the anode's side, graphite is usually used as the anode material; in rare cases, lithium titanium oxide (LTO) is used. Moreover, the first cells with the addition of very small amounts of silicon to the graphite are offered by the cell manufactures. A mixture of carboxymethyl cellulose (CMC) and styrene butadiene rubber (SBR) usually acts as a binder within the anodes. As shown in Figure 2 within the near and medium-term future, the following developments in next-generation lithium battery technology are expected:


**Figure 2.** Diversification of lithium battery technologies.

An overview of the materials and their potential contents in the different battery types is given in Table 1. According to Table 1, it can be concluded that with the use of the upcoming battery generations, the composition in terms of recyclables will also vary. While the Ni content will increase significantly in type I, the presence of Fe or Mn is expected in type II. In both types, the liquid electrolyte including Li conducting salt is also expected to have high potential for recovery. However, for type II, also sodium instead of Li has a significant potential. In contrast, for type III and eventually also type IV, solid electrolytes are employed for the separator and as electrolyte within the composite cathode. The solid electrolytes can have an oxide, sulfide, or polymer nature, whereby the compositions and properties can be highly variable (Table 2).




**Table 2.** Typical components and properties of separator technologies.

From these points, it is clear that next-generation technologies will include much fewer critical components, such as cobalt, but also new materials such as silicon, or even germanium. In the medium to long term, metallic lithium is expected on the anode side in case of type III and especially type IV. The use of lithium anodes will require the use of polymeric, sulfidic, and/or oxidic solid electrolytes as separator and as electrolyte on the cathode's side. However, more and more also lithium-free anode structures are shown in the literature [42,43]. In this case, the lithium from the cathode active material moves to the anode side and is deposited on the lithium-free anode layer. In order to minimize the formation of dangerous lithium dendrites, also lanthanum, titanium, zirconium or phosphorus in the solid electrolyte area are presented. Lithium is still indispensable for the time being, but the commercialization of sodium ion or other batteries in the near future is also conceivable. Overall, a further increase in the complexity of the battery cells developed and produced and the use of other economically strategic raw materials can be expected. The recovery of these should be tackled urgently in order to establish sustainable battery production with respect to the increasing sales figures.

In view of the rapidly evolving battery technologies, the best possible recycling routes for the newly developed battery cells should be determined and their recyclability assessed at an early stage. In addition to battery cells with a higher proportion of silicon, pure lithium, and/or solid-state electrolytes, the use of novel binders and fibrous additives or the increase in the adhesive strength of the electrodes can also significantly influence the recycling processes. In addition, active material mixtures are increasingly being used on the cathode's side, so that iron, among other substances, enters the metallurgical material preparation process. Moreover, a direct reconditioning of the active material gets very difficult to impossible. Sustainable recycling concepts must be evaluated in advance by process-based economic and ecological models for the entire life cycle.

Of great interest is the assessment of the coming battery generations with regard to the change of a future circular production of batteries, the recycling process itself, and the hazard potential within the recycling process. Accordingly, in the first step, the three criteria for today's LIB are briefly presented, and based on this, the potential challenges posed by the introduction of the different cell generations mentioned above are assessed.

#### **3. State of the Art Recycling Processes for Lithium-Ion Batteries**

As shown in Figure 1, different process chains and routes are developed for the recycling of today's lithium-ion batteries. Today, the most common ones are a combination of mechanical and hydrometallurgical processes as well as a combination of dismantling and pyrometallurgical processing. Casing and connection materials of the battery pack and module are removed in advance or even after comminution in the processes presented and fed, in the normal case, to the conventional recycling methods for aluminum, iron, polymers, and others.

#### *3.1. Mechanical–Hydrometallurgical Recycling Technology and Challenges*

A common recycling route to be found combines mechanical processing with direct hydrometallurgical processing of the batteries. The mechanical processing can be fulfilled in a dry or a wet mode. Moreover, before mechanical treatment, the battery system has to be deactivated by complete discharge and usually short circuiting and dismantled down to at least the module level. However, it is also reported that the dismantling goes down to the electrode level in order to separate anode and cathode materials already before the mechanical treatment.

The economically and ecologically attractive mechanical processing of the battery cells is carried out down to the level of active and inactive materials, and it can be combined with an evaporation of electrolyte components [20,27,28,44]. Discharged battery cells and modules are comminuted under inert atmosphere in a shredder process or under water [45]. Then, the components are separated by classification and sieving processes. To achieve a high separation quality of the materials, several steps can be run consecutively with different process settings [20,28]. Furthermore, some alternative methods also include electrolyte recovery steps. The pre-dried shredded material is divided in the separation process into metallic components (casing material), current collector foils (Al and Cu), black mass (Li, Co, Ni, and Mn oxides, graphite, PVDF, carbon black, and impurities) and organic components (separator) [19,25]. However, today, the mechanical processes are not able to achieve practical recycling rates of more than 95% regarding cathode active materials and the necessary purities. Subsequently, the black mass is processed without any intermediate steps in the hydrometallurgy.

A particular challenge without pyrolysis or pyrometallurgical processes is the handling of the remaining fluorine components, which can be deviated on the routes including pyrolysis or pyrometallurgical processes due to high temperatures. Hydrometallurgy is a process at low temperatures in aqueous phases and can be performed in three major steps. Leaching is the first step and describes the dissolution of metals via the usage of acids or bases. Typically, at first, the black mass is dissolved in NaOH and afterwards leached using H2SO4, with initial impurities such as iron, aluminum, and copper precipitated by small amounts of NaOH and sieving [46,47]. The purification is the second step in the process chain where the metals are separated and purified via e.g., selective chemical reactions. The last step is the final recovery of the metals or the salts. This can be done by means of e.g., crystallization, ionic precipitation, electrolytic deposition, or further methods. For example, the individual metals can be further precipitated as sulfate salts by adding NaOH or other basic agents and increasing the pH. Selectively, this can be controlled by considering the different solubilities of the metal salts [46,47]. Hydrometallurgical process steps are capable of producing high product purities. However, plants for counter-flow are larger than those used in pyrometallurgy and require a larger financial investment volume for their construction [21,48,49]. In general, important challenges are to achieve the demanded high recovery rates and, at the same time, high material purities.

An example of the mechanical–hydrometallurgical process chain is the LithoRec process [19,20,28], which is commercialized in similar form by Duesenfeld and Redux (Figure 3). The developed process has a high potential to close the loop of the circular economy in the battery production as well to reduce the CO2 footprint in the battery production and utilization phase due to the high recovery rates of the recyclables and the low energy consumption. Maximizing the recycling rate was a core objective of the projects. Among other things, a process was developed in which, depending on the active material, 85 to potentially over 95% of the lithium can be recovered by mechanical (includes drying step) and hydrometallurgical means (leaching and subsequent precipitation of the lithium).

**Figure 3.** Developed process chain of mechanical/thermal processing of the LithoRec process prior to hydrometallurgical steps [20,28].

Another example is the process used by Accurec GmbH® (Krefeld, Germany). Here, the batteries are discharged prior to mechanical processing and subjected to vacuum thermal pretreatment (pyrolysis) [44,50]. In this process step, volatile organic (electrolyte) as well

as polymeric components (separator) and halogenic compounds can be pyrolyzed at 200–400 ◦C and low pressure. In the further course, the pyrolyzed mass is subjected to mechanical processing, pyrometallurgical processing, and subsequently hydrometallurgical processing to recover also the lithium. The developed process also has a noticeable potential to close the loop of the circular economy. In addition, the company Erlos GmbH represents another interesting approach in the recycling of lithium-ion batteries, in which the battery cells are separated by type (anode, separator, cathode, casing). Subsequently, in the example, the entire cathode coating is separated from the aluminum substrate foil. Then, the cathode coating is separated into its components—active material, binder, and carbon—by wet chemical processes (leaching), and the active material is reconditioned. This step allows the active materials on the cathode and also anode side to be reused after the treatment without having to perform a resynthesis based on the purified substances, i.e., metals. However, the electrolyte, binder, and conductive additives are lost for reuse. Nevertheless, the active material chemistry is retained and cannot be adapted to new requirements [47,51].

#### *3.2. Pyrometallurgical Recycling Technology and Challenges*

Another approach for the recycling of lithium ion batteries is the pyrometallurgical (mechanical–pyrometallurgical) way. In this process, the battery cells or modules can be placed in a direct pyrometallurgical step or they can be mechanically processed in a first step (analogous to the mechanical–hydrometallurgical), and the obtained valuable black mass is fed into the pyrometallurgy. Pyrometallurgical processing involves high-temperature processes such as melting and roasting to produce battery slag [52]. A pyrometallurgical recycling process of LIBs starts with an initial heating in the temperature range of 150–500 ◦C, during which electrolyte components and organic solvents are removed. Subsequently, a high-temperature process with temperatures up to 1450 ◦C using reducing agents (graphite, coke, NaHSO4, CaCl2, or NH4Cl) is carried out in the furnace to obtain battery alloy and slag (Li2O as well as Li2CO3) as products [17,53,54]. Due to the different meting points, the metals Ni, Co, and Cu can be individually recovered. However, lithium, manganese, and aluminum get into the slag or the kiln dust. The battery alloy produced in this way contains the valuable materials (such as Co, Cu, and Ni) and can then be processed hydrometallurgically. Lithium and other battery components can only be recovered at great expense or not at all by this process. The advantage of this process technology is the comparatively high robustness against changing feed material and a comparatively small plant size with given throughput. The disadvantage is the comparatively lower overall quantity of recyclable materials. However, the critical metals are recycled at a high recovery rate. The processes generate only intermediates that have to be purified by further steps to enable reuse (e.g., further processing by hydrometallurgical steps). In addition, they show low economic efficiency when low concentrations of recyclable materials (e.g., Co, Cu, and Ni) are present [24,52,55].

The process from Umicore Valéas™ (Bruxelles, Belgium) can be cited as an example of the pyrometallurgical processing route. As a great advantage in advance can be mentioned the great robustness of this way to various types of batteries [17]. Prior to thermal processing, the batteries are dismantled, whereby plastic and metal housing parts in particular are removed, and the cells are exposed [56]. Then, these are pyrolyzed in the shaft furnace at three different process temperatures (400–1450 ◦C). The alloy obtained contains valuable materials such as Cu, Co, and Ni. In turn, the slag contains Li. Both are subsequently hydrometallurgically processed in a leaching process to recover the valuable materials contained for reuse [17].

#### *3.3. Potential Hazards of Lithium-Ion Batteries in Recycling Processes*

Battery systems of electrically powered vehicles (e.g., EV, PHEV, HEV) contain chemically stored energy. The systems contain energy quantities of up to 20–100 kWh (TESLA Model S) and reach system voltages of 300–800 V [2,57]. The associated hazards are particularly relevant when the battery is handled separately from its application, such as in the dismantling and the recycling process [28]. In recycling, due to mechanical process steps in particular, this should be of major importance. Today's and future battery generations combine builtin active materials with high energy densities and partly highly inflammable electrolytes. In normal use, particularly external factors such as short circuits (internal and external), high temperatures or mechanical deformation can trigger critical events and lead to the thermal runaway [58]. Thus, the hazards that can be caused by these faults from the battery in recycling can be divided into three main areas: (1) electrical, (2) thermal, and (3) chemical hazards. However, in the real case, they never occur alone but are usually a combination of hazards. Therefore, in general, a recycling process should be designed and engineered in such a way that the risks can be avoided as far as possible.


Uncontrolled temperature rises up to the so-called thermal runaway of the battery cells can be caused by external as well as internal short circuits if a critical temperature is exceeded, which can be generated by handling in recycling. These effects can be triggered during the process by mechanical penetration of foreign bodies, internal cell defects, or external arrester/electrode contacts [62–65]. In addition, overloading or high ambient temperatures can cause the thermal runaway to occur. This type of hazard mainly occurs during the storage and discharge of spent batteries and not during the recycling process. The thermal runaway involves the reaction of cathode, anode, and electrolyte (Figure 4). In the first step, the process runaway begins with the decomposition of the solid electrolyte interface (SEI) on the anode. At a critical temperature of about 90–120 ◦C in a battery, the chemical decomposition processes start. Under an exothermic reaction process, the formed SEI is decomposed and leads to different gaseous reaction products (e.g., carbon dioxide, ethane, ethane) [66]. The energy released from the first exothermic reactions leads to further heating of the battery cell and can dissolve the subsequent chain reaction processes. Then, the intercalated lithium begins to react with the electrolyte.

**Figure 4.** Schematic sequence of thermal runaway between different battery cell components after a critical cell temperature has been exceeded.

Electrolyte decomposition starts at approximately 200 ◦C with the formation of CO2, hydrogen fluoride, ethene, and other hydrocarbons containing fluoride [67,68]. The exothermic reactions of the embedded lithium with the binder and the decomposition of the cathode active material start at 240 ◦C and 250 ◦C, respectively [69]. In actual recycling processes, the steps that still involve electrolytes are to be regarded as particularly critical. To reduce hazards related to the electrolyte, it is advisable to remove the electrolyte as early as possible in the recycling process chain and, in an ideal case, to purify and return it back into the material cycle. However, how valuable a repeated use of the electrolyte is regarding costs and environmental protection is an open question. If a primarily mechanical process strategy is used, comminution can be carried out under an inert gas atmosphere (e.g., nitrogen), followed by evaporation of the electrolyte at moderate temperature increases and/or at negative pressures [20,28] or even by further extraction methods [70,71]. Then, the evaporated electrolyte components can be condensed out as in the Lithorec process. In the case of upstream pyrolysis, the electrolyte can already be removed in advance at temperatures of approximately 200–400 ◦C in specialized ovens (Accurec GmbH®). The comminution takes place downstream. The pyrometallurgical process route can handle the electrolyte removal even more easily. Due to the high temperatures during the processing of the battery materials (up to 1450 ◦C) and the addition of reducing agents, the electrolyte can be safely removed from the cell materials [17]. Moreover, at temperatures above about 650 ◦C also, the binder PVDF can be decomposed and removed under the challenge of handling hydrofluorocarbons. In summary, an improperly handled battery cell represents some potential hazards. However, with appropriate and orderly process control, these risks can be reduced to a minimum, and safe process control should be ensured.

#### **4. Circular Economy in the Context of Battery Production**

Establishing closed material cycles for batteries on the basis of scalable production and recycling technologies is a central component for CO2-reduced or CO2-neutral battery cell production and thus for electromobility, the provision of energy in household and handicraft appliances, as well as for the stationary storage of renewable energies [7,72,73]. As a matter of fact, closed material cycles in batteries are the only way to convert carbon-based energy sources into sustainably generated electrical energy in ecological, economic, and social contexts. Consequently, a circular economy for traction batteries, i.e., lithium-ion battery systems is demanded by the European Union as written in the Battery Directive 2006/66/EC and [13] as well as the Circular Economy Initiative Deutschland (CEID) of the National Academy of Science and Engineering in Germany [14]. In addition, many research studies are being done on this topic while highlighting the challenges that still exist and need to be overcome [17,21,52,72]. Such closed circles improve the sustainability of lithium-ion batteries; they especially decrease the carbon footprint, decrease the material

costs, and ensure secondary material resources for Europe [73,74]. However, in general, when closing material cycles, it should be noted that the energy used is in proportion to the products. With regard to recycling, mechanical treatments are energy-wise recommended, as they are less energy-intensive than metallurgical processes. Examples of specific consumption parameters of energy are electrical energy and wastewater. Pyrometallurgical processing requires 4.68 MJ of electrical energy per kg of battery; hydrometallurgical processing requires 0.125 MJ of electrical energy per kg of battery [75,76]. In addition, approximately 3.76 L of wastewater are produced per kg of battery during hydrometallurgy. However, as described above, a combined process strategy is necessary for good product properties and qualities [29].

A possible implementation of such a closed-loop production is shown in Figure 5. After the utilization phase, batteries are mechanically disassembled, whereby the safety of the processes is an important aspect of development. Depending on the level of detail of the mechanical process chain, components such as copper and aluminum foil can already be separated. Pyro- or hydrometallurgical processes must be applied for further purification at the latest after the active materials, the so-called black mass, have been exposed. Depending on the process, graphite and binder can also be separated. Finally, metal salts are to be precipitated and re-synthesized to produce new active materials [48,52]. Alternatively, the anode and cathode active materials can be reconditioned by means of purification processes, lithium enhancement, and functionalization [47,51,77–79]. Direct reconditioning has the advantage of being a fast and less energy-consuming process, but it cannot directly adapt the cathode chemistry to new developments. For short life cycle phases or rejects from production (e.g., losses of battery cells during formation), direct reconditioning can be very attractive [80].

**Figure 5.** Exemplary approach for the cycle of circular battery production with the presentation of both established processing routes, direct hydrometallurgical and upstream pyrometallurgical processing after previous mechanical processing.

In order to realize a circular battery cell production, a number of technological and organizational prerequisites must be created:

• Almost 100% of end-of-life batteries must be collected and recycled at the latest after any second-life application [73]. The expected lifetime of batteries in the automotive sector is at least 8 years, and, thus, in the mean, it tends to be more than 10 years. In the future, the lifetime will probably increase further. However, this value is highly dependent on the loads (fast charging, temperatures, etc.) [81]. Before recycling, it is important to check whether a second-life application can be reasonable.


In order to get an overview of the technological difficulties, an example is given below to obtain a recycling rate of 95% from a recyclable material in five process steps. If each of these process steps has a yield of 99%, a total yield of 95.09% can just be obtained with all five steps. Thus, it is very important to use few very good process steps in the reprocessing to generate the highest possible yield.

#### **5. Perspective on Recycling and Circular Economy of Future Battery Generations**

The focus in future developments is a highly energy-efficient multi-material recovery with recovery rates of at least more than 90%, most probably more than 95%, which should be adaptable to variations of respective input streams and current output demand. This results in the requirements for future recycling processes, which must either have a high degree of flexibility or which are focused on certain battery types, which in this case have to be sorted efficiently. In addition, the process routes to be developed must be specialized with regard to the individual components to be purified, e.g., electrolyte, electrode components, and active materials. To achieve this, combinations of mechanical, thermal, and chemical process steps have to be used and interconnected. These processes have to be designed also with regard to novel battery materials and chemistry (e.g., solid-state batteries). They should over a long term replace the current energy-intensive processes for classical LIB. This will render future recycling processes significantly more environmentally friendly and less energy-intensive.

Various challenges and requirements for recycling processes, closed material cycles (circular economy), and safety aspects arise from the large-scale use of the four battery types expected to dominate the future as mentioned in Section 1.

**Battery type I (Si-containing anodes, Ni-rich cathodes, liquid electrolyte):** Closedloop production is economically attractive because of the high nickel content, and regeneration of the anode materials should be very feasible both ecologically and economically. The recycling processes known today, using hydrometallurgy alone or a combination of pyroand hydrometallurgy, should be very well able to process these battery cells, which will be available on the market in the near future. However, the recycling process chains must ensure high recovery rates of 95% and higher for critical metals and most possible lithium. For active materials with high nickel and low cobalt content and already high specific capacities such as NMC 811, also a reconditioning of the active material can be attractive because a further increase in specific capacity due to new material developments is probably relatively small. These reconditioned active materials could also be used for

the production of battery type II in the near future, especially if the performance is slightly reduced due to the multiple usage. Regarding safety, battery type I cells will be more sensitive to mechanical damage and external short circuits due to the high nickel content and the presence of nanoparticulate silicon. This increases the risk of thermal runaway and resulting fires.

**Battery type II (Graphite anodes, Mn- and Fe-rich cathodes, liquid electrolyte):** This type of battery, which is mainly used in stationary applications and low-cost mobility concepts, can be well recycled with the existing processes such as the previous one. The preferential use of Mn and Fe leads to a low-cost batteries and thus, economically less attractive recycling processes. Therefore, regulations have to be set up to ensure that these batteries are recycled at the EoL. Due to the high risk of contamination, cells with active materials containing iron must most probably be separated from batteries of other types even before recycling. Otherwise, complex purification processes must be used to separate the transition metals (Ni, Co) from the iron especially in case of hydrometallurgical process routes. For such batteries, the reconditioning of the active materials is probably also a cost-efficient and sustainable option for material recovery.

**Battery type III (Lithium or lithium-free anodes, Ni-rich cathodes, solid electrolyte):** The solid-state electrolytes used in this type of battery will lead to obstacles in establishing a closed-loop economy [35]. It makes it difficult to recover the individual materials used in the battery in high degrees of purity. On the anode's side, thin lithium foils or, in the future, metallic lithium deposited on 3D structures (depending on the transfer of current research results to industry) or even lithium-free 3D-structures are used. These anodes require that the mechanical treatment of the battery cells be performed in an inert gas atmosphere. Furthermore, this results in additional equipment and work safety requirements during processing. This is also due to additionally required hydrometallurgical steps of solving and recovering lithium in the form of salt (LiOH), which leads to the formation of gaseous H2. When considering the possible recycling processes for cathodes and separators, a distinction must be made between the solid-state electrolytes used:


**Battery type IV (Lithium anodes, S/C-containing cathodes, liquid or solid electrolyte):** Apart from lithium, copper, and aluminum, no further valuable materials are

used in lithium–sulfur batteries. This results in the task of efficiently separating these three materials from the mechanically treated battery mixture. Presumably, the recovery of sulfur and carbon for direct use in a battery is not practical, since there are other inexpensive and reliable sources of sulfur and carbon that promise less effort and the needed higher purity. Accordingly, a closed-loop production of this cell type does not necessarily make sense economically and probably also ecologically according to the current state of knowledge. If solid instead of liquid electrolytes are used, the challenges described for battery type III apply additionally.

In conclusion, with the exception of battery type II, all other battery types that potentially lead to a higher energy density in Wh/L (battery types I and III) or an increased specific energy in Wh/kg (Li-sulfur battery) cause additional challenges with regard to circular economy and closed material cycles, recycling processes as well as safety in the handling and recycling of batteries. From the current state of research and industrial implementation, already established or partially established processes can be adapted and expanded to meet the challenges of the coming battery generations. Nevertheless, this can be ambitious for some battery types (i.e., battery type III/IV) with regard to the required high recycling rates of the total battery of 70% and higher or 95% of individual metals and make the development of new process steps necessary. Today's established recycling processes allow a recycling rate for individual materials of up to more than 90% in some cases, depending on the combination of the processes. However, if for example the recovery of the transition metals such as nickel and cobalt is maximized by a mainly pyrometallurgical process, a recovery of graphite as an anode material is not practical; i.e., the graphite serves as an energy source for the heating. Therefore, depending on the regulations, a certain process route can be worthwhile or not useful at all. Processes routes based on only one process type (e.g., only mechanical) enable at least lower overall recovery values. Thus, with the position today, it can be assumed that almost all battery materials can be recycled and reused within new batteries. However, an open question is how far the substances have to be purified so that there is no effect on the electrochemical performance, and if also a reconditioning of the active materials, especially also the cathode materials, is possible on large scale, as it has been shown on a lab scale. This question is the subject of actual research and can hopefully be quantitatively answered in the near future. However, it is expected that the original performance will be achieved in any case when the current active material is reprocessed to the original material purity via complex processes similar to the ones applied for primary materials [65,66,68]. For types I and III, this is also expected for the cathode materials, as the reprocessing processes will be very similar. For type II, a good performance retention result can also be expected for the cathode, since these battery chemistries and materials are already well researched. For type IV, reuse cannot be reasonably implemented from today's economic perspectives. In addition, direct reconditioning of the cathode material can be used to recycle type II and, in some cases, type I materials. These materials are subject to a longer period of use and application than new and even higher-energy materials in example for type III. With regard to the graphite-containing anode in particular (types I and III), recycling is not yet an option from an economic point of view. However, technically, there are approaches that show a reuse with the same performance [86]. Since types II and IV contain a lithium anode, recycling is an economical option for reuse here.

It becomes clear that the diversification and new developments demand a proper separation of battery types before recycling, for example by a transnational network of dismantling and sorting locations, and flexible and high sophisticated recycling processes with case-wise higher safety standards than today. Moreover, for the low-cost batteries, recycling of the batteries becomes economically unattractive, so that legal stipulations become important. However, in general, it must be still secured that closing the material cycle for all battery types with suitable processes is achieved to secure the supply of raw materials and also to further advance new developments. Last but not least, it can be stated that in absolute percentage values, a relatively small step has to be realized until we meet

the targets, but it will be enormously demanding to achieve the last percentage points in the quotas especially if more than only the transition metals shall be recovered.

**Author Contributions:** Writing, S.D., J.K.M. and P.M., with support from A.K.; project supervision, A.K. with support from P.M. All authors have read and agreed to the published version of the manuscript.

**Funding:** We have received financial support for the preparation of this manuscript from the German Research Foundation and the Open Access Publication Funds of Technische Universität Braunschweig.

**Data Availability Statement:** Not applicable.

**Acknowledgments:** We acknowledge support by the German Research Foundation and the Open Access Publication Funds of Technische Universität Braunschweig.

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

#### **References**


### *Article* **Task Planner for Robotic Disassembly of Electric Vehicle Battery Pack**

**Martin Choux \*,†, Eduard Marti Bigorra †,‡ and Ilya Tyapin**

Department of Engineering Sciences, University of Agder, 4604 Kristiansand, Norway; edumabi@gmail.com (E.M.B.); Ilya.tyapin@uia.no (I.T.)

**\*** Correspondence: martin.choux@uia.no

† These authors contributed equally to this work.

‡ Current address: Northvolt AB, 112 47 Stockholm, Sweden.

**Abstract:** The rapidly growing deployment of Electric Vehicles (EV) put strong demands on the development of Lithium-Ion Batteries (LIBs) but also into its dismantling process, a necessary step for circular economy. The aim of this study is therefore to develop an autonomous task planner for the dismantling of EV Lithium-Ion Battery pack to a module level through the design and implementation of a computer vision system. This research contributes to moving closer towards fully automated EV battery robotic dismantling, an inevitable step for a sustainable world transition to an electric economy. For the proposed task planner the main functions consist in identifying LIB components and their locations, in creating a feasible dismantling plan, and lastly in moving the robot to the detected dismantling positions. Results show that the proposed method has measurement errors lower than 5 mm. In addition, the system is able to perform all the steps in the order and with a total average time of 34 s. The computer vision, robotics and battery disassembly have been successfully unified, resulting in a designed and tested task planner well suited for product with large variations and uncertainties.

**Keywords:** robotic disassembly; electric vehicle battery; task planner

#### **1. Introduction**

As the adoption rate for electric vehicles (EV) is now accelerating worldwide, EV Lithium-Ion Batteries (EVBs) repurposing or recycling volumes are expected to be larger in 5–10 years (120 GWh/year available by 2030 [1]) and legislation will likely demand higher collection and recycling rates as for example in the newly proposed regulation of the European parliament and of the council concerning batteries and waste in December 2020. Today, the automotive Lithium-Ion Batteries (LIBs) dismantling process is mainly carried out manually and the use of robotics in this process is limited to simple tasks or human assistance [2]. These manual processes are time consuming and must be done by highly skilled personnel. As a direct consequence, the manual total disassembly of Li-ion EVBs might not be profitable and would be stopped at an optimal level, i.e., partial disassembly, that achieves maximum profit while decreasing the environmental impact [3]. In comparison, automated systems are more robust, have a lower-cost, reduce injuries and/or sickness, make the workplace more attractive for those hard-to-recruit-and-retain skilled workers, and are best suited for up-scaling to high-volumes. Therefore, fully automated disassembly of EVBs is inevitable. The main challenges for the success of the automated systems in dismantling are the variations and uncertainties in used products [4]. These challenges in the robotic disassembly of Electrical and Electronic components in electrical vehicles have been presented in article [5] where the need for cognitive systems is identified to enhance the effectiveness of automated disassembly operations. In the case of automated disassembly of EV batteries, advances in Computer Vision (CV) and cognitive robotics offer promising tools but this topic remains an open research challenge [6]. The

**Citation:** Choux, M.; Marti Bigorra, E.; Tyapin, I. Task Planner for Robotic Disassembly of Electric Vehicle Battery Pack. *Metals* **2021**, *11*, 387 . https://doi.org/10.3390/met11030387

Academic Editor: Alexandre Chagnes

Received: 25 January 2021 Accepted: 21 February 2021 Published: 26 February 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/).

disassemblability of industrial batteries, as described in articles [7,8], can be improved either by modifying their design and increasing its standardisation, for example redesigning a battery module to make it remanufacturable [9], or by developing new technologies to ease and eventually remove some of the challenges, as for example, making the recognition of fasteners an easy task. This second route, i.e., making the disassembly process smarter and more efficient through better cognitive capabilities is the one chosen in this research, motivated by the fact that many EV battery innovations are emerging making the design of modules and packs prone to rapid changes.

Over the last 15 years, research has been conducted on the recycling of Lithium-Ion Batteries (LIB) cells, mostly focusing on the mechanical and metallurgical recycling processes [10]. However none of the described recycling methods is integrating robotic disassembly in their pre-processing of EVBs, i.e., processes which does not alter the structure of the LIB cells, and the mechanical or pyrometallurgical processes start with EVB cell modules as input. However, a large portion of metals to be remanufactured or recycled comes from housings (pack and module), electrical wire and connectors. For a Nissan Leaf first generation, the weight of the cells alone represents only 60% of weight of the total battery pack [6]. The disassembly process has been extensively studied in the literature, as shown in the survey [11] where disassembly processes of Waste of Electrical and Electronic Equipment (WEEE) are also present [5,12]. In article [13] authors presented an automatic mechanical separation methodology for End-of-Life (EOL) pouch LIBs with Z-folded electrode-separator compounds (ESC). Customised handling tools were designed, manufactured, and assembled into an automatic disassembly system prototype. While this aspect is still an active research field, the focus is now shifting towards automated solutions to support the whole recycling chain. Industrial solutions for the automated disassembly of battery-operated devices have been implemented, but they are limited to specific and often small-sized products. Apple has implemented an automated disassembly line for Iphone6 [14], however the process is not flexible nor adaptive and can only disassemble one model phone in perfect conditions. In article [15], cooperative control techniques are developed and demonstrated on the robotic disassembly of PC. In article [16], a vision system to identify components for extraction and simple robotic processes are used to disassemble printed circuit boards. Using visual information to automate the disassembly process is further developed with the concept of cognitive robotic systems [17] and is applied for disassembly of Liquid Crystal Display (LCD) screens [18].

Building upon the existing work in this area, this paper aims to improve the design of the task planner responsible for automatically generating the disassembly plan and sequences without precise a priori knowledge of the product to dismantle. The developed functions are presented and the results are validated through experiments conducted on a Hybrid Audi battery pack.

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

#### *2.1. Task Planner Design*

The experimental setup is composed of IRB4400 robot (ABB, Zürich, Switzerland), IRBT4004 track (ABB, Zürich, Switzerland), and Zivid One 3D camera (Zivid, Oslo, Norway) mounted on the robot arm, all connected to a PC with Ubuntu and running the Robot Operating System (ROS), and a A3 Sportback e-tron hybrid Li-ion battery pack (Audi, Ingolstadt, Germany). More information about the connection setup and use of the ROS as a middleware can be found in article [19], whereas a complete description of the Audi battery pack and its disassembly sequences has been presented in article [3]. The task planner proposed in this paper and shown in Figure 1 analyses the information provided by the vision system based on Zivid camera, makes decisions regarding dismantling actions and sends a predefined path to the industrial robot's controller.

**Figure 1.** Task planner concept.

The main loop of the task planner is organised as the following: takes 2D and 3D images, detects and identifies components, finds the component's positions in the world reference frame, defines an order of operations, removes the components, and repeats this actions until the goal state is reached (refer to steps A–E respectively in Figure 2).

**Figure 2.** Task planner main loop.

#### 2.1.1. Image Capturing (A)

The task planner moves the robot into several predefined poses to ensure that the system is able to observe different parts of the LIB pack. In order to reach better accuracy, up to eight pictures are required, especially when screws placed in the lateral sides of the battery are present.

#### 2.1.2. Object Detection (2D Image) (B)

Different components in the image such as screws, battery modules, connecting plates, and Battery Management System (BMS) are detected using the YOLO (You Only Look Once) algorithm [20] which also provides the bounding box positions of all constituting components. The output of the object detection procedure as shown in Figure 3 is a file containing all detected objects, labels and corresponding coordinates.

The positions of the detected objects in different images are merged using a weighted mean of each positions.

**Figure 3.** The YOLOv3 algorithm takes as input the 2D image of the EV Lithium-Ion Battery (EVB) pack, detects the components, finds the bounding box coordinates and class probabilities, and store the information in a text file. The labels shown on the picture have one color for each class.

#### 2.1.3. Decision Making (C)

In order to set a sequence of the removal operations, only one image is used, which is taken with lower camera angle with respect to a horizontal plane. The images with a higher inclination are used for object detection only. A list of the component positions in the removal order is created by adding the detected screw positions and following by a computer vision analysis of each specific component. Based on the probability of being over the other components the remaining parts are added in the list in a correct disassembly order.

#### 2.1.4. Position Calculation (D)

At this point the system defines the positions of the objects in 2D camera coordinates (pixels). In order to move the robot to those positions, they must be converted into 3D points. The 2D object coordinates in camera frame are transformed into 3D coordinates using the depth information of the 3D vision system. Then, the captured positions are known, so that the world reference position of the objects can be obtained. This action is done for all images. Once all the positions from the different points of view are found the nearby points (representing the same object or component) are merged.

#### 2.1.5. Robot Communication and Removal Operation (E)

Once the order of the operations and all the positions are known, the last step is to be proceed in dismantling. The task planner calls the removal operation of every single component. Details on the design of the removal operation are not presented in this paper, since a scope is limited to moving the robot arm onto the calculated positions, where the selected removal operation based on the classification of the component is achieved.

#### *2.2. Main Functions*

The main functions and scripts used by the task planner connect the computed information by the mean of tests and loops as shown in the complete flow chart in Figure 4.

**Figure 4.** Task planner flow chart.

#### 2.2.1. Function: *main()*

Fist of all, the main function *main()* whose flow chart is shown in Figure 4, declares and initialises all the variables used and transferred in and between the subsequent functions, as for example the screws, connecting plate, Battery Management System (BMS), or module positions and number. Then, it reads the file containing the classes names (classes.names), and saves them into a dictionary *(dict\_comp)*. The creation of this dictionary aims to allow the system to have access to the nomenclature in order to relate the detected classes with their name and characteristics necessary for further analysis.

Once the dictionary is created, the function starts its principal loop. Note that this loop keeps running until the dismantling is completed. The first action of the loop is to move the robot to predefined positions and take 3D images, i.e., XYZ + color (RGB) + quality (Q) for each pixel, of the battery pack. Next, once the algorithm has been trained, it runs the object detection to detect the components placed in these images, using the YOLOv3 algorithm.

After the detection, classification and pose estimation of the objects in the image frame are done. The function analyses each taken image, finds different components and their characteristics from Dict\_comp, and converts the positions of the objects from the image base frame to the camera reference frame. Following this step, a sub-process establishes priorities to enable the decision-making operation on what component should be removed next. The respective functions are named *what\_component* and *has\_comp\_over* and are further described in the next section. Thus, when the detected positions have been converted into the camera frame, they are further transformed into the world reference frame coordinates. At this stage, since several 3D pictures of the same components have been captured, the resulting multiple positions of the same components are merged, which increases the position accuracy. The output of this function is the positions of the different components in world reference frame. The components are then placed in order of removal preference and finally, the task planner's main function runs the removal operations for all the detected component before starting the main loop again if the goal state is not reached.

#### 2.2.2. Cognitive Functions: *what\_component()* and *has\_comp\_over()*

The function *what\_component()* flow chart is shown in Figure 5 and the function is responsible for the decision making. The function analyses the detected objects, and decides the order of the removal operations using the computer vision based sub-function *has\_comp\_over()* . The sub-function *has\_comp\_over()* flow chart is shown in Figure 6, and the function establishes different probabilities of a specific component to have a component over.

The inputs for the function *what\_component()* are:


**Figure 5.** Function *what\_component* flow chart.

**Figure 6.** Function *has\_comp\_over* flow chart.

First of all, the function creates the variable *compon\_rem* (array). The aim of this array is to contain the list of the components positions in the removal order. Given the nature of the disassembly process, the first components to be removed are screws, thus, their positions are the first to be added to the *compon\_rem* array.

To add the rest of the components and decide the order of the operations, the function *what\_component* runs over the list co\_tot analysing each component using the function *has\_comp\_over*.

Essentially, *has\_comp\_over* realize a computer vision analysis of each specific component, and returns the probability of being over the other components. Thus, the components are ordered from more probability to be over to less probability.

The inputs for the sub-function *has\_comp\_over* are:


The function *has\_comp\_over* runs a nested for loop over the list containing all the components, analyzing the ones that are intersecting the component that is being inspected in the outer for loop.

To realise this comparison, the full image is converted into grey-scale and equalised. The output image contains a better distribution of the intensities maintaining the relevant image information [21].

After obtaining the equalised image, the function crops the image into the area of interest, in this case, the area of the analysed component. Then, segmentation is applied to the window, binarising it using an Otsu threshold. In the Otsu method, the threshold is determined by minimizing intra-class intensity variance [22].

When the window has been binarized, the mean of the pixel intensities in the intersection area is compared to the means of the intersecting components areas excluding the intersection. The component for which these two means (intersection alone and component area excluding intersection) are the most similar is the most probable to be on the top layer, i.e., over all the others, and hence to be removed first. An example with two overlapping components is shown in Figure 7. In this example, component 2 is over component 1 because:

$$||\mathbf{f}(I) - \mathbf{f}(A\_2 - I)|| < ||\mathbf{f}(I) - \mathbf{f}(A\_1 - I)||\tag{1}$$

where *f*(*A*) is a function calculating the mean of pixel intensity in area A.

**Figure 7.** Component 2 with total area *A*<sup>2</sup> is overlapping component 1 with total area *A*1. The intersection area is *I*.

After repeating this procedure for all the components, the function returns the maximum value, max(p\_list), of the calculated probabilities of having another component overlapping the analyzed one.

#### 2.2.3. Merging the Pictures: *merge\_detection()*

The function *merge\_detection* aims to merge positions of the components (for this version only the screws) commonly detected in one or more images. The function flow chart is shown in Figure 8.

**Figure 8.** Function *merge\_detection flow chart.*

The inputs of the function *merge\_detection* are:


The *merge\_detection()* function runs a "for in range" loop to find positions of components in the world reference frame. It calls the *WR\_pos* function, where the positions are stored in the variable *rem\_component\_WR* including repeated positions.

In the next step the function merges the repeated positions to find the output list (*rem\_component\_WR\_filt*). In order to filter the points, the function runs over the *rem\_component\_WR* array adding not-repeated components to the filtered list. The function considers two components as one, when the x and y distances are lower than 1cm, and defines the final position as the mean of both points.

#### **3. Results**

The aim of this section is to validate the order suggested by the task planner and to characterise its performance regarding time and accuracy. The Audi A3 Sportback e-tron Hybrid Li-ion Battery Pack serves as the case study. The description of the EVB pack and its components as well as the disassembly process of the battery are detailed in article [3] whereas Table 1 presents the composition, i.e., relative weight of each components and materials. For safety reason when testing the concept, all modules have been manually discharged separately. However, when integrated in the pilot plant, the EVB packs will be discharged prior complete disassembly at a certain state of charge depending if the modules are to be repaired, re-purposed, re-manufactured, or recycled. In addition, damaged EVBs that represent high risk for thermal runaway or gas emission will be sorted out. These last two steps are outside the scope of the present work.


**Table 1.** Audi A3 Sportback e-tron battery pack constructive components and materials.

#### *3.1. Object Detection Results*

Figure 9 shows the results of the YOLOv3 algorithm implementation, where the red filled boxes indicate the position of the connective components, the blue-filled boxes indicate the screws positions, the pink-filled boxes indicate the position of the BMS and the black-filled boxes indicate the position of the battery modules.

**Figure 9.** YOLOv3 output results: all screws, modules, transverse covers, and BMS are detected, classified, and localised.

#### *3.2. Time Analysis*

The timings of the operations realized by the task planner, i.e., image capture, object detection, decision making, and motion to estimated pose, have been recorded on 20 repetitions with the physical setup, resulting in the mean times summarized in Figure 10. During experiments, the speed of the industrial robot has been reduced to 25% speed for safety reasons. The expected timing at production speed (100% speed) are shown in black and red in Figure 10.

**Figure 10.** Timing summary at 25% speed in grey and expected timing at full speed in black and red.

#### 3.2.1. Image Capture (Mean Time: 29.1 s)

In this case, the image capture process refers to the robot movement into the image taking positions and image capturing. In order to implement the capturing, the process has been divided into seven different actions. The first action refers to the robot model loading, and the rest refer to the robot movements and image captures, see Figure 11.

**Figure 11.** Image capturing actions.


#### 3.2.2. Object Identification (Mean Time: 4.8 s)

In this stage the YOLO algorithm is applied to three taken images to detect and identify different components on 2D images. Mean time: 4.8 s. In this study 3 images are analysed but up to 8 images are required to increase an accuracy.

#### 3.2.3. Data Analysis and Decision Making (Mean Time 9.2 s)

Data analysis and decision-making refers to calculating the object positions in the world coordinate frame and define the optimal path for the operations. Mean time: 9.2 s.

3.2.4. Move to the Desired Positions (Mean Time: 13.1 s)

The robot approaches the components to perform the removal operation. First, the robot rapidly moves to a safety position displaced thirty centimetres in the *Z*-axis above the object and then moves to the desired location. After that, the robot moves back to the safety position. The timings for this operation have been divided into six sub-processes.


#### *3.3. Decision-Making: Optimal Path*

The decision-making of the system (the order of the removal operations) affects directly onto the dismantling time. For this reason, the aim of this section is to analyse the order suggested by the task planner for the Audi A3 Sportback e-tron Hybrid Battery Pack.

#### 3.3.1. Optimal Dismantling Plans

After manually dismantling and analysing the battery pack, the optimal dismantling plan is to first remove the screws; the second step is to remove the two connective components and the battery management system (the order of the removal operations for these three elements is not critical); and finally to remove the four battery modules in a arbitrary order.

#### 3.3.2. Dismantling Plans Proposed by the System

A set of tests have been carried out under different conditions (i.e., different orientations, different ambient lights conditions, etc.), the system has given a good response.

It has been observed, within the proposed dismantling plans, that the system follows the guidelines defined in Section 3.3.1. Because the BMS and the two connective components (left and right) are not overlapping, the system is proposing two different plans that are equivalent. These are referred as the (A) and (B) plans and are illustrated in Figure 12.

In the (A) plan, the system begins removing screws. The screws are always the first components to be removed. Afterwards, the system removes one connective component (in some tests the left one in other tests the right one), the BMS, and the other connective component in that order. Finally, it removes the battery modules. See Figure 12. In the (B) plan, screws are still the first components to be removed. Then, the system proposes to remove the BMS and the connective components in that order. As in the previous plan, the battery modules are removed the last. Thus, the main difference observed between these two plans is that in the plan (B) the BMS is removed after the screws instead of connective plates. This has no impact on the final disassembly process.

**Figure 12.** Dismantling plans (**A**) or (**B**) proposed by the task planner.

#### *3.4. Accuracy*

To analyse the system's accuracy, a 3D printed pointer has been used as Tool Center Point (TCP). The tool consists of a thin 25 cm long bar with a sharp end. With the tool mounted, the task planner has been run in debug mode. For safety reasons, the robot TCP has been moved 3 cm above in the *Z*-axis (word base frame) in order to avoid collisions with the battery pack in case of failure. Some of the tests are shown in Figure 13. In the

majority of the cases, the system has an accuracy of (<5 mm). The accuracy has been measured by the mean of a laser beam attached to the red bar and pointing towards the target position, whereas the distance of the laser pointer to the target position is measured with a caliper.

**Figure 13.** Accuracy tests showing the red pointer 3 cm above one detected screw. Four different views of the same position.

#### **4. Discussion**

In this paper the proposed objectives have been achieved. Different research areas in computer vision, robotics, circular economy and electr(on)ic components (battery) disassembly have been successfully unified.

The assigned main hardware elements such as an industrial robot, 3D camera and PC have been interconnected to carry out the principal system tasks, such as object detection, pose estimation, decision-making, and robot displacement. Therefore, the system is able to recognise the dismantling object main components, to find their position, and to move the robot to the defined positions in a specific order. Lab tests have been used to validate the designed task planner. In this case, experiments were limited to Audi LIB pack, however, a similar procedure might be applied to any EV battery pack. Compared to other disassembly processes as reviewed by Zhou et al. [11], the proposed task planner relies on state-of-theart 3D camera system with high accuracy and does not require Computer Aided Design (CAD) models of the battery pack and its components. This presents a great advantage since EoL products are often different than their original CAD models, due to possible maintenance, deformations, or corrosion. Recognising the model and date of production of the EVB to be disassembled will help to determine a first disassembly sequence based on a self-updating database, but the system must also be flexible and robust enough to handle the above-mentioned variations or in the case of new or unrecognised model. Therefore, combined with reinforcement learning and machine reasoning algorithms the proposed disassembly framework will be able in future developments to learn how to disassemble new battery pack models, if not by itself, with only limited information from the human operator. The concept of cognitive robotics in disassembly has been developed and validated on End-of-Life treatment of LCD screen monitors [17]. However, some challenges remained as (1) the too high processing time making the process economically infeasible, (2) the remaining need for human assistance, and (3) the too high inaccuracy of the vision system leading to low success rate. This paper demonstrates that the recent advances in 3D vision system, fast object detection and localisation algorithms, as well as

task planner design place EVBs with inherent uncertainties and large variations in design as a good candidate for achieving eventually autonomous and complete disassembly through the cognitive robotic concept.

The proposed task planner for disassembly of EVB pack into modules can also be extended in future work to a deeper level of disassembly, i.e., to battery cell level or even to the cell components (cell casing, electrodes, electrolyte, separator) which will increase the concentration of active materials in the subsequent steps for battery recycling and hence reduce the complexity and energy consumption of the pyrometallurgical and hydrometallurgical processes. Removing manual operations in the pre-processing stages will move the optimum disassembly level determined in the article [3] deeper toward complete disassembly when still considering techno-economic and environmental constraints.

The algorithm You Only Look Once (YOLO) is implemented to detect and find the components placed in the dismantling arena. The results show that the algorithm performs well, giving expected results and detects main components. For example the presence of screws can be distinguished from the presence of screw holes where the screw has been removed. The developed vision system can hence also be used to validate removal operations. The information extracted from the object detection was used in a pose estimation to find coordinates of components, where 2D images and the YOLO results have been matched with the 3D data sets. In future versions of the task planner, object detection and pose estimation might be realised directly in 3 dimensions based on the point cloud data with techniques such as complex-Yolo [23] or DeepGCNs [24]. However, higher processing time or computing resources are to be expected.

When validating the task planner and measuring the timings, the robot model has been loaded every time that the robot had to move. Thus, in future stages the robot model should be loaded just once, at the beginning making the disassembly sequence at least 9.6 s faster. Moreover, the ROS main has been run in manual mode at 25% of the maximum speed of the robot. In automatic mode, i.e., at 100% speed, the total time of the disassembly sequence, excluding the removal operations time, is expected to decrease from 53.6 s to 34 s.

Using an eye-in-hand configuration performs well, and has some advantages (i.e., only one camera is needed), but it presents some drawbacks too. In an industrial application, the continuous moves and removal operations could have negative consequences like unexpected collisions of the camera with the environment or causing miscalibrations.

The efforts in future stages of the research should be focused on instrumentation and tool design for the dismantling system. It is also essential to detect flexible bodies such as high-voltage wires, and wires transmitting data. Thus, the direction of the research on the object detection and pose estimation part should concern how to find a feasible solution for such the objects and remove them.

#### **5. Conclusions**

A new task planner has been designed for the disassembly of electric vehicle Li-ion battery packs, with as main objective to increase the flexibility and robustness of the system. Lab tests have been used to validate the designed task planner based on a Audi A3 Sportback e-tron hybrid Li-ion battery pack. The results obtained in the tests demonstrate that the obtained solution is able to recognise which component to remove first and the complete disassembly plan without a priori knowledge of the disassembly strategy and battery CAD models. This method is therefore well suited for product with large variations and hence increases the disassemblability. The achieved performances measured in term of accuracy, time to generate the disassembly plan and success rate validated the task planner concept and its ability to make autonomous decision. Further testing on a larger set of EV battery packs with other geometries and connections and addition of learning capabilities will be needed to further increase the robustness of the proposed method and the technological readiness level. However, the results already cast a new light on the use of automation in the EV LIB batteries disassembly process by bringing the technology one step closer to eventually fully automated operations and hence redefine the optimum level of disassembly for the batteries to enter the subsequent stages of recycling and metal recovery.

The experience in this field could also be adapted to be used for other dismantling processes and opens new doors and research challenges to other fields directly related to robotics.

**Author Contributions:** Conceptualization, E.M.B., M.C. and I.T.; methodology, E.M.B., M.C., and I.T.; software, E.M.B.; validation, E.M.B., M.C., and I.T.; formal analysis, M.C., I.T., and E.M.B.; resources, M.C. and I.T.; writing—original draft preparation, M.C., E.M.B., and I.T.; writing—review and editing, M.C., E.M.B., and I.T.; visualization, E.M.B., M.C. and I.T.; supervision, M.C., I.T.; project administration, E.M.B., M.C., and I.T.; funding acquisition, M.C. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by The Research Council of Norway grant number 282328.

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

**Informed Consent Statement:** Not Applicable.

**Data Availability Statement:** Not Applicable.

**Acknowledgments:** Special thanks go to BatteriRetur AS for the battery pack used for experiments. This work contributes to the research performed at TRCM (Priority Research Center Mechatronics at the University of Agder, Norway).

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

#### **References**


MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Metals* Editorial Office E-mail: metals@mdpi.com www.mdpi.com/journal/metals

MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel: +41 61 683 77 34

www.mdpi.com

ISBN 978-3-0365-5926-1