A framework of digital technologies for the circular economy: Digital functions and mechanisms

Funding information S~ao Paulo Research Foundation, Grant/Award Number: 2019/23655-9; EPSRC Internet of Food Things Network Plus, Grant/Award Number: EP/R045127/1; Royal Society, Grant/Award Number: NMG\R1\191115 Abstract Digital technology is regarded as providing a promising means of moving production and consumption towards the circular economy. However, it is still unclear which functions of digital technologies are most useful to improving circularity, and how these functions could be used to enhance different circular economy strategies. This paper aims to address this knowledge gap by conducting a systematic literature review. After examining 174 papers, creating 782 original codes and 259 secondround codes, the study identifies 13 critical functions of digital technologies which are most relevant to circular economy strategies. The paper then proposes a framework which reveals seven mechanisms of how these digital functions can enhance different circular economy strategies. The framework also reveals which combinations of the digital functions and circular economy strategies have already been studied extensively as well as where there may be gaps. This indicates which digital functions are more mature in terms of possible implementation for circular economy as well as what missing links there are in the empirical and theoretical research. The study advances the synergies between digital technologies and the circular economy paradigm through the lens of digital functions. The proposed framework and mechanisms build a theoretical foundation for future research, and we highlight five research areas for further studies. This study also provides a structured way for managers to explore the appropriate digital functions for their CE strategies, so as to identify required digital technologies and new value creation through digital functions.

. It has gained increasing attention from governments, practitioners, and researchers (Korhonen et al., 2018;McDowall et al., 2017). It addresses the creation of a resourceeffective and resource-efficient economic system mainly through intentionally narrowing, slowing and closing material-and energyflows World Economic Forum, 2016).
At the same time, emerging digital technology (DT), such as the internet of things (IoT), big data analytics (BDA), artificial intelligence (AI), and 3D-printing, has been radically changing the way products are made, delivered, sold, and consumed (Lasi et al., 2014). Known as Industry 4.0, the new industrial stage not only changes the manner of production but also causes versatile organizational transformation (Vaidya et al., 2018). With the emerging technologies, devices can communicate with other devices and services over the internet to achieve a diversity of goals (Whitmore et al., 2015), such as automated manufacturing, home automation, and smart waste management.
There is an increasing interest in the potential of DT in moving production and consumption towards CE (Awan et al., 2021).
Implementing DTs is considered a promising means to overcome barriers to the CE transition World Economic Forum, 2016). It can provide CE opportunities for the manufacturing industry, such as retrofitting equipment, increasing workers' efficiency and motivation, building a smart factory based on resource efficiency, and designing closed-loop manufacturing process chains (Stock & Seliger, 2016).

Current research provides insights into the interactions between
DT and CE from various perspectives. Some articles have developed frameworks linking specific DTs to various CE strategies (Ingemarsdotter et al., 2019;Kristoffersen et al., 2020;Nobre & Tavares, 2020a;Okorie et al., 2018;Rosa et al., 2020). Others link DTs to other concepts, such as lean manufacturing (Chen et al., 2020;Ciliberto et al., 2021) and sustainable development (Furstenau et al., 2020;Liu et al., 2020;Sun et al., 2021) and then link these concepts to CE. Awan et al. (2021) focus on IoT tools and implementation practices, while another stream looks at the implementation challenges (Liu et al., 2021;Lobo et al., 2022).
Despite the growing interest, the theoretical understanding of the mechanism whereby DT can be used to implement CE strategies is still inadequate and underdeveloped (Cagno et al., 2021). While some research focuses on one or a limited set of DTs, such as IoT and BDA (Ingemarsdotter et al., 2019;Mboli et al., 2020;Nobre & Tavares, 2020a;Reuter, 2016), other research focuses on a specific situation, such as supply chain management, remanufacturing, and recycling (Kerin & Pham, 2020;Sarc, 2021;Yadav et al., 2020). Kristoffersen et al. (2020) proposed a smart CE framework for exploiting IoT-and BDA-based business analysis for CE implementation and innovation. It focuses on data and business analytics but did not address the possibilities of AI and automation (Ellen MacArthur Foundation, 2019;Fraga-Lamas et al., 2021;Lutje et al., 2020). Therefore, the mechanisms of DTs enabling CE strategies still need further development. It needs not only to provide a holistic, strategic CE transformation view, but also to investigate how the combination of DTs enable CE from an operative perspective (Cagno et al., 2021).
The combined adoption of different DTs can achieve various functions. Existing studies, however, have not clarified which functions would be most helpful to improving CE. Additionally, the understanding of the mechanism of using DTs to enable CE strategies remains unclear. We aim to fill the gap by investigating the digital functions and the underlying mechanisms that can support firms to implement CE strategies. We intend to address the following research questions: • RQ1: What are the main functions of DTs for the circular economy?
• RQ2: How can these digital functions be used to implement circular economy strategies?
The study adopts a systematic literature review method to answer the two questions and proposes a framework to address the integration of DT for CE and reveal the underlining mechanisms. We examined 174 papers and created 782 original codes, from which we identify 13 digital functions of DT that can be used to improve CE, under three categories (RQ1). We also propose a DF4CE framework to explain how these digital functions could be used to enhance different CE strategies (RQ2). The framework reveals the "maturity level" of the DT-CE mechanisms, based on the frequency with which they are covered in the relevant literature. In this way, the study advances the theoretical understanding of synergies between DTs and CE, building a comprehensive theoretical framework that covers overall CE strategies. The proposed framework provides a structured way for managers to explore the appropriate digital functions for their CE strategies, so as to identify required digital technologies and new value creation through digital functions. The results, in addition, identify future research needs and highlight the directions available for investigating specific digital functions for a particular CE strategy.
The remainder of this paper is structured as follows. The next section describes the methodology of our systematic literature review and the coding process. Section 3 summarises the results of the bibliometric analysis. The coding results are interpreted and clustered into digital functions to answer RQ1 in Section 4, while Section 5 explains the mechanisms and maturity level for RQ 2. Section 6 discusses the implications and possible future research agendas. Finally, Section 7 draws some conclusions arising from the research.

| METHODOLOGY
A systematic literature review approach (Denyer & Tranfield, 2009;Tranfield et al., 2003) is employed in this paper to analyse current research on using DTs for the CE paradigm. The approach locates the existing studies and selects and analyses them following a rigorous and well-defined research protocol. It generates an unbiased overview of current studies, with an audit trail for all the research steps (Denyer & Tranfield, 2009;Thomé et al., 2016). On this basis, the systematic literature review is more likely to increase research validity and reliability than a traditional literature review. As shown in Figure 1, the study begins with a descriptive analysis of the selected papers. Then we analyse the papers within the guidelines provided by qualitative coding method so as to answer our research question and address future research agendas. Each step was recorded in detail to ensure the process was replicable and transparent.

| Identifying and refining the review scope
In this phase, we define the review scope so that it aligns with the proposed research questions (Denyer & Tranfield, 2009). The need for focusing the RQs closely lies in the vast number of different topics associated with DTs, ranging from specific technologies, such as IoT and BDA (Cai et al., 2016), to generalised concepts such as digitalisation and I4.0 (Daú et al., 2019;Lukac, 2015). The need also arises from the wide application of CE in a variety of different areas.
We chose to mainly study three DTs: IoT, BDA, and AI. The choice of IoT and BDA was made because these are considered the most promising technologies for CE (Cwiklicki & Wojnarowska, 2020). IoT refers to the inter-networking of physical items that enable objects to collect and exchange data (Oztemel & Gursev, 2020), while BDA is the application of a collection of advanced techniques and technologies to the analysis of massive data sets, aimed at obtaining meaningful insights (Ghasemaghaei et al., 2015;Mikalef et al., 2018;Russom, 2011). BDA coupled with IoT can track and share product lifecycle data to reduce waste, enhance waste recovery and connect waste management practices (Esmaeilian et al., 2018). As for AI, this has been attracting growing attention in CE research recently. It can provide a fast and agile learning process for data analysis (Kaplan & Haenlein, 2019;Kristoffersen et al., 2020), which allows faster and more flexible actions based on larger data sets, hence creating new possibilities for CE (Ellen MacArthur Foundation, 2019). In order to uncover detailed insights on the digital functions of the DTs, our study focuses on the digital aspect of Industry 4.0, rather than the ones with a general and perhaps superficial coverage of Industry 4.0 concepts.

| Selecting studies
The literature examined in this paper was searched and selected by following the rules from Briner and Denyer (2012) and Denyer and Tranfield (2009). The search for studies was conducted in EBSCOhost, Scopus, ProQuest, ScienceDirect, and Web of Science. The search string was constituted by ("digit*" OR "Internet of Things" OR "IoT" OR "Big data" OR "artificial intelligence" OR "AI" OR "industry 4.0") AND ("circular economy" OR "circularity"). Both journal papers and published conference papers were included to mitigate the publication bias (Briner & Denyer, 2012). Only publications written in English were included. The material searches were conducted in September 2021, extracting 1626 papers after eliminating duplicates. We defined a series of inclusion and exclusion criteria to select the final papers listed below.
Inclusion criteria: • Focuses on both DTs and CE • Addresses one or more technologies within scopes (IoT, big data, and AI) • Addresses either the biological or technical sides of CE Exclusion criteria: • Technical papers focusing on modelling, optimisation, algorithms or developing a specific DT • Non-English papers. These were automatically excluded from the search in five of the databases described above.
F I G U R E 1 Systematic literature review process [Colour figure can be viewed at wileyonlinelibrary.com] Two researchers then screened the title and the abstracts of 1626 papers based on the criteria. After this, the remaining 303 papers' main text was closely examined. One hundred seventy-four papers were finally chosen for our review process.

| Literature analysis and synthesis
The study reviewed the papers in two steps: descriptive analysis and content analysis. The first one aimed to reveal research trends among the 174 papers. The journals publishing most papers, the papers published each year, and the most frequent keywords were analysed as suggested by Yang and Tate (2012). The results are presented in Section 4.
The second part of this literature review is constituted by indepth qualitative content analysis, in which we have coded the reviewed papers to extract relevant information for pattern detection (Miles et al., 2014). We have applied the two cycle coding method suggested by Saldaña (2015). We established a conceptual structure to cluster the refined codes to reveal the patterns and mechanisms that can answer research questions.

| Building the structure of conceptual framework
To answer the first research question, we need to identify the key digital functions that have been or can be used to improve CE in the literature. Here, the digital function is regarded as one which enables digital technologies to deliver smart services (Allmendinger & Lombreglia, 2005). To answer the second research question, the identified digital function should be associated with different CE strategies. Inspired by the IoT-CE cross-section heat map (Ingemarsdotter et al., 2019), we built a structure for analysis, with digital functions as the y-axis and CE strategies as the x-axis (see Figure 4).
A number of previous studies have proposed CE or DT related frameworks of one kind or another. Okorie et al. (2018), for instance, proposed a framework that utilizes the technology life cycle concept. It focuses on business analytics. Among the existing frameworks, only a few of them have directly addressed digital functions (Ingemarsdotter et al., 2019) or similar lenses. They constitute, for this reason, an inadequate foundation on which to analyse the digital function as it relates to CE.

| y-axis: Digital function category
Our y-axis includes three categories: data collection and integration, data analysis, and automation. This is based on the smart CE framework and the analytics and knowledge hierarchy (Kristoffersen et al., 2020;Siow et al., 2018) because data lie at the centre of the chosen technologies, as indicated in Figure 2. Data collection and integration is defined as the first process that provides data and information. Data serve, then, as the base of the knowledge hierarchy. It is mostly collected from physical and virtual sources. Information is constituted by the interpreted data within specific contexts (Siow et al., 2018). It is mostly generated by descriptive analytics in the data integration process.
The second category is data analysis, and the third is automation.
Data analysis builds on the first process to generate knowledge and wisdom. "Knowledge" here refers to diagnostic analytics with understanding and meaning, while "wisdom" refers to discovered, predictive, or prescriptive insights. We added "automation" as a third category of digital function to represent the physical processes. This category captures the self-organized robotics control and decisionmaking process that occur without human interference (Liebrecht et al., 2021).
F I G U R E 2 Digital function categories (adapted from Kristoffersen et al., 2020, andSiow et al., 2018) [Colour figure can be viewed at wileyonlinelibrary.com] 2.3.2 | x-axis: Circular economy strategy For the x-axis, the existing research of CE-DT do not currently provide a commonly agreed categories of CE strategy. Cagno et al. (2021) and Jabbour et al. (2018) (Blomsma et al., 2019). Lastly, Bag, Gupta, and Kumar (2021) found that a higher degree of I4.0 implementation can create higher 9R strategy-based manufacturing capabilities. This finding statistically validated the applicability of the 9R framework for DT-CE research in the manufacturing industry. The definition of the elements comprising the 9R framework is laid out in Figure 3.
We then established the digital function for the circular economy (DT4CE) framework ( Figure 4) with three digital function categories on the y-axis and nine CE strategies on the x-axis. Our review aimed to identify the relevant digital functions from codes and sort them according to the three categories. Then we referred to the evidence and cases from literature and mapped the codes to the x-axis and y-axis accordingly. The intersections of the x-axis and y-axis form various combinations of using a specific digital function to enhance a specific CE strategy. We use the size of dots to represent the frequency of related codes. The higher frequency indicates a higher maturity level of using a specific digital function for a specific CE strategy.

| Coding process
All selected papers were coded in the first cycle coding following a predefined coding protocol to minimize the bias in the literature analysis and synthesis phase. The first cycle coding resulted in 782 codes, including 441 with theoretical evidence and 341 with empirical evidence. Before the second cycle coding, the codes were modified and merged to improve data quality (Saldaña, 2015), from which 102 codes with conceptual evidence and 157 codes with empirical evidence emerged.
In the second cycle of coding, we clustered the refined codes to reveal patterns and meanings. As exemplified in Table 1,  Journal of Cleaner Production

Business Strategy and The Environment
- Others This provides an answer to our second research question, as will be explained in Section 5.

| DESCRIPTIVE ANALYSIS
As shown in Table 2, the research topic covered in this article is one which has been attracting increasing interest since 2017. The Regarding the number of documents published by author (Table 3) We examined the co-occurrence of author keywords to understand how the research topics associated with circular economy and digital technologies are related to each other. The commonality of author keywords was built on the basis of keywords appearing at least five times in a paper. Figure 3 shows the 13 keywords identified and the connections between them. The six most frequently used keywords were: circular economy, Industry 4.0, sustainable development/ sustainability, digitalization/digital technologies, IoT (or Internet of Things), and big data analytics. One hundred four papers, about 60% of the total, used "circular economy" as keywords, making it clear that this was central to the research of the authors. The second most frequent keyword was industry 4.0 with 76, followed by sustainable development/sustainability with 41 occurrences.
In Figure 5, three main clusters are identified. The first one focuses on the combination of circular economy with digital technologies like the internet of things, big data, and artificial intelligence. The second cluster is related to industry 4.0 and remanufacturing, sustainable manufacturing, recycling, and waste management. The third one covers discussion of circular business models with sustainable development and digitalization. In this cluster, the papers highlighted digitalized servitization, also known as the product-service system  DTs, especially, carry almost twice as many empirical codes as conceptual codes. The AI codes are mainly drawn from papers published after 2020, and most are related to generating wisdom through data analysis.
The most frequently discussed digital functions for CE were collection, monitoring, tracking and tracing, and optimisation. Auto-plan, auto-control, and assess clearly based largely on empirical evidence.
Auto-control has the most concentration of empirical codes, with 16 of these against three conceptual codes. Collect, monitor, and forecast, however, have a similar number of empirical and conceptual codes.
The detail of each digital function is discussed below in the corresponding categories.

| Data collection and integration
Data collection, together with data integration, forms the basis for data analysis. Data collection provides a massive amount of data as the basis of analytical functions (Siow et al., 2018). Data integration extracts information from raw data through aggregation, interpretation, selection, and sorting of various data (Dalamagas et al., 2020;Kristoffersen et al., 2020). Although data integration is vital for demonstrating information (Chen et al., 2020), it is rarely discussed as a single function of IoT, BDA, or AI in the reviewed literature. Collect and share are the functions in this category, as exemplified below.

| Collect
Data collection generates and gathers data from various heterogeneous sources (Kristoffersen et al., 2020). It is the fundamental support for other functions in the framework (Ranta et al., 2021;Siow et al., 2018). IoT provides the internet connection for infrastructure to transmit generated data into a central system (McEwen & Cassimally, 2013;Rossi et al., 2020). This system enables data collection with embedded sensors to measure real-time data and report information with minimum human interaction (Zacharaki et al., 2020).
Data sources range from companies' internal processes to external supply chain partners and customers (Ranta et al., 2021). The collected data makes possible the analytical and automatic functions, such as monitoring, optimization, and innovation.

| Share
Data sharing can increase information availability and support data analysis for optimization and real-time control. Shared data can be unprocessed raw data or integrated information. Raw data can be shared by granting access to databases or allowing external data collection, such as collecting data from customers via personal devices.
The information shared between companies is often inferred or transformed from data (Kristoffersen et al., 2020).
Additionally, sharing data can strengthen the existing interconnection and create new communication and collaboration. The use of DT can greatly improve the information exchange between different production processes, including production machines, automatic warehouses, and other devices facilitating value creation (Garcia-Muiña et al., 2018). Digital data platforms can combine data collection and data sharing functions for analysis or autonomous use. It can gather information from various sources either within or across business boundaries . Compared with monitoring, this function emphasises the changes relating to a specific item, enabling its traceability in the value chain.

| Data analysis
For example, tracking the locational change of spare parts makes them easier to retrieve for remanufacturing (Ranta et al., 2021). One typical technology is the RFID tags that store and carry the information of items. They travel with items throughout the value chain, so that people can retrieve items' information by scanning the tags (Garrido-Hidalgo et al., 2020).

| Detect
This is the function that identifies deviation in performance or abnormal characteristics in items or processes (Enyoghasi & Badurdeen, 2021;Pagoropoulos et al., 2017). Possible application of detecting involves identifying differences in object's appearances for end-of-life solutions. For example, computer algorithms can decide whether to repair, recycle, or discard items based on the results from examining the returned product for scratches and defects, or detecting different materials in wastes (Turner et al., 2020;Zhang et al., 2019). Another application of detection is to identify abnormal process behaviours through monitoring, making it possible to stabilize performances (Ingemarsdotter et al., 2021;Tiwari et al., 2021).

| Assess
Assessment can be targeted on physical objects such as materials, end-of-life recovery of products, machine efficiency, and conceptual entities such as environmental impact Zhang et al., 2020). Its results can support a variety of different objectives such as revealing hidden patterns, reducing cost, and facilitating predictive maintenance.

| Connect
DTs offer the opportunity of establishing multi-actor connection networks. Customer and reverse logistic service providers, and recycling centres, can be connected through smart devices and digital platforms. Connection through digital platforms can foster industrial symbiosis by connecting factories in different industries (Birat et al., 2021). Such connection can also accelerate the collection of used devices, so that they can be delivered to recycling centres (Lawrenz & Leiding, 2021). The connection also gives customers more access to the production, design, and recycling processes, laying the basis for customer-centred production and service (Huynh, 2021).

| Forecast
This is the function that enables future events to be predicted, based on past and present data. It is often applied to predict the demand trends for products, materials and critical service parts (Boone et al., 2017). Another popular application is the predictive maintenance that forecasts need for maintenance by calculating the product failure tendency (Jabbour et al., 2018;Kerin & Pham, 2020;Morella et al., 2020). Furthermore, AI-based digital twin technology can simulate different predictive maintenance options to find out the best solution (Zacharaki et al., 2020).
The predictive function can also applied in estimating the byproduct potential of production which can be used as materials in other industries, as also in determining whether or not a product can be recycled (Ghoreishi & Happonen, 2020a

| Optimize
Optimization can be achieved at all lifecycle stages, such as in the design of products and services, production procedures, logistic operations, customers' usage behaviour, and the reuse or recycling of products. The goal of optimization is often to improve performances and reduce negative impact, such as increasing efficiency and reliability in the production system while reducing emissions and energy con-

| Automation
The third category applies DTs to enable automation and support robotics (Goering et al., 2018). It refers to the independent process of operating, acting, or self-regulating without human intervention (Liebrecht et al., 2021;Nof, 2009). Automation can enable decentralized decision-making, self-configuration, and self-optimize so as to enhance flexibility, strengthen resilience, and reduce disturbance (Kamble et al., 2018;Lee et al., 2015

| Auto-control
Auto-control is related to autonomous systems and intelligent robotics. Both can complete operational tasks without human intervention.
On the one hand, autonomous systems can monitor processes and systems with minimum human interaction (Zacharaki et al., 2020). IoT

| Auto-plan
Auto-plan uses AI for decentralized decision-making without human interference on both the predictive and prescriptive levels. Such intelligent systems can generate personalized disassembly plans for each part based on cognition information and shared databases . Condition-based monitoring can automatically schedule maintenance interventions and generate information on necessary material requirements before an actual breakdown (Kristoffersen et al., 2020;Rossi et al., 2020). Other possibilities include the automatically generation of procurement plans, end-of-life strategy decisions, and feasible logistics routes (Bag, Wood, et al., 2020;Mboli et al., 2020;Rossi et al., 2020).

| THE FRAMEWORK OF DIGITAL FUNCTIONS FOR CIRCULAR ECONOMY
We categorised the codes in relation to the DT functions discussed in reviewed papers into corresponding circular strategies. The results are mapped onto the DF4CE framework ( Figure 6). The size of the circles represents the number of second-round codes. It shows that in literature, these digital functions are most frequently used for reduce, followed by rethink, and recycle. We further analysed the codes and their contexts in literature and proposed seven underlying mechanisms as follows.

| Mechanism 1-Recycling: Digital functions empower the reverse supply chain
We identified this mechanism as using digital functions to empower the reverse supply chain through building a faster, more precise, and automated recycling system. The recycling strategy benefits from most of the digital functions covered in this article. Sort and classify seems to be the most useful function for recycling, followed by innovate and detect. Sort and classify is almost exclusively related to recycling, while forecast has not yet been linked with this mechanism.
We found that the digital functions used to improve the efficiency of linear manufacturing processes can be largely applied in the reverse supply chain in a similar way. In the forward supply chain, DTs are often applied for improving production efficiency and optimising logistics planning to increase companies' profits. Similarly, companies can use IoT to collect data from the reverse supply chain, such as the locations and conditions of used products and wastes. These data can be analysed to improve the reverse supply chain efficiency, especially through accelerating the recycling process and increasing material recovery (Dev et al., 2020;Rajput & Singh, 2019;Vetrova & Ivanova, 2021).
This similarity can be observed when applying monitor and track and trace functions on the reverse supply chain. Companies have increasingly used DT to monitor the recycling processes for minimising wastes (Kerdlap et al., 2019), to track and trace items for planning route, scheduling waste collection and transport (Rajput & Singh, 2019), as well as reducing emission and energy consumption. In addition, we identified the digital functions applied in reverse supply chain that do not often exist in linear supply chain. In reverse supply chain, for instance, sensors are often used to measure wastes and waste bins to facilitate recycling (Fatimah et al., 2020). DT empowers the recycling system to detect damages on used products, such as rust and scratches, and therefore support disassembly for end-of-life products . The optimization algorithm evaluates the end-of-life decisions, such as whether to refurbish, remanufacture, or recycle (Mboli et al., 2020;Zacharaki et al., 2020).
Our study discovered that sort and classify might be designated for increasing the recycling efficiency. In this occasion, technologies are used for separating wastes, deciding whether an item should be reused, repaired, refurbished, remanufactured, recycled, or disposed.
Such sorting and classifying process involves not only using digital calculation to decide on the end-of-life options, but also applying automatic robotics to physically separate wastes. These decisions can derive from other digital functions, for instance, detecting damage levels or material compositions (Ellen MacArthur Foundation, 2019;Laskurain-Iturbe et al., 2021;Sarc, 2021;Wilts et al., 2021).

| Digital functions for extending lifespan of product and its parts
Extending lifespan of product and its parts is mainly achieved through repurpose, remanufacture, refurbish, reuse and repair. Digital functions in the data analysis category are the most frequently used, while the one in automation category are rarely mentioned.
Analysing data for matching the products' supplies and demands is especially important for extending lifespans. IoT can support this purpose by collecting and sharing data between users, technicians, service providers, and potential second-hand buyers. DT-based tools, procedures, and platforms can accelerate the exchange of information between offers and demands. BDA then supports in generating personalised after-sale services . Among these CE strategies, repairing and remanufacturing are the most frequent ones in terms of adopting DT for lifespan extension. Altogether, we identified four mechanisms for using digital functions to extend product lifespan as follows.

| Mechanism 2-Repurpose: Digital functions foster industrial symbioses
The repurpose mechanism focuses on using DTs to foster industrial symbiosis, in which wastes or by-products generated in one industry can be converted into production resources for other industries.
Since companies often lack the knowledge for such a cross-sectoral exchange of waste, material and service, exchanging information among multiple industries can help them discover new opportunities.
One proposed digital solution for industrial symbiosis was the information exchange platform for regional industries (Dalamagas et al., 2020). It can collect data on the location, type, and quantity of input materials and waste (Song et al., 2017) and share them between traditionally separated industries in the same region (Acerbi et al., 2020;Kristoffersen et al., 2020;Zeiss et al., 2020). The data shared on information platforms can support real-time waste-toresource matching, so as to reduce the uncertainty of by-product availability and ensure the appropriate quality for further exploitation (Birat et al., 2021;Dalamagas et al., 2020;Kristoffersen et al., 2020).
These actions result in innovating industrial symbiosis links among companies in the same region (Zeiss et al., 2020).

| Mechanism 3-Remanufacture: Digital functions support remanufacturing activities
The remanufacturing mechanism applies DTs to retrieve parts that is still functioning from the non-functional products, so as to repair and rebuild the parts into new products that are similar to the original ones. Ten out of 13 digital functions discussed in this article were related to this mechanism, which covered all the functions under data analysis category and data collection and integration category. However, none of the functions under the automation category were yet to be linked with remanufacturing.
Digital functions can support the whole remanufacturing process, involving identifying useful parts, disassembling and repairing them, reselling and rebuilding them in a new product. For instance, DTs can detect a product's wear by means of embedded sensors in the products or through data shared by customers (Kerin & Pham, 2019). Track and trace provide insights into the availability and condition of used products and spare parts (Ingemarsdotter et al., 2020;Subramoniam et al., 2021). The information can enable companies to make personalized remanufacturing processes that minimizes wastes and material consumption , to optimize process efficiency as referred elsewhere (Kerin & Pham, 2020;Zacharaki et al., 2020), or to secure the spare parts sourcing, which can often be a significant challenge for industry (Dev et al., 2020;Garrido-Hidalgo et al., 2020;Rosa et al., 2020).

| Mechanism 4-Repair: Digital functions enable predictive and prescriptive maintenance
The repair mechanism uses DTs to extend machines and products lifespan through predictive and prescriptive maintenance, which means carrying out customized maintenance tasks on devices before the actual breakdown happens. It was frequently discussed in the reviewed papers, in which monitoring and forecasting were often mentioned. However, sharing and connecting does not receive enough attention, nor does the functions in the automation category.
Digital systems can support failure detection, condition-based maintenance, and automatic task scheduling for products and machines (Akkad & Bányai, 2021;Bressanelli et al., 2018a;Kristoffersen et al., 2020;Nobre & Tavares, 2020b). If the maintenance is accurately predicted with a prescription of required actions, the maintenance worker can improve the repair efficiency, in addition to reducing unnecessary services and the on-site visits (Ghoreishi & Happonen, 2020a;Ingemarsdotter et al., 2020). Another suggestion was to connect customers with companies for product maintenance.
For companies, learning each product's wear and tear from customers can help them arrange more personalised remanufacturing and parts' replacement , while customers can start maintaining their tools with companies' technical support (Ranta et al., 2021).

| Mechanism 5-Reuse: Digital functions support reselling and sharing used products
This mechanism uses DTs to support reusing products through reselling and sharing used products, such as second-hand market.
Used products which are in good condition can extend their lifespan with second-hand users. Digital functions mainly support relocating used products through finding the reusable products, then selling them to the second-hand customers. Track and trace, monitor, and collect were considered useful for the finding and provide information about products that are ready for reuse, such as the products in-use data, the real-time condition, and their location (Vetrova & Ivanova, 2021). Innovate, unlike the former, mainly supports the trading process. Developing new DT-based tools, platforms, systems, and services can efficiently connect potential second-hand buyers to the products because these technologies accelerate the information exchange process (Cagno et al., 2021;Rocca et al., 2020).

| Digital functions for smart product use and manufacturing
5.3.1 | Mechanism 6-Reduce: Digital functions improve energy and resource efficiency Our study suggests that the reduce mechanism is often related to improving energy and resource efficiency at all product lifecycle stages.
It is the most frequently discussed topic in our reviewed literature, in which sense it is clearly seen as the most feasible way to adopt DT for CE. Improving production efficiency and energy efficiency are proven benefits of Industry 4.0 (Mohamed, 2018;Oztemel & Gursev, 2020) and a sustainable value driver. One direct impact of improved process efficiency is less waste in materials and energy. Monitoring, optimization, and auto-controlling are the most applied digital functions. Forecast, track and trace, and collect also contribute strongly to this mechanism.
Logistics optimization is one of the central topics (Cagno et al., 2021) covered in the reduce strategy. It leads to reduced fossil fuel consumption for transportation. Collecting and sharing data like trucks' location and movement, loads, shipment, and product condition supports optimizing routes and loading. In addition, technologies help optimizing the real-time waste collection route leads to an increasing rate of recycling (Akkad & Bányai, 2021;Garrido-Hidalgo et al., 2020;Rossi et al., 2020).
Increasing production efficiency is another well-discussed topic.
In the short term, real-time monitoring can support quick and interactive equipment management with greater energy efficiency (Reis et al., 2021). Self-control robotics and systems are deployed to increase efficiency and reduce human conduct failings (Laskurain-Iturbe et al., 2021). In the middle to long term, tracking and tracing makes analysis and planning more precise, reducing waste from faulty conduct (Wegner-Kozlova & Guman, 2020). Optimization, forecast, and innovative functions further improve the efficiency.
End-of-life efficiency is also discussed. Apart from the valuable insights generated from data analysis, as it does in other stages, disassembly efficiency benefits from AI and control technologies. For example, detecting defects helps AI to decide disassembly levels, choosing end-of-life solutions, and sorting wastes (Sassanelli et al., 2021). Furthermore, AI can connect waste generators and collectors so as to increase waste recovery efficiency (Kurniawan et al., 2021). Automatic sorting and classifying wastes not only improves recycling efficiency but also reduces human labour.
5.3.2 | Mechanism 7-Rethink and refuse: Digital functions support product design, manufacturing, and use This mechanism uses DTs to focus on fundamental changes in product and service design, production processes, and user behaviour, so as to abandon the wasteful behaviour and replaces non-renewable materials with recycled or renewable ones. It benefits more from digital functions in the data analysis, collection and integration category rather than the automation category. Collet, monitor, innovate, and optimize seem to contribute the most for rethinking and refusing the unsustainable practices, while detect and track and track were not directly relevant to this mechanism in the reviewed papers.
In the design phases, DT provides information and accelerates the development of prototypes. On one hand, building a product biography can support CE design, which is "a composite of trajectories rather than following a linear path from design to manufacture and disposal" (Spring & Araujo, 2017;Spring & Araujo, 2017, p. 27). Data collected and shared at all product lifecycle stages can elaborate a systematic and comprehensive product biography, which can be used to support the circular design of product (Kerin & Pham, 2020) or product-service systems (Ingemarsdotter et al., 2020;Kerin & Pham, 2020;Yang & Evans, 2019;Yang, Smart, et al., 2018). On the other hand, AI can test different design models' stability and quality, accelerating the design process with the use of fewer prototypes (Getor et al., 2020;Ghoreishi & Happonen, 2020b). It can be used to find suitable renewable raw materials or recycled materials, replacing the non-renewable ones (Morella et al., 2020;Ranta et al., 2021).
DT can also be used to reshape the production process. The strategic analysis can integrate environmental impact factors into identifying novel value creation opportunities, such as production with renewable resources (Ranta et al., 2021). Embedded lifecycle assessment indicators can also reduce the environmental impact of the production process (Birat et al., 2021).
As for changing the usage behaviour, DTs provide the connectivity that brings customers closer to companies. For customers, it can encourage them to change their non-circular habitual usages and adopt sustainable behaviour. For instance, they can have more information about recyclable products and their environmental impact (Huynh, 2021). At the disposal stage, customers can receive advice on where to discard the waste for end-of-life solutions (Kurniawan et al., 2021). As for companies, they can increase recycling rates by developing innovative services and offerings to customers (Kristoffersen et al., 2020). They can also help customers reduce careless behaviour in product use by monitoring of product conditions and customer activities (Kintscher et al., 2021;Rossi et al., 2020).

| Maturity level of the digital functions mechanisms for circular economy
The maturity level of each specific digital function for each specific CE strategy can be interpreted from the size of their circles in the DF4CE framework ( Figure 6). The size represents the number of second-round codes, hence the amount of related discussions in prior studies. More discussion provides a more comprehensive understanding of how to realize the digital functions for CE strategies. In other words, a larger circle means a higher maturity level of relevant discussion.
Among seven mechanisms, reduce (M6)-digital functions improve energy and resource efficiency seems to have a significantly higher level of maturity than the rest. Overall, this study found 36% of the second-round codes relating to this mechanism, among which monitoring, optimizing, and auto-control were the most related digital functions. Recycling (M1) and rethink and refuse (M7) have the second level of maturity, with 15% and 16% second-round code respectively related to them. The least discussed mechanism is the repurpose (M2)-digital functions foster industrial symbioses, which appears to be an emerging topic in recent years. The recycling mechanism (M1) and repairing mechanism (M4) appears to rely heavily on specific digital functions, as 28% of the codes that are grouped under recycling represent the digital function of sort and classify. Similarly, 35% of the codes under repairing belongs to the forecast function.
There are still missing links in prior researches. From the CE aspect, our study found no digital functions that are linked to the recovering strategy (e.g., recovering energy by incineration of material).
Only two codes suggest the application for refurbishing old products. Additionally, the research shows that the customer has become a vital actor in CE transformation because DTs bring customer data into business decision-making, such as the design phase (Vetrova & Ivanova, 2021). It also provides customers with feedback on environmental impacts to raise customers' sustainable awareness (Akkad & Bányai, 2021 Some of the limitations in existing studies, as far as this factor is concerned, are worth making explicit. First, the DT terms need to be made consistent across the field. For example, some reviewed literature sees IoT as a means of supporting the collection, storing and processing of data, and even enabling fully autonomous systems to be created (Basso et al., 2021;Rossi et al., 2020). Others see the role of IoT as a connector between different technologies (Rocca et al., 2020;Turner et al., 2020). Second, the digital function terms also have different characteristics.
For example, "detect" can refer to finding undiscovered symbiotic links (Dalamagas et al., 2020), or to identifying differences in objects such as scratches and materials. Third, some digital function terms can have overlapping meanings. Our review, having synthesized 782 original codes, defined "monitoring" as a function that covers real-time updates on processes in a fixed position, while "tracking & tracing" covers a specific moving item's past and current changes. However, some literature uses these terms differently, such as tracking the environment , monitoring used product parts (Ada et al., 2021), and monitoring and tracking products (Vetrova & Ivanova, 2021; Wegner-Kozlova & Guman, 2020).

| Innovating technology applications for the circular economy
Our study has revealed the need of innovative DT applications for closing the loop, especially for extending products and parts lifespan.
Our findings suggest that current DT implementations for CE are often a readaptation of existing practices that aim to increase productivity and efficiency (Frank et al., 2019), which have limited contribution in closing the material loop. Among the end-of-life solutions, recycling attracts more attention than the others. However, the emphasis remains on improving efficiency in the reverse supply chain (Furstenau et al., 2020). Because CE requires a radical change of our current production and production patterns (Merli et al., 2018;Reim et al., 2021), we argue that the existing DT applications are insufficient to support the CE paradigm. Further research should create new ways to fill in the missing links in the DF4CE framework discussed in this study.

| Advancing management research
Our research reveals that the DT-CE have positive links to industrial symbiosis and value creation, which raise the need to examine their relationship with other management practices, such as business model innovation (Evans et al., 2017;Tunn et al., 2021) and business ecosystem (Kanda et al., 2021). For example, since DTs enhance communication and establish a collaboration network, future research could explore the structure and dynamics of change in circular business models and business ecosystem caused by DT implementation (Hofmann et al., 2022). Future research can also investigate the rebound effects of the smart circular economy, such as its social impact or the increased energy consumption of DT implementation (Lange et al., 2020). Hence, the definitions require further validation in academic and practical studies.