Digital Twin: An Added Value for Digital CONWIP in the Context of Industry 4.0

Abstract

Despite technological progress and a large amount of research on Industry 4.0, digital transformation remains a complex process that most manufacturers are hesitant to invest in. Interest in digital Kanban, for example, remains low compared with traditional Kanban, which is widely used. This applies to the other card-based production control systems, including CONstant Work-In-Process (CONWIP), which is the focus of this paper. In an industrial context where digitization and Industry 4.0 are the main trends, one may wonder why traditional CONWIP is preferred to digital CONWIP. Following a praxeological approach (i.e., study of practice and instrumentation), this article explores the strengths and weaknesses of the CONWIP practice, in both its paper and electronic versions, while taking into account the human dimension. The aim is to motivate potential CONWIP users to implement it in its digital mode and to show them how a Digital Twin-based solution can overcome the managerial problems that arise with digitization while enabling improved performance. As an illustration, experience feedback from several companies using Digital Twin with CONWIP is provided.

1. Introduction

In the fourth industrial revolution (Industry 4.0) era, many companies have decided to digitalize their manufacturing processes to improve their performance and remain competitive in an industrial world where all manufacturers seek to take advantage of new opportunities. The ability to store large amounts of data and use sensors and wireless real-time connected objects technologies offer promising digitalized solutions in various fields of industrial management, including production planning and control. Ref. [1] argued that production planning and control systems can be revolutionized by Industry 4.0. Specifically Card-based Production Control Systems (CPCSs), using digital control boards, are a crucial factor for practical, flexible, and adaptive computerized manufacturing systems. This paper focuses on one of the easiest CPCSs to apply, called CONWIP. It consists of controlling production lines by maintaining a constant number of Work-In-Process (WIP) through circulating a constant number of cards. However, Kanban being the best known and most studied CPCS, this article regularly refers to it to explain how CONWIP is different from Kanban or to describe findings confirmed for Kanban and that also remain valid for CONWIP. That said, Ref. [2] revealed different opportunities for Kanban in Industry 4.0 (improved demand assessment, dynamic, and more efficient milkruns, and shortened cycle times). Indeed, digital Kanban increases the manufacturing system’s efficiency by overcoming traditional Kanban’s limitations, mainly caused by the shortcomings of exploiting physical boards and handling physical cards. However, despite the positive effects of digitalization, companies’ interest in digital Kanban (and in Digital CONWIP) remains low as they continue to choose the traditional solution of using paper cards and boards [3].

Aside from investment considerations, this may be justified by these companies’ concerns about possible misappropriation issues that can result from heavy digitalization. These risks can also be witnessed in Digital CONWIP (DCONWIP), which like digital Kanban, involves replacing the physical cards with an electronic display system and the paper board with a digital screen for better visual management of the production line. Unlike traditional CONWIP, DCONWIP is very little explored, and almost no research exists in the literature on its application in a real industrial context. This is further evidence that companies seem to prefer the paper version of CONWIP over the electronic version. With that being said, two questions arise:

-

Why do manufacturers prefer not to engage in CONWIP digitalization despite the widespread Industry 4.0 trend?

-

How can CONWIP digitalization be improved so that manufacturers will more likely accept it?

In addressing these questions, our goal is not to prove the cost-effectiveness of DCONWIP or to improve its operational efficiency (e.g., by reducing production time), but rather: (i) to alert manufacturers to some of the strategic managerial problems that may come with the CONWIP digitalization and more importantly, (ii) to propose a potential solution to these problems based on Digital Twin (DT), an Industry 4.0 tool that virtualizes an object (or a physical process or system) and accurately reflects its operating state and behavior [4].

The research methodology we adopt is that of praxeology, which focuses on the theory of practice and the study of human actions and behavior [5]. In the context of management science, the praxeological approach involves studying practices within an organization and understanding how humans interact with them to create a structured synthesis of knowledge [6] to help organizations become more effective in practice [7]. For our research, this amounts to studying the praxis of the CONWIP system in a stepwise approach that reflects the evolution of CONWIP use (cf. Figure 1) through:

-

An analysis of the practice of CONWIP in its traditional version (highlighting the identified benefits and limitations);

-

An analysis of the practice of CONWIP in its digital version (highlighting the identified benefits and limitations);

-

An analysis of how the practice of CONWIP can be improved through the use of a DT-based solution (highlighting the potential benefits).

This article is organized as follows: In Section 2, we conduct a literature review, starting with traditional CONWIP; then digitalized CONWIP, which addresses the problems encountered in conventional CONWIP; and finally DT, which aims to solve the new issues brought by DCONWIP. In Section 3, we introduce the concept of Combined DT/DCONWIP, explain how DT can address each of the DCONWIP limitations, and present experience feedback from using a Combined DT/DCONWIP in several companies. In Section 4, we conclude by recalling the main findings of this article and its scientific contribution before discussing some research perspectives regarding the application of DT to CONWIP.

2. Literature Review

The CONWIP literature is explored in this section following a stepwise approach (i.e., a new solution is sought for each problem arising from an old solution). That said, we first review CONWIP in its traditional version and highlight its limitations (Section 2.1). Then, we discuss CONWIP’s digitalization (Section 2.2), a solution that addresses these limitations. Finally, we examine the DT concept (Section 2.3), a solution we propose to address some of the problems encountered in DCONWIP.

2.1. Traditional CONWIP

CONWIP was inadvertently invented by Mark Spearman in 1985 while investigating the correlation between work release rate and cycle time in a factory [8]. Introduced in 1990 by [9], CONWIP aims to control the production lines by maintaining constant WIP. Its basic principle involves circulating a constant number of physical cards (i.e., CONWIP cards) [9] (cf. Figure 2).

Concretely, each Production Order (PO) sent to the production line must be associated with a CONWIP card (authorization to produce). Once it has passed through the entire production line, the PO leaves the line and releases the accompanying CONWIP card. This card is then placed on a paper board (i.e., CONWIP board) that, in addition to indicating card availability, provides an instantaneous global view of the workflow [10]. If all cards are allocated, the new POs must wait at the beginning of the production line. This prevents the entire production line from being saturated with excessive WIP [3]. Moreover, Ref. [11] stated that CONWIP is a Pull system that shares the advantages of Push systems (i.e., systems that control starts rather than WIP [8]) in controlling inventory levels while, at the same time, better accommodating the uncertainties of dynamic environments in which Kanban does not successfully operate. Indeed, while Kanban requires two types of cards for each part at each workstation, which makes it impractical for systems with hundreds of parts, CONWIP cards are generic. They are associated with an item when they enter the production system at the beginning of the line and become generic again at the end of the line when they are released [12]. This avoids using multiple cards, making CONWIP more suitable for production systems that manufacture many different products while being more flexible and, most importantly, easier to implement. CONWIP also provides better performance for certain aspects. Ref. [13] stated that it provides high throughput, short cycle time, and low WIP. Moreover, when associated with Material Requirements Planning (MRP), or any other planning tool, CONWIP allows piloting the production lines in hybrid push/pull mode by selecting from MRP production orders, the product references, and the quantities to be produced. According to [14], this would: (i) avoid the major risk of push flow, which is to clog the workshops with too many production orders; (ii) keep the WIP and the line throughput time constant; (iii) and ensure that the actual lead time of the workshop is synchronized with the lead times set in the Enterprise Resource Planning (ERP). In addition to ERP, CONWIP engages several other production management concepts, including that of Visual Management (VM), a management system that “attempts to improve organizational performance through connecting and aligning organizational vision, core values, goals, and culture with other management systems, work processes, workplace elements, and stakeholders” [15]. Indeed, through the CONWIP board, VM allows not only to visualize the queue of PO waiting for an available CONWIP card, but also the position of POs already in the workshop. This makes it possible to monitor the WIP on the production line and check each workstation’s workload for timely intervention.

While all these advantages for improved production control have been observed in real industrial contexts, they have also given rise to new problems, mainly induced by the inefficiencies resulting from the physical manipulation of paper cards. As [16] stated, manual handling of Kanban cards (but also of CONWIP cards) can lead to their loss or misplacement, which would cause immediate problems in just-in-time production. Indeed, handling dozens of physical cards (that need to be designed, printed, and laminated) is not easy for the operators. It takes much effort to educate and train them on how to use traditional CONWIP and a lot of discipline from them to keep using these paper cards. Moreover, Kanban being vulnerable to sudden fluctuations in demand, a manual adjustment of the used card number, and an update of Kanban card information proves difficult [17]. This also applies to traditional CONWIP, for which it is complicated to determine the exact number of cards needed in advance. Another major problem with card handling comes from irregularities that may occur with the movements of the cards. In Kanban, for example, these irregularities generally arise when cards are not moved at the exact moment the materials are consumed [18], whereas in CONWIP, they can occur when a card does not join the available cards board the moment its PO leaves the production system. In practice, all these problems negatively affect the production system’s efficiency. To address them, many researchers consider digitalization as the best solution.

2.2. Digitalization of CONWIP

Despite the impressive amount of research conducted on traditional CONWIP on the one hand, and digitalization, smart factory/smart supply chain, and Industry 4.0 on the other, DCONWIP remains relatively unexplored. Digitizing CONWIP is an option that Spearman et al. mentioned in their first paper on CONWIP in 1990, indicating that physical cards can take the form of electronic signals without really going into details. In DCONWIP, the physical cards are replaced by a digital signage system using advanced Information Technologies (ITs) (e.g., barcodes, scanning devices, Radio Frequency Identification Technology (RFID), and electronic messages) and the CONWIP’s paper board is replaced by a digital screen for better visual management of the production line.

Research conducted in the scientific databases of EBSCO, ScienceDirect, and Google Scholar using several combinations of keywords (Digital CONWIP OR “Electronic CONWIP” OR “E-CONWIP”/CONWIP AND “Digital transformation” OR “Digitization” OR “Digitalization” OR “Digitisation” OR “Digitalisation”/“CONWIP” AND “Industry 4.0”/CONWIP AND “IT” …) confirm to us that, apart from [3], no papers focus on DCONWIP and its application in industry. The majority of papers found in this research (e.g., [9,19,20]) only mention that CONWIP can also exist in a digital form. Ref. [21] added that for manufacturing systems with existing IT systems and infrastructure for monitoring WIP, sophisticated tools such as electronic cards and boards can be used while being color-coded for teams so that when a card is sent upstream, the team that last completed an item can be identified. Some papers considered DCONWIP in an incidental manner, focusing more on its enabling technologies, specifically RFID technology, which according to [22,23] improves the visibility and traceability of products throughout the supply chain and enhances the efficiency of some operational processes (e.g., tracking and shipping). To improve the performance of multi-echelon inventory systems, the authors compare CONWIP, which they call RFID-enabled CONWIP, with other RFID-based strategies such as Push and (s, S) inventory policy. Other papers gathered “Industry 4.0” and “CONWIP” in their keywords, but upon further analysis, it is not the DCONWIP that is emphasized. Instead, these papers address how to improve the performance of a CONWIP system in the context of Industry 4.0, a dynamic and variable context given the increasing customization of products that comes with it. For example, to deal with the increased variability that occurs in current production systems, Ref. [24] proposed a tool for estimating the performance of a production line using a Deep Learning Neural Network. This tool is used to estimate the throughput of a CONWIP production line with fixed levels of variability and WIP in the system. Following the same aim of ensuring a high level of product customization while reducing response time, Refs. [25,26] introduced a decentralized production control system for scheduling items in a CONWIP production line, considering a predefined WIP level.

To the best of our knowledge, Ref. [3] are the only ones who have shown interest in DCONWIP and its implementation in the industrial world. The workshops where DCONWIP was implemented were already familiar with visual communication for production planning and control. The DCONWIP boards were created by simply transferring the paper version to a digital screen in the form of digital tickets [3]. This requires using Industry 4.0 technologies (RFID systems, Real-Time Location Systems (RTLSs), Sensing systems, etc.) to link the sensors’ data to virtual models that help monitor the physical processes. The CONWIP digital screen displays two panes, a left pane that shows the PO queue and a right pane that indicates the PO progress state regarding the different workstations [3]. These workstations are displayed through successive columns following the envelope-routing principle defined by [27] as a sequence of manufacturing operations in which each reference that passes on the line finds its way by always going forward and possibly skipping some operations. Digital transformation of CONWIP consists, therefore, in representing the PO in digital form at the different stages of production while maintaining a constant WIP by imposing a limited number of virtual CONWIP tickets.

Digitizing CONWIP overcomes almost all the adverse effects of manually handling CONWIP paper cards. Indeed, with DCONWIP, the design, printing, and lamination of physical cards would no longer be necessary, and the problem of lost or misplaced physical cards would be eliminated. Moreover, with a digital CONWIP board, cards no longer must be moved manually, as their virtual movement to the production order waiting room is automatically triggered [3]. This eliminates the discrepancy between the card’s position on the board and the PO’s position on the shop floor. In addition, having the CONWIP board in a digital form allows it to be replicated on different screens distributed throughout the shop floor. This reduces the movement of workers between workstations. Moreover, Ref. [3] stated that people whose workstations were physically distant from the traditional CONWIP board would interact more with the DCONWIP system. Overall, CONWIP’s digitalization enables greater transparency, visibility, and traceability of movements in the system.

Nevertheless, despite all these positive aspects of DCONWIP, its use remains limited compared with the traditional solution. The research was conducted to explore the limitations of DCONWIP in the context of Industry 4.0. This was performed by considering the electronic databases included in Google Scholar, using as keywords different combinations formed from “CONWIP” and the synonyms of the words: “limits” and “digitalization”. This research has resulted in thousands of papers, but keeping only those that examine the adverse effects of CONWIP digitization, very few papers have been identified (see Table 1). Most of these papers mentioned one of the main limitations, namely the high-cost factor [28]. Employing affordable IT is one of the most challenging issues in achieving the intended impact of Industry 4.0. The authors of [29] argued that digitalization involves increased costs and that hidden costs may occur even after the completion of the digital transformation. The investment is not just in Industry 4.0 technologies. Many researchers consider that digitizing production lines and thus controlling these digitized lines requires hiring competent workers with highly developed skills. According to [30], workers’ skills and qualifications are one of the most challenging aspects for organizations that want to drive their supply chain digitally. These workers must be capable of managing the specific technologies of Industry 4.0 that involve new tasks of high complexity [30]. The authors of [31] considered the lack of specialized skills and training as one of the challenges in adopting digital Kanban (which is also true for DCONWIP) and specified that digital Kanban is only suitable for teams whose members have overlapping skills, so everyone contributes. This is reminiscent of [3], who explained that some workers might interact less with the CONWIP’s digital control board, as with digital Kanban. This risk of poor appropriation of IT tools by workers is indeed one of the main factors in the failure of digital transformation [32]. Reasons that can negatively affect the appropriation process of new technologies include:

-

The anxiety of workers at the thought of being replaced by machines. The worry of machines taking over and replacing humans is prevalent in workers’ minds. Ref. [33] conducted a detailed study of how digital transformation affects human well-being and system performance before, during, and after transformation. They conducted interviews with 35 workers from ten different companies that decided to embark on digitizing their production lines. These interviews testify that workers who question their skills and competencies fear losing their jobs and are anxious about working with new digital technologies. Ref. [34] considered that current manufacturing jobs are at a high risk of being automated, which can significantly decrease the number of workers.

-

The reluctance of workers towards change. Often unfamiliar with digital technology, these workers are apprehensive about working in new ways that disrupt their established habits and take them entirely out of their comfort zone. Changes in work methods are a great source of stress for workers [33]. These authors reported that several workers in the studied companies resigned right after the management team announced their plans to invest in digitalization. Managing to motivate team members is one of the main challenges that [35] identified regarding adopting digital Kanban.

-

The uncertainty of decisionmakers about the success and efficiency of digital transformation. Ref. [33] pointed out that decisionmakers believe that partially developed solutions decrease the overall system performance and that new digital technologies cause uncertainty. These decisionmakers cannot rely on old answers within the new manufacturing environment [36]. In other words, the uncertainties about the benefits of digitalization result from a lack of proven business cases justifying the investments. This leaves decisionmakers lacking the confidence and courage to push for a radical transformation [37].

Another limitation related to the digitalization of Kanban (which again also applies to DCONWIP) was identified [38] by looking at several types of virtualized Kanban boards (web-based Kanban board, smartphone-controlled Kanban board, and computer-aided Kanban board). They state that the process of virtualizing the physical system is not enough and that a real-time visualization system is needed to ensure that relevant information is displayed in real-time. This overcomes another challenge often associated with the move to the digital world, namely the risk of data falsification and poor validity and quality of the captured and shared data [39,40]. Indeed, combining CONWIP with a real-time data collection system would avoid decision-making based on obsolete data. This is consistent with the findings of [3], who indicated that, by simply virtualizing CONWIP, it is challenging to estimate the delivery date of PO and to perform shop floor scheduling (i.e., a tool for allocating tasks and verifying the timely completion of the given work [3]). In contrast, this scheduling is very important since CONWIP applies to several successive operations, mobilizing several consecutive resources [41]. Indeed, having a workshop schedule would make it possible to anticipate the arrival of a PO on a workstation in the future and thus prepare tooling, organize the workstation, and identify a qualified person for the reference to be manufactured [3].

Table 1. CONWIP digitalization challenges.

CONWIP Digitalization Issues	Corresponding Challenges
Investment Issues	Need to invest in new technologies [28,29]
	Need to hire competent workers with highly developed skills [30,31]
Poor appropriation Issues	Workers’ resistance to new technologies [33,34]
	Workers’ concerns about being replaced by machines [33,35]
	Decisionmaker uncertainty about the effectiveness of digital systems [33,36,37]
Lack of Real-Time Connectivity Issues	Risks of data falsification, poor data validity and quality, and of decisions made on the basis of obsolete data [38,40]
	High difficulty in predicting delivery dates for production orders [3]
	High complexity of workshop planning [3]

Table 1. CONWIP digitalization challenges.

CONWIP Digitalization Issues	Corresponding Challenges
Investment Issues	Need to invest in new technologies [28,29]
	Need to hire competent workers with highly developed skills [30,31]
Poor appropriation Issues	Workers’ resistance to new technologies [33,34]
	Workers’ concerns about being replaced by machines [33,35]
	Decisionmaker uncertainty about the effectiveness of digital systems [33,36,37]
Lack of Real-Time Connectivity Issues	Risks of data falsification, poor data validity and quality, and of decisions made on the basis of obsolete data [38,40]
	High difficulty in predicting delivery dates for production orders [3]
	High complexity of workshop planning [3]

After highlighting the main limits of CONWIP digitalization, the following section presents the proposed solution to overcome these limits and thus increase their efficiency, i.e., the DT.

2.3. Digital Twin

2.3.1. Definition

For manufacturers who prefer traditional CONWIP over DCONWIP to commit to digital transformation, it makes sense that they must first be convinced that digitizing their CONWIP system guarantees that it will perform at least as well as its traditional version. Following this logic, the virtual model of CONWIP should be represented exactly as its physical model and behave similarly. Therefore, the virtual representation of the CONWIP system must be a kind of shadow of the physical system. This concept of a Digital Shadow is described in [42] as a digital representation of an object that has a one-way flow between the physical object and the digital object (i.e., a change in the state of the physical object results in a change in the digital object and not vice versa [42]). However, we also aim to address the limitations of CONWIP digitization regarding the risks of data falsification, poor data validity and quality, and decisions made based on obsolete data. As proposed by [38], the ideal would be to have a real-time digital system interconnected with its physical counterpart and in full and automatic bidirectional exchange. This is known in the literature as the “Digital Twin”.

The DT concept was first introduced in 2003 by Dr. Grieves in a product lifecycle management course at the University of Michigan [43]. He presented it as the virtual and digital equivalent of a physical product. Grieves later attributed it to John Vickers, a technologist at the National Aeronautical Space Administration (NASA), with whom he regularly worked [43]. NASA coined the first known definition of DT as “an integrated multi-physics, multi-scale, probabilistic simulation of a vehicle or system that uses the best available physical models, sensor updates, fleet history, etc., to reflect the life of its flying twin” [44]. DT was, therefore, initially limited to the aeronautical sector and, more specifically, to air vehicles. Soon after, DT began to expand its scope to include all products more generically, leading to its potential use in other areas, including manufacturing [45]. DT has the potential to present the virtual counterpart of not only products but all production resources. According to [46], the physical counterpart is not confined to products but also includes supply chains, systems of systems, etc. This has prompted many researchers to explore the potential impact of DT within the advanced manufacturing sector. Industry 4.0 is one of the most promising contexts where a successful application of DT can provide tangible gains in monitoring and optimizing maintenance and controlling operations [47]. There are many definitions regarding DT in the context of Industry 4.0. Some authors (e.g., [42,48,49]) have even decided to focus on analyzing and clarifying the DT definitions provided in the scientific literature. A DT can be defined as a complete and dynamic virtual representation of a physical system or process [50]. Ref. [48] reviewed all available definitions of DT and discussed why and how a concept, initially born in the aerospace sector, can be helpful in manufacturing given the advanced technologies of Industry 4.0. As a result, they ended up defining DT as “a digital representation based on semantic data models that allow running simulations in different disciplines, that support not only a prognostic assessment at the design stage (static perspective) but also a continuous update of the virtual representation of the object by a real-time synchronization with sensed data.”. In more detail, Ref. [51] defined DT as a complete digital representation of a physical asset that relies on three main components: (i) design by including static information and models reflecting the sizing and design of parts, BOMs, configuration, and layout data as well as other parameters; (ii) state by reflecting the real-time condition of the physical asset from sensor data, information from computer systems, runtime systems, or other sources of usage information; (iii) and behavior by using integrated models that describe the behavior of their physical counterparts such as multi-physics and numerical modeling, simulation, data-driven analytics, prediction, machine learning, and Artificial Intelligence (AI) techniques, etc.

Some of the main characteristics and properties ascribed to DT in the literature include: (i) context awareness [51]: DT must be aware of its context by considering, in addition to the intrinsic data of the system, the data characterizing the specific context in which it runs; (ii) fidelity [49,52]: DT must ensure a high level of fidelity, which [52] described as “the number of parameters, their accuracy, and level of abstraction that are transferred between the virtual and physical twin/environment,” so that the physical and virtual systems are perfectly synchronized, thus allowing for a more accurate simulation and optimization; (iii) autonomy [51]: DT must be able to act independently so that it can make decisions autonomously and; (iv) adaptability [51]: DT must be capable of modifying the system’s behavior during unexpected events by adjusting the control parameters of the physical entity under the objectives to be achieved.

Basically, DT can be defined as an interactive, autonomous, context-aware, adaptable, and data-driven tool that can reflect the current state of the physical system, perform real-time optimizations, and predict system behaviors in response to the different events that may occur. To meet the purposes of this paper, which include reviewing the existing literature on the application of DT to the CONWIP card-based control system and proposing a DT-based solution, we chose to simply address this definitional aspect of DT. That being said, to view more in-depth analyses of DT, we redirect the reader to the following articles: Refs. [52,53] for the characteristics and concepts of DT, Refs. [42,52] for its related knowledge gaps and required future research areas, Refs. [42,54] for its challenges, Refs. [42,53] for its enabling technologies and applications, and Ref. [47] for the implications of its implementation for already existing DT as well as its degree of integration.

2.3.2. Application of the DT to CONWIP

To examine how DT can add value to DCONWIP, we explore the available research on DT in relation to DCONWIP in the ScienceDirect database using «Digital Twin AND “CONWIP”» as KEYWORDS, «Review articles and Research articles» as DOCUMENT TYPE, and «Engineering and Decision Sciences» as SUBJECT AREAS. The English language papers published from 2003 onwards (the year of the appearance of the DT concept) are no more than 7.

The first observation to be drawn from this research is that the concept of DT concerning DCONWIP is highly under-explored. The second observation is that only [55], among the seven papers, focuses on applying DT to CONWIP-controlled production systems. The authors of [55] stated that the applications of DT to CPCS are very limited and believe that no author has used DT to improve the order release process in CONWIP-driven production lines. Their objective in this paper, which they describe as “an initial work to start exploring this kind of problem”, is to fill this gap by developing a new order release model that improves the performance of CONWIP by pairing it with its DT. It is a three-principal component-based model (see Figure 3).

-

An interface between the physical and digital systems provides real-time data collection on the state of the production process;

-

A DT consists of a simulation software and a tool for analyzing the simulation results. The simulation approach is used to test and predict the performance of the CONWIP system based on two main outcomes: the WIP level and the prediction of the average inter-output time of the orders;

-

An intelligent system to decide on the adjustments to be made when needed (e.g., increase the number of CONWIP cards, decrease it, or do nothing). These decisions can be made using an artificial intelligence system that relies on a reinforcement learning algorithm.

This approach falls under Argyris’ Double Loop Learning theory [56], according to which, based on feedback, organizational systems and management rules can be modified (e.g., by adjusting the card number) if necessary to improve performance. In [55], it was not tested in a real industrial context but experimented through two workshop examples: a Job Shop and a Flow Shop, each composed of five workstations and five queues. The results showed that the proposed approach can reduce the level of WIP without affecting the production throughput.

It is important to emphasize that the two workshops mentioned in [55] are fictional in nature, as integrating a DT into a real industrial context is a complex endeavor. This process requires a meticulous approach that cannot be achieved instantly but demands time and must be carried out progressively and cautiously to minimize risks and ensure a smooth transition to an optimized system. Therefore, it is crucial to consider the specificities of each company and adapt the DT accordingly, proceeding step by step and advancing to the next phase only once the success of the previous stage has been fully established. In this context, Ref. [42] asserts that implementing a DT involves going through three levels: (i) the Digital Model level, which represents a digital version of a physical object without automatic data exchange between the physical and digital models; (ii) the Digital Shadow level, which signifies a one-way relationship between the physical and digital objects, where changes in the physical object affect the digital representation, but not vice versa; and (iii) the Digital Twin level, which is achieved when there is bidirectional data flow between the physical object and its digital counterpart, enabling complete integration and fully autonomous decision-making.

Overall, this literature review reveals that production control based on a CONWIP system is very little explored in theory and even less in practice. This does not, however, question the strong potential of DT, which has proven to be of great use in the manufacturing industry in general. This is mainly due to its real-time capability and predictability, which, when combined with virtualization, meets the key requirements of smart factories.

3. Combined Digital Twin/Digital CONWIP

3.1. Definition

Based on the definitions of DCONWIP and Digital Twin, the Combined Digital Twin/DCONWIP (CD_TD_C) can be defined as the virtual representation of a production line driven by CONWIP using Industry 4.0 technologies (RFID systems, Real-Time Location Systems (RTLSs), sensing systems, etc.) that enable bi-directional data exchange between physical and virtual systems. This CD_TD_C should, in our opinion, be based on two main principles:

-

Maintain a strong resemblance between the physical system and its virtual counterpart. To avoid the misappropriation risks usually experienced when facing overly digitized tools, it is essential that the representation and functioning of the virtual CONWIP remain identical to that of the traditional one. Therefore, the virtual CONWIP board and cards should resemble their paper counterparts, and how these cards are manually manipulated and in which the board is being interpreted should not radically change after the digital transformation either. That said, instead of opting for complete digitization, it is advisable to maintain, through touch technology, the involvement of the workers by allowing them to manipulate the digital tickets manually, yes, but through a Touch Screen.

-

Maintain real-time connectivity between the physical system and its virtual counterpart: CONWIP digitalization refers to controlling a production line through its virtual prototype to relieve human operators of non-ergonomic, repetitive, and heavy workloads (e.g., designing, printing, and laminating CONWIP cards). This is what [57] referred to as physical automation. As for CONWIP cognitive automation, described by [58] as a process to assist or replace stressful and repetitive mental tasks with the automatic processing of large data streams, simulation approaches are typically used. However, real-time connectivity with the physical system is a key advantage of DT over pure simulation. This is because, rather than tracking the current and past states of the physical system, simulation models predict the future states of a physical system based on a set of initial assumptions [49]. This real-time interconnectivity, to be achieved through connected devices such as sensors, controllers, actuators, networks, and software [57], is the major contribution of DT to DCONWIP.

3.2. Contribution

In this section, we highlight the contributions of DT in response to the limitations of DCONWIP presented earlier (see

Table 2. Contributions of DT in response to DCONWIP’s limitations.

CONWIP Digitalization Issues	Potential Contribution
Need to invest in new technologies.	Economic profitability of this solution is not assessed in this paper (although, given its potential benefits, a substantial return on investment can be expected).
Need to hire competent workers with highly developed skills.	By maintaining high similarity between the physical system and its virtual counterpart, this solution does not require specific skills or extensive training for workers. Moderate training should be provided.
Workers’ resistance to new technologies.	By enabling workers to manipulate CONIWP cards and boards as they used to, but simply through a touch screen: - Their familiarity with controlling WIP is preserved, leaving them less reluctant to digital transformation. - Their concerns about being replaced by machines would be allayed as the touch functionality calls for their participation.
Workers’ concerns about being replaced by machines.
Decisionmaker’s uncertainty about the effectiveness of digital systems.	By making WIP control basically unchanged while going virtual, decisionmakers would no longer be so doubtful and uncertain about the success of virtual CONWIP.
Risks of data falsification, of poor data validity and quality, and of decisions made on the basis of obsolete data.	By ensuring real-time connectivity, this solution provides an accurate representation of the physical system state, real-time data collection and retrieval, and decision-making based on updated data.
High difficulty in predicting release dates for production orders.	By providing reliable predictions from continuously updated data, this solution allows, through a simulation approach: - To forecast the delivery date of a PO by recording the simulated time at the end of the last operation for that PO. - To construct the Gantt diagram of the PO trajectories in the future depending on their routings, their last known real position, and the queues.
High complexity of workshop planning.

). The contribution of DT regarding investment issues is not highlighted in this article. However, it is worth mentioning that investing in projects that entail such radical changes as digital transformation is not really a limitation. Investment is a general issue for any project involving new technologies. Indeed, significant investments are required to implement Industry 4.0, but it is clear that all manufacturers aspire to smart factories because using smart, digitized solutions portends a substantial return on investment.

By making sure that controlling flows on the shop floor is almost the same, whether it is with traditional or digital CONWIP boards, decisionmakers would no longer be as doubtful and uncertain about the success of DCONWIP because if the traditional solution is performing well, there is no reason for the digitized solution not to. Moreover, operators will no longer feel reluctant about this digital transformation thanks to the close mimicry between physical and DCONWIP. Furthermore, their familiarity with piloting the WIP is preserved by enabling them to manipulate the CONWIP cards and boards through a touch screen. Workers’ worries about being replaced by machines would also be alleviated thanks to the touch functionality of CONWIP’s digital boards. Indeed, with the digital touch boards, not all tasks will be fully automated. Operators will have to intervene in the WIP control by, for example: (i) retrieving the PO from the MRP and importing them into the digital board by dragging them on the touch screen; (ii) reordering the PO in the waiting room according to production priorities and needs; (iii) introducing the PO into the panel representing the shop floor upon their release; (vi) reinstating the PO that has been suspended due to a problem in the production line, etc. It should also be added that since the tactile manipulation of the virtual cards is identical to the physical movement of the cards, the CD_TD_C does not necessarily require specific skills or extensive training for the workers. However, a minimum of training must be provided. Workers need to have sufficient information about the changes, understand the new digital technologies and their future role in the organization, and obtain the training and education they need to perform their jobs properly enough to avoid errors, misalignments, and reproducibility problems.

On the other hand, the real-time interconnectivity functionality overcomes the other drawbacks of the virtual CONWIP, namely risks of data falsification, poor data validity and quality, decisions made based on obsolete data, great difficulty in predicting the delivery dates of PO, and the high complexity of scheduling the workshops. Indeed, as already mentioned, DT is based on the principle of fidelity in that it ensures the exact representation of the physical system state, the retrieval and collection of real-time data, and the decision-making based on updated data. In addition, with its predictive capability, based on actual, up-to-date data, the CD_TD_C can:

-

Provide an attractive computational solution to the problem of predicting PO delivery dates. Using a simulation approach that considers the individual routing of each PO and the capabilities of each workstation, it will be possible to predict the delivery date of a PO by recording the simulated time at the end of the last operation for that PO [3]. Coupling DCONWIP with DT will ensure that this prediction considers the current situation on the shop floor in terms of workstation loads, WIP levels, and queues, which will considerably improve the accuracy of the forecasted delivery dates.

-

Construct the Gantt diagram of the PO trajectories in the future depending on their routings and their last known current position [3]. To do so, the CD_TD_C can use the simulation approach to simulate the progress of PO on the different operations represented through the successive columns of the CONWIP board.

Obviously, DT shows excellent potential for increasing the overall efficiency of DCONWIP. To better illustrate the benefits of DT, which go beyond the simple virtualization of physical cards, we present in the following the impact of implementing DT within a CONWIP-driven production line in real industrial contexts.

4. Case Study

To gain a deeper understanding of the case study, it is essential to begin with a brief overview of the steps involved in gradually implementing a CD_TD_C system (Section 4.1). This will be followed by a concise contextual description of the Production Systems (PSs) being monitored by the CD_TD_C system (Section 4.2). Subsequently, we will present noteworthy feedback from the practical experiences (Section 4.3).

4.1. Steps in the Transition from DCONWIP to CD_TD_C

The establishment of a PS monitoring system using CD_TD_C cannot be achieved from scratch. It is a process that involves several steps (cf. Figure 4) and relies on the interactions between the PS and the external environment (customers, ERP system controlling interconnected PSs, etc.).

The initial step is the implementation of a DCONWIP, which provides real-time information about the utilization of the PS’s resources, production progress, and the functioning of the card-based control system through a digital board. This concept aligns with the aforementioned Digital Shadow concept.

The second step involves coupling the DCOWIP with a simulator that describes the Production System (PS), the production processes, and the control system. This simulator generates a projection of the PS’s evolution based on an initial state, forecasted arrivals in the PS (orders, materials, etc.), and the control rules in use. This forecast enables the creation of a dynamic digital board and verifies compliance with certain commitments (such as delivery dates). The simulation allows for the completion of the current digital board by forecasting potentially detrimental deviations (to be defined) that should trigger corrective actions, albeit with a slight delay.

The automation of decision-making, inherent in the concept of DT, cannot be accomplished from scratch (except perhaps in closed PSs where a cybernetic approach can be utilized). It is necessary to build a knowledge base consisting of numerous decisions made in response to similar problems, as the desired outcome often relies on the combination of multiple actions whose exact impact may not be accurately reflected in the chosen model. Therefore, the second step involves systematic human intervention.

The third step entails the partial (or complete) automation of decisions. The knowledge base can be utilized to handle standard issues through accurately programmable decisions. Artificial Intelligence (AI), particularly deep learning, can also be leveraged to automate the decision-making process. However, due to the diverse range of decision-making contexts and their potential evolution, the knowledge base underlying AI is likely to be limited, and the relevance of the suggested decisions will need to be periodically assessed. Therefore, several methodological challenges remain to be resolved before a fully developed version of DT can be safely employed.

This clarification of the step-by-step approach to achieving a complete DT in actual production systems constitutes a methodological contribution of this paper.

4.2. Context of the Case Study

In this case study, an experiential feedback approach is utilized to investigate the usage of CD_TD_C in 14 industrial companies operating across various sectors. These sectors include traditional mechanical manufacturing (four companies), electronic card manufacturing (three companies), industrial product repair (three companies), precision watchmaking (two companies), additive manufacturing (one company), and industrial parts certification (one company). The companies vary in size, ranging from small and medium-sized enterprises to large industrial groups. In each company, CONWIP is responsible for managing a specific portion of the plant flow. The workshops range in size from six to twenty-four individuals. The PSs managed by CONWIP are integrated into supply chains that involve multiple PSs managed in different ways and may share certain resources. To mitigate conflicts, a pre-allocation of these shared resources is implemented, allowing for a certain level of decision-making autonomy.

In all the cases examined, the manufacturing or repair processes exhibited a significant level of variability. This variability stemmed from various factors, including the diversity of references to be manufactured, the instability of the product mix, or the unpredictability of input sequences (as observed in the repair scenario). In each case, the implementation of CONWIP aimed to reduce work-in-process (WIP) and enhance control and lead time forecasting. Other management methods that aim to streamline flows, such as Takt time or Kanban, were not feasible due to the extensive product variety and demand instability.

Seven of these cases involved the implementation of the digital board for CONWIP, following an initial phase of experimentation with the CONWIP Pull Flow method using a traditional manual board. In the remaining cases, CONWIP was directly deployed in digital mode, as these workshops were already accustomed to visual communication boards for production planning and control. As a result, the first step has been achieved for all the studied PSs. The implementation of the second step of CD_TD_C is either completed or in an advanced testing phase.

A qualitative and quantitative evaluation of the introduction of DCONWIP in six plants, which has been validated by the participating companies, can be accessed at the following URL: https://www.wipsim.fr/en/success-stories/, (accessed on 15 October 2022).

4.3. Experience Feedback Results

As mentioned earlier, the traditional manual control solution in the workshops presents several challenges, including the following:

(i)

Inconsistency between the position of the cards on the CONWIP board and the actual position of the work-in-progress (WIP);

(ii)

Difficulty in accurately predicting the delivery dates of production orders (POs);

(iii)

Complexity in creating a workshop schedule.

The virtualization of the CONWIP board significantly minimizes the disparity between the real and assumed positions of the cards, given that the card movement is automatically synchronized with the manufacturing declarations recorded in the ERP or MES. In such cases, and under the condition that progress is accurately documented in the ERP/MES, the digital CONWIP board accurately and consistently reflects the advancement of the production orders (POs). However, one drawback observed in practice is that users tend to interact less with the digital board, which poses a risk of limited adoption and assimilation of this solution, similar to E-Kanban.

The last two challenges have been addressed through the implementation of the CD_TD_C simulation, which incorporates a digital board that provides updated forecasts. These forecasts assist in recalibrating delivery dates and workshop planning. Nearly all the case studies documented on https://www.wipsim.fr/en/success-stories/ have successfully advanced to the second stage of transitioning from DCONWIP to CD_TD_C.

Prior to digitization, accurately estimating the delivery date using a traditional CONWIP table or manual workshop planning was extremely challenging. This was due to the variability in throughput time for each production order (PO), as it depended on the queues encountered at each workstation. Since each PO had a different routing path, manual prediction by the production planner was time-consuming and prone to inaccuracies. The simulation integrated into the CD_TD_C system has proven to be a valuable tool in predicting PO delivery dates in the analyzed cases. It eliminates the need for the planner to manually calculate delays, which used to consume a significant amount of time (up to one day per week in a specific observed case), and significantly improves the accuracy of the forecasted delivery dates for each PO.

Transitioning to CONWIP-type control involves abandoning the construction of a scheduling plan since tasks are allocated progressively based on queue levels and priority production orders (POs). However, maintaining a forward-looking perspective provided by planning is desirable in order to anticipate the arrival of POs at workstations in the future. This includes activities such as tooling preparation, workstation organization, and identifying qualified personnel for upcoming manufacturing references. The simulation integrated into the CD_TD_C automatically generates an updated Gantt chart, taking into account the routing, queues, and the most recent known position of the POs. The resulting workshop planning serves as an output of the planning process rather than an input. In practice, it functions as a digitization tool that is extensively consulted in the field and provides in-depth analysis for sector managers.

5. Conclusions

This article sheds light on an intriguing and under-researched subject. The digitalization of the CONWIP system remains a relatively unexplored topic, and the integration of this digitalization with the DT technique is even less studied. Our case study presents a highly promising research perspective, as it demonstrates the step-by-step implementation of this association within open productive systems.

Additionally, we have identified a unique set of methodological challenges when integrating a simulator with the DCONWIP system, enabling proactive issue anticipation. These challenges pertain to a new class of proactive decision-making problems, distinct from reactive approaches. Addressing these challenges is crucial as they fall outside the scope of DT. Neglecting them renders the DT approach ineffective. Generalizing DTs is only possible if further research is conducted on the modeling used by the DT and proactive decision-making. It is important to note that the diverse range of contexts in which DT may be applied will likely require context-specific solutions rather than generic ones.

Furthermore, to the best of our knowledge, there is no praxeological research existing on improving the practice of controlling a production line through CONWIP (or any other CPCS) while considering the human dimension. This research work fills this gap by structuring knowledge that:

☐

Raises awareness among potential CONWIP users about the significance of going digitization, explaining how card virtualization overcomes issues encountered in traditional CONWIP while offering additional benefits such as automatic card creation and advancement and information replication across different locations on the shop floor;

☐

Alerts potential users to new organizational challenges that may arise before, during, and after the digitalization of the CONWIP system;

☐

Encourages potential users to go beyond virtualizing CONWIP by creating its DT while ensuring two main specificities:

• Displaying the DT on a touch screen, allowing operators to manipulate virtual cards in a manner similar to physically moving paper cards, thus addressing resistance to digitalization;
• Establishing perfect synchronization between CONWIP and its DT to enable multiple functionalities (e.g., predictive information calculation) and facilitate quick and autonomous decision-making, as expected from the DT concept.

From a sustainability perspective, our paper contributes to the global agenda of combating climate change and promoting sustainable development. Climate change is a pressing global emergency that requires transformative action to ensure a sustainable and resilient future. The Paris Agreement [59] and the United Nations’ 2030 Agenda [60] for Sustainable Development set targets for reducing global greenhouse gas emissions and promoting sustainable economic growth and inclusive industrialization. Our paper aligns with goals eight and nine of the sustainable development agenda, emphasizing the importance of sustainable economic growth, productive employment, resilient infrastructure, and innovation.

Finally, there are several areas for improvement and further research. Conducting a comparative study to examine the performance of the CONWIP system before and after the implementation of DT would be valuable. This analysis can evaluate factors such as WIP reduction, increased throughput, meeting delivery deadlines, etc., enabling the quantification of economic gains and assessing the profitability of this Industry 4.0 solution. Additionally, surveys and investigations involving operators and managers would provide valuable insights into their behavior and feedback regarding the implementation of CD_TD_C.