The Application of Artiﬁcial Intelligence in the Effective Battery Life Cycle in the Closed Circular Economy Model—A Perspective

: Global pollution of the environment is one of the most challenging environmental problems. Electronic-based population and anthropogenic activity are the main reasons for dramatically increasing the scale of waste generation, particularly battery waste. Improper battery waste disposal causes harmful environmental effects. Due to the release of heavy metals, battery waste affects ecosystems and health. We are faced with the challenge of effective battery waste management, especially recycling, to prevent the depletion of natural resources and maintain ecological balance. Artiﬁcial Intelligence (AI) is practically present in all areas of our lives. It enables the reduction of the costs associated with various types of research, increases automation, and accelerates productivity. This paper reviews the representative research progress of effective Artiﬁcial Intelligence-based battery waste management in the context of sustainable development, in particular, the analysis of current trends, algorithm accuracy, and data availability. Finally, the future lines of research and development directions of human-oriented Artiﬁcial Intelligence


Introduction
The growing standard of living, inextricably linked with the growing demand for electronics, is driving the market for the recycling of spent batteries. The global battery recycling market has been valued at USD 10.21 billion in 2021 and is projected to reach USD 18.96 billion by 2030, with a compound annual growth rate (CAGR) of 7.12% over the forecast period. In the case of lithium-ion batteries (LIBs), the market has been valued at USD 6.55 billion by 2028, with a CAGR of 18.5% over the forecast period [1]. The battery market has rapidly grown, where only in 2019, around 205,000 tons of portable batteries and accumulators were placed on the market in the EU. The production of electric vehicles (EVs) is also increasing and, consequently, the number of batteries produced. Another issue contributing to this is that the battery used in an electric car, which has lost 20 percent of its capacity, is considered to be used up. Recalled EV batteries pose a significant threat to the environment because they contain heavy metals in many cathode materials and toxic and corrosive electrolytes [2]. Among them, cobalt accounts for 60 percent of the cost of cathodes [3]. At the same time, approximately 100,000 tons of battery waste were collected and recognized as recyclable. It also turned out that the collection of used portable batteries and accumulators is lower than the sale [4]. The rules for monitoring battery waste, taking into account goals of waste collection and the effectiveness of recycling according to the formulation, choices of additives and solvents, time and speed of premixing powders and slurries, coating speed and comma spacing, and time and evaporation temperature. Here, methods based on Artificial Intelligence (AI), in particular, Machine Learning (ML), can significantly simplify this process and reduce its cost while they are operating on multidimensional data sets [17]. The first important issue is to collect a large amount of reliable data on which the algorithms can perform the optimization. Incorrect assumptions and unreliable data will lead to unreliable results. Some guidance on designing suitable AI-based methods is applied to estimate the state of battery charge [18,19] and predict the battery life cycle [20,21], or LIB electrode manufacturing [22].
The State of Charge (SOC) depends on several factors, such as temperature, ageing, cell unbalancing, hysteresis characteristics, self-discharge, and charge/discharge rate. It plays an important role in predicting EVs' driving race and optimal charge control, which are crucial in reducing the carbon footprint. It can be estimated using various methods based on Artificial Intelligence, but each disadvantage is the accuracy and availability of data. The estimation of SOC requires applying the algorithm to describe the battery's remaining capacity, which was described in the study [19]. In the paper [23], a simple deep neural network combined with a Kalman filter was used to estimate the SOC of the battery. In [24], the fuzzy logic methodology was used for this purpose, which analyzed the data coming from impedance spectroscopy and/or coulomb counting techniques. A genetic algorithm was used to evaluate the various types of batteries [25,26]. Genetic algorithms provide less estimation error (5 times smaller) compared to fuzzy logic ones. The support vector machine (SVM) was used to establish the relationship of the SOC to the Ni-MH battery's voltage, current, and temperature [27]. Thus, the paper [18] proposed a recurrent neural network (RNN) with long short-term memory (LSTM) for the estimation of SOC in the case of LIB. The algorithm was based on measured voltage, current, and temperature. In [28], the dependence on ambient temperatures is included. In turn, in the study [29], convolutional neural networks (CNN) and RNN were used to predict. This approach enables the prediction of SOC with a maximum mean average error under 1% and a maximum root mean square error under 2%, based on discharge profiles. It provides a reasonable estimation of nonlinear relationships between SOC and measurable variables. Recently, hybrid methodologies to estimate SOC were investigated in the study [30]. In work [31], an adaptive extended Kalman filter was proposed. Thus, hybrid techniques have the potential to multiply the advantages of individual components and thus enable a more accurate SOC estimation.
On the other hand, the study [20] shows that Machine Learning-based techniques can predict the battery life cycle with a 4.9 percent test error using the first five cycles, considering the evolution of the discharge voltage courage. In the paper [2], the cognitive digital twin batteries' design and development were shown. This Artificial Intelligencebased digital creation enables research to optimize the entire life cycle of a battery. In [20], cycle life prediction models were proposed. As input data the cycle lives of batteries ranging from 150 to 2300 using 72 different fast-charging conditions have served. In research [22], Artificial Intelligence-based tools, in particular based on a decision tree, deep neural network, and SVM to predict correlations between LIB properties and manufacturing parameters, were proposed. It took into account the characteristics of the electrode, namely the active material mass loading and porosity. It turned out that SVM links high accuracy of prediction (above 90 percent) with the possibility of graphical analysis of the results. A huge effort has been made to understand and experimentally validate the batteries, which are working with constant current, voltage, and temperature, while there is still a gap in the case of the batteries, which are working in severe, hot, wet, and rainy conditions. Here, the surrogate battery models can be helpful, and they can be used as an input dataset to the battery optimization process [32]. Thus, Artificial Intelligence can help increase the sustainability of batteries.
Artificial Intelligence can also be applied as an effective tool for the analysis of the material characteristics of battery [33] and the LIB failure mode [34]. In the study [33], as a training set, public battery cycling data, which contains 124 LiFePO 4 /graphite cells being cycled to end-of-life [31], was used. It turned out that to predict the battery properties with high accuracy, only single-cycle data are needed. The interesting solution for evaluating the residual energy of lithium-ion batteries (LIBs) based on Artificial Intelligence, in particular genetic programming, was presented in the paper [35]. The quantitative results determined the relationship between stress and capacity and can provide an optimized recycling strategy for batteries applied to electric vehicles, which is extremely important, while current generations of batteries link active materials with high energy densities with highly inflammable electrolytes.
In the paper [36], the dependence between the properties of the anode and cathode, manufacturing control parameters, the intermediate product characteristics, and the final cell performance with Machine Learning techniques were investigated. The Gradient Boosted Decision Trees (GBT) and random forest were applied to predict positive and negative electrodes as well as characteristics of the half-cell. As input data, a comma bar gap, coating speed, and coating ratio have been used. It turned out that the electrodes, including mass loading and thickness, can be predicted with an accuracy of 93.48 percent and the cell capacity with an accuracy of 91.75 percent.

Artificial Intelligence in Waste Management, Including Battery Waste Management Systems
Artificial Intelligence-based algorithms can solve various issues of information processing, including pattern recognition, classification, clustering, dimensionality reduction, image recognition, natural language processing, and predictive analysis. Recently, Artificial Intelligence was also applied in waste management [37,38], providing the opportunity to link waste management, joint supervision and collection process, and safety.
Another important issue in waste management is connected with the efficiency of the cleaning process, while Artificial Intelligence can also support waste collection schedules. The intelligent trash cans can send data, such as the presence and volume occupied by garbage, using the Internet. In the paper [39], a waste collection system based on location intelligence and applying graph optimization algorithms as a part of Smart City (Copenhagen, Denmark) was proposed. The proposed solution returns the data concerning trash level collected by the embedded sensors to the server over the Internet, which optimizes the collection routes and sends this information to workers. In this study, input data were: waste level of trashcans, which come from 3046 trashcans, and available open data about the city of Copenhagen, Denmark. On their basis, the optimal schedule of waste collection from individual places is determined, taking into account the optimization of the driving distance of the daily routes based on the Shortest Path Spanning Tree (SPST) to calculate the minimum driving distance between points and a genetic algorithm to predict the minimal driving distance between the points, is determined.
The identification, localization, and size determination of waste are based on image recognition techniques. In the study [40], based on images, the determination of the location and the degree of filling of the containers with the use of four Laws Masks and a set of support vector machine (SVM) classifiers with 99.8 percent accuracy was proposed. The containers were classified into three groups, i.e., empty, partially full, or full. The assignment to a particular group determined the garbage collection schedule. Input data were in the form of pictures of bins and the nearest neighborhoods (800 × 600 pixels), including 60 rotated and 160 unrotated. As a training set, unrotated pictures were used, while during testing of the solution proposed, both unrotated and rotated pictures were. All pictures were converted into grayscale and subjected to the automatic edge detection procedure. The bin position of the image was detected with Hough line detection and cross-correlation. It turned out that the algorithms proposed are robust against bin shift and rotation. In the research [41], the classification of electrical and electronic waste from trash pictures using the deep learning convolutional neural network (CNN) was presented. The proposed solution provides efficiency of 97 percent. As input data, pictures of refrigerators, washing machines, and television sets (three classes) in the RDG format (128 × 128 pixels) were taken. The training set includes 160 pictures (60 for each class), while the testing set includes 30 pictures (10 for each class). The pictures of waste are sent to the server, where they are subjected to the object recognition procedure. Once the waste is identified and located, waste collectors can plan for efficient collection. The systems can recognize three categories of e-waste, namely: refrigerators, washing machines, and monitors or TV sets. In the paper [42], convolutional neural networks were used to identify hazardous recyclable materials, such as batteries, syringes, and nonhazardous waste, with an accuracy of 90 percent. Datasets, including three categories (i.e., batteries, syringes, and nonhazardous waste), in the number of pictures taken in front of a white background with moderate lighting (512 × 384 pixels): 23, 91, and 1984, respectively. Artificial Intelligence-based algorithms are also involved in trash control in institutions, for example, universities [38]. This system combines linear regression (LR) with Machine Learning techniques. Dijkstra's algorithm optimizes the path for waste collection based on historical data. It operates on data containing information about the current state of filling the bin, i.e., the level of waste and bin position. The pictures were collected for 4 months during the academic year.
Thus, waste management can be treated as a multi-hierarchical clustering problem. In the paper [43], the concept of an AI-based classification of medical waste, e-waste, and toxic atmosphere pollutants, taking into account real-time indicator conditions such as daily waste and strain, was proposed. This system contains three modules: the input module (responsible for defining the essential trash characteristics), the second level module (description of the toxic patterns), and the community module. The general idea of the system is derived from LCA, MCA, and Extended Producer Responsibility. It enables e-waste tracking, taking into account the safety of the whole process. In the case of the application of Artificial Intelligence, this can reduce the duration of the assessment process by at least 35 percent.
Waste management, in particular solid waste, is an important issue, taking into account the negative impact on human health and the environment [44,45]. For an efficient waste management system, Artificial Intelligence has great potential [46]. According to the research analysis presented in [47][48][49], the reduction of waste through recycling helps to achieve a circular economy. The prediction of an accurate waste amount, mass, and type is crucial in waste management. Thus, in the paper [50], the convolution neural network was used to predict the waste mass. In the study [51], artificial neural networks and the Machine Learning framework (MLDPAF) were applied to the effective planning of waste management, including the prediction of waste amount and effectiveness of waste collection. The research [51] shows an attempt at waste management on an academic campus. In turn, in the paper [52], the concept of an effective construction waste management system was proposed.
Another issue connected with waste management strategies is waste amount prediction. In the study [53], multi-layer perceptron artificial neural networks (MLP-ANN) were used for the verification of annual waste production, including municipal, commercial, construction, and demolition waste. For the forecast, the data, which contain solid waste datasets deposited at Askar Landfill in Bahrain between 1997 and 2016, were used. It turned out that artificial neural networks enabled the estimation of the future-proof generation of different types of waste with high accuracy. In the paper [54], the comparison of different artificial neural networks, i.e., adaptive neuro-fuzzy inference systems, discrete wavelet theory artificial neural networks (DW-ANN), discrete wavelet theory-adaptive neuro-fuzzy inference systems (DWT-ANFIS), and genetic algorithms, for the amount of waste prediction has been made. This study covered two data streams, namely, data that come from governmental, semi-governmental, and private publications from the period of 1993-2011 and data that come from field surveys. It turned out that the most accurate forecast was delivered by a genetic algorithm. In the study [55], four options were used to estimate the ability of intelligent systems algorithms to predict monthly amounts of waste generated-support vector machines (SVM), adaptive neuro-fuzzy inference systems (ANFIS), artificial neural networks (ANN) and k-nearest neighbors (kNN). It has been shown that AI can be successfully used to estimate the amount of generated waste, and the best results were obtained for the ANFIS (most accurate peak forecasts) and kNN (monthly average waste prediction) systems. The medical waste generation rate was estimated in [56] based on multiple linear regression, artificial neural networks, fuzzy logic-artificial neural networks, support vector regression, least squares support vector regression, and fuzzy logic-support vector regression. It turned out that in the case of hospital solid waste, the higher accuracy was provided by fuzzy logic-support vector regression.

Artificial Intelligence in the Waste Sorting
Waste sorting, i.e., the process of separating waste into different types, plays a crucial role in the closed circular economy model [57]. The available sorting methods can be divided into two groups: manual sorting and automated/mechanical sorting, with the application of robotic technology or a combination of these two types [58]. In the case of solid waste, the manual approach prevails [59]. To provide an automatic waste sorting system based only on pictures of waste in [60], convolution neural networks and support vector machines were applied. The system classified waste into three groups, namely plastic, paper, and metal. It operates on colored images in png format (256 × 256 pixels). It turned out that support vector machines provided higher efficiency than convolution neural networks. An interesting approach was proposed in the paper [61], placing RFID tags on packages that would enable the identification and classification of individual plastic packages, for example, using Artificial Intelligence. The recycling robot ZRR2 from ZenRobotics in Finland [62] was the first attempt to apply such a solution in practice [63]. It has built-in computer vision and deep learning algorithms. The robot enables the automatic separation of selected waste from solid construction and demolition waste. In the study [64], the ZRR robot was applied to the sorting of municipal household waste streams. In this case, the main limitation in the application of the system is the protection of personal data from households. In turn, for the already collected waste the identification to sort them into two groups, i.e., glassware and plastics, based on a convolution neural network was proposed in the paper [65]. The input data was gathered with an RGB camera, i.e., 103 pictures of waste (50 glassware, 53 plastics). To increase the amount of data, image enhancement was done to the training set. After identification, the gripper sorting robot separated the waste into two groups. In the study [66], an Artificial Intelligence-based, especially hierarchical deep learning, algorithm was applied to waste detection and classification in food trays. As input, the Labeled Waste in the Wild dataset was used, which contains 1002 RGB pictures of used food trays (3456 × 4608 pixels) that have been taken with several different smartphones. Some of the objects shown in the photos were not wasted. These pictures were used to label the shape and material of the visible waste. In the paper [67], to distinguish nails and screws in construction waste, a region-based convolutional neural network was applied. The COVID-19 pandemic also revealed the need for automatic sorting of medical waste, including polyethene terephthalate (PET) waste from the pandemic period. In the study [68], the support vector machine with an accuracy of 96.5 percent was proposed for this purpose.
Since improper segregation of biomedical waste causes health hazards, the application of Artificial Intelligence to their disposal and recycling seems to be a reasonable solution [69]. One of the directions is smart bins [70]. In the first step, they have the possibility to identify and segregate waste. Next, they choose a suitable disposal method and transfer it to recycling. In turn, Abeygunawardhana et al. [71] presented the concept of such smart bins for sorting the most common solid waste-metal, glass, and plastic. Their proposal uses the convolutional neural network (CNN) system-waste is segregated using image processing and machine learning algorithms. The use of CNN made it possible to obtain a 70 percent validation accuracy with a loss of 0.03.

Artificial Intelligence in Battery Waste Recycling
Effective and environmentally friendly waste management is one of the biggest problems in the whole world. Waste processing and recovery are crucial elements in waste management systems [72,73]. One of the crucial parts of battery waste management is the recycling process [74]. Lithium-ion batteries can be recycled using various methods, including pyrometallurgical, hydrometallurgical, and biological recycling to recover valuable metals [75][76][77]. Figure 1 shows a schematic diagram of the management system and waste recovery methods for the current batteries (including LiBs).
lution [69]. One of the directions is smart bins [70]. In the first step, they have the possibility to identify and segregate waste. Next, they choose a suitable disposal method and transfer it to recycling. In turn, Abeygunawardhana et al. [71] presented the concept of such smart bins for sorting the most common solid waste-metal, glass, and plastic. Their proposal uses the convolutional neural network (CNN) system-waste is segregated using image processing and machine learning algorithms. The use of CNN made it possible to obtain a 70 percent validation accuracy with a loss of 0.03.

Artificial Intelligence in Battery Waste Recycling
Effective and environmentally friendly waste management is one of the biggest problems in the whole world. Waste processing and recovery are crucial elements in waste management systems [72,73]. One of the crucial parts of battery waste management is the recycling process [74]. Lithium-ion batteries can be recycled using various methods, including pyrometallurgical, hydrometallurgical, and biological recycling to recover valuable metals [75][76][77]. Figure 1 shows a schematic diagram of the management system and waste recovery methods for the current batteries (including LiBs). The pyrometallurgical approach is based on the high-temperature treatment of the battery waste in a wide range of temperatures in the furnace. During this process, the decomposition of organic materials occurs, and new alloys are formed [78]. It is an effective way to recover metals such as cobalt, nickel, and copper, while lithium, manganese, and aluminum get into slag or kiln dust. As a consequence, lithium, manganese, and aluminum can be extracted with a large financial outlay in another process. During this process, semi-finished products are produced, which, to be reusable, have to be subjected to further purification. The huge disadvantage of the pyrometallurgical process is the small number of recyclable materials and low efficiency in the case of low concentrations of recyclable materials [78,79]. The pyrometallurgical process is quite simple and does not cause any operational problems, but it causes air pollution and requires a lot of energy [80]. Moreover, there is no need for sorting or reduction of battery size [81,82]. Low energy consumption and high recycling efficiency are the hallmarks of hydrometallurgy processes [83]. Hydrometallurgical methods of recovering metals from used batteries most often mean acid leaching, which is based mainly on the application of strong inorganic The pyrometallurgical approach is based on the high-temperature treatment of the battery waste in a wide range of temperatures in the furnace. During this process, the decomposition of organic materials occurs, and new alloys are formed [78]. It is an effective way to recover metals such as cobalt, nickel, and copper, while lithium, manganese, and aluminum get into slag or kiln dust. As a consequence, lithium, manganese, and aluminum can be extracted with a large financial outlay in another process. During this process, semi-finished products are produced, which, to be reusable, have to be subjected to further purification. The huge disadvantage of the pyrometallurgical process is the small number of recyclable materials and low efficiency in the case of low concentrations of recyclable materials [78,79]. The pyrometallurgical process is quite simple and does not cause any operational problems, but it causes air pollution and requires a lot of energy [80]. Moreover, there is no need for sorting or reduction of battery size [81,82]. Low energy consumption and high recycling efficiency are the hallmarks of hydrometallurgy processes [83]. Hydrometallurgical methods of recovering metals from used batteries most often mean acid leaching, which is based mainly on the application of strong inorganic acids and reduction. For example, refs. [84][85][86] proposed the application of sulphuric acid and hydrogen peroxide as leaching agents due to the fact that the use of strong inorganic reagents is associated with technological problems, such as corrosion and rapid destruction of equipment, the emission of toxic vapors, and the danger of working with strong chemicals, currently. The interest of scientists is focused on the possibility of applying organic acids (e.g., acetic, citric, and DL-malic acids) in the leaching process of spent batteries [87,88]. In addition, an up-and-coming alternative to the pyrometallurgical and hydrometallurgical recovery of metals from waste batteries is the bioleaching process using microorganisms such as bacteria and fungi [89,90]. Biological methods of metal recovery allow for the reduction of the formation of secondary pollutants (including no toxic gas emissions) and, at the same time, are characterized by high efficiency, safety, and the relatively low costs of the process. However, the duration of the reaction in most cases is longer than for the acid leaching with the use of chemical reagents [91,92].
Since the recycling of metals from battery waste is a complex task, its efficiency can be improved by the application of various prediction methods, including Artificial Intelligence [93]. In the paper [94], the Machine Learning approach, including linear regression, random forest regression, AdaBoost regression, gradient boosting regression, and XG boost regression, to optimize the metal recovery of Zn and Mn from battery waste was proposed. As input, data on energy substrate concentration, pH control of bioleaching media, incubating temperature, and pulp density were used. The maximum Zn and Mn yield was the output data. It turned out that XG boost regression provided the best estimation, while linear regression was the least accurate. While the lithium-ion batteries from electric vehicles cannot be directly reused, the development of effective sorting of cells is of high importance [95]. In the study [96], the screening method for retired battery packs was shown. The support vector machine, with an accuracy of 96.8 percent, was applied. The input data come from 12 retired batteries, i.e., 240 cells, and include their capacities and resistances. It turned out that the proposed approach can reduce the time needed for sorting and four-fifths, in comparison to the manual process. In the paper [97], the sorting methods of lithium-ion batteries in large quantities were described. The degradation state of the battery was determined with X-ray radiographic scanning and digital image contrast computation. The proposed approach provides an accuracy of 79 percent. In turn, in the study [98], the Artificial Intelligence-based sorting method was applied to the recycling of unused mobile phones. As a first step, the retired batteries from mobile phones were subjected to magnetic separation, eddy current, and pyrometallurgical and hydrometallurgical processes. Next, the pictures, which were taken with purified metal, were classified with the convolutional neural network with rectified linear unit (ReLU) activation function. To increase the amount of input data, image augmentation was used.

Discussion
In e-waste management, it is extremely important to create an eco-system that enables their proper processing, recycling, and reuse. The recycling processes must be flexible and adaptable to the next generation of batteries [2]. That leads to sustainable battery cell production as a green technology of the future [99,100]. In turn, the information and communication technologies (ICTs) and Internet of Things-based concepts of waste collection were postulated [43,[101][102][103]. Since smartphones and mobile applications are commonly used in the information society [104], they can also be used in waste collection. Such applications are responsible for data collection and transfer on servers. The Artificial Intelligence-based approaches are reasonable promises in data analysis. Artificial Intelligence-based algorithms can also be helpful in the estimation of the influence of the first battery life on the recycled one, and this context, the evaluation of the different recycled materials [105]. The early prediction of the battery life cycle provides a decrease in its degradation process, while it is associated with the non-linear character of this process [106]. To enable battery monitoring in real-time cloud computing, in particular cloud storage, as well as blockchain [107,108] and the digital twin-based battery cloud management system [109,110] can be applied. Moreover, high-performance processor units can accelerate the whole computational operation, including learning processes. Based on the battery field data for low dynamics and high dynamics (capacity, nominal voltage, weight, cathodes, and anode material), and impedance, the digital twin can be used for online monitoring of the electrode level [111]. In the case of the field data, several factors affect the accuracy of the identification, including the sampling rate and sensor rate. With the increase in the sampling rate, the computational and storage costs increase. The resulting sampling rate is a compromise between the accuracy of identification and the cost incurred. Thus, Artificial Intelligence-based algorithms are the promising direction in the production, manufacturing, optimization, and monitoring of batteries [20].
It is known that Artificial Intelligence-based algorithms, in particular Machine Learning techniques, require a large amount of data to be trained and tested; however, their quality is not without significance. Datasets that are too small or contain low-quality data can lead to incorrect predictions. Thus, the first step to developing Artificial Intelligencebased algorithms is the implementation of adequate acquisition, storing, and management of the data. Providing good quality, reliable data is of high importance [17]. To ensure the amount of data that is appropriate to train the models, for example, in [18] to predict the state of battery charge, a training set was created in the laboratory by applying drive cycle loads at various ambient temperatures to a Li-ion battery and supplemented with data from the public database Panasonic 18650PF Li-ion Battery Data, which includes HPPC, drive cycles, and impedance spectroscopy tests that were performed taking into account the temperature impact [112]. Taking into account the algorithms' accuracy, the most commonly used split for training to testing data that is applied in the Artificial Intelligence-based algorithms is 90% to 10%, while in many papers in the field of waste management, a 70% to 30% split for training to testing data is considered [56,113]. The relatively small amount of data might lead to overfitting issues; to avoid this, one can combine theoretical data with experimental measurements or/and apply the algorithms that are dedicated to the small dataset, for example, the hierarchical Machine Learning approach [17]. Another issue is connected with the lack of data standardization, in particular for experimental data in the material sciences [114]. Moreover, the cross-validation of different Artificial Intelligence-based algorithms should be attached to detect overfitting.
An important aspect is also data security, especially for data from households and institutions [115]. It is connected with privacy laws and a lack of regulation, to the author's knowledge, that protects personal waste, including e-waste and battery waste. This may limit the intelligent waste management systems [116] while opening access to individual household data.
On the other hand, intelligent sorting of waste materials can also significantly decrease waste management costs and increase efficiency. Thus, manual source segregation provides a higher possibility of waste recovery, a tedious and time-consuming process. The automation process will not only improve its efficiency but also positively affect the employees' health [45].
It is also worth stressing that all waste management systems, including battery waste management, consume energy, mostly from batteries in the devices/sensor used in the solutions. At the same time, they can be replaced by renewable energy sources such as photovoltaic panels for battery charge [117].
To summarize, the algorithmic performance comparison in terms of their types and accuracy has been done in Table 1. For waste classification and sorting based on image recognition algorithms like deep neural networks, region-based convolutional neural networks, convolution neural networks, support vector machines, k-nearest neighbor, and random forest were applied. Convolution neural networks provide the highest efficiency [40,[118][119][120]. In turn, random forest, multilayer perceptron neural networks and convolution neural networks were also successfully applied to waste management [50,121], while the support vector machine and naive Bayes provide lower accuracy. For a forecast of the number of annual waste generation AI-based algorithms, including support vector machine, fuzzy logic-support vector regression, adaptive neuro-fuzzy inference system, artificial neural network, k-nearest neighbors have been proposed. It turned out that the highest accuracy of prediction can be obtained using an adaptive neuro-fuzzy inference system [55]. For the optimization of the metal recovery from battery waste using an automated neural network, linear regression, random forest regression, AdaBoost regression, gradient boosting regression, and XG boost regression were applied. It turned out that in the case of Zn recovery, the highest accuracy provides automated neural network [113], while in the case of Mn recovery, XG boost regression was used [94]. Moreover, the optimization linear regression provide the wrong prediction, namely a negative value for non-negative property, in the case of the metal recovery of Zn from battery waste.

Conclusions
Despite the recent developments in battery production and battery waste management, there is still a need to fill the gap between laboratory studies and real-world applications. In this review, we presented the recent capabilities of Artificial Intelligence, in particular Machine Learning and computational intelligence to develop algorithms, which can support effective battery life cycles, from projecting and production to battery waste management and recycling in accordance with the model of the closed circular economy. It turned out that convolution neural networks provide the most effective IT tools to evaluate the classification and sorting of waste [40,[118][119][120], in particular, battery waste based on image recognition algorithms as well as waste management [50,121], while the adaptive neurofuzzy inference system provides the prediction of annual waste generation with higher accuracy [55]. In effective recycling, a high percentage of recycled raw materials is an important issue. The prediction of it can also be successfully made with an Artificial Intelligence-based approach [94,113]. Some of the recommendations for future research are: 1. Since the development of efficient Artificial Intelligence-based algorithms strictly provides an adequate amount of good-quality data, the development of a standardized public database, which will contain battery data [17], for example, sleep data-national sleep research resource (NSRR, the database that contains physiological signals and clinical data [130]). In turn, many prediction algorithms can be based in part on experimental data from published papers and technical reports. Here, also involving the text mining methods to extract and analyze the results and conclusions already obtained will be good practice [131,132]; 2. The analysis of the relationship between manufacturing factors and anode and cathode parameters, taking into account other battery manufacturing processes such as slurry mixing and calendaring based on an Artificial Intelligence approach [36]; 3. The development of the battery multiscale models [133]; 4. The development of sorting and separation technologies supported by Artificial Intelligence, which can increase the efficiency of recycling [134]; 5. The development of local recycling or pretreating, which can reduce the cost of hazardous LIBs transport [135]; 6. The increase in the accuracy of waste detection and identification based on Artificial Intelligence, in particular Machine Learning and computational intelligence [136]; 7. Increasing the speed of data transfer, which is especially important in the part of waste management, in particular battery waste, related to efficient collection [137]; 8. The security of the data, while the intelligent waste management systems open access to individual household data [115].
Moreover, strict legal regulations must also be established in the area of battery waste, including greater producer responsibility. The recycling capacity has been developed mostly in places with significant recycling regulations, like China [138]. Moreover, the European Union has suggested legislation in the area of battery collections, labeling, and recycling [139].